LibreChat/.github
Danny Avila fb1322401d
🧪 test: Restore Two Rotted Integration Suites and Run Every One in CI (#15866)
* 🧪 test: Restore Two Rotted packages/api Integration Suites

MCPReinitRecovery.integration.test.ts mocked @librechat/data-schemas down to
its logger, so all eight tests threw once MCPServersRegistry started calling
scopedCacheKey (#15028). The mock now spreads the real module under the logger
override; ServerConfigsDB stays mocked, so the suite still never touches Mongo.

secrets.integration.spec.ts drove langfuse.secretKey through the generic admin
config API, which has stripped the langfuse section since #14108 moved it to
the dedicated Langfuse connection API. The generic-write matrix drops that row,
and a langfuse block covers the current contract against a real Config
collection: the stored secret stays encrypted at rest, generic reads redact it,
generic patch, tombstone and field-delete writes leave it untouched, and a full
base-config replacement carries it forward.

* 👷 ci: Run Every packages/api Integration Suite in CI

test:ci excludes every path containing "integration.", but the jobs meant to
run those suites selected by folder: the cache workflow only
src/{cache,cluster,mcp,stream}, the agents workflow only
src/agents/*.integration.spec.ts. Thirteen suites ran in no job, including
every *.integration.test.ts.

test:integration selects *.integration.(spec|test).ts anywhere under src and
replaces test:agents-integration in the (renamed) Integration Tests workflow.
The cache scripts also accept .test.ts, stream selects by suffix alone, and
core picks up src/middleware. Every suite test:ci excludes is now selected by
exactly one CI script.

* 🧪 test: Read the Tool Snapshot in the Concurrent Reinit Simulation

The concurrent reinitialize test copied reinitMCPServer but still called
fetchTools(), which returns [] when tools/list fails. Concurrent reinspections
of a recovered server can each succeed, and a later write replaces the app
connection an earlier request was already handed, so that request's tools/list
fails. The test failed in 9 of 25 iterations on Linux, and in the first CI run
that included the suite. Since #14686 production reads the snapshot and keeps
cached tools when it is incomplete; the simulation now does the same, and still
requires every complete snapshot to list both tools and at least one to complete.

A deterministic test pins the property that tolerance relies on, without
depending on the race: a newer config write replaces the held app connection,
whose snapshot is then incomplete, while the replacement lists the real tools.
2026-09-12 16:20:22 -04:00
..
ISSUE_TEMPLATE 🆔 feat: Built-in Build Metadata for Support Triage (#12756) 2026-05-23 09:41:13 -04:00
scripts 🎯 ci: Retarget Pull Requests Opened Against main to dev (#15416) 2026-08-31 11:29:34 -04:00
workflows 🧪 test: Restore Two Rotted Integration Suites and Run Every One in CI (#15866) 2026-09-12 16:20:22 -04:00
CODE_OF_CONDUCT.md 🔗 chore: Add Stable Discord and Homepage Links (#1835) 2024-02-19 09:42:57 -05:00
configuration-release.json 📜 ci: AutomateCHANGELOG.md (#5838) 2025-02-18 08:35:43 -05:00
configuration-unreleased.json 📜 ci: AutomateCHANGELOG.md (#5838) 2025-02-18 08:35:43 -05:00
CONTRIBUTING.md 🎯 ci: Retarget Pull Requests Opened Against main to dev (#15416) 2026-08-31 11:29:34 -04:00
FUNDING.yml
playwright.yml ⚙️ ci: Bump GitHub Actions to Node.js 24 Runtimes (#14689) 2026-08-07 11:25:29 -04:00
pull_request_template.md 📐 docs: Set PR Description, Review and Completion Standards (#15753) 2026-09-08 17:00:32 -04:00
SECURITY.md ✨ v0.8.8-rc2 (#14845) 2026-09-02 22:18:48 -04:00