Run tests on a throwaway RAM-backed Postgres#201
Merged
Conversation
Each test creates/drops its own database and runs every migration, with nextest running many in parallel. Against a disk-backed cluster the CREATE/DROP DATABASE fsync storm saturates disk I/O and makes the whole machine unresponsive — hence the habit of prefixing runs with nice. scripts/ramdisk-pg.sh spins up a throwaway Postgres on tmpfs (/dev/shm) with fsync/synchronous_commit/full_page_writes off, points DATABASE_URL at it, runs the given command, then tears down. It reuses the installed initdb/pg_ctl so the server version matches the system Postgres with no container or image to manage. On a full package this took a ~54s run down to ~2s with no I/O grind. Wired up as opt-in just recipes (test-fast passes args through to nextest; fast wraps arbitrary commands); default just test is unchanged for machines without spare RAM. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add Homebrew (opt/postgresql@NN) and Postgres.app bundle paths to the fallback so the script finds initdb/pg_ctl on macOS. macOS has no /dev/shm and no default tmpfs, so it lands on disk there — but fsync is off, which is what removes the I/O grind, so it stays fast. Reword the fallback notice accordingly and point at hdiutil for a real ramdisk via CANOPY_TEST_PG_DIR. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Flip just test / test-package / test-name / test-verbose / test-e2e to run through scripts/ramdisk-pg.sh, so the throwaway tmpfs + fsync-off cluster is the default and there's no disk-fsync grind out of the box. The per-test database footprint is tiny (~40MB base + ~8MB per concurrently-live test DB, peak well under ~250MB), so RAM isn't a real constraint. just test now takes nextest args (just test -p database, just test <name>). Add test-system as the escape hatch to run against $DATABASE_URL instead — for inspecting the DB afterwards or where initdb/pg_ctl aren't available. CI is unchanged: it calls cargo nextest directly against its own Postgres service, not the just recipes. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
CI was inlining 'cargo nextest run' plus its own systemctl start postgresql + role/db creation, which diverged from 'just test' once the recipe started wrapping nextest in scripts/ramdisk-pg.sh. Point CI at the recipes instead so it exercises the exact path developers run locally. - Test job: install just, drop the Postgres-service setup, run 'just test'. The wrapper builds its own throwaway tmpfs cluster from the runner's preinstalled initdb/pg_ctl, so no system Postgres is needed. - Playwright job: drop the Postgres-service setup and wrap the test run with scripts/ramdisk-pg.sh, which exports CANOPY_E2E_ADMIN_DATABASE_URL at the throwaway cluster for the fixture's per-worker databases. Its explicit build/cache steps stay (CI caching concerns, not test-runner divergence). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Now that tests are fast (RAM-backed), there's little value in stopping at the first failure — running the whole suite surfaces every failure in one go. Bake --no-fail-fast into the test recipes (test, test-package, test-name, test-verbose, test-system); CI inherits it via just test. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The private-server tests serve the embedded React SPA through rust-embed, so a client-side route like GET /status only returns 200 when private-web/dist/ exists. Routing CI through just test set SKIP_FRONTEND_BUILD=1, so build.rs no longer built it and a fresh checkout had an empty dist/, 404ing that test. (Dev machines have a dist/ from working on the frontend, which masked it locally.) Add a Node setup + npm ci + npm run build step to the test job so dist/ is populated before just test runs. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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.
🤖 Each test creates and drops its own database (and runs every migration), and nextest runs many in parallel. Against a disk-backed Postgres the resulting
CREATE DATABASE/DROP DATABASEfsync storm saturates disk I/O and can make the whole machine grind to a halt — hence the habit of prefixing local runs withnice.This adds
scripts/ramdisk-pg.sh, which spins up a disposable Postgres on tmpfs (/dev/shm) withfsync/synchronous_commit/full_page_writesoff, pointsDATABASE_URLat it, runs the given command, then tears it all down. Nothing touches a physical disk, so there's no I/O grind and runs are dramatically faster (a full package run went from ~54s to ~2s here). It reuses the installedinitdb/pg_ctl, so the server version matches the system Postgres with no container or image to manage.What changed
scripts/ramdisk-pg.sh: the harness. Picks a free port,initdbs a throwaway cluster on tmpfs, starts it with durability off, creates the role+db, exportsDATABASE_URL(andCANOPY_E2E_ADMIN_DATABASE_URL), runs the wrapped command, and cleans up on exit. Finds the Postgres server tools on Arch, Debian/Ubuntu, Homebrew, and Postgres.app.just test(andtest-package/test-name/test-verbose/test-e2e) now run through it by default.just testtakes nextest args, sojust test,just test -p database, andjust test <name>all work.just test-system [args]is the escape hatch to run against$DATABASE_URLinstead — for inspecting the DB afterwards, or whereinitdbisn't available.justrecipes rather than inliningcargo nextestplus its own Postgres-service setup, so it exercises the same path developers run locally.Notes
/dev/shmand no default tmpfs, so it falls back to disk there — still fast (fsync is off), just not RAM-backed unless you pointCANOPY_TEST_PG_DIRat a real ramdisk.initdb/pg_ctl), not just thepsqlclient.