We all write unit tests but the challenge we face at times is that the unit under test might be dependent on other components. And configuring other components for unit testing is definitely an overkill. Instead we can make use of Mocks in place of the other components and continue with the unit testing. To show how one can use mocks, I have a Data access layer(DAL), basically a class which provides an API for the application to access and modify the data in the data repository. I then unit test the DAL without actually the need to connect to the data repository. The data repository can be a local database or remote database or a file system or any place where we can store and retrieve the data. The use of a DAL class helps us in keeping the data mappers separate from the application code. Lets create a Java project using maven. mvn archetype:generate -DgroupId=info.sanaulla -DartifactId=MockitoDemo -DarchetypeArtifactId=maven-archetype-quickstart -DinteractiveMode=false The above creates a folder MockitoDemo and then creates the entire directory structure for source and test files. Consider the below model class for this example: package info.sanaulla.models; import java.util.List; /** * Model class for the book details. */ public class Book { private String isbn; private String title; private List authors; private String publication; private Integer yearOfPublication; private Integer numberOfPages; private String image; public Book(String isbn, String title, List authors, String publication, Integer yearOfPublication, Integer numberOfPages, String image){ this.isbn = isbn; this.title = title; this.authors = authors; this.publication = publication; this.yearOfPublication = yearOfPublication; this.numberOfPages = numberOfPages; this.image = image; } public String getIsbn() { return isbn; } public String getTitle() { return title; } public List getAuthors() { return authors; } public String getPublication() { return publication; } public Integer getYearOfPublication() { return yearOfPublication; } public Integer getNumberOfPages() { return numberOfPages; } public String getImage() { return image; } } The DAL class for operating on the Book model class is: package info.sanaulla.dal; import info.sanaulla.models.Book; import java.util.ArrayList; import java.util.Arrays; import java.util.Collections; import java.util.List; /** * API layer for persisting and retrieving the Book objects. */ public class BookDAL { private static BookDAL bookDAL = new BookDAL(); public List getAllBooks(){ return Collections.EMPTY_LIST; } public Book getBook(String isbn){ return null; } public String addBook(Book book){ return book.getIsbn(); } public String updateBook(Book book){ return book.getIsbn(); } public static BookDAL getInstance(){ return bookDAL; } } The DAL layer above currently has no functionality and we are going to unit test that piece of code (TDD). The DAL layer might communicate with a ORM Mapper or Database API which we are not concerned while designing the API. Test Driving the DAL Layer There are lot of frameworks for Unit testing and mocking in Java but for this example I would be picking JUnit for unit testing and Mockito for mocking. We would have to update the dependency in Maven’s pom.xml 4.0.0 info.sanaulla MockitoDemo jar 1.0-SNAPSHOT MockitoDemo http://maven.apache.org junit junit 4.10 test org.mockito mockito-all 1.9.5 test Now lets unit test the BookDAL. During the unit testing we will inject mock data into the BookDAL so that we can complete the testing of the API without depending on the data source. Initially we will have an empty test class: public class BookDALTest { public void setUp() throws Exception { } public void testGetAllBooks() throws Exception { } public void testGetBook() throws Exception { } public void testAddBook() throws Exception { } public void testUpdateBook() throws Exception { } } We will inject the mock BookDAL and mock data in the setUp() as shown below: public class BookDALTest { private static BookDAL mockedBookDAL; private static Book book1; private static Book book2; @BeforeClass public static void setUp(){ //Create mock object of BookDAL mockedBookDAL = mock(BookDAL.class); //Create few instances of Book class. book1 = new Book("8131721019","Compilers Principles", Arrays.asList("D. Jeffrey Ulman","Ravi Sethi", "Alfred V. Aho", "Monica S. Lam"), "Pearson Education Singapore Pte Ltd", 2008,1009,"BOOK_IMAGE"); book2 = new Book("9788183331630","Let Us C 13th Edition", Arrays.asList("Yashavant Kanetkar"),"BPB PUBLICATIONS", 2012,675,"BOOK_IMAGE"); //Stubbing the methods of mocked BookDAL with mocked data. when(mockedBookDAL.getAllBooks()).thenReturn(Arrays.asList(book1, book2)); when(mockedBookDAL.getBook("8131721019")).thenReturn(book1); when(mockedBookDAL.addBook(book1)).thenReturn(book1.getIsbn()); when(mockedBookDAL.updateBook(book1)).thenReturn(book1.getIsbn()); } public void testGetAllBooks() throws Exception {} public void testGetBook() throws Exception {} public void testAddBook() throws Exception {} public void testUpdateBook() throws Exception {} } In the above setUp() method I have: Created a mock object of BookDAL BookDAL mockedBookDAL = mock(BookDAL.class); Stubbed the API of BookDAL with mock data, such that when ever the API is invoked the mocked data is returned. //When getAllBooks() is invoked then return the given data and so on for the other methods. when(mockedBookDAL.getAllBooks()).thenReturn(Arrays.asList(book1, book2)); when(mockedBookDAL.getBook("8131721019")).thenReturn(book1); when(mockedBookDAL.addBook(book1)).thenReturn(book1.getIsbn()); when(mockedBookDAL.updateBook(book1)).thenReturn(book1.getIsbn()); Populating the rest of the tests we get: package info.sanaulla.dal; import info.sanaulla.models.Book; import org.junit.BeforeClass; import org.junit.Test; import static org.junit.Assert.*; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.when; import java.util.Arrays; import java.util.List; public class BookDALTest { private static BookDAL mockedBookDAL; private static Book book1; private static Book book2; @BeforeClass public static void setUp(){ mockedBookDAL = mock(BookDAL.class); book1 = new Book("8131721019","Compilers Principles", Arrays.asList("D. Jeffrey Ulman","Ravi Sethi", "Alfred V. Aho", "Monica S. Lam"), "Pearson Education Singapore Pte Ltd", 2008,1009,"BOOK_IMAGE"); book2 = new Book("9788183331630","Let Us C 13th Edition", Arrays.asList("Yashavant Kanetkar"),"BPB PUBLICATIONS", 2012,675,"BOOK_IMAGE"); when(mockedBookDAL.getAllBooks()).thenReturn(Arrays.asList(book1, book2)); when(mockedBookDAL.getBook("8131721019")).thenReturn(book1); when(mockedBookDAL.addBook(book1)).thenReturn(book1.getIsbn()); when(mockedBookDAL.updateBook(book1)).thenReturn(book1.getIsbn()); } @Test public void testGetAllBooks() throws Exception { List allBooks = mockedBookDAL.getAllBooks(); assertEquals(2, allBooks.size()); Book myBook = allBooks.get(0); assertEquals("8131721019", myBook.getIsbn()); assertEquals("Compilers Principles", myBook.getTitle()); assertEquals(4, myBook.getAuthors().size()); assertEquals((Integer)2008, myBook.getYearOfPublication()); assertEquals((Integer) 1009, myBook.getNumberOfPages()); assertEquals("Pearson Education Singapore Pte Ltd", myBook.getPublication()); assertEquals("BOOK_IMAGE", myBook.getImage()); } @Test public void testGetBook(){ String isbn = "8131721019"; Book myBook = mockedBookDAL.getBook(isbn); assertNotNull(myBook); assertEquals(isbn, myBook.getIsbn()); assertEquals("Compilers Principles", myBook.getTitle()); assertEquals(4, myBook.getAuthors().size()); assertEquals("Pearson Education Singapore Pte Ltd", myBook.getPublication()); assertEquals((Integer)2008, myBook.getYearOfPublication()); assertEquals((Integer)1009, myBook.getNumberOfPages()); } @Test public void testAddBook(){ String isbn = mockedBookDAL.addBook(book1); assertNotNull(isbn); assertEquals(book1.getIsbn(), isbn); } @Test public void testUpdateBook(){ String isbn = mockedBookDAL.updateBook(book1); assertNotNull(isbn); assertEquals(book1.getIsbn(), isbn); } } One can run the test by using maven command: mvn test. The output is: ------------------------------------------------------- T E S T S ------------------------------------------------------- Running info.sanaulla.AppTest Tests run: 1, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.029 sec Running info.sanaulla.dal.BookDALTest Tests run: 4, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.209 sec Results : Tests run: 5, Failures: 0, Errors: 0, Skipped: 0 So we have been able to test the DAL class without actually configuring the data source by using mocks.

This story goes back at least a decade, when I was first approached by a PHB with a question “How big servers are we going to need to buy for our production deployment”. The new and shiny system we have been building was nine months from production rollout and apparently the company had promised to deliver the whole solution, including hardware. Oh boy, was I in trouble. With just a few years of experience down my belt, I could have pretty much just tossed a dice. Even though I am sure my complete lack of confidence was clearly visible, I still had to come up with the answer. Four hours of googling later I recall sitting there with the same question still hovering in front of my bedazzled face: “How to estimate the need for computing power?” In this post I start to open up the subject by giving you rough guidelines on how to estimate memory requirements for your brand new Java application. For the impatient ones – the answer will be to start with the memory equal to approximately 5 x [amount of memory consumed by Live Data] and start the fine-tuning from there. For the ones more curious about the logic behind, stay with me and I will walk you through the reasoning. First and foremost, I can only recommend to avoid answering a question phrased like this without detailed information being available. Your answer has to be based upon the performance requirements, so do not even start without clarifying those first. And I do not mean way-too-ambiguous “The system needs to support 700 concurrent users”, but a lot more specific ones about latency and throughput, taking into account the amount of data, usage patterns. One should also not forget about the budget also – we all can all dream about sub-millisecond latencies, but those without HFT banking backbone budgets – unfortunately it will only remain a dream. For now, lets assume you have those requirements in place. Next stop would be to create the load test scripts emulating user behaviour. If you are now able to launch those scripts concurrently you have built a foundation to the answer. As you might also have guessed, the next step involves our usual advice of measuring not guessing. But with a caveat. Live Data Size Namely, our quest for the optimal memory configuration requires capturing the Live Data Size. Having captured this, we have the baseline configuration in place for the fine-tuning. How does one define live data size? Charlie Hunt and Binu John in their “Java Performance” book have given it the following definition: Live data size is the heap size consumed by the set of long-lived objects required to run the application in its steady state. Equipped with the definition, we are ready to run your load tests against the application with the GC logging turned on (-XX:+PrintGCTimeStamps -Xloggc:/tmp/gc.log -XX:+PrintGCDetails) and visualize the logs (with the help of gcviewer for example) to determine the moment when the application has reached to the steady state. What you are after looks similar to the following: We can see the GC doing its job both with minor and Full GC runs in a familiar double-saw-toothed graphic. This particular application seems to have achieved a steady state already after the first full GC run on 21st second. In most cases however, it takes 10-20 Full GC runs to spot the change in trends. After four full GC runs we can estimate that the Live Data Size is equal to approximately 100MB. The aforementioned Java Performance book is now indicating that there is a strong correlation between the Live Data Size and the optimal memory configuration parameters in a typical Java EE application. The evidence from the field is also backing up their recommendations: Set the maximum heap size to 3-4 x [Live Data Size] So, for our application at hand, we should set the -Xmx to be in between 300m and 400m for the initial performance tests and take it from there. We have mixed feelings about other recommendations given in the book, recommending to set the maximum permanent generation size to 1.2-1.5 x [Live Data Size of the Permanent Generation] and the -XX:NewRatio being set to 1-1.5 x of the [Live Data Size]. We are currently gathering more data to determine whether the positive correlation exists, but until then I recommend to base your survival and eden configuration decisions on monitoring your allocation rate instead. Why should one bother you might now ask. Indeed, two reasons for not caring immediately surface: 8G memory chip is in sub $100 territory at the time of writing this article Virtualization, especially when using large providers such as Amazon AWS make adjusting the capacity easy Both of the reasons are partially valid and have definitely reduced the need for provisioning to be precise. But both of them are still putting you in the danger zone When tossing in huge amounts of memory “just in case” you are most likely going to significantly affect the latency – going into heaps above 8G it is darn easy to introduce Full GC pauses spanning over tens of seconds. When over-provisioning with the mindset of “lets tune it later”, the “later” part has a tendency of never arriving. I have faced numerous applications running on vastly over provisioned environments just because of this. For example the aforementioned application I discovered running on Amazon EC2 m1.xlarge instance was costing the company $4,200 per instance / year. Converting it to m1.small reduced the bill to just $520 for the instance. 8-fold cost reduction will be visible from your operations budget if your deployments are large, trust me on this. Summary Unfortunately I still see way too many decisions made exactly like I was forced to do a decade ago. This leads to the under- and over planning of capacity, both of which can be equally poor choices, especially if you cannot enjoy the benefits of virtualization. I got lucky with mine, but you might not get away with your guestimate, so I can only recommend to actually plan ahead using the simple framework described in this post. If you enjoyed the content, I can only recommend to follow our performance tuning advice in Twitter.

