PacketFence
Bug Tracking System
Main | My View | View Issues | Change Log | Roadmap
  

View Issue Details [ Jump to Notes ] [ Issue History ] [ Print ]
IDProjectCategoryView StatusDate SubmittedLast Update
0001464PacketFencecorepublic2012-06-11 12:272012-08-06 13:24
Reporterobilodeau 
Assigned Toobilodeau 
PriorityhighSeverityminorReproducibilitysometimes
StatusresolvedResolutionfixed 
PlatformOSOS Version
Product Version 
Target Version3.4.0Fixed in Version3.4.0 
Summary0001464: process management issues (pid misuse in listing and killing daemons)
DescriptionProcesses started with perl ... instead of specifying directly the executable name wouldn't be listed in status (which relies on pidof) and wouldn't be killed (which relies on pkill).

It's not 100% straightforward to reproduce but anything started under the debugger (perl -d ...) is simply not properly managed (killed or listed in status) which definitely causes problems when trying to investigate odd crashes.

Also, I think this could be related to the double processes started when pfsetvlan was restarted from the Web admin and could also explain the core dumps. I remember seeing that perl was explicitly mentionned in the `ps -ef` output.

Attempted workarounds:
- Using `pkill -f` matches the entire command line of the process and fixed the killing problems. However it doesn't work on listing statuses which relies on pidof.
- Setting $0 works for pidof but sometimes didn't work with pkill but I'm not too sure now (done that last Friday before leaving..)

We could either:
- implement both changes (re-validate in lab)
- implement -f pkill fix and use `pgrep -f` instead of pidof for status (warning: output format differs slightly)
TagsNo tags attached.
fixed in git revision375c5ab
fixed in mtn revision
Attached Files

- Relationships
related to 0001478closedobilodeau pfdhcplistener not showing full start line under ps 

-  Notes
(0002756)
obilodeau (reporter)
2012-06-12 14:37

 Re-tested the stuff properly and we can avoid the risky `pkill -f` if we set $0 ($PROGRAM_NAME) to basename($0) in all daemons. I'm taking that route.
(0002757)
obilodeau (reporter)
2012-06-12 14:48

 fix available in branch fix/process-management-1464. Will hit stable before the next stable release.
(0002772)
obilodeau (reporter)
2012-06-14 12:16

 fix released in 3.4.0 yesterday
(0002848)
fgaudreault (viewer)
2012-07-19 11:18

 Regression. We lost the arguments in the ps line.
(0002872)
obilodeau (reporter)
2012-08-06 13:24

 I marked this one as resolved as the underlying issue has been resolved. I'll update the related 0001478 in a second with a target and proper priority.

- Issue History
Date Modified Username Field Change
2012-06-11 12:27 obilodeau New Issue
2012-06-12 14:36 obilodeau Status new => assigned
2012-06-12 14:36 obilodeau Assigned To => obilodeau
2012-06-12 14:37 obilodeau Note Added: 0002756
2012-06-12 14:48 obilodeau git revision => 375c5ab
2012-06-12 14:48 obilodeau Note Added: 0002757
2012-06-12 14:48 obilodeau Status assigned => resolved
2012-06-12 14:48 obilodeau Fixed in Version => +1
2012-06-12 14:48 obilodeau Resolution open => fixed
2012-06-14 12:15 obilodeau Target Version => 3.4.0
2012-06-14 12:15 obilodeau Fixed in Version +1 => 3.4.0
2012-06-14 12:16 obilodeau Note Added: 0002772
2012-06-14 12:16 obilodeau Status resolved => closed
2012-07-19 11:18 fgaudreault Note Added: 0002848
2012-07-19 11:18 fgaudreault Status closed => feedback
2012-07-19 11:18 fgaudreault Resolution fixed => reopened
2012-07-19 11:19 fgaudreault Relationship added related to 0001478
2012-08-06 13:24 obilodeau Note Added: 0002872
2012-08-06 13:24 obilodeau Status feedback => resolved
2012-08-06 13:24 obilodeau Resolution reopened => fixed

Copyright © 2000 - 2012 MantisBT Group
Powered by Mantis Bugtracker