ci: back postgres test data dir with tmpfs to avoid test timeout

[cbartz]

What

Mount the PostgreSQL service container's data directory (/var/lib/postgresql/data) on tmpfs in the Go Tests (PostgreSQL) job.

options: >-
  --health-cmd pg_isready
  --health-interval 10s
  --health-timeout 5s
  --health-retries 5
  --tmpfs /var/lib/postgresql/data

Why

The Go Tests (PostgreSQL) job has been hitting the go test timeout and failing, unrelated to the PR under test:

panic: test timed out after 15m0s
	running tests:
		TestGithubTestSuite (1m59s)
FAIL	github.com/cloudbase/garm/database/sql	900.382s

The database/sql suite creates a fresh schema and runs the full migration for every test case (OpenTestPostgresDB -> CREATE SCHEMA -> migrateDB). On the default postgres:16 service container that DDL is fsync'd to disk, and the goroutine dump confirms the suite was stuck server-side in CreateTable on IO wait.

GitHub Actions service containers can't pass server flags like postgres -c fsync=off (there's no command field), so the data directory is moved to tmpfs instead. This keeps all of Postgres's I/O in RAM, making fsync effectively free. The database is ephemeral and recreated for every job, so trading durability for speed has no downside here, and it mirrors the fsync=off config already used for local Postgres test runs.

Measured locally on the database/sql package with the same flags CI uses (-race -parallel=4): ~562s disk-backed vs ~60s on tmpfs (~9x faster), tests still passing.

[gabriel-samfira]

Hah! The psql tests finished well before the sqlite tests this time 😄

edit: Actually time is about the same. This is great!

[gabriel-samfira]

thanks!