161525 - Hang during download

Status: RESOLVED WORKSFORME

(Camino Graveyard :: Downloading, defect)

Description

[Simon Fraser [no longer active]]

I'm seeing hangs now when doing FTP downloads, which I haven't seen before. I've
seen them on a dual CPU machine; haven't tried to repro on a single CPU machine.
They've been happening for the last couple of days.

Comment 1

[Simon Fraser [no longer active]]

Attachment: Sampler report from hung chimera

Comment 2

[Simon Fraser [no longer active]]

Blocker. I can't download new builds.

Comment 3

[Brian Ryner (not reading)]

wtc, I'm guessing from the sampler report that this is some kind of pthread
deadlock, and it coincidentally happened when I checked in the patch from bug
153525 for building on Jaguar.  Any ideas?

Comment 4

[Wan-Teh Chang]

bryner: I have no idea.  Thread_0's stack in the sampler
report looks incorrect.

Comment 5

[saari (gone)]

->bryner as the current best guess

Comment 6

[Brian Ryner (not reading)]

Adjusting summary, as I just reproduced this on a single-CPU machine.

wtc: It's unclear to me exactly what the implications are of #defining
PT_NOSIGTIMEDWAIT.  It certainly causes different code to be hit in ptthread.c.
 Could this cause some kind of thrashing when switching threads, or anything
like that?

Comment 7

[Brian Ryner (not reading)]

I was also able to reproduce this problem with my original patch from bug
153525, where there should be no change at all for NSPR on OS 10.1.  This leads
me to think that the problem existed before these changes landed.  I'll do some
more testing on older builds.

Comment 8

[Greg K.]

FWIW, WorksForMe using Chimera/2002080605.

Comment 9

[sairuh (rarely reading bugmail)]

is this still happening for folx?

Comment 10

[Chris Petersen]

WFM too. I'm using a Power Mac Dual 800 mhz and haven't seen this problem in
latest builds.

Comment 11

[Simon Fraser [no longer active]]

I haven't seen this for ages. WFM?

Comment 12

[Derek Arndt]

Hum...I can't seem to reproduce this on some FTP sites I found linked off
FreeBSD.org.  I'm not having issues...

Comment 13

[Winnie Lam]

WFM by consensus

Comment 14

[Simon Fraser [no longer active]]

This has come back to bite us in the ass.

Comment 15

[Simon Fraser [no longer active]]

*** Bug 184934 has been marked as a duplicate of this bug. ***

Comment 16

[Simon Fraser [no longer active]]

It's either deadlocking, or iterating in nsPipe::GetReadSegment:

http://lxr.mozilla.org/mozilla1.0/source/xpcom/io/nsPipe2.cpp#243

It doesn't look like deadlock with another thread; the socket transport thread
is blocked in poll().

Comment 17

[brian]

I can consistently repro by trying to download the Google API's:

http://www.google.com/apis/download.html

Don't know if this is redirected to an FTP download or not.

Using 1.2.1 on Win2000 -- should this be a separate bug for non-MacOSX issues?

Comment 18

[Simon Fraser [no longer active]]

This bug is specific to Chimera on Mac OS X. Please file a separate bug for your
case.

Comment 19

[Jasper]

I can't reproduce this bug with 2003080802 build.

I'm capable of downloading builds of Camino from the ftp server
(ftp://ftp.mozilla.org/pub/camino/nightly/latest/, Comment #2?) and downloading
the Google api's using the url given in Comment #17 works just fine to. 

Comment 20

[Neil Lee [:neilio]]

WFM also, using the indicated test cases

Comment 21

[Neil Lee [:neilio]]

WFM also, using the indicated test cases: Build ID: 2003081202

Comment 22

[Chris Casciano]

I'm going to resolve WFM (again & with a little hesistation)... 

closest thing to this behavior i've seen... think late spring... downloading
nighlies regularly failed but that was due to issues with dead ftp servers
somewhere in the cluster/round robin/whatever and had different symptoms entirely