safe-global / safe-client-gateway / 37949488270

91%

push

github

fix(spaces): decide on the SES email module per app, not at import (#3556)

* perf: share the integration-test module cache and shard CI

Integration specs spent most of their run re-importing the app's module graph (about 156s of 275s locally), so the `integration` project now runs with `isolate: false` and loads the same leftover-state guard as the unit projects.

That surfaced an order-dependent `app.close()` hang. Nest's `ConfigModule` loaded the developer's `.env` into `process.env` on every app boot, and `spaces.module.ts` reads `FF_SES_EMAIL` once at import time. When the module was first imported after an app had booted, every later app ran a real SES worker that picked up a leftover `ses-email` job, called AWS with dummy credentials and never returned, so BullMQ's worker close waited forever. Every entry point already passes `node --env-file=.env`, so `ConfigModule` now sets `ignoreEnvFile: true`.

CI splits the integration job into three `vitest --shard` jobs, each with its own services and Coveralls flag.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* ci: fail fast across integration-test shards

A failing shard now cancels the other two instead of letting them run to completion, per review feedback.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* ci: let every integration-test shard finish when one fails

This reverts commit bd9093b1b. The shards split one suite by file, so their failures are independent: running them all to completion surfaces every failure in one CI run, and a flaky shard can be re-run alone instead of taking the other two with it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(spaces): decide on the SES email module per app, not at import

`spaces.module.ts` read `FF_SES_EMAIL` through `configuration()` when the file was first imported, so the decision ignored the configuration an app was built with and, with a shared module cache, stuck for every later app in the process. `AppM... (continued)

4662 of 5424 branches covered (85.95%)

Branch coverage included in aggregate %.

11797 of 12858 relevant lines covered (91.75%)

473.61 hits per line