Java Disassembled

In my last post, I talked about how false sharing due to objects being on the same cache line can cause code to run much slower. Now, I'm going to show why

matrix[fixed_index][i] = value + 1;

is slower than

matrix[i][fixed_index] = value + 1;

This slowness is a property of your CPU rather than something innate to Java but we'll use the JDK with the -XX:+PrintOptoAssembly flag and examine the generated assembler code to show the problem.

(Please note that I am not a professional x86 assembler programmer so any input would be gratefully received.)

First, some preliminaries. Most of the following was gleaned from using JDK 1.7.0 on Linux:

1. Java multi-dimensional arrays are actually arrays of arrays.

2. The second location in the memory representing an array (array reference + 1) yields information as to its type. We can show this by looking at the assembly generated by this Java:

       private void myInstanceOf(Object[] array) {
       if (array instanceof Integer[]) {
                 aPublicInt++;
       }
}

[Aside: the field aPublicInt is exactly that: an int that is publicly accessible. The reason we may need it is that Hotspot may optimize away our code if it considers that nothing noticeable changes.]

Anyway, the assembler it generates looks a little like this:

#r000 ecx   : parm 0: com/henryp/lang/SimpleClassForDisassembly:NotNull *
#r005 edx   : parm 1: java/lang/Object *[int:>=0] *
# -- Old esp -- Framesize: 16 --
#r045 esp+12: return address
#r044 esp+ 8: pad2, in_preserve
#r043 esp+ 4: pad2, in_preserve
#r042 esp+ 0: Fixed slot 0
#
000   N45: #    B1 <- BLOCK HEAD IS JUNK   Freq: 1
000       CMP    EAX,[ECX+4]    # Inline cache check
JNE    SharedRuntime::handle_ic_miss_stub
NOP
NOP
NOP

000
00c   B1: #    B5 B2 <- BLOCK HEAD IS JUNK   Freq: 1
00c       # stack bang
PUSHL  EBP
SUB    ESP,8    # Create frame
01a       MOV    EBX,[EDX + #4]
01d       NullCheck EDX
01d
01d   B2: #    B4 B3 <- B1  Freq: 0.999999
01d       CMPu   EBX,precise klass [Ljava/lang/Integer;: 0xb76eeb68:Constant:exact *
023       Jne,us B4  P=0.100000 C=-1.000000
023
025   B3: #    B4 <- B2  Freq: 0.899999
025       INC    [ECX + #16] ! Field com/henryp/lang/SimpleClassForDisassembly.aPublicInt
028
028   B4: #    N45 <- B3 B2  Freq: 0.999999
028       ADD    ESP,8    # Destroy frame
POPL   EBP
TEST   PollPage,EAX    ! Poll Safepoint

032       RET
032
033   B5: #    N45 <- B1  Freq: 1.01328e-06
033       MOV    EBP,ECX
035       MOV    ECX,#-12
03a       NOP     # 1 bytes pad for loops and calls
03b       CALL,static  wrapper for: uncommon_trap(reason='null_check' action='make_not_entrant')
# com.henryp.lang.SimpleClassForDisassembly::myInstanceOf @ bci:1  L[0]=EBP L[1]=_ STK[0]=#NULL
# OopMap{ebp=Oop off=64}
040       INT3   ; ShouldNotReachHere
040

Verbose, isn't it?

(Note that this assembler was generated when the Java method was alternately passed an Integer[] and a String[] for many iterations. The JDK will generate other assembler depending on the which paths through the method are most commonly used.)

So what does this assembly code mean? The first line says the ecx register holds a reference to this. But the second line tells us the register edx points to an the argument that is passed to this method - the Object[].

So much for the references. The interesting code starts at the label B1. Here we create the stack for this method. EBP and ESP are special "32-bit registers that are used as pointers [,] the extended base pointer (EBP) and the extended stack pointer (ESP)...

"The EBP register (sometimes called the frame pointer (FP) or local base pointer (LB)) is used to reference variables in the current stack frame. Each stack frame contains the parameters to the function, its local variables, and two pointers that are necessary to put things back the way they were: the saved frame pointer (SFP) and the return address. The stack frame pointer is used to restore EBP to its previous value, and the return address is used to restore EIP [the address of the code currently executing] to the next instruction found after the function call." [1]

Breaking this code down, we see:

    PUSHL  EBP

(Push the old frame pointer value onto the stack).

