Java

If you ask Java developers about the concept of ‘Sneaky Throws,’ I am almost sure there will be a couple of opinions that are quite differently expressed, but similar in their meaning. Some will sum it up as being able to throw checked exceptions without declaring them explicitly; others will amend that it means writing functional-style code (lambdas) and being allowed to call methods that throw checked exceptions. Most probably, it will be surely mentioned that there’s a Lombok annotation called exactly @SneakyThrows that solves the problem immediately when put on a method. Last but not least, to outline it in a more pragmatic manner, the concept allows tricking the Java compiler into treating checked exceptions as runtime exceptions. All of these are valid points of view, and to clarify the concept, this article aims to provide a straightforward yet useful approach to handling methods that throw checked exceptions. Let’s jump right in and imagine the following situation. The team is requested to enhance the currently delivered application and implement new functionalities. This obviously happens on a ‘sprint-ly’ basis. Nevertheless, the project has been successfully developed for quite a while now; it also deals with legacy code, and moreover, developers are interacting with other parts of code that were written, let’s say, in a less fortunate manner. Such an example is the class below. Java public class TwoDigitsInteger { private final Integer value; public TwoDigitsInteger(Integer value) { this.value = value; } public boolean isValid() throws NotSetException { if (value == null) { throw new NotSetException("Number value not set."); } return value >= 10 && value <= 99; } public Integer getValue() throws NotSetException { if (value == null) { throw new NotSetException("Number value not set."); } return value; } } Just as its name suggests, it models a two-digit integer number. Instances of this class are immutable; the value is set upon construction, and it declares two methods, one for reading the value — getValue() — and another one for validating it — isValid(). We’re not going to further elaborate on the quality of the code, as it helps in the experiment done. The main issue here, the plot of this article, is the fact that both methods declare a NotSetException as they might throw it under certain circumstances, and even that might be fine unless this Exception hadn’t been a checked one. Java public class NotSetException extends Exception { public NotSetException(String message) { super(message); } } One option (and definitely the one worth taking into account) is to profit and consider the moment a good opportunity to refactor this ‘legacy’ code and at least make the Exception a runtime one. A few unit tests can be written (in case these are missing), then the implementation improved, and focus can be moved on the newly requested features. Nevertheless, for the sake of the experiment in this article, it’s assumed the TwoDigitsInteger class is kept as it currently is and the Exception remains checked. Exception Function Let’s consider a very simple scenario: there is a collection of TwoDigitsIntegers and the intent is to create a string expression that outlines the sum of the numbers. Java List<TwoDigitsInteger> numbers = List.of(new TwoDigitsInteger(10), new TwoDigitsInteger(25), new TwoDigitsInteger(37)); If writing the code as in the test below, Java @Test void sumExpression() { String result = numbers.stream() .map(TwoDigitsInteger::getValue) .map(String::valueOf) .collect(Collectors.joining("+")); Assertions.assertEquals("10+25+37", result); } the Java compiler will complain, saying — Unhandled exception: com.hcd.utilities.NotSetException – as the getValue() method declares a checked Exception and obviously it cannot be used inside a stream. To solve the issue, a try-catch is needed, which makes the code quite difficult to read (and ugly). Not to mention that we’re modifying the state of the joiner as we loop the collection. Java @Test void sumExpression1() { StringJoiner joiner = new StringJoiner("+"); for (TwoDigitsInteger number : numbers) { try { joiner.add(String.valueOf(number.getValue())); } catch (NotSetException e) { throw new RuntimeException(e); } } String result = joiner.toString(); Assertions.assertEquals("10+25+37", result); } In order to overcome this and allow having a fluent API even in situations where checked Exceptions are present, the following ExceptionFunction interface is created. Java @FunctionalInterface public interface ExceptionFunction<T, R, E extends Exception> { R apply(T t) throws E; } It is general enough; it represents a function that accepts one argument (of type T), produces a result (of type R) and when applied, an Exception subclass (of type E) might be thrown. Implementers shall define a single method, which effectively applies the function. Additionally, the following class is defined. Java public final class ExceptionWrapper { public static <T, R, E extends Exception> Function<T, R> apply(ExceptionFunction<T, R, E> function) { return t -> { try { return function.apply(t); } catch (Exception e) { throw new RuntimeException(e); } }; } ExceptionWrapper() { throw new UnsupportedOperationException("No need to be called."); } } When the ExceptionWrapper#apply() method is called, in case an Exception is thrown, it is wrapped into a RuntimeException one and thrown further irrespective of the type of the initial one (the checked Exception case is obviously covered as well, so we’re good). The ExceptionFunction passed as a parameter represents the initial call that is wrapped to overcome the problem. The previously discussed test is modified to use the ExceptionWrapper#apply() method. Not only does it now compile and run successfully, but the code readability is definitely improved. Java @Test void sumExpression() { String result = numbers.stream() .map(ExceptionWrapper.apply(TwoDigitsInteger::getValue)) .map(String::valueOf) .collect(Collectors.joining("+")); Assertions.assertEquals("10+25+37", result); } Exception Predicate Let’s now consider another straightforward scenario, one in which we want to count only the valid two-digit integers that are found in a designated range. Also, for the sake of this experiment, it’s assumed the previous TwoDigitsInteger class is used. As in the previous case, the following piece of code that would do the job doesn’t compile because of the same reason – Unhandled exception: com.hcd.utilities.NotSetException — as the isValid() method declares a checked exception, and it cannot be used inside a stream. Java long count = IntStream.range(0, 150) .mapToObj(TwoDigitsInteger::new) .filter(TwoDigitsInteger::isValid) .count(); Again, assuming the TwoDigitsInteger is needed, one would have to loop through the numbers, check them in a try-catch for checked NotSetExceptions as isValid() declares it, then pack the Exception as a RuntimeException one and throw it further, finally count the valid number. This is already way too complicated even when only enumerating the steps in natural language. To be able to keep the API fluid and use streams when performing checks that declare checked Exception, the next interface is declared. Java @FunctionalInterface public interface ExceptionPredicate<T, E extends Exception> { boolean test(T t) throws E; } It represents a predicate (a boolean-valued function) of one argument that might throw an Exception subclass. The method evaluates the predicate on the given argument and returns true if the input argument matches, or false otherwise. In addition, the following method is added to the ExceptionWrapper class, very similar to the apply() one. Java public static <T, E extends Exception> Predicate<T> test(ExceptionPredicate<T, E> predicate) { return t -> { try { return predicate.test(t); } catch (Exception e) { throw new RuntimeException(e); } }; } When called, it effectively applies the provided predicate. In case an Exception is thrown, it is wrapped into a RuntimeException one and thrown further. The initial code can now be rewritten as below and successfully compiled and executed. Java @Test void count() { long count = IntStream.range(0, 150) .mapToObj(TwoDigitsInteger::new) .filter(ExceptionWrapper.test(TwoDigitsInteger::isValid)) .count(); Assertions.assertEquals(90, count); } Takeaways Although simple and to-the-point, the presented solution comes in very handy, especially when dealing with functions that declare checked Exceptions and are further used in the code that we produce. For sure, other ready-to-use alternatives already exist, an example being the Lombok @SneakyThrows annotation. Personally, I have very rarely included the Lombok library in any of my projects and as Java introduced the records, this becomes even more unlikely to happen in the future. That being said, the structures described in this article are very helpful, lightweight, and easy to understand and use when needed. ExceptionWrapper, ExceptionFunction and ExceptionPredicate source code is part of the asentinel-orm open-source project. To use it, one may either declare the Maven dependency in their pom.xml file (version 1.72.2 is the latest at the moment of this writing) XML <dependency> <groupId>com.asentinel.common</groupId> <artifactId>asentinel-common</artifactId> <version>1.72.2</version> </dependency> or use it directly if considering there’s too much overhead to include the whole library. Resources [1] – asentinel-orm open-source ORM project is here [2] – the picture was taken at ‘Harry Potter Warner Bros. Studios’, near London

By Horatiu Dan CORE

In a previous article, Working with Spreadsheets in Java: A Practical Overview, we walked through the common scenarios where Java applications need to interact with spreadsheets and the categories of tools available for the job. One of the factors mentioned there was support for modern Excel formulas — a topic that deserves more space than a single bullet point. Java applications interact with Excel more often than most teams plan for: file uploads from finance, calculation logic authored in a workbook, reporting exports back to business users. The files these users produce today are not the same as the files they produced five years ago. Excel 365 and Excel 2021 introduced a new formula model, and workbooks authored in those versions routinely use it. Depending on which library you use, those formulas may evaluate correctly, fail silently with stale cached values, or throw exceptions at recalculation time. This article goes deeper on that topic: what dynamic arrays and spill behavior are, what the new function set looks like, why supporting them is technically difficult, and what Java developers should look for when evaluating whether a library handles them correctly. Why You're Seeing Them in Real-World Workbooks Dynamic arrays spread rapidly because they eliminate many of the helper columns, copied formulas, and Ctrl+Shift+Enter array formulas that older Excel workbooks depended on. Workbooks become shorter, easier to audit, and easier to maintain. As organizations migrate to Microsoft 365, these newer formulas increasingly appear in spreadsheets exchanged with Java applications, even when the application itself hasn't changed. What Changed: One Formula, Many Values Before dynamic arrays, formulas that returned multiple values generally required a pre-sized array range and legacy array-formula syntax. Dynamic arrays changed this by allowing a single formula to return a variable-sized array and automatically spill into neighboring cells. For example: =UNIQUE(A1:A6) Entered in one cell, this returns the full list of distinct values from A1:A6. The result fills as many cells as there are distinct values. If the source data changes and the number of unique values changes, the spill range automatically grows or shrinks. This is not just a new function. It is a change in the evaluation model itself. The Mechanics of Spill A spilled formula produces a region of cells with specific roles: The anchor cell is the one cell that contains the formula. It "owns" the result.The spilled cells are the neighboring cells that display the additional values. They do not contain formulas of their own; they mirror slices of the anchor's result. You can reference the entire spilled range from another formula using the # operator. The # reference is not a fixed cell range such as A1:A6; it refers to whatever range the anchor currently spills into. If A1 contains =UNIQUE(...) and the result spills into A1:A6, then =COUNTA(A1#) counts the values in the entire spilled range. If the spill range grows or shrinks, the # reference adjusts automatically. #SPILL! errors. If the range a formula needs to spill into is blocked by an existing value, a merged region, or an Excel Table, the formula cannot spill, and the anchor cell shows #SPILL! instead of a result. Clearing the obstruction allows the formula to complete. Implicit intersection with @. Older Excel silently reduced arrays to single values in many contexts. Modern Excel returns the full array unless the formula uses the @ prefix. For example, =A1:A10 entered in a cell in modern Excel spills the values from A1:A10, while =@A1:A10 applies implicit intersection and returns the value corresponding to the formula's row. Files migrated from older Excel versions often contain automatically inserted @ prefixes to preserve their original behavior. The New Function Family The modern functions commonly associated with Excel's dynamic-array model can be grouped into three broad categories. It is worth understanding the grouping because the groups behave differently. Group 1: Language Features These are not really functions in the traditional sense. They add expression-level constructs to Excel's formula language. LET binds names to intermediate values inside a formula, so you can write =LET(total, SUM(B2:B100), tax, total*0.1, total+tax) instead of repeating SUM(B2:B100) three times.LAMBDA defines a reusable function inside a workbook. Combined with named ranges, LAMBDA effectively adds user-defined functions without VBA.ISOMITTED is used inside LAMBDA to detect whether an optional argument was supplied. These functions do not inherently produce a spilled array. LET returns the result of its calculation, which may itself be an array. Group 2: Dynamic Array Functions These are the functions people usually mean when they talk about "the new Excel functions." These functions are designed to return arrays, and when their results contain multiple values, Excel can spill those results into neighboring cells. UNIQUE returns distinct values from a range.SORT and SORTBY return sorted arrays.FILTER returns rows that match a condition.SEQUENCE generates a sequence of numbers.RANDARRAY generates an array of random numbers. Array-shaping functions form a subset of this group. They take arrays as input and return reshaped arrays: CHOOSECOLS, CHOOSEROWS, DROP, EXPAND, HSTACK, VSTACK, TAKE, TOCOL, TOROW, WRAPCOLS, WRAPROWS. TEXTSPLIT also fits here — it splits a string into an array. Other functions, including BYROW, BYCOL, MAP, REDUCE, and SCAN, build on the same dynamic array model. Group 3: Scalar Functions Added in the Same Era Dynamic arrays are primarily an evaluation model; modern functions are a collection of functions that take advantage of, or coexist with, that model. Group 3 functions were introduced as part of the broader set of modern Excel functions, but they are not themselves primarily array-producing functions. XLOOKUP and XMATCH are modern replacements for VLOOKUP and MATCH. They normally return a single value, though they can return an array when passed an array of lookup values.TEXTAFTER and TEXTBEFORE return substrings.VALUETOTEXT and ARRAYTOTEXT convert values to text (ARRAYTOTEXT takes an array as input but returns a single string). These are often lumped in with dynamic array functions because they arrived together, but their evaluation model is closer to VLOOKUP than to UNIQUE. Why This Is Hard for a Formula Engine Supporting these features is not just a matter of adding new function names to a list. The dynamic array model requires substantial changes to the evaluation engine itself. A traditional one-cell-at-a-time formula model is not sufficient to implement dynamic arrays. An engine must be able to represent a formula whose result has a variable shape and propagate that result across multiple cells. A modern engine has to handle four additional concerns: Array-shaped results. A formula's return value may be a 2D array whose dimensions depend on the input data. =UNIQUE(A1:A100) returns a different number of rows depending on how many unique values the range contains. The engine must determine the result shape at evaluation time, not at parse time. Spill range tracking. The engine must reserve the cells the formula spills into and prevent other content from occupying them. When something occupies a spill target, the anchor must return #SPILL! rather than overwrite the obstruction. The reserved region must also update when the shape of the result changes. Downstream references. Expressions like A1# refer to the entire spilled range. When the shape of the anchor formula changes, every downstream reference must be re-evaluated with the new dimensions. This makes the dependency graph more dynamic than in a one-value-per-cell model. Implicit intersection compatibility. Older Excel silently collapsed arrays to single values in many contexts. Modern Excel returns the whole array. When files authored in older Excel are opened in modern Excel, @ prefixes are inserted automatically to preserve original behavior. An engine that reads modern .xlsx files needs to honor the @ operator, or the imported formulas will produce different results. Adding these behaviors to an engine designed around the one-formula-one-value model is a substantial rewrite, not an incremental feature addition. This is part of why support across the Java ecosystem has been uneven. What Java Developers Should Check Support for these capabilities varies significantly across Java spreadsheet libraries. Some engines were originally designed around traditional one-cell-one-result evaluation and only implement subsets of the modern Excel model. Others have extended or redesigned their evaluators to support dynamic arrays. Rather than relying on feature lists, it is worth validating behavior against workbooks representative of your own application. If your application needs to evaluate modern Excel formulas, the following checks are worth running before committing to a library. Test with a file containing a spilled formula. Create a small .xlsx with =UNIQUE(A1:A100) or =SORT(A1:A100) in a cell. Load it in your candidate library and try to recalculate the anchor cell. A library that supports dynamic arrays will return the array; one that does not will typically throw an exception or return only the first value. Check for the # spill operator. In the same file, add another cell containing =COUNTA(A1#) where A1 is the anchor. This tests whether the library understands spilled range references, which is a separate capability from evaluating the anchor formula itself. Test the @ operator. Add =@A1:A10 in a cell and check whether the library correctly returns the value at the current row rather than the full array. Files migrated from older Excel routinely contain @ prefixes; a library that doesn't handle them will produce different results than Excel. Test with LET and LAMBDA. Write a formula like =LET(total, SUM(A1:A100), total * 1.1) and check both evaluation and .xlsx round-trip. Test LET and LAMBDA independently. Parsing, preserving, and evaluating these functions are separate capabilities, so a library that can read or write the formula text may not necessarily be able to evaluate it correctly. Test round-trip. Save the workbook, reopen it in Excel, and check that the formulas still produce correct results. Some engines strip modern constructs on save. Check what happens on failure. When a library encounters a function it does not implement, does it raise an exception, return an error value, or silently fall back to the cached value from the file? Silent fallback is the most dangerous behavior because it masks the problem during development and only fails in production when the data changes. Conclusion Excel's formula language has changed more in the last few years than in the two decades before it. Dynamic arrays, spill behavior, and the new function set are not experimental — they are standard in Excel 365 and Excel 2021, and they show up in workbooks that Java applications routinely have to process. For Java developers, the practical implication is that "Excel formula support" is no longer a single property that a library either has or doesn't have. There are several distinct capabilities involved, and libraries vary widely on each. As covered in the previous article, the Java spreadsheet landscape spans open source libraries such as Apache POI, commercial headless engines, and embedded spreadsheet components like Keikai. Whichever category fits your use case, the checks above are a reasonable way to verify that a candidate library handles modern Excel behavior against the workbooks your real users produce.

