Feature #828
closed
Added by Sage Weil about 13 years ago.
Updated about 13 years ago.
Description
The google-perftools cpu profiler isn't collecting any samples. I think it's because of the signal masks on our threads. To test, use the profiler branch, start the mds, watch ceph -w for messages, and
ceph mds tell 0 cpu_profiler_start /tmp/foo
ceph mds tell 0 cpu_profiler_status
Currently you see 0 samples, which you should be on the order of 100 HZ.
- Assignee set to Greg Farnum
- Assignee changed from Greg Farnum to Colin McCabe
This still isn't working for me on current master (with and without the last SIGPIPE patch).
2011-03-03 15:10:16.937699 log 2011-03-03 15:10:12.136513 mds0 10.0.1.252:6803/29611 4 : [INF] cpu_profiler: starting logging to prof.1, r=0
2011-03-03 15:10:18.894852 pg v14: 24 pgs: 24 active+clean+degraded; 2505 KB data, 259 GB used, 157 GB / 439 GB avail; 21/42 degraded (50.000%)
2011-03-03 15:10:22.018129 log 2011-03-03 15:10:19.273968 mds0 10.0.1.252:6803/29611 5 : [INF] cpu_profiler is not enabled
2011-03-03 15:10:22.018129 log 2011-03-03 15:10:19.273992 mds0 10.0.1.252:6803/29611 6 : [INF] cpu_profiler enabled start_time 1299193799 profile_name prof.1 samples 0
is what i get after
fatty:src 03:10 PM $ ./ceph mds tell 0 cpu_profiler start prof.1
2011-03-03 15:10:12.135335 mon <- [mds,tell,0,cpu_profiler,start,prof.1]
2011-03-03 15:10:12.136218 mon2 -> 'ok' (0)
fatty:src 03:10 PM $ ./ceph mds tell 0 cpu_profiler status
2011-03-03 15:10:19.272683 mon <- [mds,tell,0,cpu_profiler,status]
2011-03-03 15:10:19.273654 mon1 -> 'ok' (0)
- Translation missing: en.field_position set to 2
- Status changed from New to Resolved
- Translation missing: en.field_story_points set to 1
- Translation missing: en.field_position deleted (
2)
- Translation missing: en.field_position set to 2
- Translation missing: en.field_story_points changed from 1 to 3
- Translation missing: en.field_position deleted (
4)
- Translation missing: en.field_position set to 5
Also available in: Atom
PDF