    SUB    ESP,8    # Create frame

(Allocate 8 bytes on the stack. The stack starts high in the memory space and moves towards zero as more stack space is allocated so we SUBtract 8 bytes from the ESP).

01a       MOV    EBX,[EDX + #4]
01d       NullCheck EDX

(Move the contents of the second memory space of the array - remember, EDX points to our array - into the EBX register. The Nullcheck is not an assembly instruction. It appears to be pseudo code as it has the same address - 01d - as the next line.)

01d       CMPu   EBX,precise klass [Ljava/lang/Integer;: 0xb76eeb68:Constant:exact *
023       Jne,us B4  P=0.100000 C=-1.000000

(Is EBX the same as the constant that represent an Integer array? If the unsigned (,us) comparison concludes they are not equal, go to B4. Otherwise...)

025       INC    [ECX + #16] ! Field com/henryp/lang/SimpleClassForDisassembly.aPublicInt

(add one to this.aPublicInt which occupies the memory address 16 bytes after the memory address of this. Remember, the first line tells us that ECX points to this).

Either way, we come to B4 where we undo all the work of allocating stack space and RETurn.

3. The third location in the memory holding an array (array reference + 2) yields information as to its length. We can show this by looking at the assembly generated by this Java:

       private static int myArraySize(Object[] array) {
       return array.length;
}

The assembler generated here looks like this (ignoring all the stack frame manipulation etc that we saw in the previous example):

#r000 ecx   : parm 0: int[int:>=0]:exact *
.
.
.
00e     MOV    EAX,[ECX + #8]
011     NullCheck ECX

Here, ecx is the register that represents the argument (there is no reference to this in the code as you'll notice the Java method is static). We put the contents of the address space 2 memory slots (2 * 4 bytes = 8) after the array reference itself into EAX. This is the register that by convention holds the return value of any call.

Returning to our matrix navigation code, let's start with this method:

       private int[][] myMatrixNavigation() {
        int size = 19;
        int matrix[][] = new int[size][27];
        for (int i = 0 ; i < size ; i++) {
                matrix[i][7] = 11;
        }
        return matrix;
}

And run it enough times for the JDK to generate assembly language. It generates assembler that looks like:

       SUB    ESP,24   # Create frame
  XOR    EBP,EBP
  MOV    ECX,precise klass [[I: 0xa062f798:Constant:exact *
  MOV    EDX,#19
  MOV    EDI,#27
  MOV    [ESP + #0],EDI
  NOP     # 1 bytes pad for loops and calls
  CALL,static  wrapper for: _multianewarray2_Java
  # com.henryp.lang.SimpleClassForDisassembly::myMatrixNavigation @ bci:6  L[0]=_ L[1]=#19 L[2]=_ L[3]=_
  # OopMap{off=52}
  #checkcastPP of EAX
  MOV    EDX,#19

B2: #        B27 B3 <- B1 B5         Loop: B2-B5 inner stride: not constant pre of N159
  CMPu   EBP,#19
  Jnb,u  B27

B3: #        B28 B4 <- B2
  MOV    EDI,[EAX + #12 + EBP << #2]
  MOV    ECX,[EDI + #8]
  NullCheck EDI

B4: #        B26 B5 <- B3
  CMPu   ECX,#7
  Jbe,u  B26

B5: #        B2 B6 <- B4
  MOV    [EDI + #40],#11
  INC    EBP
  CMP    EBP,#1
  Jl,s  B2        # Loop end

B6: #        B16 B7 <- B5
  SUB    EDX,EBP
  AND    EDX,#-4
  ADD    EDX,EBP
  CMP    EBP,EDX
  Jge,s  B16
  NOP     # 6 bytes pad for loops and calls

B7: #        B28 B8 <- B6 B15        Loop: B7-B15 inner stride: not constant main of N96
  MOV    ECX,[EAX + #12 + EBP << #2]
  MOV    EDI,[ECX + #8]
  NullCheck ECX

B8: #        B23 B9 <- B7
  CMPu   EDI,#7
  Jbe,u  B23

B9: #        B28 B10 <- B8
  MOV    EDI,[EAX + #16 + EBP << #2]
  MOV    [ECX + #40],#11
.
.
.