By Hawk Chen CORE

A pull request arrives. A few hundred lines of Java implementing the new discount rule: tiered thresholds, a regional exception, something about loyalty tiers that nobody can quite explain. It compiles. The tests pass. An LLM wrote it in about forty seconds. Now: who reviews it? The person who owns that rule is in commercial operations. She knows exactly which customers should get the discount and why the regional exception exists, and she cannot read Java. The person who can read Java has no idea whether the thresholds are right. He will check that the code looks reasonable, because that is the only thing he is equipped to check. So the review that happens is not the review that matters. That is the problem I keep coming back to, and it has nothing to do with how good the model is. This Is Not an Argument About Whether the Model Is Good Enough Most objections to generated code are about competence. The model hallucinates an API. It gets an edge case wrong. It writes something that works on the happy path and falls over in production. I find these arguments unconvincing because they expire. Models get better. Any position resting on today's error rate is a position with a shelf life, and people who staked one out three years ago have mostly had to retreat from it. The durable question is different. It is not how well the model writes. It is what the thing it writes is permitted to say. A model that never makes a mistake, handed Java, can still emit Runtime.getRuntime().exec(...). Not because it is malicious or confused — because that sentence is available in the language it was asked to write. Competence and authority are separate axes, and improving the first does nothing to the second. "Write it in Java" Is a Much Bigger Grant Than Anyone Means Consider what you actually authorize when you ask for a discount rule in Java. You authorize file system access. Network sockets. Reflection. Thread creation. Process execution. Every class on the classpath, including the ones that talk to your database, your payment provider, and your secrets manager. You authorize the loading of new code at runtime. Nobody intends to grant any of this. It arrives free with the language, the way a house key also opens the shed. The task needed perhaps six operations — look up an order, total it, check a customer's tier, apply a discount, log the decision, approve or refuse — and the language you handed over contains everything Java contains. That gap, between the authority the task requires and the authority the language confers, is the whole of it. It exists whether or not the model is trustworthy. It exists whether or not anyone acts on it. It is just very large, and it is not visible in the pull request. The Usual Guardrails Are Denial Lists The standard responses all share a shape. Tell the model in the prompt not to touch the file system. Review the generated code. Run static analysis and flag dangerous calls. Run it in a sandbox with a restricted security policy. Every one of these asks you to enumerate what must not happen, over a space of things that can happen which is effectively unbounded. You are writing a deny-list against a general-purpose language. You have to think of exec. Then of reflection reaching exec. Then of the dependency that shells out on your behalf. Then of the next one. We learned this lesson in security a long time ago and reached a settled answer: allow-lists beat deny-lists, because the allow-list is finite and you wrote it. Somehow, when the subject is generated code, we reach for the deny-list again. Shrink the Language, Not the Model The alternative is to stop constraining a powerful language and instead supply a small one. Give the model a vocabulary that contains exactly the operations the domain has — the six from earlier, say — and nothing else. Not a restricted Java. A different, much smaller language, whose entire vocabulary is a list your team wrote in advance, in Java, on purpose. Generated business logic then looks like this: Python PROGRAM ApproveOrder(orderId INTEGER, limit DECIMAL) RETURNS BOOLEAN DECLARE purchase Order DECLARE total DECIMAL purchase = LOAD_ORDER(orderId) total = ORDER_TOTAL(purchase) IF total > limit THEN REJECT purchase, "over limit" RETURN FALSE END IF APPROVE purchase RETURN TRUE END. LOAD_ORDER, ORDER_TOTAL, REJECT and APPROVE are not part of the language. They are Java classes somebody decided to expose. Order is a Java object the program can hold and pass and never look inside — there is no purchase.customer.account.balance here, only the operations the domain chose to have. Two things change, and the second matters more than the first. The obvious one: dangerous programs are no longer forbidden, they are inexpressible. If the model emits DELETE_ALL_ORDERS, nothing rejects it on policy grounds. The name means nothing. The program does not compile, for the same reason a typo does not compile. There is no deny-list because there is nothing to deny. The less obvious one: the commercial operations manager can read the program above. She can tell you whether the threshold is right, whether the rejection reason is the one the contract requires, whether an approval should have been logged. The review moves to the person who owns the rule. That is the review that was missing at the start of this article, and no amount of static analysis over generated Java produces it. A small language buys something else, quietly. With no data structures, one global scope, no null, and a compiler that refuses to run a program that reads a variable before it is set, entire families of subtle wrongness have nowhere to live. Not caught — absent. What It Costs, and What It Does Not Buy I would not trust this argument from someone who only listed the advantages, so here are the bills. You have to design the vocabulary. Somebody sits down and decides that the domain has ORDER_TOTAL and CUSTOMER_RISK and not forty other things. That is real work, done before the first generated line, by someone who understands the domain. And if nobody on your team can write that list, this approach will not help you. It will only show you that the list does not exist. That is worth finding out, but it is not a pleasant morning. Complex algorithms stay in Java. Business rules are algorithms too, and they belong in the small language; that is the point. But route optimization, a scoring model, anything with real computational substance belongs behind a function the small language calls. The signal is usually that you want to build up a data structure, or that you want a helper you can call from three places. Both mean you have wandered out of business logic and should walk back. The boundary bounds naming, not doing. This is the limit people miss, and overstating it is how the idea gets dismissed. A function you expose can do anything Java can do. RUN_SHELL_COMMAND is a perfectly registrable operation. The vocabulary is only as narrow as the operations you chose, and choosing them badly gets you exactly the exposure you were avoiding. There are no resource limits yet. A generated program can still loop forever. This one is a gap rather than a decision: the interpreter walks the program one statement at a time, so a step budget or a deadline is a small addition rather than a redesign, and it will go in when somebody needs it. Until then, untrusted input needs the same containment any untrusted workload needs. What you get is narrower than "safe" and more useful than it sounds: the set of things a generated program can name is finite, written down, and reviewable by a human before anything is generated at all. When I Would Still Write Java If the thing is genuinely computational, write Java. If it is a one-off that will be deleted next week, use whatever is nearest — Java, Python, a shell script — and let the model write it; do not build a vocabulary for something with a life expectancy of days. If the rules change so fast that the vocabulary would be obsolete before it settled, the overhead will not pay for itself. And if your business logic is already reviewed by people who can read it, understand it, and are accountable for it being right — you may not have the problem this solves. Plenty of teams do not. But if you are about to let a model write business rules in Java, ask the question I started with, because the answer is usually uncomfortable. Somebody is going to approve that pull request. Are they the person who knows whether the rule is correct? If not, the language is too big. I have been building a small language along these lines: BUBAS, an orchestration language for subject-matter experts, embedded in Java. The example above is real BUBAS. The idea does not require my implementation, though — the argument is about the size of the language you hand over, and you can shrink yours however you like.

By Peter Verhas CORE

Every Java developer who runs services on Kubernetes has watched this scene play out. Traffic spikes, the autoscaler adds a pod, and then everyone waits. The container is running in two seconds. The application is not ready for another twelve seconds. During those ten seconds, your existing pods absorb the extra load, latency climbs, and if things are bad enough, the autoscaler panics and adds even more pods that are also not ready. I spent years treating Spring Boot startup time as a fact of life, the way you treat weather. Then I found out the JVM has had a fix for a big chunk of it since Java 12; it works beautifully inside Docker, and almost nobody bakes it into their images. It is called Class Data Sharing, CDS for short, and this article shows you how to make your Docker build do the work Where Those Twelve Seconds Actually Go When a Spring Boot application starts, the JVM is not mostly running your code. It is loading classes. A plain REST service with Spring Web, Spring Data, and a driver or two loads somewhere between ten and twenty thousand classes before it serves its first request. For every single one of those classes, the JVM does the same ritual. Find the class file inside a jar, read the bytes, parse them, verify the bytecode is legal, and build the internal metadata structures it needs at runtime. Thousands of times. Every startup. In every pod. Here is the part that should bother you. Your container image never changes after you build it. The same jar, the same classes, the same parsing work, repeated identically in every pod that ever starts from that image. The JVM is solving the same puzzle again and again and throwing away the answer each time. CDS is the JVM saying: let me solve it once, write the answer to a file, and just memory map that file next time. What a CDS Archive Is A CDS archive is a file, usually ending in .jsa, that contains classes already parsed and verified, stored in the exact internal format the JVM uses in memory. On startup, the JVM maps this file straight into memory. No finding, no parsing, no verifying. The work was done ahead of time. You have been using CDS without knowing it. Modern JDKs ship with a default archive covering the core JDK classes, which is why java -version is fast. The step almost everyone skips is creating an archive for your application classes, all fifteen thousand of them. That is where the real win lives. The mechanism has one rule that matters for us. The archive must be created with the same JVM and the same classpath that will use it. That rule sounds annoying until you realize a Docker image is the one place in your entire infrastructure where JVM and classpath are frozen forever. Docker is not just compatible with CDS. It is the perfect home for it. The Training Run Creating the archive takes two steps. First you do a training run, where the JVM starts your application, watches which classes get loaded, and writes the list down. Then you exit, and the JVM turns that list into the archive. Since Java 13, this is pleasantly simple: Shell java -XX:ArchiveClassesAtExit=app.jsa -jar app.jar Run the app, let it come up, stop it, and app.jsa appears. From then on you start the app like this: Shell java -XX:SharedArchiveFile=app.jsa -jar app.jar There is an obvious question here. The training run wants to actually start the application, and inside docker build there is no database, no message broker, nothing to connect to. A Spring Boot app that cannot reach Postgres will crash during training. Spring Boot 3.3 solved this neatly. Setting one property makes the application run through its entire startup sequence, create all bean definitions, and then exit just before touching the outside world: Shell java -Dspring.context.exit=onRefresh -XX:ArchiveClassesAtExit=app.jsa -jar app.jar The application loads nearly everything it will ever load, writes the archive, and exits cleanly with no infrastructure needed. This is exactly what a Docker build stage can do. The Dockerfile Here is the complete picture: a multi-stage build where the image trains itself: Shell FROM eclipse-temurin:21-jdk-alpine AS build WORKDIR /build COPY . . RUN ./mvnw -B package -DskipTests # Explode the jar so the classpath is stable RUN java -Djarmode=tools -jar target/app.jar extract --destination /app FROM eclipse-temurin:21-jre-alpine AS runtime WORKDIR /app COPY --from=build /app /app # Training run: start the context, record classes, exit RUN java -Dspring.context.exit=onRefresh \ -XX:ArchiveClassesAtExit=/app/app.jsa \ -jar /app/app.jar ENV JAVA_TOOL_OPTIONS="-XX:SharedArchiveFile=/app/app.jsa" ENTRYPOINT ["java", "-jar", "/app/app.jar"] Two details in there deserve a closer look. The extract step unpacks the fat jar into a folder with the dependencies laid out as plain files. CDS is picky about the classpath being identical between training and real runs, and a fat jar with nested jars inside it makes that fragile. The exploded layout keeps the classpath boring and stable, which is exactly what CDS wants. On Spring Boot 3.2 and older, the same idea works through the layertools jarmode instead. The training run happens as a RUN instruction, which means it executes once at build time on your CI server. Every container that ever starts from this image inherits the archive for free. You did the class loading homework once, in the build, and ten thousand pod starts copy the answer. What You Get Numbers vary with how heavy your application is, but the pattern is consistent. A typical Spring Boot 3 web service that started in 10 to 12 seconds lands somewhere between 5 and 7. The JVM portion of startup shrinks dramatically, and as a bonus, the archive is memory-mapped and shared, so if you run several JVMs on one node, they share those pages and total memory drops too. You can verify the archive is actually being used, which I recommend, because CDS fails silently by design. If something mismatches, it just quietly falls back to normal class loading: Shell docker run --rm my-service -Xlog:class+load=info | head -5 Classes loaded from the archive say source: shared objects file. If you see jar paths instead, the archive is being ignored, and the log will usually tell you why. The usual culprit is a classpath that differs from training, even by one entry. One honest caveat. The training run exercises startup, not your traffic. Classes that only load when a specific endpoint gets hit for the first time are not in the archive, so those first requests still do normal loading. The archive covers the framework and wiring, which is most of the cost, but it is not a magic warm-up for everything. Why This Beats the Alternatives You Have Heard Of Whenever container startup time comes up, someone mentions GraalVM native images, and native images are impressive. Millisecond startup is real. But they come with a price list: long build times, a closed-world assumption that fights with reflection, some libraries that simply do not work, and a different runtime profile you have to learn to debug. CDS costs you five lines of Dockerfile. Your application is still a completely normal JVM application. Same debugging, same profilers, same libraries, same behavior, just faster out of the gate. For most teams, that trade-off is not even close. It also stacks with what is coming. Project Leyden's AOT cache in Java 24 and beyond is essentially this same idea grown up, caching not just parsed classes but resolved linkage and compiled code. The Dockerfile pattern you build today, a training run at build time producing a cache file shipped in the image, is exactly the shape Leyden uses. Learning it now means the future is a flag change. The Takeaway Your Docker image is immutable. Your JVM does expensive, perfectly repeatable work on every startup. Those two facts fit together like puzzle pieces, and a training run inside docker build is where they connect. One extra build step, and every pod your autoscaler ever creates comes up in half the time. The next time you watch a rollout crawl because pods take forever to go ready, remember that the answer was hiding inside the build all along.

