Community
Participate
Working Groups
build 3.2RC7 When i launch eclipse 3.2RC7 with the monster workspace (more than 1800 projects) the Java Tooling initialization take a long time, around 1 minutes. Sometimes the initialization is slower and take more than 5 minutes I will attach some thread dumps of the 5 minutes case.
Created attachment 44145 [details] thread dump This zip file contains 3 thread dump stack01 is around 2 minutes after the beginning of the java tooling initialization stack02 is around 40 secondes later stack03 is around 40 secondes later
Created attachment 44256 [details] CPU Snapshots This is CPU snapshots made with Yourkit monster-slow-2006-06-13.cpu is a snapshot of a slow initialization with CompletionEngine.java opened monster-noCU-2006-06-13.cpu is a snapshot of a less slow initialization without opened compilation unit. I am not sure that the opened compilation unit is really the cause of the slowness.
Hi, I'm on eclipse 3.2 released, and the Java Tooling initialization seem blocked at 59% during 5 minutes before finalized. I have no file opened, only 4 projects opened on 35 (less than 1800). But all my project have been created on a 3.1 eclipse version. I have the problem only with one PC (512Mo Ram, XP Family), I have another PC (1Go ram XP Pro) without problem (same workspace). Perhaps this note would help to resolve the bug. (In reply to comment #2) > Created an attachment (id=44256) [edit] > CPU Snapshots > This is CPU snapshots made with Yourkit > monster-slow-2006-06-13.cpu is a snapshot of a slow initialization with > CompletionEngine.java opened > monster-noCU-2006-06-13.cpu is a snapshot of a less slow initialization without > opened compilation unit. > I am not sure that the opened compilation unit is really the cause of the > slowness.
see also bugs 161704 and 171653, related to the InitializeAfterLoad operation. Since the above bugs have been integrated, may you please have another go and try to reproduce this issue with a build > 3.3 M6.
I looked closely at the snapshots and here are what can we say about this performance issue: 1) monster-noCU-2006-06-13.cpu: total time=30,243ms + 27,579ms (91.2%) in JavaCore.initializeAfterLoad + 22,932ms (83.2%) in BasicSearchEngine.createWorkspaceScope() Most of this is time (72.6%) is spent in org.eclipse.pde.internal.core (problem tracked down by bug 179685): + 5,979ms (18,406-12,427) (26.1%) in RequiredPluginsInitializer.initialize(IPath, IJavaProject) + 10,657ms (46.5%) in RequiredPluginsClasspathContainer.getClasspathEntries() + 4,516ms (19.7%) in SearchEngine.searchAllTypeNames This point has been addressed in 3.3 M6 (see bug 171653) 2) monster-slow-2006-06-13.cpu: total time=388,126ms + 236,069ms (60.8%) in Main for JDT/UI views refresh + 127,121ms (32.6%) in Worker.run() dispatched as follows: + 68,568ms (53,9%) in JavaCore.initializeAfterLoad + 66,244ms (96.6%) in BasicSearchEngine.createWorkspaceScope() Same conclusion than in snapshot (1) + 32,817 (25.8%) in JavaBuilder.build Note that ratio to global time is only 8.5% but it seems interesting to investigate if something could not be done in this area. Typically, why JavaProject.getExpandedClasspath is called twice with different params but consumes the same time approximatively... + 18,776ms (14.8%) in OverrideIndicatorLabelDecorator.decorate + 18,336ms (97.8%) in SourceType.newSupertypeHiearchy + 16,743ms (91.2%) in SearchEngine.createWorkspaceScope Same conclusion than in snapshot (1) + 1,582ms (8.4%) in TypeHiearchy.compute Ratio to global time is only less than 1% => nothing to gain here... + 24,334ms (6.3%) in reconciling strategy This time show be improved with fix for bug 179684
Martin, what's your mind about the line in comment 5: + 236,069ms (60.8%) in Main for JDT/UI views refresh
Most of the UI time should be gone now (since 3.3 M3) with the fix for bug 157639. It would be good to run the same test again on a current version.
Kent, what's your mind on comment 5 JavaBuilder point?
Do we have more up to date numbers which have the same issues? I thought the builds on startup were removed with Jerome's changes that no longer 'touch' projects.
(In reply to comment #9) > Do we have more up to date numbers which have the same issues? > > I thought the builds on startup were removed with Jerome's changes that no > longer 'touch' projects. > Currently not, but David should replay this test with last I-build soon...
I did the test with the same workspace and with two different computers. The results are Pentium M 1.7GHz ram 2.0GB WINXP 3.2 RC7 - 1m45s or 13s 3.3 I20070403-1110 - always 10s Pentium 4 2.0GHz ram 768MB WINXP 3.2 RC7 - always 5m 3.3 I20070403-1110 - always 25s now performances seem good, we could close this bug as WORKSFORME.
agreed
Do you happen to have CPU snapshots again? It's always interesting to have them to get an understanding where time is spent.
Initialize Java Tooling never exceeds 10s while starting monster workspace. Verified for 3.3 M7 using build I20070427-0010.
