ci: build linux-aarch64 releases on Ubicloud ARM (#8996)

This commit is contained in:
hector
2026-10-27 21:18:05 +08:00
committed by GitHub
parent a56104206e
commit 59bc6d7c94
2 changed files with 21 additions and 11 deletions
+4 -4
View File
@@ -7,7 +7,7 @@
This repository is public. For a public repository, GitHub's standard hosted runners (`ubuntu-latest`, `windows-latest`, `macos-latest`) cost nothing and have no minute cap, whatever plan the organisation is on. GitHub bills only two things: larger hosted runners, which have custom labels and are not used here, and standard runners in private repositories.
Every other label in this repository is capacity RustFS pays for by the hour whether or not a job is running well:
Every other label in this repository is capacity RustFS pays for. ARC pods and lab machines accrue cost while the machine exists. A Ubicloud label is metered for the minutes a job holds its VM:
| Label | Paid by | What it is | Use it for |
|---|---|---|---|
@@ -15,7 +15,7 @@ Every other label in this repository is capacity RustFS pays for by the hour whe
| `windows-latest`, `macos-latest` | nobody | GitHub-hosted standard runners | Windows and macOS targets |
| `sm-standard-2` | RustFS | ARC pod, Ubuntu 24.04, minimal image: no `gh`, no C toolchain; Docker only on the `dind-` label | only a job that a hosted runner cannot run (see the placement rule) |
| `sm-standard-4` | RustFS | ARC pod, Ubuntu 24.04, 14 CPUs and 28 GB (the label is not the core count), large disk | workspace-scale Cargo builds |
| `sm-standard-4-arm` | RustFS | ARC pod, aarch64; reached through the `build.yml` matrix, so `actionlint.yaml` does not list it | native aarch64 release builds |
| `ubicloud-standard-16-arm` | RustFS | Ubicloud VM, alias of `ubicloud-standard-16-arm-ubuntu-2404`: Ubuntu 24.04 arm64, 16 vCPU, 48 GB, 300 GB disk. Reached through the `build.yml` matrix, so `actionlint.yaml` does not list it | native aarch64 release builds (`linux-aarch64-musl`, `linux-aarch64-gnu`) |
| `dind-sm-standard-2` | RustFS | ARC pod with Docker-in-Docker | image builds that need `buildx` and a registry push |
| `smoke-testing`, `pf-testing` | RustFS | long-lived lab machines | the packaged functional chain and performance suites |
@@ -40,7 +40,7 @@ A hosted runner is also the only choice for a few things the pods cannot do: `mo
- Every `shared-key` has exactly one writer. `cache-warm.yml` owns `ci-dev`, `ci-feat-rio`, `ci-feat-proto` and `ci-uring`; every other `./.github/actions/setup` call states `cache-save-if` explicitly, and `scripts/security/check_cache_save_if.sh` fails the build when one does not.
- All readers of a key run on the same image as its writer. Keys that live on hosted runners say so in their name (`ci-uring` is warmed on `ubuntu-latest`; `ci-e2e-distributed-hosted`, `e2e-upgrade-server-hosted`, `e2e-upgrade-tests-hosted`).
- Moving a job between pools therefore means renaming its key. The new key is empty until a run from `main` (a scheduled run, or a manual dispatch of that workflow from `main`) saves it; until then that lane does a cold build.
- Moving a job between pools therefore means renaming its key. The new key is empty until a run from `main` (a scheduled run, or a manual dispatch of that workflow from `main`) saves it; until then that lane does a cold build. The `build.yml` aarch64 release legs append `-ubicloud-arm` for this reason. `cache-warm.yml` does not write those keys.
- `CARGO_BUILD_JOBS` and every other `CARGO*`, `CC*`, `CFLAGS*`, `CXX*`, `CMAKE*` and `RUST*` variable is hashed into the key. `ci.yml` sets `CARGO_BUILD_JOBS` at step level for that reason; promoting it to job level rotates the key for every lane that shares it.
## Where self-hosted minutes leak
@@ -62,7 +62,7 @@ gh api "repos/rustfs/rustfs/actions/runs/<run_id>/jobs?per_page=100" \
--jq '.jobs[] | "\(.conclusion)\t\(.labels | join(","))\t\(.started_at)\t\(.completed_at)\t\(.name)"'
```
Only rows whose label starts with `sm-`, `dind-`, `smoke-` or `pf-` cost anything. At the October 2026 review, a full pull-request run of `ci.yml` spent about 150 self-hosted minutes, a third of it in jobs that never compiled; treat that as the scale of what a placement mistake costs, not as a current figure.