Any input/output operation, be it accessing a file, handling an HTTP request, or a database connection, is based on 3 fundamental system concepts — file descriptors, kernel memory, and heap size. This article discusses how modern languages help developers handle behind-the-scenes file descriptor, kernel memory, and heap management. These three concepts are major bottlenecks for scaling. 1. File Descriptors A file descriptor is just a positive number that is used by the kernel to identify any open input/output stream or connection. It is defined by the kernel for a process. The following file descriptors are defined by default for a process: 0 – Standard Input (stdin)1 – Standard Output (stdout)2 – Standard Error (stderr) Any subsequent I/O operation gets the next available integer as file-descriptor. The file descriptor value can be adjusted by using the ulimit -n command in Linux. Each application, whether it is a web server written in Java Spring Boot, an API server written in Go using net/http and gorilla-mux, or a Python Flask app, is a single process. Each process has only 1024 file descriptors defined by default. That means each application can perform only 1024 I/O operations simultaneously. This seems like an amazing concept when we talk about scaling our application or API server. As many times as we come across this question — how can we scale our API server or web application to handle 100k or 1 million requests per second? This is where our modern languages play their role very beautifully behind the scenes to enable developers to develop the application to handle such scale. 2. Kernel Memory At a lower layer than file descriptors, when an incoming TCP connection hits the network card, the Linux kernel performs a 3 Way TCP handshake for that connection. The handshake lifecycle includes the states: SYN -> SYN-ACK -> ACK. The number of requests equal to the defined file descriptor value are processed immediately, assigned a file descriptor, and forwarded to the application for further processing. When FDs are exhausted, the Kernel maintains a queue for requests waiting for FDs to become available so your application can process them. The same thing happens when a request is processed, and the response is ready to be sent back to the client. This queue is maintained within RAM by read buffers(rmem) and write buffers(wmem). The size of buffers is defined in memory by the kernel and is dynamic, depending on network throughput, round-trip time, and memory pressure. The kernel network memory is non-paged, i.e cannot be swapped to disk. It’s a big bottleneck as it directly depends on physical memory. For example, if there are 100,000 open connections and each connection holds an average of 128KB of kernel memory, it comes to 12.8GB of physical RAM. This is clearly a kernel overhead, and it doesn’t show up in JVM heap metrics or Go runtime statistics. rmem and wmem buffers are governed by kernel parameters defined in /proc/sys/net/ipv4/ 3. Heap Size When TCP connections are assigned file descriptors and kernel memory is reserved, they enter user space, which is the memory managed by the application runtime — Java JVM, Node.js V8 Engine, Python interpreter, Go runtime, etc. Each connection stores objects in the heap within three categories: Connection metadata – Keep-alive timers, IP State, Socket Wrappers, etc.Cryptographic session context – handshake caches, cipher states, TLS/SSL keys, etc.Serialized payload buffers – response queues, JSON strings, ORM entity maps, etc. A connection that is encrypted via TLS takes a lot more space in the heap compared to a regular connection. For an encrypted connection, the application has to save symmetric keys, cipher contexts, session tickets, etc. onto the heap. A regular TCP socket object in the heap consumes 2KB to 5KB of space, whereas a TLS 1.3 socket object consumes 20KB to 100KB of heap space. If an API maintains 10,000 idle TLS connections, it will consume 200MB to 1GB of heap space. When an application runs, the runtime asks the kernel for memory space as the application creates objects. The application keeps creating objects, and the kernel keeps reserving memory for those objects; this is called the heap. The maximum heap size can be defined by different programming languages at runtime; for example, in Java, -Xmx4g reserves 4GB for the heap. The operating system promises to provide that much memory as heap space for the application, but it doesn’t reserve it all at once. As the application creates objects, the kernel continues to reserve memory. When objects are marked as done, the garbage collector removes them from the heap. When an incoming request hits our API server, the application uses heap space to convert raw bytes to the application-specific data structure. Once the application finishes processing the request and returns the response, those objects in the heap become unreachable or dead. When the garbage collector sweeps those objects to reclaim that memory, it doesn’t return the memory immediately; instead, the JVM or Go runtime keeps that freed memory in its internal pool. If a new HTTP request arrives within 1 millisecond, the runtime assigns the required memory from the free memory in the pool. Now imagine 10,000 new requests arriving at the same time, each with 2MB of raw bytes, and the runtime trying to allocate heap for the objects; the app instantaneously uses 20GB of memory. This is called GC thrashing, as the runtime rapidly creates required objects in the heap faster than the GC can clean them. The garbage collector is an application thread itself; when the heap gets 80%-90% full, the garbage collector panics and consumes 100% of CPU cores to scan millions of memory pointers to find dead objects. The runtime, like the JVM or Node.js garbage collector, may stop other code execution while it reorganizes the memory. So, how do runtimes like Go and the JVM handle GC thrashing? Go follows a simple strategy – avoid creating objects on the heap. The fastest GC collector is the one that has nothing to collect. The Go compiler compiles the application to see if variables outlive their functions. If a struct is used only inside a function, Go pushes the struct to the stack instead of the heap, and the stack pointer just drops when the function returns. The memory is reclaimed in 1 CPU cycle without even involving the garbage collector. If Go does have to clean the heap, its GC runs concurrently along with other goroutines and is broken into several micro pauses. Go provides sync.Pool to help developers to reuse heap memory while creating objects. For example to instead of creating millions of []bytes for JSON parsing for every new request, developers can use sync.Pool as follows: Go // Instead of creating a new buffer for every HTTP request: var bufferPool = sync.Pool{ New: func() any { return new(bytes.Buffer) }, } func handleRequest(w http.ResponseWriter, r *http.Request) { buf := bufferPool.Get().(*bytes.Buffer) // 1. Grab an existing buffer from pool buf.Reset() defer bufferPool.Put(buf) // 2. Put it back when done! // Parse JSON into 'buf' without allocating new heap memory } By recycling buffers via sync.Pool, high-concurrency APIs can handle 100,000 requests/sec with near-zero new heap allocations. Java takes a different approach. Because Java applications historically create millions of short-lived objects on the heap, the JVM relies on Generational Hypotheses and Generational Collectors (like G1GC, ZGC, and Shenandoah). G1GC can be used like java -XX:+UseG1GC while running Java applications. G1GC divides the Heap memory into physical regions: Young Generation (Eden & Survivor spaces) and Old Generation. It kind of sorts objects into different regions so that it doesn't have to scan the complete heap and can clean where most of the marked objects live. We can also mention -XX:MaxGCPauseMillis=200 to tell G1 to pause the application for no more than 200ms, but this is not guaranteed. Older JVM collectors like Parallel GC used to freeze the entire application to clear the heap when full, leading to multi-second latency spikes. Modern JVMs introduce ZGC (Z Garbage Collector) and Shenandoah. ZGC uses specialized CPU pointer references to track moved objects in real time. ZGC can clean, move, and compact terabytes of heap memory concurrently while your API requests are actively running. ZGC guarantees GC pause times under 1 millisecond, regardless of whether your heap is 500 MB or multi-terabytes. Conclusion Keep track of these three core concepts — file descriptors, kernel memory, and heap size to know when to scale. 1. File Descriptor Saturation Signals File descriptors represent the system's open handles. When an application hits its FD threshold, the operating system stops accepting connections. The following are example scenarios that indicate when to scale. Check Kernel-wide statistics from /proc/sys/fs/file-nr, per process fds - /proc/<pid>/fd, Prometheus exposes process_open_fds. If it consistently breaches the 80–85% threshold, it's time to scale. You have already tuned ulimit -n and LimitNOFILE up to standard safety thresholds (e.g., 65,536 or 104,857), but process FD counts continue climbing toward the max. Network interfaces show growing SYN-to-LISTEN socket counts and drops in netstat -s under the listen queue overflow metric. 2. Kernel Memory Pressure Signals Because TCP receive (rmem) and transmit (wmem) buffers are non-paged, they cannot overflow onto disk swap. When kernel network memory fills up, the OS drops packets. Below are the scenarios related to kernel memory breach. Check /proc/net/sockstat under TCP: inuse and matching /proc/sys/net/ipv4/tcp_mem thresholds. Netstat counters (netstat -s | grep -i retrans) show a sharp rise in TCP Retransmission rates (>1–2%). Latency spikes occur because the kernel is dynamically shrinking socket buffers down to tcp_rmem minimums (4 KB) to avoid running out of physical RAM, throttling TCP window sizes. 3. Heap Size & Garbage Collection (GC) Thrashing Signals When user-space heap allocations outpace the garbage collector's ability to sweep dead objects (like parsed JSON payloads or session states), application performance collapses. The runtime (JVM or Go) spends more than 15–20% of its total CPU time running GC sweeps (go_gc_cpu_fraction or JVM GC CPU utilization). In Go, metrics show the pacer triggering Mark Assist, stealing CPU time from worker goroutines to help clean up memory. You can check the runtime package /cpu/classes/gc/mark/assist:cpu-seconds metrics to see if GC is asking for more help from CPU. In Spring Boot, you can use Actuator and Micrometer to expose relevant endpoints to monitor the threshold values.

“...premature optimization is the root of all evil…” Donald Ervin Knuth Introduction "Premature optimization is the root of all evil." Most software engineers know this, attributed to Donald Knuth, author of The Art of Computer Programming and one of the most influential figures in computer science. Many have also picked up the practical conclusion that followed: "let's make it work first, fix performance later." After all, it's easier to add another EC2 instance than to find the root cause. But here is what Knuth actually wrote: "We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil. Yet we should not pass up our opportunities in that critical 3%." A little different, isn't it? The second sentence is almost never quoted — and that is convenient, because it turns a careful statement into a simple excuse. Sometimes for laziness. Sometimes because people assume that optimization means sacrificing readability: cryptic bit manipulation, obscure tricks, code that only the author understands at 2 am. I believe Knuth was indeed warning against that kind of optimization. But that assumption is wrong more often than people think. Good, clean code is frequently efficient code too — not by accident, but because choosing the right tool for the job tends to be both clearer and faster. The examples in this article are proof of that. Scope This article focuses on simple, cheap, and foolproof tips that can be applied universally — regardless of your architecture, framework, or domain. In my experience, they carry virtually no risk of making things worse. Architecture, design, networking, database connectivity, threading — these are deliberately out of scope. Not because they are unimportant, but because they are context-dependent. The right answer depends on your specific system, and each of these topics deserves its own article. Examples String Operations We are all familiar with built-in JDK string utilities like: equals(), startsWith(), endsWith(), contains(): Java s1.equals(s2); s1.startsWith(s2); s1.endsWith(s2); s1.contains(s2); Unfortunately, JDK provides only one function for case-insensitive comparison: Java s1.equalsIgnoreCase(s2) There are no functions for case-insensitive startsWith(), endsWith(), contains(). So, often we combine toLowerCase() or toUppserCase() with startsWith(), endsWith(), contains(): Java s1.toLowerCase().startsWith(s2.toLowerCase()); s1.toLowerCase().endsWith(s2.toLowerCase()); s1.toLowerCase().contains(s2.toLowerCase()); A little verbose and null-prone, but just fine if not on the critical path. However, this technique might cause some performance problems. Do not forget that String is an immutable class, so instead of just a char-to-char comparison between two strings, we create two additional strings that then must be garbage-collected. Considering that String is a wrapper over a char array, the memory allocation may become expensive. The solution is to use case-insensitive utilities provided by different libraries, e.g., Apache Lang3: Java startsWithIgnoreCase(s1, s2); endsWithIgnoreCase(s1, s2); containsIgnoreCase(s1, s2); Or, starting from version 3.18.0: Java Strings.CI.startsWith(s1, s2); Strings.CS.startsWith(s1, s2); Where CI exposes case-insensitive and CS — case-sensitive utilities. Many people like regular expressions and use java.util.Pattern class sometimes, not where it is really necessary. For example: Java Pattern.compile("^prefix.+suffix$").matcher(s).find() Instead of: Java s.startsWith("prefix") && s.endsWith("suffix") Or even: Java Pattern.compile("^prefix").matcher(s).find() instead of s.startsWith("prefix") Pattern.compile("suffix$").matcher(s).find() instead of s.endsWith("suffix") Pattern matching is significantly slower than trivial substring matching. The following table shows evaluation time for 1 million operations: Operation * 1 million times Time, ms s.equals("hello") 7 s.startsWith("hello") 6 s.endsWith("hello") 11 s.contains("hello") 24 s.toUpperCase().startsWith("HELLO") 65 s.equalsIgnoreCase("hello") 5 Pattern.compile("hello").matcher(s).find() 238 pattern.matcher(s).find() 31 What can we see from this table? Performance of equals() and startsWith() is similarendsWith() is 2 times more expensivecontains() is 4 times more expensive than equalsChanging case followed by startsWith() is 10 times (!) more expensiveCase-insensitive comparison functions do not have any performance penaltiesSearching for a substring using a precompiled pattern is about 20% more expensive than using a plain contains() method. Compiling the pattern and using it is almost 10 times more expensive than the plain contains() method. So next time you reach for Pattern.compile(), it is worth pausing for a second: is regex actually needed here, or is a plain string method both simpler and faster? If you really need a pattern, at least compile it in advance — better yet, declare it as a private static final class member. Collections Let’s assume that we want to know whether a given list contains the specific element: Java list.contains("red"); In fact, this call invokes code like this: Java int n = list.size(); for (int i = 0; i < n; i++) { if ("red".equals(list.get(i))) { return true; } } Starting from Java 8, we have a streaming API that just hides from us the same gory details: Java list.stream().anyMatch("red"::equals); This is perfectly fine when the list is short, changes frequently, or is searched only occasionally. But if the list is large, stable, and searched repeatedly, a HashSet is the right tool — offering average O(1) lookup instead of O(n). If you cannot change the original data structure, converting it once at initialization time and searching the Set from that point forward is almost always worth it. If both the guaranteed element order and the fast lookup are needed, we can either hold duplicated data structures — a list for ordering and a set for search or just use LinkedHashSet, which solves both problems. Another common case is case-insensitive search. We already saw above that the combination of toLowerCase() or toUpperCase() with comparison significantly reduces the performance. This can be solved by using TreeSet with custom comparator, e.g. String.CASE_INSENSITIVE_ORDER: Java Set<String> set = new TreeSet<>(String.CASE_INSENSITIVE_ORDER); This gives you a sorted, case-insensitive set with no extra allocations - and the same approach works for TreeMap when your data is key-value pairs. Enum Lookups Everyone knows that an enum entry can be found by its name using a built-in method valueOf(s). However, what to do if the given string is lowercase while enum entries following the naming convention are called using capital letters? Some people use a combination of toUpperCase() and valueOf() that work just fine but have the penalty we discussed above. However, very often people prefer to create a special field representing a “custom” name, so the simple enum like: Java enum Color { RED, GREEN, BLUE } Turns into: Java enum Color { RED("red"), GREEN("green"), BLUE("blue"), … } Let’s mention that this design has at least two disadvantages: Duplicate data: The custom name is the same as a built-in but in a different case, which can be solved much more easily. This allows using really custom names that, according to my experience, in most cases are not needed and just create so-called “edge cases” that, in turn, in most cases are just a signal of bad design and might cause a lot of “stupid” bugs. However, let’s continue. How do people often use this custom name? Java public static Color ofColor(String color) { return Arrays.stream(values()) .filter(c -> c.color.equals(color)) .findFirst() .orElseThrow(() -> new IllegalArgumentException("No enum constant %s.%s".formatted(Color.class.getName(), color))); } The implementation looks pretty nice, but this approach means that each call of ofColor() iterates over the list. Yes, in most cases enums are not huge, so the list is short, but anyway, why do this if we can just create a map from the custom name to the enum entry once during initialization and then use it with O(1) complexity? The following example solves both problems at once: it uses a case-insensitive map where the key is the standard name() of the enum entry during initialization: Java private static final Map<String, Color> colors = Arrays.stream(values()).collect(toMap(Enum::name, e -> e, (existing, replacement) -> replacement, () -> new TreeMap<>(CASE_INSENSITIVE_ORDER))); So, now the method ofColor() becomes trivial: Java public static Color ofColor(String color) { return Optional.ofNullable(colors.get(color)) .orElseThrow(() -> new IllegalArgumentException("No enum constant for " + color)); } One can argue that a map-based implementation is not always possible because sometimes the lookup criteria are too complex to be reduced to a simple key. Although I agree in general, I can say in turn that in many (if not in most) cases this is still possible. So far, the lookup key was a simple string. But what if the search criteria is a range rather than an exact value? Consider a more physically accurate model of colors as ranges of electromagnetic waves. Java public enum Color { BLUE(450, 495), GREEN(495, 570), RED(620, 750); …} How to implement the method ofWaveLength(int waveLength)? The straight-forward way is to iterate over the values of the enum and compare the given wave length with the range for each entry, i.e. implement O(n) search. But we can do better using NavigableMap, which is designed exactly for this kind of range query: Java private static final NavigableMap<Integer, Color> wavelengthMap = Arrays.stream(values()) .collect(Collectors.toMap( color -> color.minNm, color -> color, (existing, replacement) -> existing, TreeMap::new )); Unfortunately, the search method is not as trivial as in the previous example, but still very simple and fast: Java public static Color ofWaveLength(int nm) { return Optional.ofNullable(wavelengthMap.floorEntry(nm)) .map(Entry::getValue) .filter(value -> nm <= value.maxNm) .orElseThrow(() -> new IllegalArgumentException("No enum constant for wavelength: " + nm + " nm")); } Now, let’s compare the performance. Operation * 1 million times Time, ms valueOf(s) 34 valueOf(toUpperCase(s)) 78 Iteration with equals() 40 Color.ofColor() iteration 166 Color.ofColor() map 20 Color.ofWaveLength() map 32 The table shows that: As expected, toUpperCase() reduces performance twiceIteration with call of equals is a little bit more expensive than valueOf() although the enum has only three members and will grow linearly as the enum grows. The more members enum has, the more time iteration takes. Map-based implementation is even faster than one based on the built-in valueOf(). Stream-based iteration (ofColor() iteration) is surprisingly slow. Stream setup overhead (boxing, lambda dispatch, spliterator initialization) is non-trivial for tiny collections Pre-Intitialization The principle here is: do not do something several times if you can do it once. The most trivial example is string or numeric constants: Java private static final String FILE_NAME = "config.json"; private static final int MAX_VALUE = 10_000; However, the same principle applies to heavier objects — and that is where it really matters. Let’s take a look at logging. Most people are used to writing the following “magic” line at the beginning of each class (unless we use Lombok’s @Slf4j annotation): Java private static final Logger logger = LoggerFactory.getLogger(MyClass.class); Are all these modifiers (private static final) really needed? Some people try to save typing time: Java private final Logger logger = LoggerFactory.getLogger(MyClass.class); Moreover, if the logger is not static, we can do even more: Java private final Logger logger = LoggerFactory.getLogger(getClass()); This line looks better because it is error-proof: the class here is not hard-coded, so this line can be copied as-is from one class to another or inherited from the base class. So, what’s the problem? The problem is that retrieving the correct logger is potentially expensive due to synchronized registry lookups. Doing this on every instantiation adds up. A friend of mine told me that once in the company where he worked, this change in some critical path improved performance so much that they managed to reduce the AWS cluster by about one hundred large EC2 machines. The same rule applies to pattern compilation. As the benchmark table showed, compiling a pattern on every method call is nearly ten times slower than reusing a precompiled one. The result of Pattern.compile() should always be stored in a static final field. The only exception is the case when the regular expression is generated dynamically, but we should do our best to avoid such a design. Very often we have to format or parse dates. Traditionally I used SimpleDateFormat. What can be more obvious than this: Java private static final String FORMAT = "yyyy-MM-dd HH:mm:ss"; private static final DateFormat format = new SimpleDateFormat(FORMAT); Frankly speaking, I did this many times following the principle I stated above: there is no reason to create the instance every time we need it if we can create it only once. The problem is that SimpleDateFormat is not thread-safe, so sharing the same instance among different threads can cause the problem. Even worse: we can live with this bug for years without knowing about it, since it only happens under high load and in some cases can just produce slightly wrong results that can be lost in an ocean of valid data. So, should we create instances of SimpleDateFormat every time we need it and cause CPU and GC to work hard? Fortunately, starting from Java 8, we can use DateTimeFormatter instead: Java private static final DateTimeFormatter formatter = DateTimeFormatter.ofPattern(DATE_FORMAT); This class is thread-safe, so we can share its instance among different threads and get consistent results. Conclusion We started with a quote that is almost always cited incomplete. Knuth never said ignore performance — he said don't sacrifice clarity for speculative gains, while reminding us not to pass up opportunities in that critical 3%. The examples in this article live in that 3%. None of the performance issues described here should ever appear in production code. They are not hard to avoid — they require no profiler, no benchmarking framework, no architectural discussion. Just the habit of reaching for the right tool. And that habit pays off. Choosing equalsIgnoreCase() over toLowerCase().equals() is cleaner and faster. A static final logger is simpler and cheaper. A pre-built enum map is more readable and O(1). Good code and efficient code are not in conflict here — they are the same code. The only thing required is the habit of pausing for a second and asking: am I doing this n times when once would do? All code examples from this article are available on Gist.