Hugely abbreviated, it does something like this:

        XOR    EBP,EBP

(Set EBP to 0. Exclusive-or of anything on itself will be 0. This is an optimisation. EBP will be our loop counter - equivalent to i in the Java code above).

        CALL,static  wrapper for: _multianewarray2_Java

(Call JVM code to create our array. Remember that the register EAX is used to return a value from a call so it's this that points to our array).

        MOV    EDI,[EAX + #12 + EBP << #2]

(Move the ith element of the outer array into EDI. Remember that the second memory location of an array (+4) is its type, the third (+8) its length so +12 is the first element).

        MOV    [EDI + #40],#11

(This element is itself an array. So, bearing in mind that the array reference + 4 is the type, the array reference + 8 is the length, the array reference + 12 the first element and so on, the array reference + 40 must be the 7th ((7 * 4) + 12) element. We set the contents of this memory address to 11 as our Java code says we should).

Note that this memory address is likely to be far away from the reference of the first array (stored in the EAX register). Compare that to the assembly code generated by this Java method:

       private int[][] myMatrixNavigation2ndIndice() {
           int size = 19;
           int matrix[][] = new int[27][size];
           for (int i = 0 ; i < size ; i++) {
                   matrix[7][i] = 11;
           }
           return matrix;
   }

This is very similar to before but this time we're fixing the 2nd index of the matrix, not the first. I'll spare you all the gory assembler, but the relevant code boils down to this:

        MOV    EDI,[EAX + #40]

(Once more, EAX stores the reference to our array and the +40 is used to access the 7th element).

        MOV    [EDI + #12 + EBP << #2],#11

(The 7th element is itself an array. So, we traverse this array once more with EBP as our index, setting the memory contents to 11, just as our Java code says we should).

Now the thing to note here is that if this code immediately above is multi-threaded, all threads will traverse a contiguous chunk of memory, that is [EDI + 12 + EBP] since EBP is just our integer index.

In the first piece of Java code where the first index is fixed, a thread will be writing to some element in the array that is found at [EDI + 12 + EBP] not the location [EDI + 12 + EBP] itself. It is much less likely that this memory is going to be close to the memory location for any other value of EBP (in fact for a sufficiently large array I think it might be impossible). Since cache lines are typically only 64 bytes, contention for the same cache line is much less likely and so the first piece of Java code is faster.

[1] The Art of Exploitation - Jon Erickson.

False Sharing

If multiple threads concurrently change the values in a matrix, which is faster: working through the matrix column-by-column or row-by-row? And why is there a difference?

Let's say we have a square, 2-dimensional matrix of ints that we model like so:

int SIZE    = 1000;
int ROWS    = SIZE;
int COLUMNS = SIZE;

int[][] matrix = new int[ROWS][COLUMNS];

Technically, Java doesn't have multi-dimensional arrays. It has arrays of arrays. But for our purposes, let's define the first index as the row and the second as the column.

Now, let's say that at the same time our many threads change all the values in a given column with the following code:

  // change all the values in a fixed column of our matrix
 for (int i = 0 ; i < ROWS ; i++) {
    int value = matrix[i][fixed_index];
    matrix[i][fixed_index] = value++;
 }

and after this code has finished running, we change it slightly so that now the many threads change all the values in a given row.

    // change all the values in a fixed row of our matrix
   for (int i = 0 ; i < COLUMNS ; i++) {
       int value = matrix[fixed_index][i];
       matrix[fixed_index][i] = value + 1;
   }

This time, we've fixed the "row" index and let the "column" index vary so all values in a given row change.

The results on my MacBook Pro (Intel Core 2 Duo) running 100 threads each performing 1000 iterations look like this:

Run____fixed column____fixed row
1__________209_____________91___
2__________194_____________78___
3__________201_____________92___
4__________199_____________60___
5__________228_____________63___

[mean time in ms. Apologies for the formatting. Blogger keeps trying to eat whitespace].

