Memory-infra heap profiler: WorkerPool threads should be grouped together in the stack |
||||
Issue descriptionBackground context: go/memory-infra Just realized a drawback of having Thread names as roots of the heap profiler stack: tasks from WorkerPool gets fragmented into each individual thread of the pool. See attached screenshot. That case, specifically, is not ideal. There is little value in separating the threads of the workerpool (while I share the usefulness of separating the threads in general). Can we do something special for WorkerPool and just call it [WorkerPool] and not [WorkerPool/1234] ?
,
Feb 10 2017
I can remove the numbers from thread names. Will give it a try
,
Feb 10 2017
Yes, I remember someone complained about it too. I really want to fix how we deal with thread names. Especially the bit where we go do prctl(PR_GET_NAME) and end up with truncated name on Android.
,
Feb 10 2017
https://codereview.chromium.org/2683913002/ will fix some thread name issues and start profiling earlier. So, prctl will be used rarely. Lets try to debug more after that is fixed.
,
Feb 10 2017
Yeah sure. It's just the whole thread name thing is very hacky now. I would prefer something cleaner, which would require more cooperation from ThreadNameManager, pools, etc.
,
Mar 3 2017
,
Aug 1
|
||||
►
Sign in to add a comment |
||||
Comment 1 by primiano@chromium.org
, Feb 10 2017