Goose — the open-source, Rust-based AI developer agent from Block (donated to the Linux Foundation’s Agentic AI Foundation) — interacts natively with your local development environment via the Model Context Protocol (MCP). In this tutorial, you will learn how to build stateless, cloud-native Java microservices using Quarkus LangChain4j and expose them as governed MCP extensions that Goose can discover and run seamlessly. Autonomous AI coding agents like Goose go far beyond simple code autocompletion. Built in Rust for speed and portability, Goose runs on your local machine, inspects files, runs terminal commands, and uses tools over MCP to automate complex engineering tasks. However, when developers want an AI agent to query enterprise microservices, trigger database migrations, or fetch internal API metrics, writing custom local scripts or ad-hoc wrappers is brittle and dangerous. The solution is to build a stateless MCP Tool Server in Java using Quarkus LangChain4j. Quarkus provides near-zero startup time and low memory footprint, while LangChain4j makes exposing @Tool methods via standard MCP HTTP/JSON-RPC trivial. Architecture: How Goose Integrates With Quarkus MCP Markdown ┌────────────────────────────────────────────────────────┐ │ Goose AI Agent (Rust Runtime) │ │ (Local CLI / Desktop App / ACP Server) │ └───────────────────────────┬────────────────────────────┘ │ Model Context Protocol (MCP) │ JSON-RPC over Stateless HTTP ▼ ┌────────────────────────────────────────────────────────┐ │ Quarkus LangChain4j MCP Server │ │ - @Tool Annotations & Bean Validation │ │ - Reactive SmallRye Mutiny Execution │ │ - GraalVM Native Image Ready │ └───────────────────────────┬────────────────────────────┘ │ Reactive Clients ▼ Enterprise APIs / Databases / Dev UI Goose Agent (Client): Executes on the developer machine, orchestrating LLM tool loops via MCP.MCP HTTP Transport: Goose sends structured tool calls to the Quarkus backend as stateless HTTP POST requests using standardized MCP methods (tools/list, tools/call).Quarkus Microservice: Validates parameters with Jakarta Bean Validation, executes reactive business logic, and returns structured data to Goose. Step 1: Configuring Dependencies in Quarkus Create a new Quarkus project or update your pom.xml to include quarkus-langchain4j-mcp and Reactive: Note: Find the completed demo application here: https://github.com/danieloh30/governed-mcp-tools.git. XML <dependencyManagement> <dependencies> <dependency> <groupId>${quarkus.platform.group-id}</groupId> <artifactId>${quarkus.platform.artifact-id}</artifactId> <version>${quarkus.platform.version}</version> <type>pom</type> <scope>import</scope> </dependency> </dependencies> </dependencyManagement> <dependencies> <dependency> <groupId>io.quarkus</groupId> <artifactId>quarkus-arc</artifactId> </dependency> <dependency> <groupId>io.quarkus</groupId> <artifactId>quarkus-rest-jackson</artifactId> </dependency> <dependency> <groupId>io.quarkiverse.mcp</groupId> <artifactId>quarkus-mcp-server-http</artifactId> <version>2.0.0.CR2</version> </dependency> <dependency> <groupId>io.quarkus</groupId> <artifactId>quarkus-hibernate-validator</artifactId> </dependency> <dependency> <groupId>io.quarkus</groupId> <artifactId>quarkus-junit</artifactId> <scope>test</scope> </dependency> </dependencies> Step 2: Implementing Hardened MCP Tools We will create a Customer Services MCP Tool that Goose can call when an engineer asks: "Goose, check the database status for customer CUST-4091 and fetch their recent telemetry." By placing @Tool annotations on CDI beans, Quarkus LangChain4j automatically registers the class as an MCP server endpoint: Embedded Javascript @ApplicationScoped public class CustomerServiceTools { @Tool(description = "Retrieve the current account status, service tier, and primary deployment region for a given customer.") public Uni<CustomerStatusResponse> getCustomerStatus( @ToolArg(description = "Customer ID formatted as CUST-XXXX") @NotNull @Pattern(regexp = "^CUST-[0-9]{4,8}$") String customerId) { CustomerStatusResponse response = switch (customerId) { case "CUST-4091" -> new CustomerStatusResponse("CUST-4091", "ACTIVE", "ENTERPRISE_TIER", "US-EAST-1"); case "CUST-2187" -> new CustomerStatusResponse("CUST-2187", "ACTIVE", "BUSINESS_TIER", "EU-WEST-1"); case "CUST-7734" -> new CustomerStatusResponse("CUST-7734", "SUSPENDED", "STARTER_TIER", "AP-SOUTH-1"); default -> new CustomerStatusResponse(customerId, "NOT_FOUND", "UNKNOWN", "UNKNOWN"); }; return Uni.createFrom().item(response); } @Tool(description = "Retrieve recent health-check logs and diagnostic metrics for a specified availability zone.") public Uni<List<String>> getZoneHealthLogs( @ToolArg(description = "Zone identifier, e.g., US-EAST-1") @Size(max = 20) String zoneId) { return Uni.createFrom().item(List.of( "[" + zoneId + "] CPU utilization: 42% (healthy)", "[" + zoneId + "] Memory pressure: 31% (normal)", "[" + zoneId + "] Network I/O: 1.2 Gbps ingress / 0.8 Gbps egress", "[" + zoneId + "] Disk IOPS: 12,400 read / 8,300 write (within SLA)", "[" + zoneId + "] Active connections: 18,230 (capacity: 50,000)", "[" + zoneId + "] Last incident: none in past 72 hours" )); } @Tool(description = "Track the current status, item count, and estimated delivery for an enterprise order.") public Uni<OrderStatusResponse> getOrderStatus( @ToolArg(description = "Order ID formatted as ORD-XXXXXXXX") @NotNull @Pattern(regexp = "^ORD-[0-9]{8}$") String orderId) { OrderStatusResponse response = switch (orderId) { case "ORD-20240815" -> new OrderStatusResponse("ORD-20240815", "SHIPPED", 12, "$48,750.00", "2024-08-22", "US-EAST-1"); case "ORD-20240901" -> new OrderStatusResponse("ORD-20240901", "PROCESSING", 5, "$12,300.00", "2024-09-10", "EU-WEST-1"); case "ORD-20241003" -> new OrderStatusResponse("ORD-20241003", "DELIVERED", 28, "$134,500.00", "2024-10-08", "AP-SOUTH-1"); default -> new OrderStatusResponse(orderId, "NOT_FOUND", 0, "$0.00", "N/A", "UNKNOWN"); }; return Uni.createFrom().item(response); } @Tool(description = "Retrieve SLA compliance metrics including uptime, latency, and violation count for a service.") public Uni<SLAComplianceResponse> getSLACompliance( @ToolArg(description = "Service identifier, e.g., api-gateway, auth-service") @NotNull @Size(max = 40) String serviceId) { SLAComplianceResponse response = switch (serviceId) { case "api-gateway" -> new SLAComplianceResponse("api-gateway", 99.97, "45ms", 99.99, 0, "2024-Q3"); case "auth-service" -> new SLAComplianceResponse("auth-service", 99.82, "120ms", 99.95, 3, "2024-Q3"); case "data-pipeline" -> new SLAComplianceResponse("data-pipeline", 98.50, "340ms", 99.80, 12, "2024-Q3"); case "notification-hub" -> new SLAComplianceResponse("notification-hub", 99.91, "78ms", 99.97, 1, "2024-Q3"); default -> new SLAComplianceResponse(serviceId, 0.0, "N/A", 0.0, -1, "N/A"); }; return Uni.createFrom().item(response); } ... } Step 3: Enabling the MCP Extension in application.properties Configure your Quarkus MCP server settings: Properties files quarkus.mcp-server.server-info.name=customer-tools quarkus.mcp-server.server-info.version=1.0.0 quarkus.mcp-server.http.root-path=/mcp quarkus.log.category."io.quarkiverse.mcp".level=DEBUG Launch Quarkus in dev mode: Shell ./mvnw quarkus:dev Step 4: Connecting Goose to Your Quarkus MCP Server Goose can be extended with any MCP server over stdio or HTTP. Configure Goose by editing its YAML configuration file or using the Goose CLI. Option A: Using the Goose CLI Register the Quarkus MCP server directly in your terminal: Shell goose extension add customer-tools \ --type http \ --uri http://localhost:8080/mcp Option B: Editing ~/.config/goose/config.yaml Add the Quarkus backend to your Goose extensions configuration: YAML extensions: customer-tools: enabled: true type: http uri: http://localhost:8080/mcp headers: Content-Type: "application/json" Step 5: Testing the Developer Workflow Launch Goose via CLI or the Desktop App: Shell goose session Prompt Goose: Developer: "I'm debugging customer CUST-4091. Use customer-tools to fetch their account tier, and then check the health logs for their primary region." Frontend UI: Developer: Choose one of the Tool explorers. Select the “Run tool” button on the right panel. Verify the audit events. What Happens Under the Hood Discovery: Goose sends an HTTP POST /mcp JSON-RPC tools/list request. Quarkus responds with JSON schema definitions derived from getCustomerStatus and getZoneHealthLogs.Tool Invocation 1: Goose parses the prompt, formats a tools/call JSON payload with {"customerId": "CUST-4091"}, and posts it to Quarkus.Execution and validation: Quarkus executes Hibernate Bean Validation. Since CUST-4091 matches ^CUST-[0-9]{4,8}$, it runs getCustomerStatus and returns primaryRegion: US-EAST-1.Tool Invocation 2: Goose sees US-EAST-1, triggers getZoneHealthLogs("US-EAST-1"), receives the green health metrics, and summarizes the complete diagnostic report back to you in the CLI. Summary and Next Steps By wrapping Java business logic in Quarkus LangChain4j @Tool beans, you give local AI developer agents like Goose secure, validated access to enterprise backend systems. However, when hundreds of developers run local Goose agents against shared backend microservices in production, connecting them directly creates security and governance risks. Coming up in Part 2: We will introduce agentgateway — the Linux Foundation data plane proxy —to sit between Goose and Quarkus. We will configure OAuth2/OIDC authentication, fine-grained tool-level RBAC, and rate limiting to harden our enterprise AI infrastructure.