As you can see, changing all the values in one of our rows is much faster than changing all the values in one of our columns.

Remembering that there are really no rows and columns in Java (we're actually modeling our matrix using arrays of arrays), the easy way to remember this is the 3 Fs: when it comes to changing the values along a give index of a 2-D array, Fixing the First is Fastest.

This phenomena is due to a mechanism called "false sharing" (there is a good podcast by Scott Meyers on the matter here).

False sharing slows concurrent changes to RAM. It occurs when one thread is writing to memory that is close to where another thread is writing. Robert Love defines it as when there are "two or more objects mapping to the same cache line despite existing at different addresses in memory" (Linux Kernel Development 3r Edition).

The maximum size of memory in which these two or more addresses need to be for this to happen is called a cache line. The exact size is dependent on the architecture but it is typically 64 bytes.

What's happening in the above example is that ints in the column of our matrix are evidently stored close together - so close there is some cache line collisions occurring - and that threads are contending to write to that piece of RAM.

I'll write another post as to how we can prove this.

Synchronicity

I had a debate over a beer with a friend about volatile arrays. I forgot who was trying to prove what but I did come across an interesting quirk.

We were talking about how adding the volatile keyword to a reference means that threads will always see the same thing once the reference had been set. But what if that reference were an array? Would the threads agree on the elements of the array? (The answer is "not necessarily" as pointed out in my blog last year).

But I had trouble demonstrating this with some simple, multi-threaded code. Annoyingly, my threads kept agreeing on the elements of the array. Maybe it was the chip architecture, I lazily thought.

Then it occurred to me. I was putting some System.out.println statements in my code to help me try to solve the problem. This, of course, was the problem. The reason why is that by introducing this logging, the threads are synchronizing on the same object - and as I mentioned in another blog post, this ensures "all threads see the most up-to-date values of shared mutable variables [when] the reading and writing threads ... synchronize on a common lock" (Java Concurrency in Practice, p37).

So, what was introducing this mysterious common lock? Why, System.out.println itself! The println method is synchronized so both threads synchronize on the common lock of the System.out object.

This reminded me of a time when a colleague was wondering why adding System.out.println appeared to be altering the behaviour of a piece of code. Stumped, I just thought it might have introduced a timing issue to his multi-threaded code. Now I know better.

But it begs the question: how much code is out there that is not properly multi-threaded with only some ad hoc logging holding it together?

Optimising Collection Access

I had an interview question recently that asked me to write code to access as efficiently as possible a collection while multiple threads intensively read and write to it.

Giving the caveat that everything should be subject to stress tests anyway, I went ahead and used a java.util.ConcurrentHashMap. I knew this is a very efficient piece of code and I wouldn't have to worry about synchronization.

When I got home, I wondered about my answer and started to put together JMeter tests to establish some empirical data. JMeter provides a lot of the plumbing for stress tests that makes it superior to just writing your own class.

[Incidentally, if you do write your own executable classes for stress tests, don't forget to pause for 4 seconds before running your code. Any object created in the first 4s of the JVM's life is not eligible for an optimization called Biased Locking (see this email from a Sun engineer). Biased Locking is an optimization for the use case where a lock is mostly sought by just one thread. In this case, the favoured thread finds attaining the lock is cheap.]

First, if you wish to plug your code into JMeter, it's helpful to have a superclass that implements the JMeter interface:

package com.henryp.stress;



import org.apache.jmeter.protocol.java.sampler.JavaSamplerClient;

import org.apache.jmeter.protocol.java.sampler.JavaSamplerContext;

import org.apache.jmeter.samplers.SampleResult;



public abstract class AbstractSamplerClient implements JavaSamplerClient {



public SampleResult runTest(JavaSamplerContext javaSamplerContext) {

 SampleResult result = new SampleResult();

 result.sampleStart();



 doSample();



 result.sampleEnd();

 result.setSuccessful(true);

 result.setResponseCodeOK();

 result.setResponseMessageOK();



 return result;

}



protected abstract void doSample();



}

Then, subclass it. For my ConcurrentHashMap, it looks something like this:

package com.henryp.stress;



import java.util.Map;



import org.apache.jmeter.config.Arguments;



public abstract class AbstractMapSamplerClient extends AbstractSamplerClient {



protected static Map concurrentHashMap;

protected final MapPopulator mapPopulator = new MapPopulator();