Only rows whose label starts with `sm-`, `dind-`, `ubicloud-`, `smoke-` or `pf-` cost anything. At the October 2026 review, a full pull-request run of `ci.yml` spent about 150 self-hosted minutes, a third of it in jobs that never compiled; treat that as the scale of what a placement mistake costs, not as a current figure.
## Checklist for a new or moved job
+17 -7
View File
@@ -181,12 +181,18 @@ jobs:
selected="all"
fi
# linux-aarch64-* need a real aarch64 host (placement rule: hardware)
# so the packaged binary and console smoke execute the artifact.
# ubicloud-standard-16-arm is that host: Ubuntu 24.04 arm64, alias of
# ubicloud-standard-16-arm-ubuntu-2404 (16 vCPU, 48 GB). linux-x86_64-* stay on the
# ARC sm-standard-4 pool. GNU rows keep cross=true so cargo-zigbuild
# pins the glibc 2.31 floor on the same architecture; musl stays a
# native cargo build.
all='{"include":[
{"target_id":"linux-x86_64-musl","os":"sm-standard-4","target":"x86_64-unknown-linux-musl","cross":false,"platform":"linux","rustflags":"","features":"","glibc":""},
{"target_id":"linux-aarch64-musl","os":"sm-standard-4-arm","target":"aarch64-unknown-linux-musl","cross":false,"platform":"linux","rustflags":"","features":"","glibc":""},
{"target_id":"linux-aarch64-musl","os":"ubicloud-standard-16-arm","target":"aarch64-unknown-linux-musl","cross":false,"platform":"linux","rustflags":"","features":"","glibc":""},
{"target_id":"linux-x86_64-gnu","os":"sm-standard-4","target":"x86_64-unknown-linux-gnu","cross":true,"platform":"linux","rustflags":"","features":"pyroscope","glibc":"2.31"},
{"target_id":"linux-aarch64-gnu","os":"sm-standard-4-arm","target":"aarch64-unknown-linux-gnu","cross":true,"platform":"linux","rustflags":"","features":"pyroscope","glibc":"2.31"},
{"target_id":"linux-aarch64-gnu","os":"ubicloud-standard-16-arm","target":"aarch64-unknown-linux-gnu","cross":true,"platform":"linux","rustflags":"","features":"pyroscope","glibc":"2.31"},
{"target_id":"macos-aarch64","os":"macos-latest","target":"aarch64-apple-darwin","cross":false,"platform":"macos","rustflags":"","features":"pyroscope","glibc":""},
{"target_id":"windows-x86_64","os":"windows-latest","target":"x86_64-pc-windows-msvc","cross":false,"platform":"windows","rustflags":"","features":"","glibc":""}
]}'
@@ -244,7 +250,11 @@ jobs:
with:
rust-version: stable
target: ${{ matrix.target }}
cache-shared-key: build-${{ matrix.target }}${{ matrix.glibc && format('-glibc-{0}', matrix.glibc) || '' }}
# rust-cache keys include OS and architecture but not the runner image.
# The Ubicloud aarch64 legs therefore use their own suffix so they
# cannot restore a cache written on a different arm64 image.
# x86_64, macOS, and Windows keys stay unsuffixed.
cache-shared-key: build-${{ matrix.target }}${{ matrix.glibc && format('-glibc-{0}', matrix.glibc) || '' }}${{ matrix.os == 'ubicloud-standard-16-arm' && '-ubicloud-arm' || '' }}
# main only. A cache saved on refs/tags/X is scoped to that tag: no
# other tag, no main run and no PR can restore it, so every release
# cycle wrote up to 12 entries of 1-2GB (preview tag plus final tag,
@@ -764,9 +774,9 @@ jobs:
exit 0
fi
# The self-hosted Linux runners do not ship the aws CLI (GitHub-hosted
# images did). Install the official CLI bundle on demand so R2 uploads
# survive a fresh runner without relying on a distro package.
# ARC pods (the x86_64 legs) do not ship the aws CLI. Ubicloud arm64
# images follow the GitHub runner image and often do. Install the
# official CLI bundle when it is missing so R2 uploads work on either.
if ! command -v aws >/dev/null 2>&1; then
echo "aws CLI not found on runner; installing..."
if command -v apt-get >/dev/null 2>&1; then