python / python/cpython

Add runtime control for `profiling.sampling`

未關閉
#145,411 1 則留言 1 個 reaction 已指派 0 人 在 GitHub 檢視

還沒有人認領這個 Issue。

stdlib topic-profiling type-feature
主要語言
Python
星號
77.2k
分支
36k
PR 合併指標
PR 指標待擷取

描述

Feature or enhancement

Proposal:

The idea is to add an optional control channel (socket? named pipe?), so another process can start, stop etc. profiling without killing/restarting the profiler completely. Right now, the only option is to interact with it via TUI mode or just stop it via Ctrl-C.

While working on https://github.com/psf/pyperf/pull/230 I ran into the following problem: some parts (eg: warming up) of benchmarking should not be profiled, as they do not add any value. My only option was sending SIGINT signal.

Generally, the current flow is nice for headful sessions but it could be improved for (especially long-running) headless sessions.

perf record --control does it in a similar way:

       --control=fifo:ctl-fifo[,ack-fifo], --control=fd:ctl-fd[,ack-fd]
           ctl-fifo / ack-fifo are opened and used as ctl-fd / ack-fd as
           follows. Listen on ctl-fd descriptor for command to control
           measurement.

       Available commands:

       •   enable : enable events
       •   disable : disable events
       •   enable name : enable event name
       •   disable name : disable event name
       •   snapshot : AUX area tracing snapshot).
       •   stop : stop perf record
       •   ping : ping
       •   'evlist [-v|-g|-F] : display all events
               -F  Show just the sample frequency used for each event.
               -v  Show all fields.
               -g  Show event group information.

Specifically:

  • We could start with just enable and disable initially. snapshot next but it'd be lovely to dump without loosing accumulated data. Or maybe even set_rate.
  • Possible transport layers could be Unix sockets (works great on Linux) or named pipe. One option might be portable, another not?
  • Possble new arguments like --control unix:/tmp/tachyon-ctl.sock or --control fifo:/tmp/tachyon-ctl, similar to what perf record --control does.

It could be also ommited from the documentation initially, so we don't limit ourselves too early.

Has this already been discussed elsewhere?

No response given

Links to previous discussion of this feature:

No response

Linked PRs
  • gh-149958

貢獻指南

開啟貢獻指南

從這裡開始

  1. 先讀完整個 Issue,再讀專案的貢獻指南。
  2. 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
  3. Fork 儲存庫,在一個分支上完成修改。
  4. 送出 Pull Request,並在描述裡引用這個 Issue 編號。

研究方向

首先追蹤現有的 profiling.sampling 流程,以及 TUI 模式和 SIGINT 目前如何控制 profiling。將提議的啟用/停用控制流程與 perf record --control 進行比較,包括 Unix socket 或 named pipe 選項以及提議的 --control 引數。完成表示定義一個執行時控制介面,支援無需重新啟動 profiler 的無頭 profiling。

由索引模型根據 Issue 內容生成。

評估

技術堆疊
python
領域
performance, tooling
Issue 類型
功能
難度
5/5
預估耗時
一週以上
活躍度
停滯
描述清晰度
需要釐清
新手友好度
20/100

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。