@Override

public Arguments getDefaultParameters() {

  return mapPopulator.getDefaultParameters();

}



@Override

protected void doSample() {

  for (Map.Entry entry : concurrentHashMap.entrySet()) {

      // NoOp while I think of something to do here

  }

}



@Override

public synchronized void setupTest(JavaSamplerContext javaSamplerContext) {

  if (concurrentHashMap == null) {

      System.out.println("setupTest: " + javaSamplerContext);



      concurrentHashMap = mapPopulator.makePopulatedMap(javaSamplerContext);

  }

}



@Override

public void teardownTest(JavaSamplerContext javaSamplerContext) {

  concurrentHashMap = null;

}

}

The MapPopulator is just a utility that instantiates a Map of a user-defined type and populates it with a user-defined number of elements. Setting the default parameters in that class looks something like:

.

.

.

protected static final int    DEFAULT_COLLECTION_SIZE = 100;

protected static final String COLLECTION_SIZE_KEY     = "COLLECTION_SIZE_KEY";



public Arguments getDefaultParameters() {

   System.out.println("getDefaultParameters");



   Arguments arguments = new Arguments();



   Argument argument = new Argument();

   argument.setName(COLLECTION_SIZE_KEY);

   argument.setValue("" + DEFAULT_COLLECTION_SIZE);

   arguments.addArgument(argument );



   return arguments;

}

.

.

.

To make JMeter pick up my classes, I put this in $JMETER_HOME/bin/jmeter.properties:

search_paths=/Users/henryp/Documents/workspace/TestMaven/target/TestMaven-0.0.1-SNAPSHOT.jar

This way, I only had to execute Maven to compile my code and rebuild the JAR. It would be nice to hot-deploy the code but I don't know if JMeter supports this. So, instead I have to restart JMeter each time I change the code :-(

My code that iterates over the entry set of a ConcurrentHashMap was producing a throughput of about 25 000/s when using 100 read-threads while another 10 write-threads added an element to the collection with a constant throughput of 100/s (I've not shown the code for this).

Not bad. But could it be improved easily? Let's try using a java.util.List (perhaps surprisingly, ArrayList and LinkedList gave me similar figures) and use read/write locks to synchronize access. So, the read-code looks something like this:

 static ReadWriteLock readWriteLock;



 @Override

 protected void doSample() {

     Lock readLock = readWriteLock.readLock();

     try {

         readLock.lock();

         for (String aValue : list) {

             // no-op

         }

     } finally {

         readLock.unlock();

     }

 }

and the write-code looks something like this:

 private final AtomicInteger callId = new AtomicInteger();

.

.

.

 @Override

 protected void doSample() {

     Lock writeLock = readWriteLock.writeLock();

     try {

         writeLock.lock();

         String aValue = this.toString() + callId.getAndIncrement();

         SingleListReaderWithRWLocksSamplerClient.list.add(aValue);

     } finally {

         writeLock.unlock();

     }

 }

The subsequent JMeter test (100 read thread, 10 write threads with a throughput of 100/s) indicated that this was about twice as fast as using ConcurrentHashMap.

What's more, it seemed to suffer a far smaller degradation in performance when I increased the throughput of write-threads to 1000/s.

Conclusion: they're both pretty nifty and perform at the same order of magnitude. Your mileage may vary but it seems java.util.concurrent.locks.ReadWriteLock is a more efficient approach.

Tomcat/Maven plugin

My consultancy contract has come to an end so now I have time to attend my much neglected blog.

In the last few months, I have been Maven-izing and modularizing a legacy application. (Architectural tip: err on the side of being too modularized. It's much easier to fuse modules together than to split them.)

Anyway, one of the jobs was to improve the developer experience by having a web app running via Maven. The advantages include the project working straight out of the box with no awkward manual configuration. But it was only partially working with the Tomcat Maven Plugin. Some parts of the page were not loading because of NullPointerExceptions in our code. And yet, when I pointed a standalone version of Tomcat at the same exploded WAR, everything worked.