"And let Reform her columns roll. With thunder peal, and lightning flash..." - Ignis, "The Genius of Liberty" Vol III No. 2 Introduction In the past couple of articles we've seen how a Kanban board is able to help in the attainment of transparency and the stabilization of an agile team. Today we'll see if we can resolve one of the most common queries that result from this usage: how does a team decide which columns should appear on the board for tracking the progress of work items? The simplest case...and why it may not be enough When we set up a Kanban board in the last article, there were only three columns - or to use the correct term, "stations". These were a "Backlog" station (essentially a "To Do" list of work that has not yet been started), a station for showing which work is "In Progress", and a finally a station for representing the work that has been completed. You can't get much simpler than that, and it begs the question as to why you would want to make it more complicated. In practice however, there are at least two situations in which this minimalist approach will be found wanting: A team isn't cross-trained, and its members effectively work in skill silos. Consequently we can expect dependencies between team members, some of whom may become blocked while waiting for others to complete their part of the work. The incurral of this wasted time will not be apparent if it is all considered to be "work in progress". For example, the team may be split into developers and testers, and bottlenecks may arise as work passes between them. We may need to break Work In Progress down into further stations in order to expose this waste more fully. Bottlenecks arise due to constraints in the workflow, which is a different problem. In this situation a team might be fully cross-trained, and none of its members become blocked waiting for another. Rather, waste arises because the work itself is inefficiently staged. This often happens with activities like development, review, and test. For example if two people are required for a review, but only one is needed for development and test, then a bottleneck may well occur. Work will build-up awaiting review due to contention for these resources and the value of the investment in effort will start to depreciate. Again the incurral of this waste will not be apparent if it is all considered to be "work in progress". More stations are needed to expose it. Adding further stations These then are the two key things to consider when choosing additional stations. We're out to expose waste caused by work silos, or by the inappropriate staging of activities. Either of these can introduce constraints and become the source of bottlenecks in a value stream. Sometimes blockages can occur due to a dependency on something that must be done outside of the team. When this happens, it implies that the team are not fully in control of their own process, and consequently are unable to meet their own Definition of Done. They don't have all of the skills or resources needed. This is a problem and a contra-indication to agile practice. If it happens it's essential to make the dependency clear so that it can be challenged and removed. We may therefore choose to have an "externally blocked" column on the board to expose problems of this nature. It isn't really a station, because it doesn't represent a state in which value is added. Rather, it shows that an item has stalled within the value stream and that the team are not in a position to provide remedy. Another option is to place a red, day-glo sticky note on the ticket highlighting the seriousness of the problem. This is a clear signal that an impediment has occurred...that is to say, in Lean-Kanban terminology, it is an andon flag. In this case the flag shows that a major blockage has arisen and needs resolving. Challenging the boundaries Now we need to turn our attention to the boundaries of the board. There are two principal areas we should look at. Firstly, on the leftmost side of the board, we can see the work that a team inducts into its "Backlog" prior to actioning it. Secondly, on the rightmost side, we can see the work that the team considers to be "Done". These two boundaries are very often a source of waste. To understand why, just consider how backlogs are often allowed to grow without effective limit, and at how completed work may be permitted to accumulate in a "Done" column. These stations may not represent work in progress as far as the team is concerned, but it would be foolish to deny that they are batches too. After all, they are still part of the value stream. They represent inventory that is depreciating in value, or relevance, until something useful is done with it all. It behooves us to query the waste that is incurred, and to ask how the size of these batches may be constrained. Specifically: How can work be inducted into a backlog with minimal accumulation and delay? How can value be delivered to consumers as soon as work is completed? In short, what can be done to "lean" these process boundaries, so that inventory in the team's part of the value stream enters and exits in a "just-in-time" fashion? We can answer these questions by improving transparency still further. This can mean the refinement of the "Backlog" and "Done" columns into other, more finely-grained stations. For example, work might be building up in a Product Backlog because it is not being triaged appropriately, or perhaps because acceptance criteria are insufficiently well defined. We might be able to expose these problems by replacing a backlog with "Triaged", "Accepted", and "Ready" stations. At the other side of the board, completed work may be building up in the "Done" column because a release cannot yet be made. Additional stations such as "System Integrated", "In User Acceptance", and "Awaiting Release" could add clarity here. Removing stations The simple, 3 column board we started has now exploded into a behemoth of perhaps ten columns or more. This may seem like an excessively complex structure for a workflow and a casual observer may criticize it for being fundamentally unagile. After all, inventory should either be work in progress by an agile team, or it will be awaiting their attention or have already been completed. The criticism is a valid one but we need to bear one thing in mind: these stations are there to expose problems. Only once transparency has been attained can we hope to provide remedy. The bottlenecks, along with the diagnostic stations we added to reveal them, can then be removed. Conclusion Knowing how many "columns" to include on an agile board, and what they should be, is something of a black art to many agile teams. In this article we've looked at the issues involved in making this decision. The board of a fully cross-trained team should be elegant in its simplicity, but when problems arise we must be prepared to do some digging in order to root out their causes.

So far, we have just put the data in and out. And we have had a pretty good track record doing so. However, what do we do with the data now that we have it? As you can expect, we need to read it out. Usually by specific date ranges. The interesting thing is that we usually are not interested in just a single channel, we care about multiple channels. And for fun, those channel might be synchronized or not. An example of the first might be the current speed and the current engine temperature in a car. They are generally share the exact same timestamps. An example of out of sync is when you have a sensor on a rooftop measuring rainfall, and another sensor in the sewer measuring water flow rates. (Again, thanks to Dan for helping me with the domain). This is interesting, because it present quite a few interesting problems: We need to merge different streams into a unified view. We need to handle both matching and non matching sequences. We need to handle erroneous data, what happens when we have two reading for the same time for the same sensor? Yes, that shouldn’t happen, but it does. I solved this with the following API: public class RangeEntry { public DateTime Timestamp; public double?[] Values; } IEnumerable results = dts.ScanRanges(DateTime.MinValue, DateTime.MaxValue, new[] { "6febe146-e893-4f64-89f8-527f2dbaae9b", "707dcb42-c551-4f1a-9203-e4b0852516cf", "74d5bee8-9a7b-4d4e-bd85-5f92dfc22edb", "7ae29feb-6178-4930-bc38-a90adf99cfd3", }); This API gives me the results in the time order, with the same positions as the ids requested for the values. With nulls if there isn’t a value matching the value from that time in that particular sensor channel. The actual implementation relies on this method: IEnumerable ScanRange(DateTime start, DateTime end, string id) All this does it provide the entries all the entries in a particular date range, for a particular channel. Let us see how we implement multi channel scanning on top of this: private class PendingEnumerator { public IEnumerator Enumerator; public int Index; } private class PendingEnumerators { private readonly SortedDictionary> _values = new SortedDictionary>(); public void Enqueue(PendingEnumerator entry) { List list; var dateTime = entry.Enumerator.Current.Timestamp; if (_values.TryGetValue(dateTime, out list) == false) { _values.Add(dateTime, list = new List()); } list.Add(entry); } public bool IsEmpty { get { return _values.Count == 0; } } public List Dequeue() { if (_values.Count == 0) return new List(); var kvp = _values.First(); _values.Remove(kvp.Key); return kvp.Value; } } public IEnumerable ScanRanges(DateTime start, DateTime end, string[] ids) { if (ids == null || ids.Length == 0) yield break; var pending = new PendingEnumerators(); for (int i = 0; i < ids.Length; i++) { var enumerator = ScanRange(start, end, ids[i]).GetEnumerator(); if(enumerator.MoveNext() == false) continue; pending.Enqueue(new PendingEnumerator { Enumerator = enumerator, Index = i }); } var result = new RangeEntry { Values = new double?[ids.Length] }; while (pending.IsEmpty == false) { Array.Clear(result.Values,0,result.Values.Length); var entries = pending.Dequeue(); if (entries.Count == 0) break; foreach (var entry in entries) { var current = entry.Enumerator.Current; result.Timestamp = current.Timestamp; result.Values[entry.Index] = current.Value; if(entry.Enumerator.MoveNext()) pending.Enqueue(entry); } yield return result; } } We are getting a single entry from each channel into the pending enumerators. Then, we collate all the entries that share the same time into a single entry. We use the Index property to track the actual expected index of the entry in the output. And we handle duplicate times in the same channel by outputting multiple entries. Testing this on my 1.1 million records data set, we can get 185 thousands records back in 0.15 seconds.

JDave, Concordion, Easyb, JBehave, Cucumber are all compared here briefly for your convenience.

it's very important to learn hadoop by practice. one of the learning curves is how to write the first map reduce app and debug in favorite ide, eclipse. do we need any eclipse plugins? no, we do not. we can do hadoop development without map reduce plugins this tutorial will show you how to set up eclipse and run your map reduce project and mapreduce job right from your ide. before you read further, you should have setup hadoop single node cluster and your machine. you can download the eclipse project from github . use case: we will explore the weather data to find maximum temperature from tom white’s book hadoop: definitive guide (3rd edition) chapter 2 and run it using toolrunner i am using linux mint 15 on virtualbox vm instance. in addition, you should have hadoop (mrv1 am using 1.2.1) single node cluster installed and running, if you have not done so, would strongly recommend you do it from here download eclipse ide, as of writing this, latest version of eclipse is kepler 1. create new java project 2. add dependencies jars right click on project properties and select java build path add all jars from $hadoop_home/lib and $hadoop_home (where hadoop core and tools jar lives) 3. create mapper package com.letsdobigdata; import java.io.ioexception; import org.apache.hadoop.io.intwritable; import org.apache.hadoop.io.longwritable; import org.apache.hadoop.io.text; import org.apache.hadoop.mapreduce.mapper; public class maxtemperaturemapper extends mapper { private static final int missing = 9999; @override public void map(longwritable key, text value, context context) throws ioexception, interruptedexception { string line = value.tostring(); string year = line.substring(15, 19); int airtemperature; if (line.charat(87) == '+') { // parseint doesn't like leading plus // signs airtemperature = integer.parseint(line.substring(88, 92)); } else { airtemperature = integer.parseint(line.substring(87, 92)); } string quality = line.substring(92, 93); if (airtemperature != missing && quality.matches("[01459]")) { context.write(new text(year), new intwritable(airtemperature)); } } } 4. create reducer package com.letsdobigdata; import java.io.ioexception; import org.apache.hadoop.io.intwritable; import org.apache.hadoop.io.text; import org.apache.hadoop.mapreduce.reducer; public class maxtemperaturereducer extends reducer { @override public void reduce(text key, iterable values, context context) throws ioexception, interruptedexception { int maxvalue = integer.min_value; for (intwritable value : values) { maxvalue = math.max(maxvalue, value.get()); } context.write(key, new intwritable(maxvalue)); } } 5. create driver for mapreduce job map reduce job is executed by useful hadoop utility class toolrunner package com.letsdobigdata; import org.apache.hadoop.conf.configured; import org.apache.hadoop.fs.path; import org.apache.hadoop.io.intwritable; import org.apache.hadoop.io.text; import org.apache.hadoop.mapreduce.job; import org.apache.hadoop.mapreduce.lib.input.fileinputformat; import org.apache.hadoop.mapreduce.lib.output.fileoutputformat; import org.apache.hadoop.util.tool; import org.apache.hadoop.util.toolrunner; /*this class is responsible for running map reduce job*/ public class maxtemperaturedriver extends configured implements tool{ public int run(string[] args) throws exception { if(args.length !=2) { system.err.println("usage: maxtemperaturedriver "); system.exit(-1); } job job = new job(); job.setjarbyclass(maxtemperaturedriver.class); job.setjobname("max temperature"); fileinputformat.addinputpath(job, new path(args[0])); fileoutputformat.setoutputpath(job,new path(args[1])); job.setmapperclass(maxtemperaturemapper.class); job.setreducerclass(maxtemperaturereducer.class); job.setoutputkeyclass(text.class); job.setoutputvalueclass(intwritable.class); system.exit(job.waitforcompletion(true) ? 0:1); boolean success = job.waitforcompletion(true); return success ? 0 : 1; } public static void main(string[] args) throws exception { maxtemperaturedriver driver = new maxtemperaturedriver(); int exitcode = toolrunner.run(driver, args); system.exit(exitcode); } } 6. supply input and output we need to supply input file that will be used during map phase and the final output will be generated in output directory by reduct task. edit run configuration and supply command line arguments. sample.txt reside in the project root. your project explorer should contain following ] 7. map reduce job execution 8. final output if you managed to come this far, once the job is complete, it will create output directory with _success and part_nnnnn , double click to view it in eclipse editor and you will see we have supplied 5 rows of weather data (downloaded from ncdc weather) and we wanted to find out the maximum temperature in a given year from input file and the output will contain 2 rows with max temperature in (centigrade) for each supplied year 1949 111 (11.1 c) 1950 22 (2.2 c) make sure you delete the output directory next time running your application else you will get an error from hadoop saying directory already exists. happy hadooping!

