Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 0 additions & 10 deletions .github/runs-on.yml
Original file line number Diff line number Diff line change
Expand Up @@ -38,13 +38,3 @@ runners:
family: ["c7g", "m7g"]
image: ubuntu24-full-arm64
volume: 200gb:gp3:750mbs:4000iops

# Benchmarks. Pinned to one instance family and spot: false — a run that
# lands on a different CPU, or gets interrupted and retried elsewhere, is
# not comparable with the history it is being charted against.
bench-x64:
cpu: 8
family: ["c7a"]
image: ubuntu24-full-x64
volume: 120gb:gp3:500mbs:4000iops
spot: false
85 changes: 0 additions & 85 deletions .github/workflows/bench.yml

This file was deleted.

30 changes: 0 additions & 30 deletions .github/workflows/ci.yml
Original file line number Diff line number Diff line change
Expand Up @@ -108,36 +108,6 @@ jobs:
{ echo '## Spec'; echo; echo '```'; head -20 spec.log; echo '```'; } \
>> "$GITHUB_STEP_SUMMARY"

# Not benchmarks: every harness here carries a positive control (GENERATE_UUID
# uniqueness; per-connection catalogs that MUST fail; a gRPC status that must
# come from our handler rather than the transport) and exits nonzero when the
# control does not behave, so a broken build cannot quietly pass them.
concurrency-harness:
name: concurrency harnesses
needs: shim
runs-on: runs-on=${{ github.run_id }}/runner=zig-x64/extras=s3-cache
steps:
- name: Harden the runner (Audit all outbound calls)
uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
with:
egress-policy: audit

- uses: runs-on/action@284bff869d4a825f778ac7c38e50f1e31ef198fd # v2.6.0
- uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
- id: shim
uses: ./.github/actions/shim
with:
platform: linux-x64
- uses: ./.github/actions/zig
- run: zig build analyzer-threads -Drelease=true -Dshim-dylib=${{ steps.shim.outputs.path }}
- run: zig build shared-catalog-threads -Drelease=true -Dshim-dylib=${{ steps.shim.outputs.path }}
# Same contract: the Storage Read transport answers UNIMPLEMENTED, and
# three controls (an unknown service must reach our handler too, a second
# call on one channel must be answered, and the listener must be gone
# after stop()) mean a build that merely links cannot pass. It binds port
# 0, so it cannot collide with anything else on the runner.
- run: zig build grpc-smoke -Drelease=true -Dshim-dylib=${{ steps.shim.outputs.path }}

# The Python client suite ported from Recidiviz/bigquery-emulator, run the
# way downstream fixtures consume the emulator: testcontainers against the
# Docker image, built here from the same static `dist` binary a release
Expand Down
3 changes: 1 addition & 2 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -90,8 +90,7 @@ Intentional, explicit control over memory allocation improves process stability.

| Workflow | Trigger | What it does |
|---------------|-----------------------------|---------------------------------------------------------------------------------------------------------------------------|
| `ci.yml` | PR, push | ephor + `zig fmt`, unit tests, the spec corpus under `--strict`, the concurrency harnesses, and a `gen-proto` drift check |
| `bench.yml` | push, `bench` label, manual | `gsql-bench --json` charted by github-action-benchmark on the `benchmarks` branch |
| `ci.yml` | PR, push | ephor + `zig fmt`, unit tests, the spec corpus under `--strict`, and a `gen-proto` drift check |
| `release.yml` | `v*.*.*` tag | four platform tarballs, a multi-arch GHCR image, and a GitHub release |

Runners are powered by [RunsOn](https://runs-on.com) on EC2 (see `.github/runs-on.yml`).
Expand Down
5 changes: 2 additions & 3 deletions docs/BUILD.md
Original file line number Diff line number Diff line change
Expand Up @@ -211,8 +211,7 @@ See `docs/COVERAGE.md` before quoting a number.
### Harnesses with positive controls

Not benchmarks, and not part of `test`: each carries a control that must fail,
and exits nonzero when it does not, so a broken build cannot quietly pass. CI
runs all three at `-Drelease=true`.
and exits nonzero when it does not, so a broken build cannot quietly pass.

| Step | Arguments | What it checks |
|--------------------------|-----------------------------------------------------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
Expand Down Expand Up @@ -245,7 +244,7 @@ are not comparable.
| Step | Arguments | What it measures |
|----------------|---------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| `bench` | `[queries_per_size] [catalog_sizes...]` (default 2000 over 1, 50, 500) | Per-query cost as a function of catalog size, with a fresh `CREATE TABLE` between timed queries so the catalog delta is never empty. |
| `gsql-bench` | `[--time MS] [--iters N] [--filter SUBSTR] [--corpus DIR] [--phases] [--markdown] [--json]` | The googlesqlite corpus vendored under `testdata/bench_corpus`, reported in the same shape as upstream's RESULTS.md. `--json` is what `bench.yml` charts; `--markdown` goes to the run summary. Default budget 1000 ms per query. |
| `gsql-bench` | `[--time MS] [--iters N] [--filter SUBSTR] [--corpus DIR] [--phases] [--markdown] [--json]` | The googlesqlite corpus vendored under `testdata/bench_corpus`, reported in the same shape as upstream's RESULTS.md. Default budget 1000 ms per query. |
| `codec-bench` | `[rows]` (default 200000) | Value codec cost with real rows. Needs `-Dcodec-timers=true`. |
| `tz-bench` | `[rows]` (default 200000) | Per-row cost of named-timezone resolution, the A/B point for the ICU tzdata integration. |
| `load-bench` | none | Whether a delete forces work proportional to catalog size. Run-only, not installed. |
Expand Down
Loading