Enable logical replication slots (wal_level=logical + pgoutput fixes)

[riderx]

Problem

Fixes #880 — PGlite cannot use logical replication slots.

Two separate things were broken:

1. wal_level defaults to replica, so pg_create_logical_replication_slot() fails with logical decoding requires "wal_level" >= "logical".
2. Even with wal_level=logical (via custom startParams), consuming a slot with pgoutput crashed the wasm instance: the -sMAIN_MODULE=2 build dead-code-eliminates the 46 core symbols that only pgoutput.so imports (defGetStreamingMode, logicalrep_write_*, GetRelationPublications, …), because pgoutput is built from src/backend and — unlike pgxs extensions — never got a .imports file feeding the exported-functions list. The first decode call hit an undefined table entry (TypeError: r is not a function), and execProtocolRawSync silently swallowed it, leaving the session returning empty results for every subsequent query until Postgres PANICs with ERRORDATA_STACK_SIZE exceeded.

Changes

• defaultStartParams: add -c wal_level=logical so replication slots work out of the box. (WAL volume impact is negligible for an embedded single-user database.)
• execProtocolRawSync: rethrow exceptions that are not an emscripten ExitStatus instead of swallowing them. A trap or missing-symbol call means the wasm instance state can no longer be trusted; surfacing the error beats silent corruption. Known exit statuses (including the POSTGRES_MAIN_LONGJMP error-recovery path) behave exactly as before.
• Tests (tests/replication-slots.test.ts): slot create/list/drop, decoding inserts/updates/deletes through pg_logical_slot_peek_binary_changes/get_binary_changes with pgoutput, clean error recovery when decoding options are invalid, and slot persistence across a dumpDataDir/loadDataDir cycle.
• Docs: note the new default in the startParams section of api.md.

Companion PR

The symbol-export fix lives in the submodule: electric-sql/postgres-pglite#79. The submodule pointer in this PR points at the head of that PR (fetchable from the upstream repo via refs/pull/79/head), so a clean wasm:build of this tree includes the fix; it should be re-bumped to the merge commit once #79 lands.

Verification

With the rebuilt wasm, all 5 new tests pass and the full existing suite passes (267 tests, no regressions):

✓ tests/replication-slots.test.ts (5 tests) 4613ms
Test Files  1 passed (1)
     Tests  5 passed (5)

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4aaa87118e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Bump postgres-pglite for pgoutput exports

This enables wal_level=logical and documents logical slot support, but the commit does not include the required wasm-side symbol export fix: git ls-tree shows both this commit and its parent still pin postgres-pglite to 7b4ee5086055dc5e54ae1e13e487888249438e68, and the root wasm:build script builds the release wasm from that submodule. In a clean build of this tree, pgoutput still lacks the exported core symbols described in the commit message, so calls like pg_logical_slot_get_binary_changes(..., 'pgoutput') will still hit the missing-symbol crash instead of delivering changes; please include the submodule bump before advertising/enabling this feature.

Useful? React with 👍 / 👎.

[riderx]

Good catch — addressed in 6d55f6c: the submodule now points at the head of electric-sql/postgres-pglite#79 (04f1bea), which contains the pgoutput symbol-export fix. That commit is served by the upstream repo via refs/pull/79/head, so git submodule update can fetch it by SHA (verified with a git fetch --depth 1 <upstream> 04f1bea from a fresh clone). A clean wasm:build of this tree now produces a main module that exports all 46 symbols pgoutput.so imports, and the new replication-slot tests pass against it. Once #79 merges, this pointer should be bumped to the merge commit.

[riderx]

hey @tdrz i would love your input on that, i'm really looking for logical replication in pglite for a very specific case where i need to replicate in many regions.
If you see anything i could inprove i'm up for the job, also if you can allow CI to run it's nice