Bugzilla – Full Text Bug Listing
| Summary: | NPE on commit after connection lost when using ReconnectingSessionConfiguration | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Product: | [Modeling] EMF | Reporter: | Sebastian Paul <sebastian.paul.ext> | ||||||||
| Component: | cdo.net4j | Assignee: | Caspar D. <caspar_d> | ||||||||
| Status: | CLOSED WORKSFORME | QA Contact: | |||||||||
| Severity: | normal | ||||||||||
| Priority: | P3 | CC: | caspar_d | ||||||||
| Version: | 4.0 | ||||||||||
| Target Milestone: | --- | ||||||||||
| Hardware: | All | ||||||||||
| OS: | All | ||||||||||
| URL: | http://www.eclipse.org/forums/index.php?t=msg&goto=660535&S=1290accdc8c19348495aa655b82c3309 | ||||||||||
| Whiteboard: | |||||||||||
| Attachments: |
|
||||||||||
Created attachment 192165 [details]
The test which shows the problem
I haven't yet been able to reproduce the problem with your testcases.
This may be because we're on different platforms. I'm on Linux. Are
you on Windows?
In any case, what's important is that it's the user app's responsibility
to avoid making calls on the protocol (or calls on the session that will
trigger calls on the protocol) while connection recovery is in
progress. The app must register a listener with the session object, and
listener for CDOSessionRecoveryEvent.Type.STARTED and FINISHED, and not
make any calls between those events.
From a quick look at your testcases, I get the impression there's no
code that does this, and so my guess is that
The reason things work this way, is that otherwise a RecoveringSession
would have to override every session method to wrap the super invocation
with something like
if (weHappenToBeDoingRecoveryRightNow())
{
// Throw ex? Or suspend caller?
}
else
{
super.doTheNormalThing();
}
Please comment further if I misunderstand.
I see I left a sentence unfinished. What I meant to say was:
> From a quick look at your testcases, I get the impression there's no
> code that does this, and so my guess is that
your testcode proceeds prematurely with another attempt to commit.
Thanks for the quick response. I have fixed the test, now it succeeds. I am still wondering why reconnect takes so long (10sec). I tried tweaking this via the session configuration, but without any effect. Created attachment 192181 [details]
Fixed test (succeeds)
Created attachment 192248 [details]
Fixed tests (Caspar's version)
I had to make a few changes to your tests to make them work
reliably in my environment. Specifically, I replaced the
boolean 'recoveryInProgress' with a latch, and added a
check to see whether the real proxy process is actually still
running when we attempt to stop it. The tests all pass on my
platform as well now.
(In reply to comment #4) > I am still wondering why reconnect takes so long (10sec). I tried > tweaking this via the session configuration, but without any effect. I've looked into this a bit, and I get the impression that the actual recovery is very fast, but that it is slowed down somehow by the restart of the proxy process. What makes me say this, is that if I set a breakpoint in your #notifyEvent handler, and keep it suspended there for 10 seconds or so, while allowing the main thread to restart the proxy, then when I resume the thread later the recovery is almost instantaneous. That is, the recovery FINISHED event is fired less than a second later. We can discuss further, but I'll flag this issue as RESOLVED/WORKSFORME because the original issue as described in the summary has been addressed without changing the recovery logic. Closing. |
Build Identifier: 4.0.0.v20110303-0827 When using ReconnectingSessionConfiguration, a NullPointerException is thrown when committing a transaction after the connection was interrupted. For this, I have attached a JUnit test. The test "testConnectionLostWithRecovery" succeeds. But the test "testTransactionWithLostConnectionAndRecovery" fails: org.eclipse.emf.cdo.util.CommitException: java.lang.NullPointerException at org.eclipse.emf.internal.cdo.transaction.CDOTransactionImpl.commit(CDOTransactionImpl.java:1067) at org.eclipse.emf.internal.cdo.transaction.CDOTransactionImpl.commit(CDOTransactionImpl.java:1073) at ITCdoConnection.testTransactionWithLostConnectionAndRecovery ... Caused by: java.lang.NullPointerException at org.eclipse.emf.internal.cdo.session.DelegatingSessionProtocol.getSession(DelegatingSessionProtocol.java:117) at org.eclipse.emf.internal.cdo.session.DelegatingSessionProtocol.handleException(DelegatingSessionProtocol.java:799) at org.eclipse.emf.internal.cdo.session.DelegatingSessionProtocol.commitTransaction(DelegatingSessionProtocol.java:270) at org.eclipse.emf.internal.cdo.transaction.CDOSingleTransactionStrategyImpl.commit(CDOSingleTransactionStrategyImpl.java:80) at org.eclipse.emf.internal.cdo.transaction.CDOTransactionImpl.commit(CDOTransactionImpl.java:1053) ... 40 more The NPE comes from DelegatingSessionProtocol#commitTransaction. Obviously, the call ReconnectingCDOSessionImpl(CDOSessionImpl).setSessionProtocol(CDOSessionProtocol) line: 301 is done too late (runs in a different thread). When I set a breakpoint, the other thread has the chance to set the delegate and my test succeeds. Reproducible: Always Steps to Reproduce: 1. Import the attached project 2. Run ITCdoConnection as JUnit test 3. All tests should succeed except from testTransactionWithLostConnectionAndRecovery