i got tired of doing arbitrary performance testing, so i decided to take the freedb dataset and start working with that. freedb is a data set used to look up cd information based on the a nearly unique disk id. this is a good dataset, because it contains a lot of data (over three million albums, and over 40 million songs), and it is production data. that means that it is dirty . this makes it perfect to run all sort of interesting scenarios. the purpose of this post (and maybe the new few) is to show off a few things. first, we want to see how voron behaves with realistic data set. second, we want to show off the way voron works, its api, etc. to start with, i run my freedb parser, pointing it at /dev/null. the idea is to measure what is the cost of just going through the data is. we are using freedb-complete-20130901.tar.bz2 from sep 2013. after 1 minute, we went through 342,224 albums, and after 6 minutes we were at 2,066,871 albums. reading the whole 3,328,488 albums took about a bit over ten minutes. so just the cost of parsing and reading the freedb dataset is pretty expensive. the end result is a list of objects that looks like this: now, let us see how we want to actually use this. we want to be able to: lookup an album by the disk ids lookup all the albums by an artist*. lookup albums by album title*. this gets interesting, because we need to deal with questions such as: “given pearl jam, if i search for pearl, do i get them? do i get it for jam?” for now, we are going to go with case insensitive, but we won’t be doing full text search, we will allow, however, prefix searches. we are using the following abstraction for the destination: public abstract class destination { public abstract void accept(disk d); public abstract void done(); } basically, we read data as fast as we can, and we shove it to the destination, until we are done. here is the voron implementation: public class vorondestination : destination { private readonly storageenvironment _storageenvironment; private writebatch _currentbatch; private readonly jsonserializer _serializer = new jsonserializer(); private int counter = 1; public vorondestination() { _storageenvironment = new storageenvironment(storageenvironmentoptions.forpath("freedb")); using (var tx = _storageenvironment.newtransaction(transactionflags.readwrite)) { _storageenvironment.createtree(tx, "albums"); _storageenvironment.createtree(tx, "ix_artists"); _storageenvironment.createtree(tx, "ix_titles"); tx.commit(); } _currentbatch = new writebatch(); } public override void accept(disk d) { var ms = new memorystream(); _serializer.serialize(new jsontextwriter(new streamwriter(ms)), d); ms.position = 0; var key = new slice(endianbitconverter.big.getbytes(counter++)); _currentbatch.add(key, ms, "albums"); if(d.artist != null) _currentbatch.multiadd(d.artist.tolower(), key, "ix_artists"); if (d.title != null) _currentbatch.multiadd(d.title.tolower(), key, "ix_titles"); if (counter%1000 == 0) { _storageenvironment.writer.write(_currentbatch); _currentbatch = new writebatch(); } } public override void done() { _storageenvironment.writer.write(_currentbatch); } } let us go over this in detail, shall we? in line 10 we create a new storage environment. in this case, we want to just import the data, so we can create the storage inline. on lines 13 – 15, we create the relevant trees. you can think about voron trees in a very similar manner to the way you think about tables. they are a way to separate data into different parts of the storage. note that this still all reside in a single file, so there isn’t a physical separation. note that we created an albums tree, which will contain the actual data. and ix_artists, ix_titles trees. those are indexes into the albums tree. you can see them being used just a little lower. in the accept method, you can see that we use a writebatch, a native voron notion that allows us to batch multiple operations into a single transaction. in this case, for every album, we are making 3 writes. first, we write all of the data, as a json string, into a stream and put it in the albums tree. then we create a simple incrementing integer to be the actual album key. finally, we add the artist and title entries (lower case, so we don’t have to worry about case sensitivity in searches) into the relevant indexes. at 60 seconds, we written 267,998 values to voron. in fact, i explicitly designed it so we can see the relevant metrics. at 495 seconds we have reads 1,995,385 entries from the freedb file, we parsed 1,995,346 of them and written to voron 1,610,998. as you can imagined, each step is running in a dedicated thread, so we can see how they behave on an individual basis. the good thing about this is that i can physically see the various costs, it is actually pretty cool here is the voron directory at 60 seconds: you can see that we have two journal files active (haven’t been applied to the data file yet) and the db.voron file is at 512 mb. the compression buffer is at 32 mb (this is usually twice as big as the biggest transaction, uncompressed). the scratch buffer is used to hold in flight transaction information (until we send it to the data file), and you can see it is sitting on 256mb in size. at 15 minutes, we have the following numbers: 3,035,452 entries read from the file, 3,035,426 parsed and 2,331,998 written to voron. note that we are reading the file & writing to voron on the same disk, so that might impact the read performance. at that time, we can see the following on the disk: note that we increase the size of most of our files by factor of 2, so some of the space in the db.voron file is probably not used. note that we needed more scratch space to handle the in flight information. the entire process took 22 minutes, start to finish. although i have to note that this hasn’t been optimized at all, and i know we are doing a lot of stupid stuff through it. you might have noticed something else, we actually “crashed” closed the voron db, this was done to see what would happen when we open a relatively large db after an unordered shutdown. we’ll actually get to play with the data in my next post. so far this has been pretty much just to see how things are behaving. and… i just realized something, i forgot to actually add an index on disk id . which means that i have to import the data again. but before that, i also wrote the following: public class jsonfiledestination : destination { private readonly gzipstream _stream; private readonly streamwriter _writer; private readonly jsonserializer _serializer = new jsonserializer(); public jsonfiledestination() { _stream = new gzipstream(new filestream("freedb.json.gzip", filemode.createnew, fileaccess.readwrite), compressionlevel.optimal); _writer = new streamwriter(_stream); } public override void accept(disk d) { _serializer.serialize(new jsontextwriter(_writer), d); _writer.writeline(); } public override void done() { _writer.flush(); _stream.dispose(); } } this completed in ten minutes, for 3,328,488 entries. or a rate of about 5,538 per / second. the result is a 845mb gzip file. i had twofold reasons to want to do this. first, this gave me something to compare ourselves to, and more to the point, i can re-use this gzip file for my next tests, without having to go through the expensive parsing of the freedb file. i did just that and ended up with the following: public class voronentriesdestination : entrydestination { private readonly storageenvironment _storageenvironment; private writebatch _currentbatch; private int counter = 1; public voronentriesdestination() { _storageenvironment = new storageenvironment(storageenvironmentoptions.forpath("freedb")); using (var tx = _storageenvironment.newtransaction(transactionflags.readwrite)) { _storageenvironment.createtree(tx, "albums"); _storageenvironment.createtree(tx, "ix_diskids"); _storageenvironment.createtree(tx, "ix_artists"); _storageenvironment.createtree(tx, "ix_titles"); tx.commit(); } _currentbatch = new writebatch(); } public override int accept(string d) { var disk = jobject.parse(d); var ms = new memorystream(); var writer = new streamwriter(ms); writer.write(d); writer.flush(); ms.position = 0; var key = new slice(endianbitconverter.big.getbytes(counter++)); _currentbatch.add(key, ms, "albums"); int count = 1; foreach (var diskid in disk.value("diskids")) { count++; _currentbatch.multiadd(diskid.value(), key, "ix_diskids"); } var artist = disk.value("artist"); if (artist != null) { count++; _currentbatch.multiadd(artist.tolower(), key, "ix_artists"); } var title = disk.value("title"); if (title != null) { count++; _currentbatch.multiadd(title.tolower(), key, "ix_titles"); } if (counter % 100 == 0) { _storageenvironment.writer.write(_currentbatch); _currentbatch = new writebatch(); } return count; } public override void done() { _storageenvironment.writer.write(_currentbatch); _storageenvironment.dispose(); } } now we are actually properly disposing of things, and i also decreased the size of the batch, to see how it would respond. note that it is now being fed directly from the gzip file, at a greatly reduced cost. i also added tracking note only for how many albums we write, but also how many entries . by entries i mean, how many voron entries (which include the values we add to the index). i did find a bug where we would just double the file size without due consideration to its size, so now we are doing smaller file size increases. word of warning : i didn’t realized until after i was done with all the benchmarks, but i actually run all of those in debug configuration, which basically means that it is utterly useless as a performance metric. that is especially true because we have a lot of verifier code that runs in debug mode. so please don’t take those numbers as actual performance metrics, they aren’t valid. time # of albums # of entries 4 minutes 773,398 3,091,146 6 minutes 1,126,998 4,504,550 8 minutes 1,532,858 6,126,413 18 minutes 2,781,698 11,122,799 24 minutes 3,328,488 13,301,496 the status of the file system midway during the run. you can see that now we increase the file is smaller increments. and that we are using more scratch space, probably because we are under very heavy write load. after the run: scratch & compression are only used when the database is running, and deleted on close. the database is 7gb in side, which is quite respectable. now, to working with it, but i’ll save that for my next post, this one is long enough already.

If you ever need to persuade management why it might be better to deploy a larger change in multiple stages and push it to customers gradually, read on. A deployment of many changes is risky. We want therefore to deploy them in a way which minimizes the risk of harm to our customers and our companies. The deployment can be done either in an all-at-once (also known as big-bang) way or a gradual way. We will argue here for the more gradual (“stepwise”) approach. Big-bang or stepwise deployment? A big-bang deployment seems to be the natural thing to do: the full solution is developed and tested and then replaces the current system at once. However, it has two crucial flaws. First, it assumes that most defects can be discovered by testing. However, due to differences in test/prod environments, unknown dependencies, and the sheer scale of a typical larger system there always will be problems that are not discovered until production deployment or even until the application runs for a while in production (whichapplies even to airplanes). The more parts have been changed, the more of these production defects will happen at the same time. A gradual deployment makes it possible to discover and handle them one by one. Second, the more complex the deployment, the higher chance of human error(s), i.e. the deployment itself is a likely source of serious defects. Some of the drawbacks of a big-bang deployment in more detail: Complexity: A big-bang deployment requires coordination of many people and “moving parts” that depend on each other, providing a huge opportunity for human mistake (i.e. there will be mistakes). Lot of time: Such a deployment requires lot of time (typically also more than planed/expected) and thus lot of downtime when users cannot use the system. Hard troubleshooting: With a network of inter-dependent parts that changed all at the same time, while perhaps also changing the infrastructure (i.e. connections between them), it is extremely hard to pinpoint the source of defects, thus considerably increasing the time to detect and correct defects while also increasing the risk of people stepping on the toes of each other and “panic fixes” that either cause more problems than they remove or are not good enough (as the rollback that sped upKnight’s downfall). Rollback is likely either impossible or equally time-consuming and risky as the deployment itself, thus increasing the impact of defects and inviting even more human errors. Impact: Deploying everything to all users at the same time means that everybody will be impacted by a potential defect/error/mistake. Long freeze: All needs to be tested together after all development is finished, which requires a lot of time while the code is frozen and no more fixes and changes can get into production for weeks. Risk mitigation The goal of a good deployment plan is to mitigate the risk of the deployment and get it to an acceptable level. There are two aspects to risk: the probability of a defect and the impact of the defect. The following table shows how the possible measures affect them: Defect probability reduction Defect impact reduction testing stepwise deployment gradual migration of users to the new version (f.ex. 1 in 1000 or particular subsets) rollback mechanism => these also lead to much lower time to detect and fix defects Practices for stepwise deployment Enable stepwise deployment: Use parallel change and other Continuous Delivery techniques to make it possible to deploy updated components independently from each other and to switch on/off new features and to switch what versions of the components they depend on are currently used. (Parallel change – keeping the old and new code and being able to use one or the other – is crucial here. Also notice that parallel change applies also to data – you will need to evolve your data schema gradually and keep both old and new one at the same time in a period of time.) Enable rollback. The previous measure – stepwise deployment – makes it also easy(ier) to roll-back the changes by switching to a previous version of a dependency or by switching back to the old code. Migrate users gradually to the new version, i.e. expose the new version only to a small subset of the users initially and increase that subset until everybody uses it. This can be done f.ex. by deploying to only a subset of servers and sending a random/particular subset of users to the new servers but there are also ways if you have only a single machine. (See f.ex. my post Webapp Blue-Green Deployment Without Breaking Sessions/With Fallback With HAProxy.) Monitoring – make sure you are able to monitor flow of users through the system and detect any anomalies and errors early, long before angry calls from the business. Tools such as Logstash, Google Analytics (with custom events from JavaScript), client-side error logging via one of existing services or a custom solution are invaluable. About these ads

