ci: probe namespace.so runners for KVM/kernel-dependent jobs - #3352
ci: probe namespace.so runners for KVM/kernel-dependent jobs#3352tomassrnka wants to merge 3 commits into
Conversation
Checks everything that blocked or nearly blocked Blacksmith (closed PR 3332): uffd UFFD_FEATURE_WP_ASYNC via a real UFFDIO_API handshake (the actual blocker - needs kernel >= 6.7), nbd driver availability, /dev/kvm, hugepages, cgroup2, swap. Temporary; deleted once the verdict is in. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
PR SummaryLow Risk Overview Reviewed by Cursor Bugbot for commit 05ca3cb. Bugbot is set up for automated code reviews on this repo. Configure here. |
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
| grep -q . /proc/swaps && [ "$(wc -l < /proc/swaps)" -gt 1 ] && echo "swap already active" || { | ||
| sudo fallocate -l 256M /swapfile-probe && sudo chmod 600 /swapfile-probe && sudo mkswap /swapfile-probe && sudo swapon /swapfile-probe && sudo swapoff /swapfile-probe | ||
| } | ||
| echo "HOST_SMOKE_OK" |
There was a problem hiding this comment.
Hugepage smoke skips allocation check
Medium Severity
The hugepage smoke writes nr_hugepages and then reports success without reading the value back. The kernel can accept that write while allocating fewer pages or none, so the job can still go green when hugepages are not actually available.
Reviewed by Cursor Bugbot for commit 84cf9dc. Configure here.
The first two runs queued while the Namespace app did not yet cover this repo - workflow_job events are not re-delivered after access changes, so a fresh push is needed. The 22.04 canary uses the exact label from their docs in case the 24.04 labels are not offered. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
There are 2 total unresolved issues (including 1 from previous review).
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 05ca3cb. Configure here.
| echo "kernel=$(uname -r) NBD=$NBD_OK KVM=$KVM_OK (WP_ASYNC verdict is the uffd handshake step)" | ||
| $NBD_OK | ||
| # KVM is required on x64 only (no arm64 runner anywhere has it) | ||
| if [ "${{ matrix.arch }}" = "x64" ]; then $KVM_OK; fi |
There was a problem hiding this comment.
Verdict ignores failed NBD check
Medium Severity
The Verdict step runs under set -x only, so $NBD_OK does not fail the step when it expands to false. On x64 a later successful $KVM_OK masks it; on arm64 the trailing if always exits 0. A missing NBD device can still produce a green job.
Reviewed by Cursor Bugbot for commit 05ca3cb. Configure here.
5aad415 to
d71980e
Compare
Each compression config runs the whole suite on one runner, where api/templates alone sets the wall clock: 853s of the 856s test step under zstd1, because go test caps every package at -parallel=4. Splitting a config across jobs is the only lever that moves that, so this adds the machinery. scripts/select-tests.sh enumerates the top-level tests from the source tree and bin-packs each package across the shards using recorded per-test times, so every shard holds a slice of every package, a newly added test always lands in exactly one shard, and the shards finish at roughly the same time. update-test-weights.go refreshes the recorded times from a run's JUnit files. SHARDS defaults to 1, so this changes nothing until someone turns the dial. Measured at 4 shards the slowest job drops 20.1m -> 11.3m, but the infra-tests pool only served ~6 concurrent runners, so the extra jobs spent their savings queueing and end-to-end got worse. Raise SHARDS alongside more capacity (#3352), not before.
|
Closing — not pursuing the namespace.so runner probe for now. The integration-suite speed-up landed via work reduction instead (#3462: 21.5m → 11.1m end-to-end, no extra runner capacity needed). |
1 participant
What
Follow-up to #3332 (Blacksmith attempt, blocked on their 6.6 kernel lacking
UFFD_FEATURE_WP_ASYNC). This probes namespace.so runners as an alternative host for the jobs still on GitHub larger runners (orchestrator unit shards + integration tests ≈ 80% of paid CI minutes).Per runner (x64
nscloud-ubuntu-24.04-amd64-8x16, arm64nscloud-ubuntu-24.04-arm64-4x8):WP_ASYNChandshake — a compiledUFFDIO_APIprobe requesting the orchestrator's exact feature set (WP_ASYNC | EVENT_REMOVE | MISSING_HUGETLBFS). This is the check that would have failed on Blacksmith; requires kernel ≥ 6.7.linux-modules-extrafallback)/dev/kvm(x64 verdict-gating), tun, cgroup2, hugepages, uffd sysctl, swapfile behaviorGreen x64 job = all hard requirements met → the migration is flipping the same three runner labels as #3332 attempted.
Note: if jobs sit in "Waiting for a runner", the Namespace GitHub app isn't installed/enabled for this repo yet — that's step 0.
🤖 Generated with Claude Code