By Daniel Oh CORE

In a chapter of The SingleStore Cookbook, there is a complete sentiment analysis pipeline using Rust compiled to WebAssembly and loaded directly into SingleStore via its Code Engine. The result was clean: one CLI command to deploy, sentiment scoring running inside the database engine alongside the data and a full stock-price-plus-headlines analytical pipeline built on top of it. Can we do the same thing in Neo4j? Neo4j has a fully documented, officially supported extensibility model that lets us write custom functions and procedures in Java and register them directly with the database engine. Java also has a port of Valence Aware Dictionary and sEntiment Reasoner (VADER), the same lexicon-based sentiment analyzer used in the SingleStore Rust implementation. The pieces are all there. The question is how well they would fit together and what the resulting pipeline would look like compared to the SingleStore Wasm approach. This article documents an experiment from start to finish: the UDF implementation, the graph schema, a complete data loading and scoring pipeline, and a full set of analytical queries. Along the way, we also discovered that Neo4j has a second path to sentiment analysis via NLP procedures, and the choice between the two turns out to be an interesting engineering decision in its own right. The goal here isn't to claim a new sentiment-analysis technique. It's to explore what Neo4j's extension model makes possible and how the result compares with the equivalent SingleStore implementation. The full source code is available on GitHub. What We Are Building Figure 1 shows how data moves through the pipeline. CSV files are loaded into Neo4j via LOAD CSV or the Python loader. As each Headline node is created, sentiment.score() is called inline in the same Cypher statement — scoring happens inside the database at ingestion time, not in a separate application step. The resulting graph is then available for the analytical queries covered later in the article. Figure 1. Pipeline data flow The pipeline mirrors the one in the SingleStore book chapter: A VADER-based sentiment function registered with the system and callable from queriesA graph containing synthetic stock price ticks and news headlinesA set of analytical queries: per-headline scoring, daily aggregation, sentiment-vs-price joins, most positive and most negative ranking, and a live consistency check For the example in this article, we'll need a local install of Neo4j, a Docker container, or a server where we can place files and restart the process. How Neo4j Extensibility Works Neo4j lets us extend Cypher with custom Java code packaged as a .jar file. This is a fully documented and supported extensibility path. Neo4j publishes official guidance on setting up a plugin project and maintains a Neo4j Procedure Template on GitHub. Neo4j provides this extensibility model for building custom extensions. There are several extension types: User-defined functions (UDFs) – take inputs, return a single value, called inline in a query like a built-in functionUser-defined aggregation functions (UDAs) – group-level aggregation, analogous to SUM or COLLECTProcedures – more flexible, can return multiple rows and perform side effects, called with CALL For our sentiment use case, a UDF is the right fit. We pass in a string and get back a map of polarity scores. In SingleStore, the equivalent was a Table-Valued Function (TVF) that returned a row set. A Neo4j UDF returning a Map<String, Double> is the closest structural equivalent. One practical note on naming is that Neo4j maintains a list of reserved and deprecated procedure namespaces, such as db.*, dbms.*, graph.* and others. These are off-limits. The sentiment.* namespace is not reserved or deprecated, so it's a safe choice. Check User-defined procedures before choosing a namespace for any new plugin to confirm it doesn't conflict with a built-in namespace. What to Know Before We Build Because a Neo4j UDF runs inside the same JVM as the database engine, it's worth understanding a few practical considerations before diving in. These are the same considerations that apply to any extension of a running JVM process — Neo4j's own plugin authors deal with them too — and being aware of them upfront makes for a smoother build experience. Memory. If a plugin allocates more memory than the JVM has available — for example, loading a very large model file or accumulating state across calls — it can trigger an OutOfMemoryError. The VADER UDF we build here loads a compact lexicon and holds no state, so this is not a concern in practice. For more complex plugins that allocate significant heap memory, Neo4j provides a preview ProcedureMemory API where we can register allocations against the configured transaction memory limits, which prevents uncapped growth from causing database restarts. Uncaught exceptions. An unhandled RuntimeException in a UDF propagates up through the Neo4j query execution engine. Good error handling in the UDF code keeps this from becoming a problem. Infinite loops and thread starvation. A UDF that hangs — waiting on a network call, deadlocked or stuck in a loop — ties up a JVM thread from Neo4j's shared pool. The VADER UDF makes no network calls, holds no state and performs a relatively small amount of computation per call, so this is not a concern here, but it matters for more complex plugins. Dependency conflicts. Because the plugin jar shares the classpath with the database engine, any library bundled into the fat jar must not conflict with libraries Neo4j already ships. This problem was encountered during development and more on that in the build section below, including a straightforward fix. Startup failures. A jar that fails to load prevents the system from starting. The solution is always to test in a development environment first, such as Neo4j Desktop or a local Docker container, before deploying anywhere more critical. Security. A Java plugin has full access to the JVM, filesystem and network. This is the same trust model as Neo4j's own plugins and is appropriate for code we've written and reviewed. For third-party plugins from untrusted sources, the same caution applies as for any third-party code running inside a critical process. AuraDB. AuraDB supports plugins provided and certified by Neo4j, such as APOC, GDS and GenAI, but not arbitrary third-party or custom jars. The Java UDF approach in this article requires self-managed Neo4j, such as Desktop, Docker or a server install. If AuraDB is the target, the Java UDF approach described here is not available; the GenAI plugin or an external service are the alternatives. None of this should discourage us from building a Java UDF. The VADER UDF we build here is small, does one thing, makes no network calls, holds no state and uses a well-tested library. The sensible approach, which applies to any plugin development, is to build and test on a local development instance first, then deploy with confidence. In Neo4j, the steps to deploy our UDF are: Build a fat jarStop the serverCopy the jar file to the server's plugins directoryAdd an allowlist entry to neo4j.confRestart the server The deployment model differs from the Wasm approach — more on that in the build and deploy section below. Setting Up the Project Prerequisites We'll need the following before starting: Java 21 – check with java -version. Java 21 is the version used by the official Neo4j plugin template and by this articleMaven 3.8+ – check with mvn -versionNeo4j 2026.06.0 – the version used for this article, running in one of the ways described below Choosing a Neo4j Install For this experiment, we'll use either Neo4j Desktop or Docker. Neo4j also supports server installs on Linux and Windows — the plugin mechanism is the same — but we did not test that path and don't provide instructions for it here. Neo4j Desktop is the easiest starting point. Download it from Neo4j for Desktop, create a new project and start a local database server. Find the exact path to the plugins directory by clicking Open folder > plugins. Docker is convenient for a clean, throwaway environment. The command below starts Neo4j 2026.06.0 with a plugins volume mounted to a local directory, which is where we'll drop the jar: Shell mkdir -p ~/neo4j/plugins ~/neo4j/data docker run \ --name neo4j-sentiment \ -p 7474:7474 -p 7687:7687 \ -v ~/neo4j/plugins:/plugins \ -v ~/neo4j/data:/data \ -e NEO4J_AUTH=neo4j/password \ -e NEO4J_dbms_security_procedures_allowlist="sentiment.*" \ neo4j:2026.06.0 With Docker we pass the allowlist as an environment variable rather than editing neo4j.conf directly. The jar goes into ~/neo4j/plugins/ on the host. Creating the Project Structure Create a new Maven project directory: Shell mkdir neo4j-sentiment-udf cd neo4j-sentiment-udf The full directory tree should look like this when finished: Plain Text neo4j-sentiment-udf/ ├── pom.xml └── src/ ├── main/ │ └── java/ │ └── sentiment/ │ └── Sentimentable.java └── test/ └── java/ └── sentiment/ └── SentimentableTest.java The sections below cover each part in turn. Next, we'll create both source directories: Shell mkdir -p src/main/java/sentiment mkdir -p src/test/java/sentiment Maven Dependencies We'll create a pom.xml file in the project root. The structure follows the official Neo4j procedure template at Neo4j Procedure Template, with three adjustments specific to this project that are explained below. XML <?xml version="1.0" encoding="UTF-8"?> <project xmlns="http://maven.apache.org/POM/4.0.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd"> <modelVersion>4.0.0</modelVersion> <groupId>org.neo4j.example</groupId> <artifactId>sentimentable</artifactId> <version>1.0.0-SNAPSHOT</version> <packaging>jar</packaging> <name>Neo4j Sentiment UDF</name> <description>VADER sentiment analysis as a Neo4j user-defined function</description> <properties> <java.version>21</java.version> <maven.compiler.release>${java.version}</maven.compiler.release> <neo4j.version>2026.06.0</neo4j.version> </properties> <!-- ADJUSTMENT 1: JitPack required for VaderSentimentJava --> <repositories> <repository> <id>jitpack.io</id> <url>https://jitpack.io</url> </repository> </repositories> <dependencies> <dependency> <groupId>org.neo4j</groupId> <artifactId>neo4j</artifactId> <version>${neo4j.version}</version> <scope>provided</scope> </dependency> <!-- ADJUSTMENT 2: VaderSentimentJava runtime dependency --> <dependency> <groupId>com.github.apanimesh061</groupId> <artifactId>VaderSentimentJava</artifactId> <version>v1.1.1</version> </dependency> <!-- Test dependencies — let neo4j-harness manage JUnit version --> <dependency> <groupId>org.neo4j.test</groupId> <artifactId>neo4j-harness</artifactId> <version>${neo4j.version}</version> <scope>test</scope> </dependency> <dependency> <groupId>org.neo4j.driver</groupId> <artifactId>neo4j-java-driver</artifactId> <version>6.0.2</version> <scope>test</scope> </dependency> </dependencies> <build> <plugins> <plugin> <artifactId>maven-compiler-plugin</artifactId> <configuration> <source>21</source> <target>21</target> </configuration> </plugin> <plugin> <groupId>org.apache.maven.plugins</groupId> <artifactId>maven-surefire-plugin</artifactId> <version>3.5.4</version> </plugin> <plugin> <artifactId>maven-shade-plugin</artifactId> <version>3.5.1</version> <executions> <execution> <phase>package</phase> <goals><goal>shade</goal></goals> <configuration> <!-- ADJUSTMENT 3: relocate commons-lang3 to avoid version conflict with Neo4j's internal copy --> <relocations> <relocation> <pattern>org.apache.commons.lang3</pattern> <shadedPattern>sentiment.shaded.org.apache.commons.lang3</shadedPattern> </relocation> </relocations> <artifactSet> <excludes> <exclude>org.neo4j:*</exclude> </excludes> </artifactSet> <shadedArtifactAttached>false</shadedArtifactAttached> </configuration> </execution> </executions> </plugin> </plugins> </build> </project> The three adjustments from the official template are called out inline as comments. Everything else — groupId convention, provided scope for the Neo4j dependency, the shade plugin structure and the test dependency pattern — follows the official guidance. Writing the UDF We'll create the file src/main/java/sentiment/Sentimentable.java and paste in the following: Java package sentiment; import com.vader.sentiment.analyzer.SentimentAnalyzer; import com.vader.sentiment.analyzer.SentimentPolarities; import org.neo4j.procedure.Description; import org.neo4j.procedure.Name; import org.neo4j.procedure.UserFunction; import java.util.Map; public class Sentimentable { @UserFunction("sentiment.score") @Description("Score a string with VADER. Returns compound, positive, negative, neutral.") public Map<String, Double> score(@Name("text") String text) { if (text == null || text.isBlank()) { return Map.of("compound", 0.0, "positive", 0.0, "negative", 0.0, "neutral", 1.0); } final SentimentPolarities polarities = SentimentAnalyzer.getScoresFor(text); return Map.of( "compound", (double) polarities.getCompoundPolarity(), "positive", (double) polarities.getPositivePolarity(), "negative", (double) polarities.getNegativePolarity(), "neutral", (double) polarities.getNeutralPolarity() ); } } The following implementation details are worth highlighting. The v1.1.1 API uses a static method — SentimentAnalyzer.getScoresFor(text) — rather than a mutable instance. This means there is no shared state between calls, which is what we want in a Neo4j UDF where multiple Cypher queries may invoke the function concurrently. The VADER lexicon is loaded internally by the library on first call and cached for subsequent calls. The @UserFunction("sentiment.score") annotation registers the method as callable from Cypher under that name. The @Name annotation on the parameter provides the argument name for Neo4j's function metadata and documentation — UDFs are always called with positional arguments in Cypher, as shown throughout this article: sentiment.score(row.headline). The return type is Map<String, Double>. In Cypher, this surfaces as a map literal, so callers can destructure it with dot notation: sc.compound, sc.positive and so on. In the SingleStore version, the TVF returned a row set and was used in a FROM clause. Here the UDF is called inline in a WITH or RETURN clause instead. Writing the Tests Following the official Neo4j procedure template pattern, we'll use neo4j-harness to spin up a lightweight embedded Neo4j instance in JUnit, register our UDF with it and run Cypher queries against it — all without deploying to a running database. This is the recommended testing approach in Neo4j's own documentation. We'll create the file src/test/java/sentiment/SentimentableTest.java and paste in the following: Java package sentiment; import org.junit.jupiter.api.AfterAll; import org.junit.jupiter.api.BeforeAll; import org.junit.jupiter.api.Test; import org.junit.jupiter.api.TestInstance; import org.neo4j.driver.Driver; import org.neo4j.driver.GraphDatabase; import org.neo4j.driver.Session; import org.neo4j.harness.Neo4j; import org.neo4j.harness.Neo4jBuilders; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertTrue; @TestInstance(TestInstance.Lifecycle.PER_CLASS) public class SentimentableTest { private Neo4j embeddedDatabaseServer; private Driver driver; @BeforeAll void initializeNeo4j() { this.embeddedDatabaseServer = Neo4jBuilders.newInProcessBuilder() .withDisabledServer() .withFunction(Sentimentable.class) .build(); this.driver = GraphDatabase.driver(embeddedDatabaseServer.boltURI()); } @AfterAll void closeNeo4j() { this.driver.close(); this.embeddedDatabaseServer.close(); } @Test void scorePositiveSentence() { try (Session session = driver.session()) { var scores = session.run( "RETURN sentiment.score('The movie was great') AS scores" ).single().get("scores").asMap(); assertTrue((Double) scores.get("compound") > 0.5); assertTrue((Double) scores.get("positive") > 0.0); assertEquals(0.0, (Double) scores.get("negative")); } } @Test void capitalizationIncreasesScore() { try (Session session = driver.session()) { var normal = session.run( "RETURN sentiment.score('The movie was great') AS scores" ).single().get("scores").asMap(); var caps = session.run( "RETURN sentiment.score('The movie was GREAT!') AS scores" ).single().get("scores").asMap(); assertTrue((Double) caps.get("compound") > (Double) normal.get("compound")); } } @Test void emptyStringReturnsNeutral() { try (Session session = driver.session()) { var scores = session.run( "RETURN sentiment.score('') AS scores" ).single().get("scores").asMap(); assertEquals(0.0, (Double) scores.get("compound")); assertEquals(1.0, (Double) scores.get("neutral")); } } @Test void nullStringReturnsNeutral() { try (Session session = driver.session()) { var scores = session.run( "RETURN sentiment.score(null) AS scores" ).single().get("scores").asMap(); assertEquals(0.0, (Double) scores.get("compound")); assertEquals(1.0, (Double) scores.get("neutral")); } } } The four tests mirror the tests we'll run manually in Neo4j Browser, but now they run automatically as part of the build. Neo4jBuilders.newInProcessBuilder() starts a lightweight embedded instance with the Sentimentable function registered; .withDisabledServer() skips the HTTP server since we only need the Bolt connection. The structure follows the official JoinTest.java pattern. Building and Deploying Step 1: Install the Maven Wrapper and build The official Neo4j procedure template uses the Maven Wrapper (mvnw), which means we only need Java installed, not a separate Maven installation. To add the wrapper to the project: Shell mvn wrapper:wrapper Then build and run the tests: Shell ./mvnw clean package Or to skip the tests during development: Shell ./mvnw clean package -DskipTests To use a globally installed Maven directly, mvn clean package -DskipTests works equally well — the wrapper is a convenience, not a requirement. Maven compiles the Java source, runs the Shade plugin and writes two jar files to target/. The one we want is sentimentable-1.0.0-SNAPSHOT.jar — the fat jar with VADER bundled inside. The original-sentimentable-1.0.0-SNAPSHOT.jar is the plain jar without dependencies, so we'll ignore it. If the build fails with a package org.neo4j.procedure does not exist error, check that the pom.xml has <scope>provided</scope> on the Neo4j dependency and that the version matches the running Neo4j instance. Step 2: Copy the Jar to the Plugins Directory Neo4j Desktop: Stop the serverOpen folder > plugins and copy sentimentable-1.0.0-SNAPSHOT.jar into that folderOpen folder > conf > neo4j.conf, find dbms.security.procedures.allowlist= and uncomment the line if it is commented outAdd sentiment.* to the end of the line Docker: Copy to the host directory mounted as /plugins: Shell cp target/sentimentable-1.0.0-SNAPSHOT.jar ~/neo4j/plugins/ Step 3: Whitelist the Function Namespace Neo4j's default dbms.security.procedures.allowlist is *, which loads all plugins. If an allowlist is configured with specific entries, any custom namespace must be included or the function will silently be unavailable — no error on startup, it simply won't exist. It's good practice to configure an explicit allowlist following the principle of least privilege. Our UDF uses only the public Neo4j procedure API, which means it doesn't require the separate dbms.security.procedures.unrestricted setting — that's only needed for extensions that access internal APIs. Step 4: Restart Neo4j Neo4j Desktop: Restart the server using the button in the Desktop UI. If Desktop shows "stopped" immediately after starting, open http://localhost:7474 directly — the server may be running before the UI reflects it. Docker: If this is the initial launch, no restart is needed — the docker run command in the Choosing a Neo4j Install section already starts Neo4j with the jar in place from the mounted plugins directory. If updating the jar after the container is already running, stop the container, replace the jar in ~/neo4j/plugins/ and then restart: Shell docker stop neo4j-sentiment cp target/sentimentable-1.0.0-SNAPSHOT.jar ~/neo4j/plugins/ docker start neo4j-sentiment The clearest confirmation that the plugin loaded correctly is to run the verification queries in step 5 below — if sentiment.score() is visible and returns results, the jar was picked up successfully. Verifying the Function We can interact with Neo4j by entering http://localhost:7474 in the browser. Step 5: Confirm the Function Loaded First, we'll check that Neo4j can see the function at all: Cypher SHOW FUNCTIONS YIELD name WHERE name STARTS WITH 'sentiment' RETURN name; Expected output: Plain Text +-----------------+ | name | +-----------------+ | sentiment.score | +-----------------+ If this returns zero rows, the jar is either not in the plugins directory, the allowlist entry is missing or misspelled or Neo4j was not fully restarted. Step 6: Run the Tests Run the following tests: Cypher RETURN sentiment.score('The movie was great') AS scores; Expected output: JSON { neutral: 0.4230000078678131, negative: 0.0, positive: 0.5770000219345093, compound: 0.6248999834060669 } Now we'll test that VADER's capitalization awareness is working: Cypher RETURN sentiment.score('The movie was GREAT!') AS scores; Expected output: JSON { neutral: 0.36899998784065247, negative: 0.0, positive: 0.6309999823570251, compound: 0.7289999723434448 } The compound score rises with the capitalized GREAT!, exactly as in the Wasm version. For the examples we tested, the Java port produces scores consistent with the Rust crate used in the book chapter. Now, we'll test the null guard. Passing an empty string should return a neutral result rather than an exception: Cypher RETURN sentiment.score('') AS scores; Expected output: JSON { neutral: 1.0, negative: 0.0, positive: 0.0, compound: 0.0 } If all three return the expected values, the UDF is working and we're ready to build the graph schema and load data. Designing the Graph Schema The graph model for this pipeline has three node labels, as