Some Eclipse Foundation services are deprecated, or will be soon. Please ensure you've read this important communication.

Bug 343815

Summary:	[performance] Huge regression in FullSourceWorkspaceModelTests#testCloseProjects
Product:	[Eclipse Project] Platform	Reporter:	Satyam Kandula <satyam.kandula>
Component:	Resources	Assignee:	Szymon Brandys <Szymon.Brandys>
Status:	RESOLVED FIXED	QA Contact:
Severity:	major
Priority:	P3	CC:	daniel_megert, jarthana, Olivier_Thomann, srikanth_sankaran, Szymon.Brandys
Version:	3.7	Keywords:	performance
Target Milestone:	3.7 M7
Hardware:	PC
OS:	Windows XP
Whiteboard:
Bug Depends on:
Bug Blocks:	313891

Description Satyam Kandula

2011-04-26 06:23:55 EDT

There is around 300% performance regression in FullSourceWorkspaceModelTests#testCloseProjects. This is visible on both the windows machine and the linux machine. It seems to have got introduced between 21st and 23rd April. 
Unfortunately, the performance results are not generated for these builds - I could see by directly connecting to the database.

Comment 1 Satyam Kandula

2011-04-26 08:37:48 EDT

I don't see this much of regression on my machine by just using Eclipse 3.7M7 and different versions of JDT/Core. There is a regression somewhere else probably

Comment 2 Jay Arthanareeswaran

2011-04-26 12:52:25 EDT

(In reply to comment #1)
> I don't see this much of regression on my machine by just using Eclipse 3.7M7
> and different versions of JDT/Core. There is a regression somewhere else
> probably

I see over 300% regression comparing the I20110424-2000 build with 3.7 M6. In both cases, I used the latest JDT/Core code. So, I confirm what Satyam said. There is a regression but not in JDT/Core. I will do run some more tests and see if I can find the cause.

Comment 3 Jay Arthanareeswaran

2011-04-26 13:00:03 EDT

I just copied Szymon on the bug. For his information, the test that has regression is just trying close an existing project.

Szymon, was there any recent change in that area?

Comment 4 Jay Arthanareeswaran

2011-04-27 02:53:50 EDT

I ran some more tests and now I can confirm that the regression is likely in Core/Resources. This is the exercise I did:

The test, when run with eclipse build I20110424-2000 plus JDT/Core from HEAD does show the regression - the elapsed process time is around 9.29 seconds. However, on the above setup, when I apply the org.eclipse.core.resources from v20110420, I do not see the regression. The elapsed time is only 3.36 seconds. And I do see there were changes to the project after 20110420.

Moving to Core/Resources.

Comment 5 Szymon Brandys

2011-04-27 07:48:45 EDT

The regression is caused by the fix for Bug 149121. We added a sanity check in SaveManager#saveMasterTable that checks if we are not saving wrong tree root sequence number. I could remove the part of the fix causing the regression for M7 and investigate how to improve it during RC1.

Comment 6 Szymon Brandys

2011-04-27 08:53:48 EDT

As said I reverted part of the fix for Bug 149121.

Comment 7 Srikanth Sankaran

2011-04-27 09:57:48 EDT

(In reply to comment #4)
> I ran some more tests and now I can confirm that the regression is likely in
> Core/Resources. This is the exercise I did:

Jay, Thanks a lot for quickly analyzing this and routing it to the
right component in Satyam's absence. Szymon, thanks for the quick
follow up.

Comment 8 Satyam Kandula

2011-05-04 05:10:20 EDT

(In reply to comment #6)
> As said I reverted part of the fix for Bug 149121.
Szymon, Thanks for reverting the fix. Though the regression has come down significantly, there is still 25% regression on windows and 10% regression on linux. Do you see that the retained part of the fix would have caused this?