perf_output in case of SMP: sorting based on timestamps
Open
- Dominant language
- C
- Stars
- 22.7k
- Forks
- 4.1k
- Avg merge
- 10d 4h
- Merged PRs (30d)
- 3
Description
In an SMP environment, while polling the perf rings in the order of CPU's (readers) there is a chance that we might reorder the events.
For Eg. CPU0: RING-A, CPU1: RING-B
While processing A.1 event (callback in progress), we might get B.1 and A.2 events, but BCC reports A.2 and B.1 to the callback (as it processes events based on order of CPU's).
Should we add a builtin feature to BCC to sort the events based on timestamps (or) does every user of BPF_PERF_OUTPUT should take care of this?
This is a problem if the BPF is used for doing state checks, ordering is a must for those apps.
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.