shown in Figure 2. A central Stock node connects to Tick nodes via HAS_TICK relationships and to Headline nodes via HAS_HEADLINE relationships. VADER polarity scores are stored directly on each Headline node at ingestion time, making them available to any Cypher query without recomputing. Figure 2. Graph data model Plain Text (:Stock {symbol}) -[:HAS_TICK]-> (:Tick {symbol, ts, open, high, low, close, volume}) -[:HAS_HEADLINE]->(:Headline {id, symbol, ts, headline, url, publisher, compound, positive, negative, neutral}) The Stock node acts as the join key. In SingleStore the queries join tick and stock_sentiment on (symbol, DATE(ts)); in Neo4j that same co-reference is expressed by traversing from a shared Stock node to both Tick and Headline nodes with a date predicate. The relationship replaces the foreign key. Let's now run these commands to create constraints and indexes: Cypher CREATE CONSTRAINT tick_pk IF NOT EXISTS FOR (t:Tick) REQUIRE (t.symbol, t.ts) IS NODE KEY; CREATE CONSTRAINT headline_id IF NOT EXISTS FOR (h:Headline) REQUIRE h.id IS UNIQUE; CREATE CONSTRAINT stock_id IF NOT EXISTS FOR (s:Stock) REQUIRE s.symbol IS UNIQUE; CREATE INDEX tick_symbol_ts IF NOT EXISTS FOR (t:Tick) ON (t.symbol, t.ts); CREATE INDEX headline_symbol_ts IF NOT EXISTS FOR (h:Headline) ON (h.symbol, h.ts); Loading Data and Scoring Headlines Getting the Datasets The datasets, notebook and SQL files for the original SingleStore book chapter are all publicly available in the book's GitHub repository. The two CSV files we need are in the datasets subdirectory: fictitious_stocks.csv – synthetic daily OHLCV stock prices (random-walk model, fictitious symbols)raw_fictitious_headlines.csv – programmatically generated news headlines (templates + ticker symbols + financial events) We'll download both files into our local working directory. Dataset Format fictitious_stocks.csv has seven columns. The date and Name columns are renamed to ts and symbol, respectively, to match the graph schema: Plain Text date,open,high,low,close,volume,Name 2013-01-02,743.98,756.93,736.15,745.68,9142645,BBRQ-FX 2013-01-03,764.41,779.16,757.72,765.16,1208771,BBRQ-FX ... raw_fictitious_headlines.csv has five columns that map directly to the Headline node properties: Plain Text headline,url,publisher,ts,symbol BBRQ-FX stock record revenues after analyst update,http://www.hill.net/,The Stock Chronicle,2014-10-22,BBRQ-FX ... No preprocessing is needed beyond what the loader already does, such as dropping nulls, filtering the one extreme volume outlier and sorting by date. The Python Loader The data_loader.py below reads the two CSV files and writes them into Neo4j via the Python driver. Install the dependencies first if not already done so: Shell pip install -r requirements.txt Then run the loader, substituting the actual paths to the downloaded CSV files. Also replace your_password_here with your actual password. Python # data_loader.py import pandas as pd from neo4j import GraphDatabase from tqdm import tqdm URI = "bolt://localhost:7687" AUTH = ("neo4j", "your_password_here") TICK_CSV = "fictitious_stocks.csv" RAW_CSV = "raw_fictitious_headlines.csv" driver = GraphDatabase.driver(URI, auth=AUTH) def chunks(df, size): for i in range(0, len(df), size): yield df.iloc[i:i+size].to_dict("records") # load tick data tick_df = (pd.read_csv(TICK_CSV) .dropna() .query("volume <= 2_147_483_647") .rename(columns={"date": "ts", "Name": "symbol"}) .sort_values(["ts", "symbol"])) tick_batches = list(chunks(tick_df, 1000)) print(f"Loading {len(tick_df):,} tick rows in {len(tick_batches)} batches...") with driver.session() as session: for batch in tqdm(tick_batches, desc="Ticks", unit="batch"): session.run(""" UNWIND $rows AS row MERGE (s:Stock {symbol: row.symbol}) CREATE (t:Tick {symbol: row.symbol, ts: date(row.ts), open: row.open, high: row.high, low: row.low, close: row.close, volume: toInteger(row.volume)}) CREATE (s)-[:HAS_TICK]->(t) """, rows=batch) # load headlines and score at ingestion time raw_df = pd.read_csv(RAW_CSV) raw_batches = list(chunks(raw_df, 1000)) print(f"Loading {len(raw_df):,} headline rows in {len(raw_batches)} batches...") with driver.session() as session: for batch in tqdm(raw_batches, desc="Headlines", unit="batch"): session.run(""" UNWIND $rows AS row MATCH (s:Stock {symbol: row.symbol}) WITH s, row, sentiment.score(row.headline) AS sc CREATE (h:Headline { id: randomUUID(), symbol: row.symbol, ts: datetime(row.ts), headline: row.headline, url: row.url, publisher: row.publisher, compound: sc.compound, positive: sc.positive, negative: sc.negative, neutral: sc.neutral }) CREATE (s)-[:HAS_HEADLINE]->(h) """, rows=batch) print("Done.") driver.close() Run the Python program: Shell python data_loader.py The key line is sentiment.score(row.headline) AS sc inside the Cypher. This is doing what the sentimentable(i.headline) TVF call does in the SingleStore INSERT ... SELECT — computing scores at the database level in the same operation that writes the record, with no round-trip to the application layer. One important note if we need to re-run the loader is that the script uses CREATE for Tick and Headline nodes, so running it a second time without clearing the database will create duplicates rather than overwriting. Clear the database first with the following Cypher, using the Query tab: Cypher MATCH (n) CALL { WITH n DETACH DELETE n } IN TRANSACTIONS OF 100 ROWS; The batch size of 100 is deliberate — larger values can exceed the default transaction memory limit and fail. After clearing, re-run the schema constraints and indexes before running the loader again. Alternative Loading Directly From GitHub With LOAD CSV To stay entirely within Cypher and avoid Python, Neo4j's LOAD CSV command can fetch the files directly from GitHub over HTTPS. No file copying, no import directory, no Python dependencies. Run both queries using the Query tab in order — ticks first, then headlines, since the headlines query does a MATCH on Stock nodes created by the tick query. Cypher LOAD CSV WITH HEADERS FROM 'https://...' AS row CALL { WITH row MERGE (s:Stock {symbol: row.Name}) CREATE (t:Tick { symbol: row.Name, ts: date(row.date), open: toFloat(row.open), high: toFloat(row.high), low: toFloat(row.low), close: toFloat(row.close), volume: toInteger(row.volume) }) CREATE (s)-[:HAS_TICK]->(t) } IN TRANSACTIONS OF 1000 ROWS; LOAD CSV WITH HEADERS FROM 'https://...' AS row CALL { WITH row MATCH (s:Stock {symbol: row.symbol}) WITH s, row, sentiment.score(row.headline) AS sc CREATE (h:Headline { id: randomUUID(), symbol: row.symbol, ts: datetime(row.ts), headline: row.headline, url: row.url, publisher: row.publisher, compound: sc.compound, positive: sc.positive, negative: sc.negative, neutral: sc.neutral }) CREATE (s)-[:HAS_HEADLINE]->(h) } IN TRANSACTIONS OF 1000 ROWS; LOAD CSV WITH HEADERS reads the first row as column names, so the original names (row.Name, row.date) are mapped directly to the graph property names inline — the same column renaming the Python loader does with rename(). The IN TRANSACTIONS OF 1000 ROWS batching is required for the tick file at ~600,000 rows to avoid the transaction memory limit. The same delete-before-reload rule applies here: re-running either query without clearing the database first will create duplicates. The only requirement is that Neo4j has outbound HTTPS access to reach GitHub, which is the case for Desktop and local Docker. In a network-restricted server environment the Python loader with local files is the safer fallback. Next, some example queries to test using the Query tab. Headline-Level Sentiment Cypher MATCH (h:Headline) RETURN h.symbol AS symbol, date(h.ts) AS ts, left(h.headline, 30) AS headline, round(h.positive, 3) AS positive, round(h.negative, 3) AS negative, round(h.neutral, 3) AS neutral ORDER BY h.symbol, h.ts LIMIT 10; Aggregate Sentiment by Stock and Day Cypher MATCH (h:Headline) WITH h.symbol AS symbol, date(h.ts) AS ts, avg(h.positive) AS avg_positive, avg(h.negative) AS avg_negative, avg(h.neutral) AS avg_neutral, count(h) AS num_headlines RETURN symbol, ts, round(avg_positive, 3) AS avg_positive, round(avg_negative, 3) AS avg_negative, round(avg_neutral, 3) AS avg_neutral, num_headlines ORDER BY symbol, ts LIMIT 10; Join Sentiment With Closing Price In Cypher, the shared Stock node makes the symbol join implicit and we only need a date predicate. Cypher MATCH (t:Tick)<-[:HAS_TICK]-(s:Stock)-[:HAS_HEADLINE]->(h:Headline) WHERE date(t.ts) = date(h.ts) RETURN t.symbol AS symbol, date(t.ts) AS ts, round(t.close, 2) AS close, round(h.positive, 3) AS positive, round(h.negative, 3) AS negative, round(h.neutral, 3) AS neutral ORDER BY t.symbol, t.ts LIMIT 10; Most Positive Headlines Cypher MATCH (h:Headline) RETURN h.symbol, date(h.ts) AS ts, left(h.headline, 30) AS headline, round(h.positive, 3) AS positive ORDER BY h.positive DESC LIMIT 10; Most Negative Headlines Cypher MATCH (h:Headline) RETURN h.symbol, date(h.ts) AS ts, left(h.headline, 30) AS headline, round(h.negative, 3) AS negative ORDER BY h.negative DESC LIMIT 10; In the SingleStore book, CEO scandal headlines dominated the negative ranking across multiple stocks. We see the same pattern here because the underlying VADER lexicon is identical. Validate Stored Scores Against Live UDF Calls This mirrors the consistency check from the SingleStore book, where stored stock_sentiment values were compared against a fresh JOIN LATERAL sentimentable(...) call to confirm the ingestion pipeline was deterministic. Cypher MATCH (h:Headline {symbol: 'BBRQ-FX'}) WITH h, sentiment.score(h.headline) AS live RETURN h.symbol AS symbol, date(h.ts) AS ts, left(h.headline, 30) AS headline, CASE WHEN round(h.positive, 3) = round(live.positive, 3) AND round(h.negative, 3) = round(live.negative, 3) AND round(h.neutral, 3) = round(live.neutral, 3) THEN 'match' ELSE 'not match' END AS comparison LIMIT 10; Daily Average Sentiment vs. Closing Price The CTE-style aggregation from the book translates naturally to Cypher's WITH chaining. Cypher MATCH (h:Headline) WITH h.symbol AS symbol, date(h.ts) AS ts, avg(h.positive) AS avg_positive, avg(h.negative) AS avg_negative, avg(h.neutral) AS avg_neutral MATCH (t:Tick {symbol: symbol}) WHERE date(t.ts) = ts RETURN symbol, ts, round(t.close, 2) AS daily_close, round(avg_positive, 3) AS avg_positive, round(avg_negative, 3) AS avg_negative, round(avg_neutral, 3) AS avg_neutral ORDER BY symbol, ts LIMIT 10; What We Learned The experiment was a clear success. VADER runs inside Neo4j, scores headlines at ingestion time via a simple Cypher call and all the analytical queries from the SingleStore book have direct equivalents in Cypher. For the examples we tested, the Java port produces scores consistent with the Rust crate used in the SingleStore book — although independent language ports may differ in edge cases due to differences in tokenization or floating-point handling. The graph model handles the stock-tick-plus-headlines domain naturally and in several respects the Cypher queries are more expressive than their SQL counterparts — the relationship traversal from a shared Stock node replaces a keyed SQL join in a way that reflects the actual structure of the domain rather than just being an implementation detail. The graph model is a genuine advantage for the join queries. Replacing JOIN tick ON (symbol, DATE(ts)) with a graph traversal through a shared Stock node is not just syntactic preference — it reflects the actual structure of the domain. A stock symbol connects ticks and headlines naturally as a graph entity and Cypher expresses that more directly than a keyed SQL join. In-database scoring works. Calling sentiment.score(row.headline) inside the Cypher CREATE statement means scoring and ingestion happen in the same operation, with no round-trip to an application layer. This is the same goal the SingleStore Wasm pipeline achieves and the Java UDF delivers it cleanly. The dependency conflict is a one-time fix. We hit the commons-lang3 version conflict during development and it stopped the server from starting. The fix — relocating the bundled classes to a private namespace using the Maven Shade plugin — is straightforward once we know what to look for and the solution is baked into the pom.xml in this article. There are also honest differences from the SingleStore Wasm approach. Deployment requires a restart. SingleStore uses a tool that loads a function into a live database with no downtime. Neo4j requires a jar build, a file copy, a config edit and a restart. For an initial Docker launch, the jar is picked up automatically — but any subsequent update to the jar requires a container restart. The Maven Wrapper and the clear deployment steps in this article make the process repeatable. No execution sandbox. SingleStore runs each Wasm function instance in its own isolated process with a hard memory boundary. The Neo4j UDF runs in the same JVM as the server. For a small, well-behaved plugin like the VADER UDF this makes no practical difference, but it's a meaningful architectural distinction for more complex or heavyweight plugins. Language is JVM-based. The Wasm approach accepts any language that compiles to the Wasm core spec. Neo4j's extensibility model is JVM-only. For teams that want to bring existing Python or Rust models into the database, that is worth knowing about upfront. Alternative Approaches The Java UDF is the focus of this article, but it's not the only way to bring sentiment scoring close to Neo4j data. We considered several alternatives during the experiment. Some are compelling for specific use cases and others less so. Knowing the options helps us choose the right tool for our situation. Pre-scoring outside the database. Score all headlines before loading. Add the polarity scores as columns in the CSV and load everything with LOAD CSV. Nothing custom runs inside Neo4j at all. For a batch pipeline like this one, where data are loaded once and queried many times, this is entirely practical and requires no Java knowledge. The only thing we give up is the ability to call sentiment.score() inline in Cypher at query time. For many teams this will be the right answer and it's the simplest path to a working pipeline. External microservice. Deploy a small Python or Rust service that runs VADER and exposes an HTTP endpoint. An external microservice can expose VADER through an HTTP API, with the application layer calling the service before or during ingestion. This gives us complete process isolation — a crash in the sentiment service cannot touch the database — and works with AuraDB. The tradeoff is network latency on every call and the operational overhead of running a separate service. For lower-volume or interactive use cases it's a clean, flexible pattern. Neo4j GenAI plugin. Neo4j's GenAI plugin supports calling embedding and LLM APIs — OpenAI, Azure OpenAI and compatible endpoints — directly from Cypher. It's fully managed by Neo4j, works on AuraDB and requires no Java. To use a cloud LLM for sentiment classification rather than VADER’s lexicon is a well-supported, low-friction path. The tradeoff is API cost and the opacity of a large language model compared to VADER's fully transparent, inspectable lexicon — which matters in regulated domains where we need to explain a score. GraalVM native compilation. GraalVM can ahead-of-time compile Java UDFs to native binaries, reducing JVM startup overhead and memory footprint. This is a performance optimization rather than an architectural change — the code still runs inside the Neo4j process — and adds significant build complexity for modest gain in this use case. It is worth knowing about for larger, more heavyweight plugins, but not the right choice here. Wasm runtime embedded inside a Java UDF. Theoretically, we could embed a Wasm runtime such as wasmtime inside a Java UDF and execute the VADER Wasm module from within Neo4j, getting Wasm's sandbox guarantees inside Neo4j's plugin model. It's technically feasible but no published working example appears to exist and the complexity cost is high relative to the alternatives. An interesting idea to watch, but not practical today. The table below shows how these approaches compare on the dimensions that matter most. ApproachCompute locationAuraDBLanguage choiceOperational complexityPre-score outside DBCompleteYesAnyLowExternal microserviceCompleteYes (via APOC)AnyMediumAPOC NLP (cloud API)Remote serviceNo (APOC Extended required)N/ALowGenAI pluginRemote serviceYesN/ALowJava UDF (this article)Shared JVMNoJVM-basedMediumWasm-in-Java (theoretical)Wasm sandboxNoAny (via Wasm)Very high The Java UDF sits in the middle of this table — it's uniquely capable of calling sentiment.score() inline from any Cypher query without application-layer involvement and it runs entirely within the system without external API calls or network latency. Whether that inline, self-contained capability is what our use case needs is the key question. For development, experimentation and pipelines where the data and team are well understood, it's a compelling and practical approach. For other situations, the alternatives above offer different but equally valid tradeoffs. A Second Path Is APOC NLP Procedures The two approaches differ in where the computation happens, as shown in Figure 3. With the Java UDF, the VADER lexicon is bundled in the jar and scoring runs inside the Neo4j JVM — no network call, no external dependency, no per-call cost. With APOC NLP, Neo4j orchestrates calls to an external cloud API and receives scores back over the network. That single architectural difference drives most of the tradeoffs covered in this section. Figure 3. Java UDF vs. APOC NLP Neo4j already has sentiment analysis capability — it just works quite differently and it lives not in GDS but in APOC Extended, a separate component from APOC Core. APOC's NLP procedures act as wrappers around cloud-based Natural Language APIs. The supported providers are AWS Comprehend, Azure Cognitive Services and Google Cloud Natural Language. The calling pattern is straightforward. With AWS, for example: Cypher MATCH (h:Headline {symbol: 'BBRQ-FX'}) CALL apoc.nlp.aws.sentiment.stream(h, { key: $apiKey, secret: $apiSecret, nodeProperty: 'headline' }) YIELD value RETURN h.headline, value.sentiment, value.sentimentScore; And with Azure: Cypher MATCH (h:Headline {symbol: 'BBRQ-FX'}) CALL apoc.nlp.azure.sentiment.stream(h, { key: $apiKey, url: $apiUrl, nodeProperty: 'headline' }) YIELD value RETURN h.headline, value.sentiment, value.sentimentScore; The graph variant goes one step further and writes the sentiment result back as a node property automatically, with write: true in the config map. Choosing Between the Two Java VADER UDFAPOC NLP (AWS / Azure / GCP)Where scoring runsInside Neo4j JVMExternal cloud APINetwork call per batchNoYesCost per callNo API chargeAPI pricing appliesModel qualityLexicon-based (VADER)Cloud NLP / ML modelsAuraDB compatibleNoNo (APOC Extended not available in AuraDB)Java knowledge neededYesNoOffline / air-gappedYesNoDeterministic resultsYesProvider-dependentDomain tuningLimited (lexicon)Better (ML models handle context) The Java UDF is the stronger choice when scoring volume is high, API costs matter, the text is short social-media-style content that VADER was designed for, or an offline/air-gapped environment is required. The VADER lexicon is fully transparent — we can inspect why a string received a given score, which matters in regulated domains. APOC NLP is the stronger choice when Java knowledge is limited, the text requires linguistic nuance beyond VADER’s lexicon (negation, sarcasm, domain-specific vocabulary), or cloud NLP APIs are already in use for other workloads. One important constraint applies to both: APOC NLP is part of APOC Extended, not APOC Core. AuraDB includes APOC Core by default, but APOC Extended is not available in AuraDB — so neither the Java UDF nor APOC NLP works there. The GenAI plugin or an external microservice are the practical AuraDB paths. GDS, Neo4j's Graph Data Science library, does not include text-level sentiment analysis — it's graph-algorithm-oriented. Text scoring in Neo4j is either in-database via a Java UDF or delegated to a cloud NLP service via APOC. Summary The experiment confirms that Neo4j's Java extensibility model is a capable platform for in-database compute. The VADER UDF works, the graph model is a natural fit for the stock-tick-plus-headlines domain and the analytical queries translate cleanly from SQL to Cypher — in some cases more expressively, because the relationship between prices and headlines is explicit in the graph schema rather than inferred at query time through a join predicate. The more interesting engineering question is when to use a Java UDF versus the alternatives. The answer depends primarily on four factors: Deployment model (self-managed Neo4j only for UDFs)Latency and network requirements (the UDF has none; APOC NLP and external microservices introduce both)Model sophistication (VADER's lexicon is transparent and fast but limited; cloud NLP APIs offer better linguistic coverage)Operational constraints (Java knowledge, plugin management and the restart-on-update requirement all have a cost) There is no universally correct choice — the table in the APOC NLP section lays out the tradeoffs and reasonable teams will land in different places depending on their priorities. What the article does establish is that the approach works and is officially supported. Building a plugin is documented and templated. For development, experimentation and well-understood production pipelines, it's a practical and interesting path. To go further, the official Neo4j Procedure Template is an excellent starting point, neo4j-harness makes unit testing UDFs straightforward without needing a running database instance and the full Neo4j Java Reference covers procedures, aggregation functions and the complete extensibility API in depth. The full source code is available on GitHub.

