- Wednesday, June 4, 2014

openafs@conference.openafs.org

Wednesday, June 4, 2014< ^ >

Room Configuration

Room Occupants

GMT+0
[09:56:07] shadow@gmail.com/barnowlC1E71073 joins the room
[10:08:31] shadow@gmail.com/barnowlC1E71073 leaves the room
[10:10:29] shadow@gmail.com/barnowlC1E71073 joins the room
[10:11:33] shadow@gmail.com/barnowlC1E71073 leaves the room
[13:37:22] wiesand joins the room
[14:04:21] kaduk joins the room
[15:49:13] gorgo joins the room
[16:03:12] wiesand leaves the room
[16:46:41] meffie joins the room
[19:17:52] <kaduk> Any tips for debugging if I think I'm leaking conns and/or calls?
[19:23:03] <meffie> maybe commit e453f49b ?
[19:24:05] <meffie> or something similar
[19:25:00] <kaduk> (I'm looking at the vlserver in particular.)
[19:32:38] <kaduk> AFAICT rxi_FreeCall() is never being called.
[19:38:18] <kaduk> And in fact, where call->refCount is compared with zero to conditionalize a rxi_FreeCall call, the refCount is regularly negative.
[19:55:16] <kaduk> git log -S causes me to look funnily at the direction of 20034a8
[19:55:16] gorgo leaves the room
[19:58:13] <kaduk> In particular, the parts where the return value of rxevent_Cancel() is not checked.
[20:00:40] <Jeffrey Altman> rxevent_Cancel never returns leaving the event object in place.  The return value is used to indicate whether the event was canceled before it fired.
[20:02:15] <kaduk> Okay.
[20:04:39] <kaduk> I wonder whether I want to try to reproduce in stock master (with my rxgk patches reverted) before staring at the code.
[20:05:08] <Jeffrey Altman> refcnt debugging sucks.   you need to instrument all of the increments/decrements, generate a log and assert when it goes negative, then walk the log to analyze each increment/decrement to see where something is decremented twice or not incremented.    refcnt leaks are in some ways harder but figure out what a reasonable number is for the particular reference and then assert when that is exceeded.
[20:05:23] <Jeffrey Altman> probably
[20:06:02] <kaduk> yeah, I guess it is not too hard to rebuild.
[20:13:10] gorgo joins the room
[20:29:25] meffie leaves the room
[20:36:18] <kaduk> Still seeing negative refcounts at 0abf5fcb.
[20:38:39] <kaduk> (Indeed, I never saw a positive refcount at rx.c:6417, ever.)
[20:39:31] <Jeffrey Altman> clearly you are looking for a missing reference increment
[20:39:39] <kaduk> Yup.
[21:19:21] <kaduk> Hah.  The first entry in the debug log is CALL_RELE() on a call with refcount 0.
[21:31:26] <jhutz@jis.mit.edu/owl> You shouldn't do that.
[21:33:42] <kaduk> One would think so...
[21:38:45] kaduk leaves the room