Using Firebug's debugger and single-stepping through Javascript code revealed nothing amiss. But stepping through code of the Tomcat embedded in the Maven plugin, I found this in org.apache.catalina.authenticator.AuthenticatorBase:

if (session != null && changeSessionIdOnAuthentication) {

Manager manager = request.getContext().getManager();

manager.changeSessionId(session);

request.changeSessionId(session.getId());

}

So, session IDs were changing on each call. Indeed, Firebug did show that each AJAX call had a different JSESSIONID.

Further investigation shows that the Maven Tomcat Plugin uses Tomcat 6.0.29 (you can tell this by looking at the plugin's pom.xml). Our non-Maven Tomcat was 6.0.14.

What could be different? The change of a patch version number suggests bug-fixes and not new functionality. And what is this changeSessionIdOnAuthentication field?

It happens that we had a bespoke security plugin which extends Tomcat's AuthenticatorBase and authenticates on every HTTP call. With the default value of changeSessionIdOnAuthentication being true, this was causing the session to be discarded on the 6.0.29 Tomcat instance and so all fields in the session object to be uninitialized, ergo the NullPointerExceptions.

The changeSessionIdOnAuthentication field was added some time between releases 6.0.14 and 6.0.29. According to the Tomcat documentation, it is used "to prevent session fixation attacks".

These attacks are when somebody sends you a link to a website for which you have an account. The link includes the JSESSIONID that the attacker has been given by the website. You enter your credentials and the HTTP session associated with this JSESSIONID now carries some flag to say you have logged in. After this, the person who sent you the link can go to the website and as long as he is using the same JSESSIONID as you, he too can see your account since the associated HTTP sessions says you are authenticated.

The changeSessionIdOnAuthentication field being set to true means that as soon as you log in, you get a new JSESSIONID and the attacker is foiled.

Since our developers don't need to worry about this, we turned the functionality off in the server.xml file that the Maven plugin is told to use with:

<host name="localhost"... >
.
.
.
<valve classname="our.implementation.of.AuthenticatorBase" changeSessionIdOnAuthentication="false"...

In praise of checked exceptions

I am swimming against the tide but I quite like checked exceptions - when done properly. I find them self-documenting and they force engineers to consider the possibility that their request cannot be serviced.

My partner, who is a lawyer but a wannabe geek, laughs when I tell her that lots of the problems I encounter at work are because somebody did not consider a certain set of circumstances. She tells me that in law, it is a very poor contract that does not cover all eventualities.

When I discuss my unpopular view with other engineers, I often hear that if you can't do anything then you have no choice but to re-throw an Exception. This is true but people often underestimate the number of options still open to them upon catching one.

For example, in the event of a database deadlock, one of the threads will immediately be rolled back. The JDBC driver will throw an SQLException but then that thread can try again.

To be fair, Java's JDBC API has left something to be desired. For most JDBC drivers, an update that violates a constraint and is never going to succeed throws the same exception as one that was deadlocked and might work upon a retry. (A class was added to the SQLException hierarchy to address this - java.sql.SQLRecoverableException - but it only appeared as late as JDK 1.6).

I am content to adopt the house-style. If my client prefers RuntimeExceptions then so be it. But an area where I feel checked exceptions should be mandatory is when one team creates an API for another.

A problem I saw recently was when the core team wrote a Java API and published it via Java RMI. A second team called that API but upon encountering a non-deterministic deadlock was surprised when the JVM barfed with an Error. It transpired that DeadlockLoserDataAccessException was not in their classpath - a Spring class that was thrown when Spring converted the SQLException to something a little more descriptive.

With checked exceptions, APIs are not only somewhat self-documenting but better define what the client needs in its classpath.

This proposal is not a panacea - objects passed over the wire might reference non-standard implementations. For instance, objects re-hydrated with Hibernate might have Hibernate implementations of standard Java collection interfaces (you might want to look at Gilead to clean up the object graph in this case).

[Interestingly, the proposal to use Gilead on this particular project also prompted a discussion on what was reasonable tech debt. Not all debt is bad - a mortgage allowed me to buy my house. Only when it becomes unmanageable is it a problem. When the team who wrote the API that passed Hibernate re-hydrated objects across the wire first started coding, they were the only clients of it. Therefore, they considered the use of Gilead as unnecessary. "Do the simplest thing that works," they said and put Hibernate in their client's classpath. The problem became a little more intractable when a second team used the API since they used a different version of Hibernate.]