This technical tip shows how developers can customize the Appearance of Pivot Table Reports inside their Android applications using Aspose.Cells for Android. Previously we have shown how to create a simple pivot table. This article further goes and discusses how to customize the appearance of a pivot table by setting its properties like Setting the AutoFormat and PivotTableStyle Types, Setting Format Options, Setting Row Column and Page Fields Format, Modify a Pivot Table Quick Style and Clearing PivotFields etc. //Setting the AutoFormat and PivotTableStyle Type //Setting the PivotTable report is automatically formatted for Excel 2003 formats pivotTable.setAutoFormat(true); //Setting the PivotTable atuoformat type. pivotTable.setAutoFormatType(PivotTableAutoFormatType.CLASSIC); //Setting the PivotTable's Styles for Excel 2007/2010 formats e.g XLSX. pivotTable.setPivotTableStyleType(PivotTableStyleType.PIVOT_TABLE_STYLE_LIGHT_1); //Setting Format Options //The code sample that follows illustrates how to set a number of pivot table formatting options, including adding grand totals for rows and columns. //Dragging the third field to the data area. pivotTable.addFieldToArea(PivotFieldType.DATA,2); //Show grand totals for rows. pivotTable.setRowGrand(true); //Show grand totals for columns. pivotTable.setColumnGrand(true); //Display a custom string in cells that contain null values. pivotTable.setDisplayNullString(true); pivotTable.setNullString("null"); //Setting the layout pivotTable.setPageFieldOrder(PrintOrderType.DOWN_THEN_OVER); //Setting Row, Column, and Page Fields Format //The code example that follows shows how to access row fields, access a particular row, set subtotals, apply automatic sorting, and using the autoShow option. //Accessing the row fields. PivotFieldCollection pivotFields = pivotTable.getRowFields(); //Accessing the first row field in the row fields. PivotField pivotField = pivotFields.get(0); //Setting Subtotals. pivotField.setSubtotals(PivotFieldSubtotalType.SUM,true); pivotField.setSubtotals(PivotFieldSubtotalType.COUNT,true); //Setting autosort options. //Setting the field auto sort. pivotField.setAutoSort(true); //Setting the field auto sort ascend. pivotField.setAscendSort(true); //Setting the field auto sort using the field itself. pivotField.setAutoSortField(-1); //Setting autoShow options. //Setting the field auto show. pivotField.setAutoShow(true); //Setting the field auto show ascend. pivotField.setAscendShow(false); //Setting the auto show using field(data field). pivotField.setAutoShowField(0); //The following lines of code illustrate how to format data fields. //Accessing the data fields. PivotFieldCollection pivotFields = pivotTable.getDataFields(); //Accessing the first data field in the data fields. PivotField pivotField = pivotFields.get(0); //Setting data display format pivotField.setDataDisplayFormat(PivotFieldDataDisplayFormat.PERCENTAGE_OF); //Setting the base field. pivotField.setBaseField(1); //Setting the base item. pivotField.setBaseItem(PivotItemPosition.NEXT); //Setting number format pivotField.setNumber(10); //Modify a Pivot Table Quick Style //The code examples that follow show how to modify the quick style applied to a pivot table. File sdDir = Environment.getExternalStorageDirectory(); String sdPath = sdDir.getCanonicalPath(); //Open the template file containing the pivot table. Workbook wb = new Workbook(sdPath + "/Template.xlsx"); //Add pivot table style Style style1 = wb.createStyle(); com.aspose.cells.Font font1 = style1.getFont(); font1.setColor(Color.getRed()); Style style2 = wb.createStyle(); com.aspose.cells.Font font2 = style2.getFont(); font2.setColor( Color.getBlue()); int i = wb.getWorksheets().getTableStyles().addPivotTableStyle("tt"); //Get and Set the table style for different categories TableStyle ts = wb.getWorksheets().getTableStyles().get(i); int index = ts.getTableStyleElements().add(TableStyleElementType.FIRST_COLUMN); TableStyleElement e = ts.getTableStyleElements().get(index); e.setElementStyle(style1); index = ts.getTableStyleElements().add(TableStyleElementType.GRAND_TOTAL_ROW); e = ts.getTableStyleElements().get(index); e.setElementStyle(style2); //Set Pivot Table style name PivotTable pt = wb.getWorksheets().get(0).getPivotTables().get(0); pt.setPivotTableStyleName ("tt"); //Save the file. wb.save(sdPath + "/OutputFile.xlsx"); //Clearing PivotFields //PivotFieldCollection has a method named clear() for the task. When you want to clear all the PivotFields in the areas e.g., page, column, row or data, you can use it. The code sample below shows how to clear all the PivotFields in data area. File sdDir = Environment.getExternalStorageDirectory(); String sdPath = sdDir.getCanonicalPath(); //Open the template file containing the pivot table. Workbook workbook = new Workbook(sdPath + "/PivotTable.xlsx"); //Get the first worksheet Worksheet sheet = workbook.getWorksheets().get(0); //Get the pivot tables in the sheet PivotTableCollection pivotTables = sheet.getPivotTables(); //Get the first PivotTable PivotTable pivotTable = pivotTables.get(0); //Clear all the data fields pivotTable.getDataFields().clear(); //Add new data field pivotTable.addFieldToArea(PivotFieldType.DATA, "Betrag Netto FW"); //Set the refresh data flag on pivotTable.setRefreshDataFlag(false); //Refresh and calculate the pivot table data pivotTable.refreshData(); pivotTable.calculateData(); //Save the Excel file workbook.save(sdPath + "/out1.xlsx");

This article presents the usage of sripted data set in the eclipse's BIRT.

what can one create during the new year and christmas holidays? as it turned down – quite enough. even if you have two kids and a bunch of family members whom you want to visit. the only thing you cannot accomplish in time is to finish an article for dzone. it takes a lot of time, nearly the entire january. by the 5th of january i had a laptop and a couple of days to spend on some development. having estimated what i can do here, i decided to create a mobile app that would work faster than the original. for this, i needed to find communicative creators of a popular app. hence, i found a “ spender ” app in the app store. it is a simple app for tracking your budget. with it, you can estimate how effectively you spend your money in the end of each month. by the 5th of january, this app was in top-10 in the russian app store. i also found their dev-story on iphones.ru. in their dev-story, the developers wrote that after completing their previous project, they had three-four free days. so, they decided to create a new app during this free time. their product manager and programmers helped them with positioning the app and its key features. this encouraged me and i began to think how to create nearly the same app in 2 days . note: the original app was updated in the middle of january, and now it looks a little different from my app. anyway, you can find its screenshots in the dev-story. i already had the experience of mobile app development using c# and cocoa. since this was my personal free time, i wanted to use it with maximum effectiveness. even if i didn’t succeed, i was eager to learn a new framework or programming language. i was working for devexpress from 2006 till2011 and have been reading their announces since i left the company. so, i knew that they created a mobile js-framework based on cordova/phonegap. they made it after i left the company, so i was curious to try it. the gartner research company reports that by august, 2013 most of the enterprise mobile software was created using phonegap or phonegap-based products (like kony ). from my consumer experience, it's far from true. maybe i was wrong? i'm not so good at html and javascript. i can create mark-up with stackoverflow.com and i can write simple selectors with jquery. i can also find the required information in their documentation. in other words, html+js was a gap in my knowledge and i was ready to fill it or gain some experience. thus, i planned to create a cross-platform application that could become an advantage over the original ios-only spender app. moreover, i wanted to spend my time in the most effective way. on the one hand, i had a potentially effective js framework, on the other – a lack of js experience. i hoped that the js framework advantages could balance my poor experience. since i like to use a vcs during development, i'll try to recover my progress. you can download complete apps here: ios , android i'm not sure i can provide public access to my repo, because it contains images i bought from fotolia and third-party libraries, each with a difference license. i'm not a lawyer, so i’d prefer not to take the risk. the most curious of you can take a look into the app bundle itself. js wasn't minified. place: tula, russia, date: january, 5, 2014 +20 minutes spent on installing node.js and cordova cli +10 minutes downloaded a template app from cordova. added a template from phonejs. created a git-repo, registered it in webstorm. added a new record to the httpd.conf in order to have an ability to debug my future app in the browser. +38 minutes changed the app namespace to "io.nikitin.thriftbox". added navigation. phonejs is an mvc-framework. each app screen is represented as a collection of html markup (views) and fabric function (viewmodel). here is how it looks at its simplest // view content and thriftbox.home = function (params) { // request parameters taken from uri return {}; // viewmodel instance }; then view and view model are bound via knockout-bindings . to be in time, i create only two screens: expense input and monthly expense report. +4 hours 20 minutes here i got stuck for the first time. i couldn't create a markup of digit buttons. the original app had a huge keyboard that looked like a calculator or dialer. i found out that it was not that easy to create such a keyboard, even using a table tag. in the iphone retina screen, 1px borders between buttons changed their colors after clicking on the buttons. on my iphone, the difference in colors was very noticeable. i had to invent how to tackle this. i tried to implement buttons using div s. but i couldn't achieve a border width of 1 px and make all buttons look equal in different screens. three hours later i gave up the idea of using divs and moved forward. +28 minutes removing a clicked button indicator on ios. ios displays a gray indicator around tapped links and objects with the onclick event handler. since i had my own indicator of a tapped object (the tapped button became darker), i didn't need the default indicator. i solved this problem using the dxaction event: was: 1 became: 1 this event is an extended variation of a "click" event: its handler supports uri navigation between views and correctly works in the scrollable area. +14 minutes the buttonpress event handler shown in the previous example now validates numbers from user input. var number = ko.observable(null); var isvalidnumber = ko.computed(function() { return number() && parsefloat(number()) > 0; }); ...... function buttonpress(button) { if (button) { if (number()) number(number() + button); else number(button); } else if (number()) number(number().substr(0, number().length - 1)); } var viewmodel = { number: number, isvalidnumber: isvalidnumber, viewshowing: viewshowing, buttonpress: buttonpress }; ..... +8 minutes added a fastclick.js , which removes a delay between tapping the screen and raising the 'click' event on phones. the mobile browser delays the raising of the click event by default to be sure the end-user will not perform a double tap. for the end-user, this looks as if the app is sluggish. you click buttons much faster than an app responds. fastclick.js handles the touchstart event and then creates all the click event process logic. btw, adding this library was a mistake; later i'll tell why. +4 minutes added a limitation to the length of user input numbers. corrected the font size for a better look-and-feel. +58 minutes added a choice of an expense category. added a scrollable pane with available categories below the input field. video . it took less time than it could be. in the phonejs component collection, i found dxtileview . it provides a kinetic scrolling with the required appearance out-of-the-box. it's not easy to implement kinetic scrolling by yourself and thus it’s great that this scrolling is enabled for ios only - android doesn't have it. it was 7:40 pm, so, i decided to continue the next day. place: tula, russia, date: january, 5, 2014 +3 hours 9 minutes storing data on a local storage. phonejs contains classes for working with data: selection, filtering, sorting, and grouping. there are several approaches to store data: odata and localstorage. i didn't want to implement a server side for a free app, and decided to use localstorage. later i found out that this was not an ideal decision. for example, when updating to ios 5.1 user data is erased , other people complained that localstorage is cleared regularly or even when shutting the device down. i didn't want to risk, so i used file api of phonegap. documentation says that this api is based on w3c file api. in fact, this means that this api differs in safari for mac os, chrome for mac os, cordova for ios and cordova for android. file api implementation is different for ios and android . e.g. android implementation doesn't contain the 'blob' class and 'window.permanent' constant. ii however implements the 'localfilesystem' and 'localfilesystem.persistent' classes. the laptop browser provides additional api for requesting an additional storage space, which mobile browsers don't provide. the available documentation for this api adds more problems. i found several articles searching by "html5 file api". and, i couldn't find an article that would cover all my questions. finally i created a new class for working with fileapi. this class supports cordova 3.3 on ios, android, and chrome 32 for mac os and windows 8. you can find it here: https://github.com/chebum/filestorage-for-phone.js/blob/master/filestorage.js you can use it as follows: // in this example i create data/records file in the documents folder of the app fs.initfileapi(1000000, true) .then(function () { var records = new fs.filearraystore({ key: "id", filename: "records" }); return records.insert({ customer: "peter" }) }) .then(function () { alert("record saved!"); }); // or use low-level api: fs.initfileapi(100000, true) .then(function() { return fs.writefile("file1", "file content") }) .then(function() { alert("file saved!"); }); +33 minutes saving the added records to the storage. category list is stored in arraystore , to simplify the selection operations. +26 minutes creating layout for the app's views. phonejs provides several layouts that are the placeholders for the views. my app's start page didn't fit into any of the available layout, so i have chosen the emptylayout. but, it doesn't provide animation effects when navigating through views. i copied the emptylayout code and added an attribute that had animation effects. +1 h. 51 min. template's about screen was redesigned to a report screen, empty by that moment. created a viewmodel that selects data for a current month. added localization date formatting for the screen caption. +59 minutes added the display of expenses grouped by categories for a current month. +28 minutes added the selection of months for which the report should be generated. end-users can tap the screen header to select the required month. +1 h. 20 min. added cordova-plugin statusbar that didn't work outof-the-box. i found that the reference to cordova.js was commented in the phonejs app template: as a result, the native part of my app didn't work. +39 minutes in the report screen, the upper part was changed to dxtoolbar . +22 minutes i discoveredwhy the dxbutton click event handler didn't work. removing the fastclick.js solved my problem, but caused a delay between tapping and event raising. i've changed the dxaction event subscription to 'touchstart'. +25 minutes formatting output strings when generating a report. at night i dreamed of crappy buttons in the application’s main screen. places: tula, vnukovo airport, date: january, 7-8, 2014 i had an early flight to budapest from vnukovo, and because i had no time in the afternoon, i gradually completed at the airport at night. as you know, it’s not very comfortable to sleep or sit in a café chair for a long time, but it turned out that programming was ok. +2 h. 5 min. in the morning, i decided to split the buttons in order to remove borders between them. i took the ios dialer keyboard as a sample. i created three keyboards. the button size changes depending on screen resolution: for 3.5'', 4'' and 5'' phones. each table cell contained a div with configured alignment. because of the lack of an incomplete vertical text alignment in html, the final css style for buttons ended to be quite complex: .home-view .buttons td div { color: #4a5360; border: 1px solid #4a5360; text-align: center; position: absolute; left: 50%; /* small buttons - default */ font-size: 26px; padding: 13px 0 13px 0; width: 52px; line-height: 26px; border-radius: 26px; margin-left: -27px; margin-top: -27px; } +1 h. 50 minutes i bought several vector icon sets on fotolia. i cut the required icons and converted them to png. it took me quite a long time, maybe, because it was 1.30 am :) +1 hour 10 minutes added a splash-screen for the app. +36 minutes created three sizes for the app icon. localized the app name for ios. +20 minutes hiding the splash screen after the app is completely loaded. +2 hours fixing multiple bugs. +2 hours creating screenshots for play store +30 minutes creating screenshots for app store +30 minutes writing an app description for two app stores. +1 h. 30 minutes submitting my app to the app store. here i faced with an issue with the app certification. my accountancy let's summarize the time i spent and divide it into categories. development: 21 hours 37 minutes graphics and texts: 8 hours 26 minutes totally: 30 hours 3 minutes as a result, i got a minimum-feature working app, though it is not as cool as the latest version of "spender". i couldn't create splitting expenses by days and income input. my app's ui could be more elegant as well. after analyzing the original 'spender' developer work, i got the following. they say that they involved four developers for three-four days. it is about 96-128 man-hours. i spent only 30 man-hours and got an app for three mobile platforms. ios and android versions are already in stores. the version for windows phone 8 requires a ui redesign. i can be proud of myself :). you can download complete apps here: ios , android

