Community
Participate
Working Groups
Created attachment 122285 [details] log from running jdt core model tests The jdt.core model tests ran out of heap space in the N20090111-2000 build. Current the settings in the eclipse.ini are -Xms40m -Xmx256m I reran the tests by themselves and the tests passed. I'm just wondering if the values in the eclipse.ini should be increased. Have you seen this in your testing? I'll attach some of the log.
On the contrary, we recently reduced the size taken by the Java model. So if the test ran out of heap space, it means that another component took more heap space. -Xmx256m is enough to run the whole IDE, so I don't think this setting should be changed. Did subsequent builds have this problem? Or was it a transient problem with this build?
It was transient problem. I reran the jdt.core model suite and the problem didn't occur. Olivier suggested that the problem may have resulted from a previous test run. So I reran all the jdt core tests that proceeded this test. The tests ran without running out of heap space. Since we can't reproduce this problem I'll close this as invalid.
Reopen as we hit this issue several times again...
Created attachment 124288 [details] Patch to activate the memory trace This patch will set on the tests memory trace and log results in a file located in /buildtests/${buildID}/eclipse-testing directory of the test machine. With this trace, we should hopefully see in which test suite the memory consumption responsible of the OOME occurs...
(In reply to comment #4) > Created an attachment (id=124288) [details] > Patch to activate the memory trace > Released in HEAD
Hi Frederic, I'm not sure which VM is being used to run these tests but why not set -XX:+HeapDumpOnOutOfMemoryError so a heap dump is genereted when the OOME is triggered. A heap should make it easier to determine what's bloated or leaking.
(In reply to comment #6) > Hi Frederic, > I'm not sure which VM is being used to run these tests but why not set > -XX:+HeapDumpOnOutOfMemoryError so a heap dump is genereted when the OOME is > triggered. > > A heap should make it easier to determine what's bloated or leaking. > Thanks Chris for the tip. I've added this VM argument to our test.xml file...
Frederic I can take a look at the dumps if/when they happen. Feel free to ping me on ST cgrinds@us.ibm.com.
I think we can consider it fixed as the problem hasn't occurred again since blocking bug 269476 has been fixed...
Verified for 3.5M7.