By Akmal Chaudhri CORE

Java Meets the Spreadsheet Apache POI has been the standard Java library for reading and writing Excel files for over twenty years. It handles the majority of everyday spreadsheet tasks well. But a growing category of real-world Excel files now contains formulas that POI's evaluator cannot execute at all. This is one of several situations Java developers hit when working with spreadsheets that are not obvious until you are already in production. Business users produce, share, and reason about data in spreadsheets. Finance teams model in Excel. Operations teams track inventory in Excel. Analysts hand deliverables to engineering as .xlsx files. Java applications end up interacting with all of it: back-office services accept Excel uploads, pricing engines run calculations that were originally authored in a workbook, reporting tools export data in a format the recipient can open in Excel without formatting problems. Despite how common these situations are, "Java + spreadsheets" is not a topic most developers think about until they hit it for the first time. This article provides a practical overview of the category: common scenarios, moving parts, available approaches, and things that tend to catch teams by surprise. Three Common Scenarios Most Java developers who work with spreadsheets fall into one of three cases. It is worth locating yourself in one of them before evaluating tools. File Exchange (Headless Import and Export) The application reads uploaded Excel files and extracts data, or generates Excel files from database contents. There is no spreadsheet UI in the application itself. This is the most common case. Examples include batch data ingestion, report generation, and integration with third-party systems that expect .xlsx. In-App Calculation (Headless Formula Evaluation) The application uses spreadsheet-style formulas as calculation logic. Business users author pricing rules, tax formulas, or allocation logic in Excel; the Java application executes those formulas at runtime, sometimes against data the users never see. This scenario is less common but appears in fintech, insurance, and enterprise resource planning. In-App Editing (Embedded Spreadsheet UI) The application renders an interactive spreadsheet in the browser, similar to Excel Online. Users view, edit, and collaborate on workbooks inside the application. This is common in reporting tools, financial modeling platforms, and any application where end users need the flexibility of a spreadsheet without leaving the application. The three scenarios have very different technical requirements. A library that fits one may be a poor fit for another. What Working With Spreadsheets Actually Involves Developers often assume spreadsheet integration is primarily about reading cell values. In practice, most production issues arise from features beyond raw data: formula evaluation, formatting fidelity, workbook structure, and modern Excel behavior. File formats: The dominant format is .xlsx (Office Open XML). Older files use .xls (binary). Simpler tabular data is often exchanged as .csv, but CSV loses formulas, multiple sheets, formatting, and cell types. Any real spreadsheet integration has to handle .xlsx. Formulas and formula evaluation: Excel files often contain formulas that reference other cells. Reading the file gives you the formula text and the last cached value. Recalculating the formula requires an evaluator that understands Excel's formula language. Libraries vary widely in which functions they implement. Modern Excel behavior: Excel 365 and Excel 2021 introduced dynamic array formulas, spill behavior, and new functions such as UNIQUE, SORT, FILTER, LET, XLOOKUP, and LAMBDA. In a dynamic array formula, a single cell can produce a whole array of values that "spill" into neighboring cells. For example, =UNIQUE(A1:A100) entered in one cell produces the full list of distinct values from that range and fills as many cells as needed. Files created in modern Excel routinely contain these constructs. Older evaluation engines usually cannot execute them. Cell formatting and styling: Number formats, date formats, colors, borders, conditional formatting, merged cells. This matters both for accurate reading (a value formatted as a percentage means something different from a raw decimal) and for export fidelity. Custom number formats such as accounting-style parentheses for negative numbers, and Excel table styles, are among the formats most likely to be lost or changed on round-trip. Charts, images, and other embedded content: Some libraries preserve these on round-trip; others silently drop them. Data validation, filters, tables, and pivot tables: Structural features that users depend on. Coverage varies significantly across libraries. Not every application needs all of this. A batch job that only reads numeric data from a fixed template needs very little. An application that lets users upload arbitrary workbooks and edit them needs almost all of it. Approaches Available in Java There is no single "Java Excel library." The landscape has several categories, each with its own tradeoffs. Apache POI The de facto standard in the Java ecosystem for headless file processing. Open source, mature, widely used. Supports .xlsx and .xls read and write, and includes a formula evaluator. POI's formula evaluator implements around 250 built-in functions; functions outside that list raise NotImplementedException at evaluation time. Dynamic array formulas and spill behavior are not supported. A minimal POI read example: Java try (Workbook wb = WorkbookFactory.create(new File("data.xlsx"))) { Sheet sheet = wb.getSheetAt(0); Cell cell = sheet.getRow(0).getCell(0); System.out.println(cell.getStringCellValue()); } The boundary is the dynamic array family: SEQUENCE, FILTER, SORT, UNIQUE and TEXTSPLIT raise NotImplementedFunctionException at evaluation time, and spilled ranges have no representation in POI's cell model at all. LET is worse still: POI's formula grammar has no notion of variable binding, so a LET formula cannot even be parsed: Java // Cell A1 contains: =LET(total, SUM(B1:B100), total * 1.1) FormulaEvaluator eval = wb.getCreationHelper().createFormulaEvaluator(); Cell cell = sheet.getRow(0).getCell(0); eval.evaluate(cell); // threw: org.apache.poi.ss.formula.FormulaParseException: // Specified named range 'total' does not exist in the current workbook. The file itself opens without error, and reading the cached value works. It is only when the application needs to recalculate that the problem surfaces. (Verified the code above with POI 5.5.1). Commercial Headless Libraries Products such as Aspose.Cells offer broader formula coverage, better format fidelity, and more complete support for advanced features (charts, pivot tables, formatting). They are usually licensed per developer or per deployment. Teams typically choose these when POI's limitations become blockers and rewriting is not an option. Embedded Spreadsheet Components Products such as Keikai (Java) and SpreadJS (JavaScript) render an interactive spreadsheet UI in the browser and coordinate with the backend. They combine file I/O, formula evaluation, and rendering in a single component. Suitable for applications where end users need to view and edit workbooks directly. Cloud Spreadsheet Services Google Sheets API and Microsoft Graph let the application outsource the spreadsheet entirely and integrate over REST. The spreadsheet lives in the cloud service; the Java application reads and writes through the API. This works well when the workbook itself is the artifact users care about, and less well when the spreadsheet needs to be embedded inside a larger application experience. These categories can also be combined. It is common to use POI for backend generation and a separate embedded component for user-facing editing. Choosing an Approach Match the approach to the scenario. For file exchange, start with Apache POI. It is free, well-documented, and adequate for a large percentage of import/export use cases. Move to a commercial headless library if you hit specific limits: modern formula evaluation, complex formatting fidelity, or performance on large workbooks. For in-app calculation, evaluate the formula coverage of your candidate libraries carefully. If the formulas that need to run come from real Excel files authored by real users, they will include functions that not every engine supports. This is where dynamic arrays and modern functions matter most: a formula containing LET or UNIQUE will not evaluate correctly on a library that does not implement them. For in-app editing, POI alone is not enough because it has no UI. You need either an embedded spreadsheet component that runs in the browser, or a cloud spreadsheet service that you integrate with. The choice depends on how tightly the spreadsheet needs to fit into your application experience, and whether user data can leave your infrastructure. The three scenarios can also stack. A single application might use POI for backend batch ingestion, a headless engine for scheduled recalculation of business rules, and an embedded component for the end-user editing screen. Things That Catch Teams By Surprise A few practical issues that tend to appear later in a project than they should. Formula coverage is not uniform. Two libraries may both advertise "Excel formula support," and both fail on different subsets of real workbooks. Modern functions (UNIQUE, SORT, FILTER, LET, XLOOKUP, LAMBDA) are the most common gap. Verify with your actual files, not with synthetic examples. Dynamic array files behave differently on different engines. A file authored in Excel 365 with =UNIQUE(A1:A100) in one cell may open correctly (showing cached values), fail to recalculate, or throw an exception, depending on the library. If your application needs to recalculate uploaded files, this matters. Cached values can mislead you. When a library cannot evaluate a formula, it often falls back to the cached value stored in the file. This masks the problem during development, because everything looks correct. It only fails when the underlying data changes and the formula needs to be re-evaluated, which is often in production, not in testing. Formatting fidelity varies. Custom number formats, conditional formatting rules, and merged cell behavior are not preserved equally across libraries. If your workbook is going back to Excel users, test the round-trip explicitly with the exact templates your business owners use. Memory and performance scale non-linearly. Loading a 100,000-row workbook is a different problem from loading a 1,000-row workbook. Some libraries hold the entire workbook in memory as a rich object model, and applications typically start hitting issues in the range of tens of thousands of rows. Others offer streaming APIs (POI's SXSSF for write, XSSF event model for read) that trade the object model for scalability. If your use case involves large workbooks, benchmark early. Conclusion Spreadsheets remain one of the most widely used data tools in business, and Java applications increasingly need to interact with them. There is no single correct approach — the right one depends on whether you are exchanging files, running calculations, or embedding a spreadsheet UI. The available options have grown in the last few years, especially for teams that need to handle modern Excel behavior such as dynamic arrays and the newer function set. Understanding the scenarios and the moving parts before picking a library, and testing with the workbooks your real users produce, will save meaningful effort later.