Every now and then we’ll come across a search problem that can’t simply be solved with plain Solr relevancy. This usually means a customer knows exactly how documents should be scored. They may have little tolerance for close approximations of this scoring through Solr boosts, function queries, etc. They want a Lucene-based technology for text analysis and performant data structures, but they need to be extremely specific in how documents should be scored relative to each other. Well for those extremely specialized cases we can prescribe a little out-patient surgery to your Solr install – building your own Lucene Query. This is the Nuclear Option Before we dive in, a word of caution. Unless you just want the educational experience, building a custom Lucene Query should be the “nuclear option” for search relevancy. It’s very fiddly and there are many ins-and-outs. If you’re actually considering this to solve a real problem, you’ve already gone down the following paths: You’ve utilized Solr’s extensive set of query parsers & features including function queries, joins, etc. None of this solved your problem You’ve exhausted the ecosystem of plugins that extend on the capabilities in (1). That didn’t work. You’ve implemented your own query parser plugin that takes user input and generates existing Lucene queries to do this work. This still didn’t solve your problem. You’ve thought carefully about your analyzers – massaging your data so that at index time and query time, text lines up exactly as it should to optimize the behavior of existing search scoring. This still didn’t get what you wanted. You’ve implemented your own custom Similarity that modifies how Lucene calculates the traditional relevancy statistics – query norms, term frequency, etc. You’ve tried to use Lucene’s CustomScoreQuery to wrap an existing Query and alter each documents score via a callback. This still wasn’t low-level enough for you, you needed even more control. If you’re still reading you either think this is going to be fun/educational (good for you!) or you’re one of the minority that must control exactly what happens with search. If you don’t know, you can of course contact us for professional services. Ok back to the action… Refresher – Lucene Searching 101 Recall that to search in Lucene, we need to get a hold of an IndexSearcher. This IndexSearcher performs search over an IndexReader. Assuming we’ve created an index, with these classes we can perform searches like in this code: Directory dir = new RAMDirectory(); IndexReader idxReader = new IndexReader(dir); idxSearcher idxSearcher = new IndexSearcher(idxReader) Query q = new TermQuery(new Term(“field”, “value”)); idxSearcher.search(q); Let’s summarize the objects we’ve created: Directory – Lucene’s interface to a file system. This is pretty straight-forward. We won’t be diving in here. IndexReader – Access to data structures in Lucene’s inverted index. If we want to look up a term, and visit every document it exists in, this is where we’d start. If we wanted to play with term vectors, offsets, or anything else stored in the index, we’d look here for that stuff as well. IndexSearcher — wraps an IndexReader for the purpose of taking search queries and executing them. Query – How we expect the searcher to perform the search, encompassing both scoring and which documents are returned. In this case, we’re searching for “value” in field “field”. This is the bit we want to toy with In addition to these classes, we’ll mention a support class exists behind the scenes: Similarity – Defines rules/formulas for calculating norms at index time and query normalization. Now with this outline, let’s think about a custom Lucene Query we can implement to help us learn. How about a query that searches for terms backwards. If the document matches a term backwards (like ananab for banana), we’ll return a score of 5.0. If the document matches the forwards version, let’s still return the document, with a score of 1.0 instead. We’ll call this Query “BackwardsTermQuery”. This example is hosted here on github. A tale of 3 classes – A Query, A Weight, and a Scorer Before we sling code, let’s talk about general architecture. A Lucene Query follows this general structure: A custom Query class, inheriting from Query A custom Weight class, inheriting from Weight A custom Scorer class inheriting from Scorer These three objects wrap each other. A Query creates a Weight, and a Weight in turn creates a Scorer. A Query is itself a very straight-forward class. One of its main responsibilities when passed to the IndexSearcher is to create a Weight instance. Other than that, there are additional responsibilities to Lucene and users of your Query to consider, that we’ll discuss in the “Query” section below. A Query creates a Weight. Why? Lucene needs a way to track IndexSearcher level statistics specific to each query while retaining the ability to reuse the query across multiple IndexSearchers. This is the role of the Weight class. When performing a search, IndexSearcher asks the Query to create a Weight instance. This instance becomes the container for holding high-level statistics for the Query scoped to this IndexSearcher (we’ll go over these steps more in the “Weight” section below). The IndexSearcher safely owns the Weight, and can abuse and dispose of it as needed. If later the Query gets reused by another IndexSearcher, a new Weight simply gets created. Once an IndexSearcher has a Weight, and has calculated any IndexSearcher level statistics, the IndexSearcher’s next task is to find matching documents and score them. To do this, the Weight in turn creates a Scorer. Just as the Weight is tied closely to an IndexSearcher, a Scorer is tied to an individual IndexReader. Now this may seem a little odd – in our code above the IndexSearcher always has exactly one IndexReader right? Not quite. See, a little hidden implementation detail is that IndexReaders may actually wrap other smaller IndexReaders – each tied to a different segment of the index. Therefore, an IndexSearcher needs to have the ability score documents across multiple, independent IndexReaders. How your scorer should iterate over matches and score documents is outlined in the “Scorer” section below. So to summarize, we can expand the last line from our example above… idxSearcher.search(q); … into this psuedocode: Weight w = q.createWeight(idxSearcher); // IndexSearcher level calculations for weight Foreach IndexReader idxReader: Scorer s = w.scorer(idxReader); // collect matches and score them Now that we have the basic flow down, let’s pick apart the three classes in a little more detail for our custom implementation. Our Custom Query What should our custom Query implementation look like? Query implementations always have two audiences: (1) Lucene and (2) users of your Query implementation. For your users, expose whatever methods you require to modify how a searcher matches and scores with your query. Want to only return as a match 1/3 of the documents that match the query? Want to punish the score because the document length is longer than the query length? Add the appropriate modifier on the query that impacts the scorer’s behavior. For our BackwardsTermQuery, we don’t expose accessors to modify the behavior of the search. The user simply uses the constructor to specify the term and field to search. In our constructor, we will simply be reusing Lucene’s existing TermQuery for searching individual terms in a document. private TermQuery backwardsQuery; private TermQuery forwardsQuery; public BackwardsTermQuery(String field, String term) { // A wrapped TermQuery for the reverse string Term backwardsTerm = new Term(field, new StringBuilder(term).reverse().toString()); backwardsQuery = new TermQuery(backwardsTerm); // A wrapped TermQuery for the Forward Term forwardsTerm = new Term(field, term); forwardsQuery = new TermQuery(forwardsTerm); } Just as importantly, be sure your Query meets the expectation of Lucene. Most importantly, you MUST override the following. createWeight() hashCode() equals() The method createWeight() we’ve discussed. This is where you’ll create a weight instance for an IndexSearcher. Pass any parameters that will influence the scoring algorithm, as the Weight will in turn be creating a searcher. Even though they are not abstract methods, overriding the hashCode()/equals() methods is very important. These methods are used by Lucene/Solr to cache queries/results. If two queries are equal, there’s no reason to rerun the query. Running another instance of your query could result in seeing the results of your first query multiple times. You’ll see your search for “peas” work great, then you’ll search for “bananas” and see “peas” search results. Override equals() and hashCode() so that “peas” != bananas. Our BackwardsTermQuery implements createWeight() by creating a custom BackwardsWeight that we’ll cover below: @Override public Weight createWeight(IndexSearcher searcher) throws IOException { return new BackwardsWeight(searcher); } BackwardsTermQuery has a fairly boilerplate equals() and hashCode() that passes through to the wrapped TermQuerys. Be sure equals() includes all the boilerplate stuff such as the check for self-comparison, the use of the super equals operator, the class comparison, etc etc. By using Lucene’s unit test suite, we can get a lot of good checks that our implementation of these is correct. @Override public boolean equals(Object other) { if (this == other) { return true; } if (!super.equals(other)) { return false; } if (getClass() != other .getClass()) { return false; } BackwardsTermQuery otherQ = (BackwardsTermQuery)(other); if (otherQ.getBoost() != getBoost()) { return false; } return otherQ.backwardsQuery.equals(backwardsQuery) && otherQ.forwardsQuery.equals(forwardsQuery); } @Override public int hashCode() { return super.hashCode() + backwardsQuery.hashCode() + forwardsQuery.hashCode(); } Our Custom Weight You may choose to use Weight simply as a mechanism to create Scorers (where the real meat of search scoring lives). However, your Custom Weight class must at least provide boilerplate implementations of the query normalization methods even if you largely ignore what is passed in: getValueForNormalization normalize These methods participate in a little ritual that IndexSearcher puts your Weight through with the Similarity for query normalization. To summarize the query normalization code in the IndexSearcher: float v = weight.getValueForNormalization(); float norm = getSimilarity().queryNorm(v); weight.normalize(norm, 1.0f); Great, what does this code do? Well a value is extracted from Weight. This value is then passed to a Similarity instance that “normalizes” that value. Weight then receives this normalized value back. In short, this is allowing IndexSearcher to give weight some information about how its “value for normalization” compares to the rest of the stuff being searched by this searcher. This is extremely high level, “value for normalization” could mean anything, but here it generally means “what I think is my weight” and what Weight receives back is what the searcher says “no really here is your weight”. The details of what that means depend on the Similarity and Weight implementation. It’s expected that the Weight’s generated Scorer will use this normalized weight in scoring. You can chose to do whatever you want in your own Scorer including completely ignoring what’s passed to normalize(). While our Weight isn’t factoring into the scoring calculation, for consistency sake, we’ll participate in the little ritual by overriding these methods: @Override public float getValueForNormalization() throws IOException { return backwardsWeight.getValueForNormalization() + forwardsWeight.getValueForNormalization(); } @Override public void normalize(float norm, float topLevelBoost) { backwardsWeight.normalize(norm, topLevelBoost); forwardsWeight.normalize(norm, topLevelBoost); } Outside of these query normalization details, and implementing “scorer”, little else happens in the Weight. However, you may perform whatever else that requires an IndexSearcher in the Weight constructor. In our implementation, we don’t perform any additional steps with IndexSearcher. The final and most important requirement of Weight is to create a Scorer. For BackwardsWeight we construct our custom BackwardsScorer, passing scorers created from each of the wrapped queries to work with. @Override public Scorer scorer(AtomicReaderContext context, boolean scoreDocsInOrder, boolean topScorer, Bits acceptDocs) throws IOException { Scorer backwardsScorer = backwardsWeight.scorer(context, scoreDocsInOrder, topScorer, acceptDocs); Scorer forwardsScorer = forwardsWeight.scorer(context, scoreDocsInOrder, topScorer, acceptDocs); return new BackwardsScorer(this, context, backwardsScorer, forwardsScorer); } Our Custom Scorer The Scorer is the real meat of the search work. Responsible for identifying matches and providing scores for those matches, this is where the lion share of our customization will occur. It’s important to note that a Scorer is also a Lucene DocIdSetIterator. A DocIdSetIterator is a cursor into a set of documents in the index. It provides three important methods: docID() – what is the id of the current document? (this is an internal Lucene ID, not the Solr “id” field you might have in your index) nextDoc() – advance to the next document advance(target) – advance (seek) to the target One uses a DocIdSetIterator by first calling nextDoc() or advance() and then reading the docID to get the iterator’s current location. The value of the docIDs only increase as they are iterated over. By implementing this interface a Scorer acts as an iterator over matches in the index. A Scorer for the query “field1:cat” can be iterated over in this manner to return all the documents that match the cat query. In fact, if you recall from my article, this is exactly how the terms are stored in the search index. You can chose to either figure out how to correctly iterate through the documents in a search index, or you can use the other Lucene queries as building blocks. The latter is often the simplest. For example, if you wish to iterate over the set of documents containing two terms, simply use the scorer corresponding to a BooleanQuery for iteration purposes. The first method of our scorer to look at is docID(). It works by reporting the lowest docID() of our underlying scorers. This scorer can be thought of as being “before” the other in the index, and as we want to report numerically increasing docIDs, we always want to chose this value: @Override public int docID() { int backwordsDocId = backwardsScorer.docID(); int forwardsDocId = forwardsScorer.docID(); if (backwordsDocId <= forwardsDocId && backwordsDocId != NO_MORE_DOCS) { currScore = BACKWARDS_SCORE; return backwordsDocId; } else if (forwardsDocId != NO_MORE_DOCS) { currScore = FORWARDS_SCORE; return forwardsDocId; } return NO_MORE_DOCS; } Similarly, we always want to advance the scorer with the lowest docID, moving it ahead. Then, we report our current position by returning docID() which as we’ve just seen will report the docID of the scorer that advanced the least in the nextDoc() operation. @Override public int nextDoc() throws IOException { int currDocId = docID(); // increment one or both if (currDocId == backwardsScorer.docID()) { backwardsScorer.nextDoc(); } if (currDocId == forwardsScorer.docID()) { forwardsScorer.nextDoc(); } return docID(); } In our advance() implementation, we allow each Scorer to advance. An advance() implementation promises to either land docID() exactly on or past target. Our call to docID() after we call advance will return either that one or both are on target, or it will return the lowest docID past target. @Override public int advance(int target) throws IOException { backwardsScorer.advance(target); forwardsScorer.advance(target); return docID(); } What a Scorer adds on top of DocIdSetIterator is the “score” method. When score() is called, a score for the current document (the doc at docID) is expected to be returned. Using the full capabilities of the IndexReader, any number of information stored in the index can be consulted to arrive at a score either in score() or while iterating documents in nextDoc()/advance(). Given the docId, you’ll be able to access the term vector for that document (if available) to perform more sophisticated calculations. In our query, we’ll simply keep track as to whether the current docID is from the wrapped backwards term scorer, indicating a match on the backwards term, or the forwards scorer, indicating a match on the normal, unreversed term. Recall docID() is always called on advance/nextDoc. You’ll notice we update currScore in docID, updating it every time the document advances. @Override public float score() throws IOException { return currScore; } A Note on Unit Testing Now that we have an implementation of a search query, we’ll want to test it! I highly recommend using Lucene’s test framework. Lucene will randomly inject different implementations of various support classes, index implementations, to throw your code off balance. Additionally, Lucene creates test implementations of classes such as IndexReader that work to check whether your Query correctly fulfills its contract. In my work, I’ve had numerous cases where tests would fail intermittently, pointing to places where my use of Lucene’s data structures subtly violated the expected contract. An example unit test is included in the github project associated with this blog post. Wrapping Up That’s a lot of stuff! And I didn’t even cover everything there is to know! As an exercise to the reader, you can explore the Scorer methods cost() and freq(), as well as the rewrite() method of Query used optionally for optimization. Additionally, I haven’t explored how most of the traditional search queries end up using a framework of Scorers/Weights that don’t actually inherit from Scorer or Weight known as “SimScorer” and “SimWeight”. These support classes consult a Similarity instance to customize calculation certain search statistics such as tf, convert a payload to a boost, etc. In short there’s a lot here! So tread carefully, there’s plenty of fiddly bits out there! But have fun! Creating a custom Lucene query is a great way to really understand how search works, and the last resort short in solving relevancy problems short of creating your own search engine. And if you have relevancy issues, contact us! If you don’t know whether you do, our search relevancy product, Quepid – might be able to tell you!