One last point: if you publish APIs and both you and your clients use Maven, you can have the pom.xml of the API module to pull in any necessary third-party libraries as transitive dependencies. However, do remember that not all teams use Maven. What's more, not all APIs use just Java.

To Infinispan and Beyond!

I've been playing for the last few weeks with JBoss's Infinispan - a distributed cache and the successor to JBossCache.

It's a nice piece of technology. But for our needs, it may be inappropriate.

It may be OK for storing users' HTTP sessions and the like. But we were thinking about storing data for which integrity is absolutely essential (bond orders that run typically into the billions of dollars). Let me run through a small number of arguments that made me think Infinispan is not for us.

The Infinispan FAQ says it uses MVCC to optimize performance by having write locks not block read locks. However, dig deeper and you see this is not MVCC as most people who are familiar with Oracle and Postgres would know it. In these database implementations, a reading thread works with a snapshot of data. If a writing thread updates that row in the database, then that's OK. The reading thread continues with its snapshot. This is great since nobody is blocking anybody and it can improve performance.

However, with Infinispan, this is not quite what is happening. It's true that write threads do not block read threads. But the data in question may be mutable.

In the Infinispan mailing lists, we read:

"[I]n this case you may well see the change before [a write transaction] commits, we don't explicitly clone or copy mutable objects ... I suppose though we could add the ability to defensively copy mutable objects, but we'd need a way of knowing which are immutable, etc. Also, this would be more expensive, depending on the size of the atomic map."

A quick-and-dirty test I wrote demonstrates this:

package org.infinispan.replication;

import static org.testng.AssertJUnit.assertNull;

import java.util.concurrent.CountDownLatch;

import javax.transaction.HeuristicMixedException;
import javax.transaction.HeuristicRollbackException;
import javax.transaction.NotSupportedException;
import javax.transaction.RollbackException;
import javax.transaction.SystemException;
import javax.transaction.TransactionManager;

import org.infinispan.Cache;
import org.infinispan.config.Configuration;
import org.infinispan.test.MultipleCacheManagersTest;
import org.infinispan.test.TestingUtil;
import org.infinispan.test.AbstractCacheTest.CleanupPhase;
import org.testng.AssertJUnit;
import org.testng.annotations.Test;

@Test(groups = "functional", testName = "replication.SyncReplLockingAtomicityPHTest")
public class SyncReplLockingAtomicityPHTest extends MultipleCacheManagersTest {

  private static final String CACHE_NAME = "testcache";

  private static final String k = "key", v = "value";

  public SyncReplLockingAtomicityPHTest() {
     cleanup = CleanupPhase.AFTER_METHOD;
  }

  protected Configuration.CacheMode getCacheMode() {
     return Configuration.CacheMode.REPL_SYNC;
  }

  protected void createCacheManagers() throws Throwable {
     Configuration cfg = getDefaultClusteredConfig(getCacheMode(), true);
     cfg.setLockAcquisitionTimeout(500);
     createClusteredCaches(2, CACHE_NAME, cfg);
  }

  public void testUpdateWhileReadLock() throws Exception {
      final Cache             cache               = cache(0, CACHE_NAME);
      final CountDownLatch    latchAfterRead      = new CountDownLatch(1);
      final CountDownLatch    latchBeforeCommit   = new CountDownLatch(1);
      final ReaderThread      readerRunnable      = new ReaderThread(k, cache, latchAfterRead, latchBeforeCommit);
      final UpdateThread      updateRunnable      = new UpdateThread(k, cache);
      updateWhileReadTX(latchAfterRead, latchBeforeCommit, readerRunnable, updateRunnable);
  }

