Bug 200593 - Delay accounting in cgroups
Summary: Delay accounting in cgroups
Status: NEW
Alias: None
Product: Process Management
Classification: Unclassified
Component: Other (show other bugs)
Hardware: All Linux
: P1 normal
Assignee: process_other
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2018-07-18 22:03 UTC by Ivan Babrou
Modified: 2018-12-12 09:26 UTC (History)
1 user (show)

See Also:
Kernel Version: 4.14.56
Subsystem:
Regression: No
Bisected commit-id:


Attachments

Description Ivan Babrou 2018-07-18 22:03:14 UTC
Currently it's possible to get task delay accounting:

* https://github.com/torvalds/linux/blob/master/Documentation/accounting/delay-accounting.txt

It's possible with either /proc/<pid>/schedstat or with getdelays program:

* https://github.com/torvalds/linux/blob/master/tools/accounting/getdelays.c

While cgroups are great for getting cpu, memory and block IO usage, they do not have delay accounting yet. Tools like cAdvisor have to iterate individual processes in cgroups to accumulate totals. Dying processes reset counters and skew the resulting data:

* https://github.com/google/cadvisor/pull/1872#issuecomment-402326304

I suggest to expose delay accounting the same way cpu stats are exposed in cgroups: as aggregate counters.

Note You need to log in before you can comment on or make changes to this bug.