dan liebster has been kind enough to send me a real world time series database. the data has been sanitized to remove identifying issues, but this is actually real world data, so we can learn a lot more about this. this is what this looks like: the first thing that i did was take the code in this post , and try it out for size. i wrote the following: int i = 0; using (var parser = new textfieldparser(@"c:\users\ayende\downloads\timeseries.csv")) { parser.hasfieldsenclosedinquotes = true; parser.delimiters = new[] {","}; parser.readline();//ignore headers var startnew = stopwatch.startnew(); while (parser.endofdata == false) { var fields = parser.readfields(); debug.assert(fields != null); dts.add(fields[1], datetime.parseexact(fields[2], "o", cultureinfo.invariantculture), double.parse(fields[3])); i++; if (i == 25*1000) { break; } if (i%1000 == 0) console.write("\r{0,15:#,#} ", i); } console.writeline(); console.writeline(startnew.elapsed); } note that we are using a separate transaction per line , which means that we are really doing a lot of extra work. but this simulate very well incoming events coming one at a time. we were able to process 25,000 events in 8.3 seconds. at a rate of just over 3 events per millisecond . now, note that we have in here the notion of “channels”. from my investigation, it seems clear that some form of separation is actually very common in time series data. we are usually talking about sensors or some such, and we want to track data across different sensors over time. and there is little if any call for working over multiple sensors / channels at the same time. because of that, i made a relatively minor change in voron, that allows it to have an infinite number of separate trees. that means that i can use as many trees as you want, and we can model a channel as a tree in voron. i also changed things so we instead of doing a single transaction per line, we will do a transaction per 1000 lines. that dropped the time to insert 25,000 lines to 0.8 seconds. or a full order of magnitude faster. that done, i inserted the full data set, which is just over 1,096,384 records. that took 36 seconds. in the data set i have, there are 35 channels. i just tried, and reading all the entries in a channel with 35,411 events takes 0.01 seconds. that allows doing things like doing averages over time, comparing data, etc. you can see the code implementing this in the following link .

In our previous post on MongoDB storage structure and dbStats metrics, we covered how MongoDB stores data and the differences between the dataSize, storageSize and fileSize metrics. We can now apply this knowledge to evaluate strategies for re-using MongoDB disk space. When documents or collections are deleted, empty record blocks within data files arise. MongoDB attempts to reuse this space when possible, but it will never return this space to the file system. This behavior explains why fileSize never decreases despite deletes on a database. If your app frequently deletes or if your fileSize is significantly larger than the size of your data plus indexes, you can use one of the methods below reclaim free space. Getting your free space back Compacting individual collections You can compact individual collections using the compact command. This command rewrites and defragments all data in a collection, as well as all of the indexes on that collection. Important notes on compacting: This operation blocks all other database activity when running and should be used only when downtime for your database is acceptable. If you are running a replica set, you can perform compaction on secondaries in order to avoid blocking the primary and use failover to make the primary a secondary before compacting it. Compacting individual collections will not reduce your storage footprint on disk (i.e., your fileSize) but it will defragment the collections you compact. Compacting one or more databases For a single-node MongoDB deployment, you can use the db.repairDatabase() command to compact all the collections in the database. This operation rewrites all the data and indexes for each collection in the database from scratch and thereby compacts and defragments the entire database. To compact all the databases on your server process, you can stop your mongod process and run it with the “–repair” option. Important notes on running a repair: This operation blocks all other database activity when running and should be used only when downtime for your database is acceptable. Running a repair requires free disk space equal to the size of your current data set plus 2 GB. You can use space in a different volume than the one that your mongod is running in by specifying the “–repairpath” option. Compacting all databases on a server by re-syncing replica set nodes For a multi-node MongoDB deployment, you can resync a secondary from scratch to reclaim space. By resyncing each node in your replica set you effectively rewrite the data files from scratch and thereby defragment your database. Please note that if your cluster is comprised of only two electable nodes, you will sacrifice high availability during the resync because the secondary is completely wiped before syncing. If your app is sensitive to downtime, we recommend a process similar to the one we use here at MongoLab which we call a “rolling node replacement.” This process replaces each node in your cluster in turn by bringing a new node into the cluster, replicating the data to that new node and removing the old node. In this way, your cluster can maintain the same level of redundancy during the compaction as during normal operations. A tip about efficiently using space usePowerOf2Sizes Setting the usePowerof2Sizes option is a proactive approach to reusing space in collections that experience frequent document moves or deletions. This option supersedes the default padding factor mechanism and reduces the impact of fragmentation within the collection by allocating additional space for each document in intervals that follow the powers of 2. Setting this option for a specific collection makes it less likely that documents in that collection need to be moved when they grow in size, less likely that a document will need to be moved more than once in its lifetime, and more likely that space left by moving documents can be reused by new or other moved documents. Thanks for reading! We hope the above strategies help guide you in evaluating options for reusing empty space in your MongoDB.

