docs(experiment): oliphaunt/pglite-oxide as a lightweight dev-mode database (WIN-2130)

[rubenfiszel]

Summary

The WIN-2130 experiment: can Windmill run against oliphaunt / pglite-oxide — embedded Postgres 17.5 (WASIX, no Docker, no install) — as a lightweight dev mode?

Yes. With the migration-bootstrap fix (merged separately in #9970, now on main), a full Windmill server+worker boots on the embedded single-connection database and runs scripts and flows:

SCRIPT: {"success":true,"completed":true,"result":"6x7=42"}
FLOW:   {"success":true,"completed":true,"result":"step-b-ok-flow-complete"}
PostgreSQL version: PostgreSQL 17.5 on wasm32-unknown-wasix … 32-bit

This PR is now just the experiment — the enabling code fix has landed on its own. It adds:

• docs/experiments/oliphaunt-lightweight-dev-mode.md — the write-up: the single-connection ≡ max_connections=1 insight, why the runtime needs only one connection, the three migration-bootstrap sites that didn't, the fix (fix(migrations): run bootstrap housekeeping on the migrator's held connection #9970), caveats, and how to reproduce/isolate.
• dev/embedded-db/ — standalone harness (deliberately not a backend-workspace member; it pulls the heavy wasmer/alpha-pinned tree that must not leak into the backend lockfile):
  • wm-embedded-db — boots embedded pglite, exports DATABASE_URL, execs a child (e.g. the windmill binary).
  • schema-compat — applies Windmill's full 1182-migration set on a single connection (~950 ms) to prove schema compatibility.
  • committed Cargo.lock with the required virtual-net = 0.702.0-alpha.3 pin (else wasmer-wasix won't compile).

It is an experiment / build-it-yourself dev tool, not a release artifact (publish = false): pglite-oxide 0.5.1 pins alpha wasmer crates and is 32-bit/single-connection, so it's not something to ship in a supported release yet.

Caveats

• Single-connection = serialized: a single-developer dev convenience, not production/load.
• 32-bit wasm backend (4 GiB ceiling). uuid-ossp is the only extension pglite lacks and Windmill doesn't need it (already stripped in OVERRIDDEN_MIGRATIONS; schema uses gen_random_uuid()).

Test plan

• cd dev/embedded-db && cargo run --bin schema-compat → all 1182 migrations apply on one connection in ~950 ms.
• cargo run -- ../../backend/target/debug/windmill (windmill from main, incl. fix(migrations): run bootstrap housekeeping on the migrator's held connection #9970) → migrations apply, server+worker boot, a script and a 2-step flow complete on PostgreSQL 17.5 on wasm32-unknown-wasix.

Fixes WIN-2130. Depends on #9970 (merged).

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying windmill with    Cloudflare Pages

Latest commit:	8e7c1ce
Status:	✅  Deploy successful!
Preview URL:	https://40b02100.windmill.pages.dev
Branch Preview URL:	https://ruben-win-2130-experiment-ol.windmill.pages.dev

View logs

[rubenfiszel]

The migration-bootstrap fix in this PR has been broken out as a standalone, independently-mergeable PR: #9970. That's the durable, broadly-useful change (it also helps connection-constrained managed Postgres / PgBouncer) and can merge on its own.

This PR (#9967) remains the full WIN-2130 experiment — the dev/embedded-db harness + write-up — kept self-contained (it includes the same fix) so it stays reproducible end-to-end. Once #9970 lands, this branch rebases cleanly onto it.