Concurrent access fails various ways and performance is bad [moved]

[lvca]

This is Issue 115 moved from a Google Code project.
Added by 2010-09-18T08:56:55.000Z by jani.laa...@gmail.com.
Please review that bug for more context and additional comments, but update this bug.
Closed (Done).

Original labels: Type-Defect, Priority-Medium, v0.9.23

Original description

I am having serious trouble accessing local database with multiple concurrent threads. Suspecting there exists serious thread safety issues that need to be fixed asap.

Using trunk @ 1414

Here are some of the issues I've seen so far..


1) Transactions are not consistent.

I have a transaction in which I save 1000 objects of class A and 1000 objects of class B but after succesful saves and commit I get different number of type A and B objects.

Here's an example run with my example code using THREAD_COUNT = 2:
Thread[main,5,main]: Creating new database
Thread[Thread-2,5,main]: THREAD START
Thread[Thread-4,5,main]: THREAD START
Thread[Thread-2,5,main]: internal=22, index=0, default=0, orole=3, ouser=3, person=1204, company=1000, 
Thread[Thread-2,5,main]: THREAD END
Thread[Thread-4,5,main]: COMMIT FAILED 100 TIMES
Thread[Thread-4,5,main]: internal=22, index=0, default=0, orole=3, ouser=3, person=1204, company=1000, 
Thread[Thread-4,5,main]: THREAD END
Thread[main,5,main]: All db write test threads finished
Thread[main,5,main]: internal=22, index=0, default=0, orole=3, ouser=3, person=1204, company=1000, 

NOTE: Issue is that it's impossible for person count to be something different than company count. Even if commit fails, it stores 204(?) person objects but not a single company objects.


2) Acquiring connection from ODatabaseObjectPool.global().acquire throws four different exceptions

Acquire throws "randomly" these:
com.orientechnologies.orient.core.exception.OStorageException: Error in creation of new cluster 'company' of type: LOGICAL
or
com.orientechnologies.common.concur.lock.OLockException: Error on creation of the new resource in the pool
or
com.orientechnologies.orient.core.exception.OCommandExecutionException: Error on execution of command: OSQLQuery [text=select from OUser where name = 'admin']
Caused by: java.lang.NumberFormatException: For input string: "2ORole"
or
com.orientechnologies.common.concur.lock.OLockException: Can't acquire lock on requested resource: local:/tmp/orient-database-test/foo

3) ODatabaseObjectTx.save is erratic throws two different exceptions

save sometimes throws these:
com.orientechnologies.orient.core.exception.OStorageException: Error in creation of new cluster 'company' of type: LOGICAL
or
java.util.ConcurrentModificationException


4) "Recovery of pending transactions" during connection acquire is logged even though I'm sure I have closed all resources properly and even rollbacked failed transactions

[OTxSegment] Started the recovering of pending transactions after a brute shutdown. Found 4736 entry logs. Scanning...


5) Performance is very bad (hangs?) during concurrent database access

I have ten concurrent threads simply storing 1000 class A and 1000 class B objects and then performing simple count query. I cannot find the way to perform this reliably. No matter which TXTYPE I choose and try to retry failed commits.  When I perform tiny Thread.sleep after failed commit, it seems my program won't finish ever.

In any case having commit to throw because db is "busy" is quite akward resolution, I cannot seen to find out the right way to cope this.



I've attached my source code here. Using Xmx1024M and Xms1024M. It's simple and short but a bit hard to read on some places because I've tried to do the best to housekeep everything right and be gentle when errors occur.


Best of luck resolving these issues, I truly hope this product matures quickly, because the values in OrientDB match my need brilliantly. Thank you for the effort so far!