test: split stack-backed e2e into a serial project and gate CI jobs by path - #6472
Draft
kanadgupta wants to merge 3 commits into
Draft
test: split stack-backed e2e into a serial project and gate CI jobs by path#6472kanadgupta wants to merge 3 commits into
kanadgupta wants to merge 3 commits into
Conversation
…arallel Every e2e project ran files strictly serially, a setting from the initial architecture commit that contradicts the flake policy in AGENTS.md and left the CLI's 32 stackless e2e files queueing behind the 10 that start a local stack. The stack-backed files are 72% of e2e time, so this split is the main lever on the worst CI shard. E2e files that start a local Supabase stack or run a Docker container now carry the *.stack.e2e.test.ts suffix and belong to a serial e2e-stack project. Plain *.e2e.test.ts files keep the e2e project with Vitest's default file parallelism. packages/stack has only stack-backed e2e, so its e2e project becomes e2e-stack. Package test:e2e:run scripts select both projects. Thirteen files are renamed with no content changes. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
About a third of recent PRs were Go-only or docs-only and still ran the full TypeScript test suite. A paths-filter job now classifies the change: docs or release-notes only skips every test job; Go-only skips unit and integration but keeps e2e, which shells out to the Go sidecar; anything under .github or a TypeScript workspace runs everything. The code-quality job is never gated. Both summary jobs fail unless the gate job succeeded, so a broken gate cannot produce a false green. Turbo's --affected was considered and rejected: its package graph does not know the CLI depends on the Go sidecar (that edge exists only in the task graph), and a gate job that installs dependencies would add over a minute before any test job could start. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Records the agreed direction for how tests run: root Vitest runs as the unit of execution, stack-backed e2e as a suffix-marked serial sub-kind, weight-class sharding, and Turbo reduced to the build graph. The ADR carries the vocabulary for these concepts; the dated plan carries the measured baseline and the three-PR sequence this PR starts. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
kanadgupta
added
the
run-ci
Contributor
Supabase CLI previewnpx --yes https://pkg.pr.new/supabase/cli/supabase@ccb2a6f539b077663b270966378dc920be432864Preview package for commit |
This was referenced Sep 4, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
First of three PRs from the test-execution plan in
docs/superpowers/plans/2026-09-04-test-execution-topology.md. Everything here works on Vitest 4; the Vitest 5 upgrade (#6457) and the root-run topology follow separately.Stack-backed e2e becomes a serial sub-kind; everything else runs in parallel. Every e2e project ran files strictly serially, a setting from the first architecture commit that contradicts the flake policy in
AGENTS.md. In the last develop run the 13 e2e files that start a local stack accounted for about 72% of e2e time while 58 stackless files queued behind them. Those 13 files now carry the*.stack.e2e.test.tssuffix and run one at a time in ane2e-stackproject; plain*.e2e.test.tsfiles keep thee2eproject with default file parallelism. Vitest schedules the serial group after the parallel one within a run, so no custom scheduling is involved. The renames have no content changes.CI test jobs are gated on the paths a PR touches. About a third of recent PRs were Go-only or docs-only and still ran the full TypeScript suite. A
paths-filterjob classifies the change: docs or release-notes only skips every test job; Go-only skips unit and integration but keeps e2e, which shells out to the Go sidecar; anything under.githubor a TypeScript workspace runs everything. The code-quality job is never gated, and both summary jobs fail unless the gate succeeded, so a broken gate cannot produce a false green.turbo --affectedwas rejected because Turbo's package graph does not know the CLI depends on the Go sidecar, and a gate that installs dependencies would delay every test job by over a minute.ADR 0024 and the plan. The ADR records the whole direction, including the vocabulary (test kind, stack-backed versus stackless, test project, standalone versus root run) and the rejected alternatives, so reviewers of the later PRs can see the rationale. The plan carries the measured baseline and the three-PR sequence.
Expected effect: the worst e2e shard's test step drops from about 7.9 minutes toward 5 minutes even under today's per-package hash sharding; balancing across packages is PR C's job.
🤖 Generated with Claude Code