ci: add an opt-in cargo --timings run to decide the sccache question (#5545)

sccache can only cache compilation units whose --emit includes link. That
covers workspace rlibs and nothing else: clippy is metadata-only, and the ~100
test binaries, the rustfs bin and every build script invoke the system linker.
So the headline claim that started this — s3select-query costing 17m13s inside
a 19m32s nextest build — does not by itself justify sccache, because cargo
prints "Compiling" when a crate starts and never when it finishes: that number
is the whole compilation tail, not one crate.

Adding --timings behind a workflow_dispatch input on Cache Warm answers it with
data instead. Read two shares off the report: workspace lib codegen against the
whole build, and s3select-query's own rlib. The plan in rustfs/backlog#1601
adopts sccache only above 50% and 25%; if linking dominates instead, the answer
is mold/lld plus split-debuginfo, which is precisely the region sccache cannot
reach.

Off by default: it doubles the ci-dev build, and this only needs running once
per question. Nothing about the normal warm path changes.

Refs: rustfs/backlog#1598, rustfs/backlog#1601
This commit is contained in:
Zhengchao An
2026-08-01 11:36:48 +08:00
committed by GitHub
parent 428fde069d
commit 48c2fcb62b
+35
View File
@@ -65,6 +65,14 @@ on:
- ".github/workflows/docker.yml"
- ".github/workflows/audit.yml"
workflow_dispatch:
inputs:
emit_timings:
description: >-
Also emit cargo --timings for the ci-dev build and upload it. Used to
decide whether sccache is worth adopting (rustfs/backlog#1601 gate).
required: false
default: false
type: boolean
permissions:
contents: read
@@ -103,6 +111,33 @@ jobs:
# e2e_test, which test-and-lint's own run excludes. The second build adds
# the e2e-test-hooks feature resolution that build-rustfs-debug-binary uses
# and that no lint lane enables.
# rustfs/backlog#1601 gate. sccache can only cache compilation units whose
# --emit includes link, so it covers workspace rlibs and nothing else:
# clippy is metadata-only, and the ~100 test binaries, the rustfs bin and
# every build script invoke the system linker. Before spending a bucket,
# credentials and a supply-chain boundary on it, measure how much of the
# build is actually rlib codegen.
#
# Read from the report: workspace lib codegen as a share of the build, and
# s3select-query's own rlib as a share. The plan adopts sccache only above
# 50% and 25% respectively; if linking dominates instead, the answer is
# mold/lld plus split-debuginfo, which is exactly the part sccache cannot
# touch. Off by default — this doubles the ci-dev build.
- name: Build ci-dev superset (with --timings)
if: inputs.emit_timings
env:
CARGO_BUILD_JOBS: "2"
run: cargo build --workspace --all-targets --timings
- name: Upload cargo timings report
if: inputs.emit_timings
uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
with:
name: cargo-timings-ci-dev
path: target/cargo-timings/
retention-days: 30
if-no-files-found: error
- name: Build ci-dev superset
env:
# Same limit ci.yml puts on its nextest step: this builds the same