MantisBT - Rosetta
View Issue Details
IDProjectCategoryView StatusDate SubmittedLast Update
0000026Rosetta[All Projects] Crashpublic2012-06-25 13:232013-03-01 09:57
ReporterAndrewLeaverFay 
Assigned ToSergey 
PrioritynormalSeveritymajorReproducibilityrandom
StatusresolvedResolutionfixed 
PlatformAll platformsOSAnyOS VersionAny
Product VersionTrunk 
Fixed in VersionTrunk 
Application(s) Affectedall
Command Line Usedpython integration.py -j24
Developer Options
Fixed in SVN Version51417
Summary0000026: integration.py hangs w/ one job that never starts
Descriptionpython integration.py -j 12 will finish 168 of the 169 tests, while the last one will never get started. Sometimes two tests will hang without ever getting started. It's likely a race condition in the integration test Queue class.

It's never the same test that's the one that fails to start, so this problem is not likely to be a problem with Rosetta.
Steps To ReproduceHappens more often when -j is high.
Additional Information
TagsNo tags attached.
Relationships
parent of 0000185resolved smlewis Is integration3.py deprecated? 
related to 0000033resolved Sergey integration tests spawned subprocesses are not killed properly 
Attached Files
Issue History
Date ModifiedUsernameFieldChange
2012-06-25 13:23AndrewLeaverFayNew Issue
2012-06-26 16:39AndrewLeaverFayAssigned To => AndrewLeaverFay
2012-06-26 16:39AndrewLeaverFayStatusnew => assigned
2012-06-26 16:41AndrewLeaverFayFixed in SVN Version => 49614
2012-06-26 16:41AndrewLeaverFayNote Added: 0000014
2012-06-26 16:41AndrewLeaverFayStatusassigned => resolved
2012-06-26 16:41AndrewLeaverFayFixed in Version => Trunk
2012-06-26 16:41AndrewLeaverFayResolutionopen => fixed
2012-08-02 12:58smlewisNote Added: 0000044
2012-08-02 12:58smlewisStatusresolved => feedback
2012-08-02 12:58smlewisResolutionfixed => reopened
2012-08-08 09:20smlewisAssigned ToAndrewLeaverFay => Sergey
2012-08-08 09:20smlewisStatusfeedback => assigned
2012-09-07 19:49SergeyNote Added: 0000093
2012-09-10 09:12smlewisNote Added: 0000095
2012-09-28 13:50SergeyFixed in SVN Version49614 => 51417
2012-09-28 13:50SergeyNote Added: 0000111
2012-09-28 13:50SergeyStatusassigned => resolved
2012-09-28 13:50SergeyResolutionreopened => fixed
2012-10-02 13:34SergeyRelationship addedrelated to 0000033
2013-03-01 09:57smlewisRelationship addedparent of 0000185

Notes
(0000014)
AndrewLeaverFay   
2012-06-26 16:41   
The subprocess module in python 2.x deadlocks on calls to Popen. This is fixed in python3.

I'm adding a new integration test launch script, integration3.py, for use in python3.
(0000044)
smlewis   
2012-08-02 12:58   
Sergey (?) stated at the dev meeting that this can be fixed in Python 2.? by using fork instead of subprocess, without the inconvenient need for Python 3.
(0000093)
Sergey   
2012-09-07 19:49   
Committed a path at r51043. Please use new command line option '--fork' when running integration.py and see if it helps. This commit also should improve cntr-c handling.
(0000095)
smlewis   
2012-09-10 09:12   
I tried testing post 51043, but couldn't get the 'bad behavior' to reproduce reliably (not that it ever did) - I was going to try to do a ten-runs-of-each comparison between integration.py, integration.py --fork, and integration3.py, but the former only failed the first time (out of ten tries) and so I didn't bother with the latter. Maybe it was non-normal load conditions (Sunday afternoon, nobody on the itest machine...)
(0000111)
Sergey   
2012-09-28 13:50   
I rewrote integration.py script using Unix 'fork' command to span subprocess. This seems to fix the issue.