At the end of last year I was playing around with running scheduled tasks to monitor a Neo4j cluster and one of the problems I ran into was that the monitoring would sometimes exit. I eventually realised that this was because a RuntimeException was being thrown inside the Runnable method and I wasn’t handling it. The following code demonstrates the problem: import java.util.ArrayList; import java.util.List; import java.util.concurrent.*; public class RunnableBlog { public static void main(String[] args) throws ExecutionException, InterruptedException { ScheduledExecutorService executor = Executors.newSingleThreadScheduledExecutor(); executor.scheduleAtFixedRate(new Runnable() { @Override public void run() { System.out.println(Thread.currentThread().getName() + " -> " + System.currentTimeMillis()); throw new RuntimeException("game over"); } }, 0, 1000, TimeUnit.MILLISECONDS).get(); System.out.println("exit"); executor.shutdown(); } } If we run that code we’ll see the RuntimeException but the executor won’t exit because the thread died without informing it: Exception in thread "main" pool-1-thread-1 -> 1391212558074 java.util.concurrent.ExecutionException: java.lang.RuntimeException: game over at java.util.concurrent.FutureTask$Sync.innerGet(FutureTask.java:252) at java.util.concurrent.FutureTask.get(FutureTask.java:111) at RunnableBlog.main(RunnableBlog.java:11) at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method) at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:57) at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43) at java.lang.reflect.Method.invoke(Method.java:601) at com.intellij.rt.execution.application.AppMain.main(AppMain.java:120) Caused by: java.lang.RuntimeException: game over at RunnableBlog$1.run(RunnableBlog.java:16) at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:471) at java.util.concurrent.FutureTask$Sync.innerRunAndReset(FutureTask.java:351) at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:178) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:178) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:293) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1110) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:603) at java.lang.Thread.run(Thread.java:722) At the time I ended up adding a try catch block and printing the exception like so: public class RunnableBlog { public static void main(String[] args) throws ExecutionException, InterruptedException { ScheduledExecutorService executor = Executors.newSingleThreadScheduledExecutor(); executor.scheduleAtFixedRate(new Runnable() { @Override public void run() { try { System.out.println(Thread.currentThread().getName() + " -> " + System.currentTimeMillis()); throw new RuntimeException("game over"); } catch (RuntimeException e) { e.printStackTrace(); } } }, 0, 1000, TimeUnit.MILLISECONDS).get(); System.out.println("exit"); executor.shutdown(); } } This allows the exception to be recognised and as far as I can tell means that the thread executing the Runnable doesn’t die. java.lang.RuntimeException: game over pool-1-thread-1 -> 1391212651955 at RunnableBlog$1.run(RunnableBlog.java:16) at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:471) at java.util.concurrent.FutureTask$Sync.innerRunAndReset(FutureTask.java:351) at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:178) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:178) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:293) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1110) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:603) at java.lang.Thread.run(Thread.java:722) pool-1-thread-1 -> 1391212652956 java.lang.RuntimeException: game over at RunnableBlog$1.run(RunnableBlog.java:16) at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:471) at java.util.concurrent.FutureTask$Sync.innerRunAndReset(FutureTask.java:351) at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:178) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:178) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:293) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1110) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:603) at java.lang.Thread.run(Thread.java:722) pool-1-thread-1 -> 1391212653955 java.lang.RuntimeException: game over at RunnableBlog$1.run(RunnableBlog.java:16) at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:471) at java.util.concurrent.FutureTask$Sync.innerRunAndReset(FutureTask.java:351) at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:178) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:178) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:293) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1110) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:603) at java.lang.Thread.run(Thread.java:722) This worked well and allowed me to keep monitoring the cluster. However, I recently started reading ‘Java Concurrency in Practice‘ (only 6 years after I bought it!) and realised that this might not be the proper way of handling the RuntimeException. public class RunnableBlog { public static void main(String[] args) throws ExecutionException, InterruptedException { ScheduledExecutorService executor = Executors.newSingleThreadScheduledExecutor(); executor.scheduleAtFixedRate(new Runnable() { @Override public void run() { try { System.out.println(Thread.currentThread().getName() + " -> " + System.currentTimeMillis()); throw new RuntimeException("game over"); } catch (RuntimeException e) { Thread t = Thread.currentThread(); t.getUncaughtExceptionHandler().uncaughtException(t, e); } } }, 0, 1000, TimeUnit.MILLISECONDS).get(); System.out.println("exit"); executor.shutdown(); } } I don’t see much difference between the two approaches so it’d be great if someone could explain to me why this approach is better than my previous one of catching the exception and printing the stack trace.

As it turns out, doing work on big data sets is quite hard. To start with, you need to get the data, and it is… well, big. So that takes a while. Instead, I decided to test my theory on the following scenario. Given 4 GB of random numbers, let us find how many times we have the number 1. Because I wanted to ensure a consistent answer, I wrote: public static IEnumerable RandomNumbers() { const long count = 1024 * 1024 * 1024L * 1; var random = new MyRand(); for (long i = 0; i < count; i++) { if (i % 1024 == 0) { yield return 1; continue; } var result = random.NextUInt(); while (result == 1) { result = random.NextUInt(); } yield return result; } } /// /// Based on Marsaglia, George. (2003). Xorshift RNGs. /// http://www.jstatsoft.org/v08/i14/paper /// public class MyRand { const uint Y = 842502087, Z = 3579807591, W = 273326509; uint _x, _y, _z, _w; public MyRand() { _y = Y; _z = Z; _w = W; _x = 1337; } public uint NextUInt() { uint t = _x ^ (_x << 11); _x = _y; _y = _z; _z = _w; return _w = (_w ^ (_w >> 19)) ^ (t ^ (t >> 8)); } } I am using a custom Rand function because it is significantly faster than System.Random. This generate 4GB of random numbers, at also ensure that we get exactly 1,048,576 instances of 1. Generating this in an empty loop takes about 30 seconds on my machine. For fun, I run the external sort routine in 32 bits mode, with a buffer of 256MB. It is currently processing things, but I expect it to take a while. Because the buffer is 256 in size, we flush it every 128 MB (while we still have half the buffer free to do more work). The interesting thing is that even though we generate random number, sorting then compressing the values resulted in about 60% compression rate. The problem is that for this particular case, I am not sure if that is a good thing. Because the values are random, we need to select a pretty high degree of compression just to get a good compression rate. And because of that, a significant amount of time is spent just compressing the data. I am pretty sure that for real world scenario, it would be better, but that is something that we’ll probably need to test. Not compressing the data in the random test is a huge help. Next, external sort is pretty dependent on the performance of… sort, of course. And sort isn’t that fast. In this scenario, we are sorting arrays of about 26 million items. And that takes time. Implementing parallel sort cut this down to less than a minute per batch of 26 million. That let us complete the entire process, but then it halts with the merge. The reason for that is that we push all the values into a heap, and there are 1 billion of them. Now, the heap never exceed 40 items, but those are still 1 billion * O(log 40) or about 5.4 billion comparisons that we have to do, and we do this sequentially, which takes time. I tried thinking about ways to parallel, but I am not sure how that can be done. We have 40 sorted files, and we want to merge all of them. Obviously we can sort each 10 files set in parallel, then sort the resulting 4, but the cost we have now is the actual sorting cost, not I/O. I am not sure how to approach this. For what is it worth, you can find the code for this here.

Security has become a great topic of discussion in the last few years due to the recent releasing of documents from Edward Snowden and the explosion of hacking against online commerce stores like JC Penny, Sony andTarget. While this post will not give you all of the tools to help prevent the use of illegally sourced data, this post will provide a starting point for building a set of tools and tactics that will help prevent the use of data by other parties. This post will show how to adopt AES encryption for strings in a Java environment. It will talk about creating AES keys and storing AES keys in a JCEKS keystore format. A working example of the code in this blog is located athttps://github.com/mike-ensor/aes-256-encryption-utility It is recommended to read each section in order because each section builds off of the previous section, however, this you might want to just jump quickly jump to a particular section. Setup - Setup and create keys with keytool Encrypt - Encrypt messages using byte[] keys Decrypt - Decrypt messages using same IV and key from encryption Obtain Keys from Keystore - Obtain keys from keystore via an alias What is JCEKS? JCEKS stands for Java Cryptography Extension KeyStore and it is an alternative keystore format for the Java platform. Storing keys in a KeyStore can be a measure to prevent your encryption keys from being exposed. Java KeyStores securely contain individual certificates and keys that can be referenced by an alias for use in a Java program. Java KeyStores are often created using the "keytool" provided with the Java JDK. NOTE: It is strongly recommended to create a complex passcode for KeyStores to keep the contents secure. The KeyStore is a file that is considered to be public, but it is advisable to not give easy access to the file. Setup All encryption is governed by laws of each country and often have restrictions on the strength of the encryption. One example is that in the United States, all encryption over 128-bit is restricted if the data is traveling outside of the boarder. By default, the Java JCE implements a strength policy to comply with these rules. If a stronger encryption is preferred, and adheres to the laws of the country, then the JCE needs to have access to the stronger encryption policy. Very plainly put, if you are planning on using AES 256-bit encryption, you must install theUnlimited Strength Jurisdiction Policy Files. Without the policies in place, 256-bit encryption is not possible. Installation of JCE Unlimited Strength Policy This post is focusing on the keys rather than the installation and setup of the JCE. The installation is rather simple with explicit instructions found here (NOTE: this is for JDK7, if using a different JDK, search for the appropriate JCE policy files). Keystore Setup When using the KeyTool manipulating a keystore is simple. Keystores must be created with a link to a new key or during an import of an existing keystore. In order to create a new key and keystore simply type: keytool -genseckey -keystore aes-keystore.jck -storetype jceks -storepass mystorepass -keyalg AES -keysize 256 -alias jceksaes -keypass mykeypass Important Flags In the example above here are the explanations for the keytool's parameters: Keystore Parameters genseckey Generate SecretKey. This is the flag indicating the creation of a synchronous key which will become our AES key keystore Location of the keystore. If the keystore does not exist, the tool will create a new store. Paths can be relative or absolute but must be local storetype this is the type of store (JCE, PK12, JCEKS, etc). JCEKS is used to store symmetric keys (AES) not contained within a certificate. storepass password related to the keystore. Highly recommended to create a strong passphrase for the keystore Key Parameters keyalg algorithm used to create the key (AES/DES/etc) keysize size of the key (128, 192, 256, etc) alias alias given to the newly created key in which to reference when using the key keypass password protecting the use of the key Encrypt As it pertains to data in Java and at the most basic level, encryption is an algorithmic process used to programmatically obfuscate data through a reversible process where both parties have information pertaining to the data and how the algorithm is used. In Java encryption, this involves the use of a Cipher. A Cipher object in the JCE is a generic entry point into the encryption provider typically selected by the algorithm. This example uses the default Java provider but would also work with Bouncy Castle. Generating a Cipher object Obtaining an instance of Cipher is rather easy and the same process is required for both encryption and decryption. (NOTE: Encryption and Decryption require the same algorithm but do not require the same object instance) Cipher cipher = Cipher.getInstance("AES/CBC/PKCS5Padding"); Once we have an instance of the Cipher, we can encrypt and decrypt data according to the algorithm. Often the algorithm will require additional pieces of information in order to encrypt/decrypt data. In this example, we will need to pass the algorithm the bytes containing the key and an initial vector (explained below). Initialization In order to use the Cipher, we must first initialize the cipher. This step is necessary so we can provide additional information to the algorithm like the AES key and the Initial Vector (aka IV). cipher.init(Cipher.ENCRYPT_MODE, secretKeySpecification, initialVector); Parameters The SecretKeySpecification is an object containing a reference to the bytes forming the AES key. The AES key is nothing more than a specific sized byte array (256-bit for AES 256 or 32 bytes) that is generated by the keytool(see above). Alternative Parameteters There are multiple methods to create keys such as a hash including a salt, username and password (or similar). This method would utilize a SHA1 hash of the concatenated strings, convert to bytes and then truncate result to the desired size. This post will not show the generation of a key using this method or the use of a PBE key method using a password and salt. The password and/or salt usage for the keys is handled by the keytool using the inputs during the creation of new keys. Initialization Vector The AES algorithm also requires a second parameter called the Initialiation Vector. The IV is used in the process to randomize the encrypted message and prevent the key from easy guessing. The IV is considered a publicly shared piece of information, but again, it is not recommended to openly share the information (for example, it wouldn't be wise to post it on your company's website). When encrypting a message, it is not uncommon to prepend the message with the IV since the IV will be a set/known size based on the algorithm. NOTE: the AES algorithm will output the same result if using the same IV, key and message. It is recommended that the IV be randomly created each time an encryption takes place. With the newly initialized Cipher, encrypting a message is simple. Simply call: byte[] encryptedMessageInBytes = Cipher.doFinal((message.getBytes("UTF-8")); String base64EncodedEncryptedMsg = BaseEncoding.base64().encode(encryptedMessageInBytes); String base32EncodedEncryptedMsg = BaseEncoding.base32().encode(encryptedMessageInBytes); Encoding Results Byte arrays are difficult to visualize since they often do not form characters in any charset. The best recommendation to solve this is to represent the bytes in HEX (base-16), Double HEX (base-32) or Base64 format. If the message will be passed via a URL or POST parameter, be sure to use a web-safe Base64 encoding. Google Guava library provides a excellent BaseEncoding utility. NOTE: Remember to decode the encoded message before decrypting. Decrypt Decrypting a message is almost a reverse of the encryption process with a few exceptions. Decryption requires a known initialization vector as a parameter unlike the encryption process generating a random IV. Decryption When decrypting, obtain a cipher object with the same process as the encryption method. The Cipher object will need to utilize the exact same algorithm including the method and padding selections. Once the code has obtained a reference to a Cipher object, the next step is to initialize the cipher for decryption and pass in a reference to a key and the initialization vector. // key is the same byte[] key used in encryption SecretKeySpec secretKeySpecification = new SecretKeySpec(key, "AES"); cipher.init(Cipher.DECRYPT_MODE, secretKeySpecification, initialVector); NOTE: The key is stored in the keystore and obtained by the use of an alias. See below for details on obtaining keys from a keystore Once the cipher has been provided the key, IV and initialized for decryption, the cipher is ready to perform the decryption. byte[] encryptedTextBytes = BaseEncoding.base64().decode(message); byte[] decryptedTextBytes = cipher.doFinal(encryptedTextBytes); String origMessage = new String(decryptedTextBytes); Strategies to keep IV The IV used to encrypt the message is important to decrypting the message therefore the question is raised, how do they stay together. One solution is to Base Encode (see above) the IV and prepend it to the encrypted and encoded message: Base64UrlSafe(myIv) + delimiter + Base64UrlSafe(encryptedMessage). Other possible solutions might be contextual such as including an attribute in an XML file with the IV and one for the alias to the key used. Obtain Key from Keystore The beginning of this post has shown how easy it is to create new AES-256 keys that reference an alias inside of a keystore database. The post then continues on how to encrypt and decrypt a message given a key, but has yet shown how to obtain a reference to the key in a keystore. Solution // for clarity, ignoring exceptions and failures InputStream keystoreStream = new FileInputStream(keystoreLocation); KeyStore keystore = KeyStore.getInstance("JCEKS"); keystore.load(keystoreStream, keystorePass.toCharArray()); if (!keystore.containsAlias(alias)) { thrownew RuntimeException("Alias for key not found"); } Key key = keystore.getKey(alias, keyPass.toCharArray()); Parameters keystoreLocation String - Location to local keystore file location keypass String - Password used when creating or modifying the keystore file with keytool (see above) alias String - Alias used when creating new key with keytool (see above) Conclusion This post has shown how to encrypt and decrypt string based messages using the AES-256 encryption algorithm. The keys to encrypt and decrypt these messages are held inside of a JCEKS formatted KeyStore database created using the JDK provided "keytool" utility. The examples in this post should be considered a solid start to encrypting/decrypting symmetric keys such as AES. This should not be considered the only line of defense when encrypting messages, for example key rotation. Key rotation is a method to mitigate risks in the event of a data breach. If an intruder obtains data and manages to hack a single key, the data contained in multiple files should have used several keys to encrypt the data thus bringing down risk of a total exposure loss. All of the examples in this blog post have been condensed into a simple tool allowing for the viewing of keys inside of a keystore, an operation that is not supported out of the box by the JDK keytool. Each aspect of the steps and topics outlined in this post are available at: https://github.com/mike-ensor/aes-256-encryption-utility. NOTE: The examples, sample code and any reference is to be used at the sole implementers risk and there is no implied warranty or liability, you assume all risks.

