chenBright opened a new pull request, #3541:
URL: https://github.com/apache/brpc/pull/3541
### What problem does this PR solve?
Issue Number: resolve
Problem Summary:
`TaskGroup` tracks per-group CPU time as a 128-bit stat (a packed
`last_run_ns` + task-type word and a `cumulated_cputime_ns` word) that a
worker updates while other threads (e.g. bvar sampling) read it
concurrently. This was implemented with `AtomicInteger128`, whose 128-bit
"atomic" load/store had no portable, guaranteed lock-free backing:
- On x86 it relied on wide SSE/AVX aligned loads/stores being atomic. This
is
not guaranteed by the ISA, because Intel and AMD do not officially
promise 128-bit
AVX load/store atomicity. It merely happens to hold on current
microarchitectures
(Skylake, Zen 2). Depending on it is relying on unspecified hardware
behavior.
- On platforms without such a wide atomic it fell back to a mutex, which
serializes
readers and blocks them behind the writer. Exactly what a
consistent-snapshot
read is meant to avoid.
### What is changed and the side effects?
Changed:
- Add `butil::Seqlock`, a general sequence lock providing lock-free,
consistent snapshot
reads around a caller-owned atomic payload.
- Replace `TaskGroup::AtomicInteger128` with an `AtomicCPUTimeStat` backed
by
`butil::Seqlock<>` over a `CPUTimeStat` payload accessed via relaxed
atomics.
Side effects:
- Performance effects:
- Breaking backward compatibility:
---
### Check List:
- Please make sure your changes are compilable.
- When providing us with a new feature, it is best to add related tests.
- Please follow [Contributor Covenant Code of
Conduct](https://github.com/apache/brpc/blob/master/CODE_OF_CONDUCT.md).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
