SQLite gives e2e tests a single shared file across all containers without
needing a heavyweight MariaDB/MongoDB/PostgreSQL service. WAL + busy_timeout
keep multi-process writers tolerable. Memory adapter was considered but its
per-process PHP arrays don't survive Swoole worker fan-out or cross-container
queue handoffs, so file-backed SQLite is the practical choice.
- registers: add 'sqlite' scheme to pool factory and worker db register,
configured with PRAGMA journal_mode=WAL, busy_timeout=5000, foreign_keys=ON
- compose: pass _APP_DB_SQLITE_PATH and mount appwrite-sqlite:/storage/sqlite
on every service that touches the main DB
- .env: switch _APP_DB_ADAPTER to sqlite, drop mariadb from COMPOSE_PROFILES
- ci.yml: add SQLite to the e2e matrix alongside MongoDB so each service
is timed against both adapters in the same workflow run
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`https://appwrite.io/install/compose` now returns a 308 redirect to the
HTML install docs (`/docs/advanced/self-hosting/installation`) instead
of serving the compose file, so the Benchmark job's "Installing latest
version" step was downloading 0 bytes and `docker compose up -d` died
with "empty compose file". This has been failing the Benchmark job on
every recent PR, not just this one.
Resolve the latest release tag via the GitHub API, then fetch the
compose file and `.env` from `raw.githubusercontent.com` at that tag.
Switched both curl calls to `-fsSL` so they fail loudly on non-2xx
responses or redirect loss instead of silently writing empty files.
ProjectWebhooks tests have shared state dependencies (e.g. index
creation must complete before assertions). Running with --functional
(parallel methods) causes flaky failures where indexes are still
'processing' instead of 'available'.
Spec generation silently produced a fatal error when a response model
string could not be resolved to a registered model object. Now throws a
clear RuntimeException in both Swagger2 and OpenAPI3 formats, for both
single and array model responses.
Also adds a CI job to run spec generation on every PR so unresolved
models are caught before merge.
Jake BarnbyandClaude Opus 4.7
Matej Bačo
Chirag Aggarwal
premtsd-code