  private void updateWhileReadTX(
          CountDownLatch latchAfterRead,
          CountDownLatch latchBeforeReadCommit,
          ReaderThread readerRunnable,
          UpdateThread updateRunnable) throws SecurityException, IllegalStateException, RollbackException, HeuristicMixedException, HeuristicRollbackException, SystemException, NotSupportedException, InterruptedException {
      Cache cache1 = cache(0, "testcache");
      Cache cache2 = cache(1, "testcache");
      assertClusterSize("Should only be 2  caches in the cluster!!!", 2);
      assertNull("Should be null", cache1.get(k));
      assertNull("Should be null", cache2.get(k));
   
      final StringBuffer  initialObj      = populateCache(cache1);
      final Thread        readThread      = new Thread(readerRunnable);
      readThread.start();
      latchAfterRead.await();
      final StringBuffer  readObject      = readerRunnable.getFromCache();
   
      final Thread        updateThread    = new Thread(updateRunnable);
      updateThread.start();
      updateThread.join();
      final StringBuffer  writeObject     = updateRunnable.getValue();
      final String        writeSnapshot   = writeObject.toString();
      final String        readSnapshot    = readObject.toString();
   
      latchBeforeReadCommit.countDown();
   
      AssertJUnit.assertSame(readObject, writeObject);
        AssertJUnit.assertFalse(readSnapshot.equals(writeSnapshot));
  }

  private StringBuffer populateCache(Cache cache) throws SecurityException, IllegalStateException, RollbackException, HeuristicMixedException, HeuristicRollbackException, SystemException, NotSupportedException {
      StringBuffer mutableObject = new StringBuffer("test");
   
      TransactionManager mgr = TestingUtil.getTransactionManager(cache);
      mgr.begin();
      cache.getAdvancedCache().lock(k);
      cache.put(k, mutableObject);
      mgr.commit();
      return mutableObject;
  }

  private class UpdateThread implements Runnable {
   
      protected final String              key;
      protected final Cache               cache;
      protected final TransactionManager  mgr;
      private StringBuffer                fromCache;

      public UpdateThread(String k, Cache cache) {
          this.key = k;
          this.cache = cache;
          mgr = TestingUtil.getTransactionManager(cache);
      }

      @Override
      public void run() {
          try {
              mgr.begin();
              getValueFromCache();
              fromCache.append(System.currentTimeMillis());
              cache.put(k, fromCache);
          } catch (Exception x) {
              x.printStackTrace();
          } finally {
              finishTx();
          }
       
      }

      private void finishTx() {
          try {
              mgr.commit();
          } catch (Exception x) {
              x.printStackTrace();
          }
      }

      private void getValueFromCache() {
          fromCache = (StringBuffer) cache.get(key);
      }
   
      public StringBuffer getValue() {
          return fromCache;
      }
   
  }


  private class ReaderThread implements Runnable {
   
      private final String key;
      private final Cache cache;
      private final CountDownLatch latchAfterRead, latchBeforeCommit;
   
      private StringBuffer fromCache;

      public ReaderThread(String k, Cache cache,
              CountDownLatch latchAfterRead, CountDownLatch latchBeforeCommit) {
          this.key                = k;
          this.cache              = cache;
          this.latchAfterRead     = latchAfterRead;
          this.latchBeforeCommit  = latchBeforeCommit;
      }

      public StringBuffer getFromCache() {
          return fromCache;
      }

      @Override
      public void run() {
          TransactionManager mgr = TestingUtil.getTransactionManager(cache);
          try {
              mgr.begin();
              fromCache = (StringBuffer) cache.get(key);
              latchAfterRead.countDown();
          } catch (Exception x) {
              x.printStackTrace();
          } finally {
              try {
                  latchBeforeCommit.await();
                  mgr.commit();
              } catch (Exception x) {
                  x.printStackTrace();
              }
          }
      }
   
  }
}

The line in bold is where the test fails. The snapshot the read-thread has is the same as the write-thread, even though it read the object from the cache first. This is simply due to the fact that both read- and write-threads have the same object reference.

Note, if the write operations were rolled back, the read-thread would still see the change it had made. In fact, the whole cache from this point on would hold incorrect data.

This is not dissimilar to Hibernate updating an object (eg, assigning a primary key or a timestamp) even when the DB transaction it was in was rolled back. I've seen this happen when there were DB deadlocks and my transaction was chosen as the victim.

There are many other reasons why I think business critical data should not be put into Infinispan. I hope to present them in upcoming blog entries.