By Hawk Chen CORE

Scaling JMS Listeners With Java Virtual Threads Event-driven architecture is widely used in enterprise systems to decouple services, absorb traffic spikes, and move work out of request paths. Java Message Service (JMS), now standardized as Jakarta Messaging, remains common in systems built around ActiveMQ, IBM MQ, Solace, TIBCO EMS, and similar brokers. Java 21 virtual threads give these systems another scaling option. A JMS listener often spends more time waiting on a database, HTTP service, cache, or file system than it spends using the CPU. Moving that blocking work to virtual threads can reduce platform-thread pressure without forcing the application into a reactive programming model. However, virtual threads do not make the broker, database, or downstream services unlimited. They also do not change acknowledgment, transaction, redelivery, or ordering semantics. A safe design combines virtual threads with bounded JMS consumer concurrency, explicit resource limits, idempotency, and production metrics. This article explains what virtual threads change for Spring JMS listeners, how to configure them explicitly, and how to avoid moving the bottleneck from the JVM into the rest of the system. The Traditional JMS Listener Model A typical queue-based flow moves messages from the broker through a Spring listener container and into a handler that calls downstream systems. Figure 1 compares how that handler work occupies platform threads with how it runs when the container's consumer-invoker tasks use virtual threads. Figure 1. Platform threads compared with virtual-thread consumer invokers in a Spring JMS listener. The container manages JMS connections, sessions, consumers, acknowledgments, and listener invocation. The handler contains the business logic: Java @JmsListener( destination = "orders.created", containerFactory = "jmsListenerContainerFactory" ) public void handle(OrderCreatedEvent event) { Customer customer = customerClient.getCustomer(event.customerId()); inventoryService.reserve(event.orderId(), customer); orderRepository.markAsProcessing(event.orderId()); } This code is easy to read, but each downstream operation may block. With platform threads, an operating-system-backed thread remains occupied while a query or network call is waiting. When enough listener threads are blocked, new messages wait even if the CPU is not saturated. The application has become thread-bound rather than CPU-bound. Before virtual threads, teams usually increased the listener thread pool, scaled out more service instances, or rewrote the flow around asynchronous or reactive APIs. Those options remain valid, but each has a cost. Larger platform-thread pools use more memory and add scheduling overhead. More instances increase infrastructure and operational work. Reactive code can scale efficiently, but it changes libraries, control flow, debugging, and error handling. What Virtual Threads Change A virtual thread is still a java.lang.Thread, but it is scheduled by the JVM rather than being permanently tied to one operating-system thread. The platform thread that temporarily runs a virtual thread is called its carrier. When a virtual thread blocks on supported I/O, the JVM can unmount it from the carrier. The carrier is then free to run another virtual thread. This lets an application maintain straightforward, sequential code while supporting many concurrent blocking operations. As Figure 1 shows, virtual threads that are waiting on supported I/O can unmount from their carriers, leaving those carriers available to execute other ready work. Virtual threads can improve throughput when platform-thread scarcity is the limiting factor. They do not make an individual database call or HTTP request faster, and they do not add CPU capacity. Good candidates include handlers dominated by: JDBC callsBlocking REST or gRPC clientsCache lookupsFile or object-storage operationsLegacy synchronous SDKsSynchronous orchestration across downstream systems Weak candidates include handlers dominated by: CPU-heavy transformationsEncryption or compressionImage or video processingMachine learning inferenceLarge in-memory aggregation The JDK guidance is to create a virtual thread per task rather than pool virtual threads. Limited resources should be protected with explicit mechanisms such as semaphores, rate limiters, connection pools, and framework concurrency settings. The JMS Detail That Changes the Design For Spring's DefaultMessageListenerContainer, a listener thread normally belongs to a consumer invoker. That invoker owns or reuses a JMS Session and MessageConsumer and may process many messages during its lifetime. Therefore, enabling virtual threads does not necessarily create one new virtual thread for every message. It places the container's consumer tasks on virtual threads. The distinction matters because raising concurrency also raises the number of active JMS consumers and sessions. Those broker-side resources are not as cheap as virtual threads. The right side of Figure 1 models this relationship explicitly: a configured consumer-invoker task runs on a virtual thread and may process multiple messages during its lifetime. This architecture is still useful. A consumer can unmount from its carrier while its handler waits on downstream I/O. But the listener container's concurrency remains the primary control over how many messages can be processed at once. Configure the JMS Executor Explicitly Spring Boot can enable virtual threads for several Boot-managed execution paths with spring.threads.virtual.enabled=true. Do not assume that this property alone proves that a JMS listener container uses virtual threads. Configure the JMS container's executor explicitly and verify it at runtime. Figure 2 separates the application wiring from the runtime flow. The explicit connection between the virtual-thread-enabled TaskExecutor and the JMS listener factory is the important step; the container's concurrency setting continues to bound active consumers and sessions. Figure 2. Explicit Spring JMS virtual-thread wiring and runtime message flow. The following example uses Java 21 or later and Spring Framework 6.1 or later. It supplies a virtual-thread-enabled SimpleAsyncTaskExecutor to the listener container factory: Java import java.util.concurrent.Executor; import jakarta.jms.ConnectionFactory; import org.springframework.beans.factory.annotation.Qualifier; import org.springframework.context.annotation.Bean; import org.springframework.context.annotation.Configuration; import org.springframework.core.task.SimpleAsyncTaskExecutor; import org.springframework.jms.config.DefaultJmsListenerContainerFactory; @Configuration(proxyBeanMethods = false) class JmsConfiguration { @Bean("jmsVirtualThreadExecutor") SimpleAsyncTaskExecutor jmsVirtualThreadExecutor() { SimpleAsyncTaskExecutor executor = new SimpleAsyncTaskExecutor("jms-vt-"); executor.setVirtualThreads(true); return executor; } @Bean DefaultJmsListenerContainerFactory jmsListenerContainerFactory( ConnectionFactory connectionFactory, @Qualifier("jmsVirtualThreadExecutor") Executor executor ) { DefaultJmsListenerContainerFactory factory = new DefaultJmsListenerContainerFactory(); factory.setConnectionFactory(connectionFactory); factory.setTaskExecutor(executor); // Example limits only. Derive these from load tests and // the safe capacity of the broker and downstream systems. factory.setConcurrency("10-100"); // Prefer transactional JMS acknowledgment when redelivery // on listener failure is required. factory.setSessionTransacted(true); return factory; } } SimpleAsyncTaskExecutor.setVirtualThreads(true) requires Java 21. Spring Framework 6.2 also added DefaultMessageListenerContainer.setVirtualThreads(true) for applications that construct the listener container directly and use its internal default executor. If a Spring Boot application uses Boot's DefaultJmsListenerContainerFactoryConfigurer, apply it before the explicit executor, concurrency, and transaction overrides so that other Boot JMS properties are retained. Virtual threads are daemon threads. In a non-web worker where no other non-daemon thread keeps the JVM alive, use Spring Boot's spring.main.keep-alive=true or an equivalent application-lifecycle mechanism. Do not rely on incidental threads created by a broker client to keep the process running. A small startup test can confirm the execution mode: Java if (!Thread.currentThread().isVirtual()) { throw new IllegalStateException( "The JMS listener is not running on a virtual thread" ); } Use this as a test or temporary diagnostic rather than performing it for every production message. Also confirm the active container factory when an application defines more than one. Bound Concurrency Around Real Capacity Virtual threads reduce thread scarcity. They do not remove resource scarcity. A listener can still be limited by: JMS sessions and consumersBroker prefetch, consumer windows, or creditDatabase connectionsHTTP client connectionsDownstream rate limitsMemory used by in-flight payloadsTransaction locksCPU A useful first estimate comes from Little's Law: Shell required concurrency ~= target throughput x average processing time If the target is 200 messages per second and the average handler time is 250 milliseconds, the initial estimate is: Shell 200 messages/second x 0.25 seconds = 50 concurrent handlers That value is only a starting point. It must be capped by the safe capacity of every dependency. If each message holds a database connection and the usable pool capacity is 30, setting listener concurrency to 100 may only create 70 additional waiters. If a payment API permits 40 concurrent requests, protect that call separately with a semaphore or rate limiter. The concurrency range 10-100 in the example means that the container can maintain a baseline and scale to a maximum. It does not guarantee that 100 is safe, and a maximum of 100 may be much too high for some brokers or workloads. Broker flow-control settings matter as well. Excessive prefetch can move a large backlog from the broker into consumers, increase the number of unacknowledged messages, and make recovery less predictable. Keep enough prefetched work to feed consumers, but avoid using prefetch as an unbounded application queue. Acknowledgment and Transactions Must Be Deliberate Virtual threads do not change message-delivery guarantees. This is especially important with Spring's DefaultMessageListenerContainer. In its default AUTO_ACKNOWLEDGE mode, the container acknowledges before listener execution, so a listener exception does not cause redelivery. If the application requires rollback and redelivery after a handler failure, use a transacted JMS session or an appropriately configured external transaction manager. A local JMS transaction covers JMS receipt and JMS sends performed through the same session. It does not automatically include a database transaction. A database commit can succeed, and the JMS commit can fail, causing the message to be delivered again. There are three common strategies: Use idempotent handlers and local transactions.Use an inbox/outbox design to make database effects repeatable and outbound publication reliable.Use JTA/XA when atomic coordination across JMS and another transactional resource is required, and its operational cost is justified. Figure 3 shows the inbox/outbox lifecycle, including the duplicate path, the separate JMS acknowledgment boundary, broker-managed redelivery, and dead-letter handling. Figure 3. Idempotent JMS processing, acknowledgment, retry, and dead-letter lifecycle. Do not treat @Transactional on a database service as proof that the JMS acknowledgment participates in the same transaction. Verify which transaction manager is active and which resources it coordinates. Make the Consumer Idempotent Redelivery can occur after broker failover, transaction rollback, application restart, timeout, or a failure between two resource commits. Higher concurrency also makes race conditions in duplicate detection easier to expose. An inbox table is a common solution. As shown in Figure 3, the application atomically inserts the message ID and applies the business changes in the same database transaction. A duplicate key follows a safe no-op path instead of repeating the business effect. The database must enforce a unique constraint on the message ID. A separate exists() check is not enough because two concurrent deliveries can both observe that the row is absent. Java @Transactional public void process(OrderCreatedEvent event) { boolean firstDelivery = processedMessageRepository.tryInsert(event.messageId()); if (!firstDelivery) { return; } orderService.apply(event); } tryInsert should use an atomic insert-if-absent operation protected by a unique key and report a duplicate without committing a separate transaction. Avoid catching a generic constraint exception if the persistence provider marks the whole transaction rollback-only. If the business update fails, the transaction should roll back both the inbox insert and the business changes. External side effects need their own idempotency strategy. For example, send an idempotency key to a payment API or persist an operation state before invoking a service that cannot participate in the local transaction. Keep Transactions and Retries Short Avoid holding a database or JMS transaction open while a slow external service retries for minutes. The risky pattern begins a transaction, calls an external API, waits and retries, and only then updates the database and commits. This can hold locks, database connections, JMS sessions, and unacknowledged messages. A virtual thread makes the waiting thread cheaper, but it does not release those resources. A safer design, illustrated in Figure 3, commits the business update and outbox record as local intent and continues asynchronously through an outbox publisher. The database update and outbox insert occur in one local transaction. A separate publisher sends pending outbox records and marks them complete. If the inbound JMS message is redelivered after the database commit, the inbox key prevents the business update and outbox insert from being repeated. Long retry delays should normally be handled with broker redelivery delay, a retry queue, or a scheduler. Sleeping a virtual thread is cheap from a carrier-thread perspective, but the listener may still hold a JMS consumer, session, transaction, and message during the delay. Classify errors before retrying: Failure typeTypical responseTransient network or dependency failureRetry with exponential backoff and jitterRate limitHonor the server's delay and reduce concurrencyInvalid message schemaSend to a dead-letter queueMissing required business dataDead-letter or route for correctionRepeated unknown failureStop after a bounded attempt count and alert Every production listener should define a maximum redelivery count, dead-letter destination, replay procedure, and owner for investigating poison messages. Do Not Detach Work From the Listener Carelessly A tempting design is to let the JMS listener receive a message, submit the real work to another executor, and return immediately. This can create more parallelism, but it can also acknowledge the message before the work finishes. It may also cross thread boundaries with a JMS Session, which is single-threaded by contract. Transaction context, error propagation, and redelivery behavior can all be lost. Let the listener container own the handler's execution unless the application deliberately implements a handoff protocol. A safe handoff usually means persisting the message or command durably before the listener returns, not merely placing a Runnable in an in-memory executor. Preserve Ordering Where It Matters Higher concurrency changes ordering behavior. Once a queue has multiple active consumers, messages can complete in a different order from the order in which the broker delivered them. Choose the ordering scope explicitly: Keep concurrency at one for strict global ordering.Partition or route messages by a business key.Serialize processing for the same key.Add sequence checks when events can arrive out of order.Design state transitions to reject stale events. Virtual threads are easiest to adopt when messages are independent or when ordering is limited to a partition or business key. For topics, do not increase consumer concurrency as if the destination were a queue. Depending on subscription configuration, additional topic consumers can receive additional copies of each message. Review durable and shared subscription semantics for the broker and container. Test the Bottleneck, Not Just the Thread Count An illustrative order-processing workload may perform one database read, two HTTP calls, one database update, and one outbound event for each message. Compare platform threads and virtual threads with: The same message corpus and payload distributionThe same acknowledgment and transaction settingsThe same database and HTTP pool limitsThe same broker prefetch or creditThe same retry and dead-letter policyA controlled concurrency ramp Measure more than throughput: metricwhat it revealsQueue depth and oldest-message ageBacklog and user-visible delayConsume rateSustainable throughputHandler p50, p95, and p99 latencyNormal and tail behaviorScheduled and active JMS consumersActual container concurrencyPlatform and virtual thread countsWhether thread pressure movedCarrier CPU and pinned-thread eventsScheduler or compatibility problemsDatabase pool utilization and wait timeDatabase saturationHTTP pool utilization and timeoutsOutbound connection pressureDownstream throttlingRate-limit pressureRedelivery and DLQ countsFailure amplificationHeap and garbage collectionCost of in-flight work Virtual threads are successful when the system sustains the required throughput with lower platform-thread pressure and without increasing timeouts, throttling, redelivery, or tail latency. If throughput rises while downstream errors rise faster, the system is not healthier. It is only delivering overload more efficiently. Diagnose Pinning and Provider Compatibility On Java 21, a virtual thread can pin its carrier when it blocks while executing certain synchronized or native code. Occasional short pinning is usually harmless. Frequent long pinning can reduce scalability. Use Java Flight Recorder's jdk.VirtualThreadPinned event or run a load test with: Shell -Djdk.tracePinnedThreads=full Do this with the actual JMS provider, JDBC driver, HTTP client, monitoring agents, and security libraries used in production. Compatibility cannot be inferred from a synthetic Thread.sleep benchmark. JDK 24's JEP 491 removes nearly all pinning caused by synchronized methods and blocks, but native or foreign-function interactions and third-party behavior still deserve testing. Decision Matrix scenariovirtual-thread fitBlocking JDBC callsStrongBlocking REST or gRPC callsStrongLegacy synchronous SDKsStrongHigh-volume, I/O-bound queue listenersStrong with bounded consumersCPU-heavy transformationWeakStrict global orderingLimitedSmall downstream capacityUseful only with strict limitsWeak acknowledgment or retry designFix delivery semantics firstNo observabilityAdd measurements first Production Checklist Before enabling virtual threads for JMS listeners, confirm that: The application runs on Java 21 or later.The JMS executor is explicitly configured and verified as virtual.Listener concurrency is capped by measured downstream capacity.Broker prefetch, consumer window, or credit is tuned.Acknowledgment and transaction behavior is documented and tested.Duplicate processing is prevented with an atomic idempotency mechanism.Retries are bounded, delayed, and classified.A dead-letter queue and replay process exist.Ordering requirements are explicit.Load tests use real drivers and representative dependencies.Queue age, tail latency, pool saturation, redelivery, and pinned-thread events are monitored. Conclusion Virtual threads are a strong fit for JMS listeners that spend much of their time waiting on blocking I/O. They let teams preserve simple, imperative Java code while reducing the platform-thread cost of concurrent message processing. The safe adoption pattern is not “turn on virtual threads and remove the limits.” It is: Put the listener container's consumer tasks on virtual threads.Bound consumer concurrency using broker and downstream capacity.Make acknowledgment, transactions, and idempotency explicit.Test with the real provider and dependencies.Measure where the bottleneck moves. When those controls are in place, virtual threads can modernize an established JMS application without requiring a reactive rewrite. They make waiting cheaper. The architecture still has to decide how much work the system can safely accept. References JEP 444: Virtual ThreadsOracle Java 21 Virtual Threads GuideSpring Framework: DefaultMessageListenerContainerSpring Framework: Processing JMS Messages Within TransactionsSpring Boot 3.2 Release Notes: Virtual Thread SupportJakarta Messaging 3.1 SpecificationJEP 491: Synchronize Virtual Threads Without Pinning