This technical tip shows how .NET developers can export Microsoft Visio diagram to XML inside their own applications using Aspose.Diagram for .NET. Aspose.Diagram for .NET lets you export diagrams to a variety of formats: image formats, HTML, SVG, SWF and XML formats: VDX defines an XML diagram. VTX defines an XML template. VSX defines an XML stencil. The Diagram class' constructors read a diagram and the Save method is used to save, or export, a diagram in a different file format. The code snippets in this article show how to use the Save method to save a Visio file to VDX, VTX and VSX. Exporting VSD to VDX VDX is a schema-based XML file format that lets you save diagrams in a format that products other than Microsoft Visio can read. It's a useful format for transferring diagrams between software applications and retaining editable data. To export a VSD diagram to VDX first create an instance of the Diagram class and call the Diagram class' Save method to write the Visio drawing file to VDX. Exporting from VSD to VSX VSX is an XML format for defining stencils, the basic objects from which a diagram is built up. When a Visio file is converted to VSX, only the stencils are exported. To export a VSD diagram to VSX first you need to create an instance of the Diagram class and then call the Diagram class' Save method to write the Visio drawing file to VSX. //The Sample code shows how to export VSD to VDX //[C# Sample] //Call the diagram constructor to load diagram from a VSD file Diagram diagram = new Diagram("D:\\Drawing1.vsd"); this.Response.Clear(); this.Response.ClearHeaders(); this.Response.ContentType = "application/vnd.ms-visio"; this.Response.AppendHeader("Content-Disposition", "attachment; filename=Diagram.vdx"); this.Response.Flush(); System.IO.Stream vdxStream = this.Response.OutputStream; //Save input VSD as VDX diagram.Save(vdxStream, SaveFileFormat.VDX); this.Response.End(); //[VB.NET Code Sample] 'Call the diagram constructor to load diagram from a VSD file Dim diagram As New Diagram("D:\Drawing1.vsd") Me.Response.Clear() Me.Response.ClearHeaders() Me.Response.ContentType = "application/vnd.ms-visio" Me.Response.AppendHeader("Content-Disposition", "attachment; filename=Diagram.vdx") Me.Response.Flush() Dim vdxStream As System.IO.Stream = Me.Response.OutputStream 'Save inpupt VSD as VDX diagram.Save(vdxStream, SaveFileFormat.VDX) Me.Response.End() //The Sample code shows how to export VSD to VSX format. [C# Code Sample] // Call the diagram constructor to load diagram from a VSD file Diagram diagram = new Diagram("D:\\Drawing1.vsd"); this.Response.Clear(); this.Response.ClearHeaders(); this.Response.ContentType = "application/vnd.ms-visio"; this.Response.AppendHeader("Content-Disposition", "attachment; filename=Diagram.vsx"); this.Response.Flush(); System.IO.Stream vsxStream = this.Response.OutputStream; //Save input VSD as VSX diagram.Save(vsxStream, SaveFileFormat.VSX); this.Response.End() //[VB.NET Code Sample] 'Call the diagram constructor to load diagram from a VSD file Dim diagram As New Diagram("D:\Drawing1.vsd") Me.Response.Clear() Me.Response.ClearHeaders() Me.Response.ContentType = "application/vnd.ms-visio" Me.Response.AppendHeader("Content-Disposition", "attachment; filename=Diagram.vsx") Me.Response.Flush() Dim vsxStream As System.IO.Stream = Me.Response.OutputStream 'Save input VSD as VSX diagram.Save(vsxStream, SaveFileFormat.VSX) Me.Response.End()

Working in and around Silicon Valley and technology, I hear people throwing around the terms “geek”, “dork”, ”nerd” and “dweeb” constantly. They’re thrown around interchangeably, in fact, which is where the problem lies. They’re not the same and knowing that matters a great deal. In fact, using the wrong term gives credit where credit it isn’t due or unfairly labels someone’s better qualities. Venn diagram Finding a Venn diagram to explain the difference was a big moment. I suddenly see how I should interview for different roles and what to look for in partners and employees. I know what to seek and what to avoid. It was an epiphany. Starting from that point, I could see career paths for each and every one (OK, except one). It breaks out like this: Geek, Nerd, Dweeb and Dork in order of value to the organization Geek – Both smart and driven but able to talk about fun things — they’re your leaders and sales people Nerd - Centered in smarts and drive, tempered by some awkwardness — they sustain your company Dweeb – Smart and awkward but probably uncommitted — they won’t stay up all night to solve a problem Dork - Least fun of the bunch — Avoid these people because they waste your time and sap your will to live These may not be everyone’s definitions, but maybe it’s time to standardize in our labels lest we use the terms insensitively. For an interesting take on these categories, I found this on Democratic Underground. Geek: Someone who spends a lot of time and energy in a certain special but conventional area, like computer programming or trouble-shooting, but not necessarily computers or technology. You can apparently have chess geeks, guitar geeks, or cooking geeks. A geek is an outwardly normal person who can relate to others in general but who has taken the time to learn specific technical skills and would rather talk about their special obsession than anything else. They are generally not athletic and enjoy sedentary pursuits like video games, comic books, being on the internet, etc. They usually dress to suit their special interest, which can be flamboyant, such as wearing a tee-shirt describing their special obsession or a hat bearing a logo of their special pursuit. Geeks can be self-confident and proud of their traits. Nerd: Someone with a great interest in academic subjects like math and science and who is socially awkward and has trouble relating to others outside of their fields of academia. Their IQ often exceeds their weight. Science fiction such as The Matrix and Star Wars or LOTR are often their cup of tea, as are hobbies like astronomy or chemistry sets. Nerds usually dress conservatively and are more interested in the mind than their outward appearance, although as both men and women they tend to be tidy, clean-cut, and hygienic. Nerds generally are self-confident in the academic setting and take pride in their intellect and band together with other nerds although their social skills outside of their academic obsession are diminished. Dork: Someone who has special interests like a geek but whose interests and obsessions are less common and odd, such as having an oddball collection of some sort like old Three Stooges bubblegum cards or an uncommon skill like yodeling. Walking talking Star Trek encyclopedic knowledge and convention dress up obsessions can be considered dorky. They can act silly at times and not care what anyone thinks. Dorks are typically more noted for their quirky personality and tend to be loners. Hygiene can sometimes be an issue. Dorks can nonetheless be self-confident and proud of the way they are because they simply don’t care what others think. Dweeb: A person who tends to be regarded as physically wimpish, intellectually challenged, and socially awkward, with little self-confidence. Dweebs tend to be obsessed with unusual pursuits like dorks (tap dancing or ant farms) but are lacking in skill, knowledge, or ability. Dweebs tend to be loners like dorks but understand their shortcomings and lack pride. Hygiene can also be an issue. I’m not a dork, nor dweeb and I don’t see myself as a nerd…I’m clearly a geek :-). Thank you to the Great White Snark, where I found the Venn diagram.

I mentioned several times that the entire point of the exercise was to just see how this works, not to actually do anything production worthy. But it is interesting to see how we could do better here. In no particular order, I think that there are at least several things that we could do to significantly improve the time it takes to sort. Right now we defined 2 indexes on top of a 1GB file, and it took under 1 minute to complete. That gives us a runtime of about 10 days over a 15TB file. Well, one of the reason for this performance is that we execute this in a serial fashion, that is, one after another. But we have to completely isolated indexes, there is no reason why we can’t parallelize the work between them. For that matter, we are buffering in memory up to a certain point, then we sort, then we buffer some more, etc. That is pretty inefficient. We can push the actual sorting to a different thread, and continue parsing and adding to a buffer while we are adding to the buffer. We wrote to intermediary files, but we wrote to those using plain file I/O. But it is usually a lot more costly to write to disk than to compress and then write to disk. We are writing sorted data, so it is probably going to compress pretty well. Those are the things that pop to mind. Can you think of additional options?