> On July 5, 2016, 1:22 p.m., Jonathan Hurley wrote: > > I don't see how this actually fixes the problem; importing gc from > > subprocess would be the same as importing it normally, no? There isn't a > > different gc module that subprocess uses. Also, after you patch it, you > > then re-import it - wouldn't that break the existing patching? > > Andrew Onischuk wrote: > "I don't see how this actually fixes the problem" - this was tested on > the instance where I could reproduce this and is a sure fix. > > Jonathan Hurley wrote: > So, let's say that it does fix the issue by always reporting that GC is > enabled. Wouldn't this then cause side effects since Popen is specifically > disabling it for a reason? > > Andrew Onischuk wrote: > popen disables gc anyway no matter if it is enabled or not. > > Andrew Onischuk wrote: > And even with fix it still disables it. The fix the for the race > condition, when it doesn't re-enable it again after the work is done. > > Jonathan Hurley wrote: > OK, so `subprocess` doesn't check before it disables it; I can see that. > However, we're changing the behavior of a function provided by the gc and we > don't know what kinds of issues this could cause on other distributions of > Python. There could be other versions which check this function and/or > turning on the gc when it's already on could cause a problem.
We support python2.6.* and python2.7.* Both have the same code there. Any other python disributions are not suported by us. - Andrew ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/49535/#review140761 ----------------------------------------------------------- On July 4, 2016, 8:45 a.m., Andrew Onischuk wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/49535/ > ----------------------------------------------------------- > > (Updated July 4, 2016, 8:45 a.m.) > > > Review request for Ambari and Dmitro Lisnichenko. > > > Bugs: AMBARI-17539 > https://issues.apache.org/jira/browse/AMBARI-17539 > > > Repository: ambari > > > Description > ------- > > *Reason of memory leak:* > Race condition in subprocess python module. > Due to this race condition at some unlucky cases python garbage collection > was disabled. > This usually happened when running alerts, as a bunch of our alerts run shell > commands and they do it in a different threads. > > *Fix for the issue:* > Synchronizing subprocess is not the best option. Since some people can still > use it without synchronization not knowing about the issue. > Also synchronizing will provide some unnecessary slowdown. So for this issue > the proposed fix is to monkey patch subprocess.gc.isenabled. > > > Diffs > ----- > > ambari-agent/src/main/python/ambari_agent/main.py 4db89f8 > > Diff: https://reviews.apache.org/r/49535/diff/ > > > Testing > ------- > > mvn clean test > > > Thanks, > > Andrew Onischuk > >
