Skip to content

fix(wrapper): bump vitest testTimeout to 15s for CI transport test#19

Merged
manojp99 merged 1 commit into
mainfrom
fix/vitest-timeout-for-ci
May 28, 2026
Merged

fix(wrapper): bump vitest testTimeout to 15s for CI transport test#19
manojp99 merged 1 commit into
mainfrom
fix/vitest-timeout-for-ci

Conversation

@manojp99

Copy link
Copy Markdown
Collaborator

Why

The publish-wrapper workflow keeps timing out on a single test in test/transport.test.ts:

test/transport.test.ts > Transport > terminate() resolves with SIGTERM signal or non-zero exit code
  → Test timed out in 5000ms

Failing runs (after PR #17 + #18 merged and the wrapper-v0.4.0 tag was placed on the new merge commit):

The test spawns sh -c "sleep 60", sends SIGTERM via terminate(), then waits for the subprocess to exit. On a clean local machine it completes in <1s. On busy GitHub Actions runners it routinely exceeds vitest's 5s default.

Fix

Bump global testTimeout to 15s in wrappers/typescript/vitest.config.ts. Comment in the config references this PR's context so future maintainers know why the default isn't 5s.

Verification

$ cd wrappers/typescript && bun test
 61 pass, 0 fail
$ cd wrappers/typescript && bun test test/transport.test.ts
 3 pass, 0 fail  (terminate() runs in ~1ms locally)

After this merges

engine-v0.3.0 and wrapper-v0.4.0 get moved to the new merge commit; pushing wrapper-v0.4.0 re-triggers the publish workflow and ships amplifier-agent-ts@0.4.0 to npm.

The publish-wrapper workflow times out on test/transport.test.ts > Transport >
terminate() resolves with SIGTERM signal, which spawns a real subprocess and
waits for SIGTERM exit. vitest's default 5s timeout is too tight on busy
GitHub Actions runners. Bumping to 15s accommodates the real-world round-trip.

Locally verified: 61/61 wrapper tests pass.

Failing workflow run: https://github.com/microsoft/amplifier-agent/actions/runs/26551192854

Generated with [Amplifier](https://github.com/microsoft/amplifier)

Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>
@manojp99 manojp99 merged commit 53f99bb into main May 28, 2026
1 of 3 checks passed
manojp99 pushed a commit to manojp99/amplifier-app-nanoclaw that referenced this pull request May 28, 2026
…3.0 (protocol 0.2.0)

Adopt the engine + wrapper coordinated fix from amplifier-agent PR #17
(protocol 0.2.0, path-based MCP config delivery) and the catalog fix from
PR #20 (openai default_model gpt-5.5 to match the bundle's
extended_thinking=true default).

Engine pin: AMPLIFIER_AGENT_REF=engine-v0.3.0 in container/Dockerfile.
The tag was moved to PR #20's merge commit so a rebuild pulls the catalog
fix without needing a separate engine version bump. Layer-cache busting
during rebuild required --no-cache (or BuildKit --no-cache-filter).
A persistent CACHEBUST arg for the Dockerfile is a separate follow-up.

Wrapper bump: amplifier-agent-ts 0.3.1 -> 0.4.0 (published with provenance
via OIDC trusted publishing on the wrapper-v0.4.0 tag).

Host + agent-runner code: drops the workarounds we'd been carrying for
the engine's MCP plumbing bugs. The engine now handles --mcp-config-path
end-to-end (validates the path, sets AMPLIFIER_MCP_CONFIG, lets the
tool-mcp module discover via its documented config priority). The
host-side bridge no longer needs to compensate.

Verified end-to-end inside the rebuilt container:
- amplifier-agent run --provider openai -y "..." returns a real reply
  through gpt-5.5 with extended_thinking=true flowing cleanly through
  loop-streaming -> provider-openai -> OpenAI API.
- Engine envelope parseable on stdout, protocolVersion=0.2.0,
  metadata block present.

Refs:
- microsoft/amplifier-agent#17 (engine + wrapper Path D)
- microsoft/amplifier-agent#18 (smoke test fix)
- microsoft/amplifier-agent#19 (vitest timeout fix)
- microsoft/amplifier-agent#20 (catalog gpt-5.5 default)
- engine tag: engine-v0.3.0 -> commit 2bd951f
- wrapper tag: wrapper-v0.4.0 -> commit 53f99bb

Generated with [Amplifier](https://github.com/microsoft/amplifier)

Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>
manojp99 added a commit that referenced this pull request Jun 3, 2026
…0.6.1) (#37)

The `test/transport.test.ts > terminate() resolves with SIGTERM signal
or non-zero exit code` test exercises actual subprocess SIGTERM
handling, which is slower on Ubuntu CI runners than on local macOS.
The 0.6.0 publish-wrapper.yml workflow run failed at the Test step
with `Error: Test timed out in 5000ms` before reaching `npm publish`.

Per-test timeout bumped from 5s to 15s. Same class of fix as
#19 fix(wrapper): bump vitest testTimeout to 15s for CI transport test
from a prior release window.

`amplifier-agent-ts@0.6.0` was never published to npm; this 0.6.1
release supersedes that aborted attempt. Consumers can install 0.6.1
directly without first installing 0.6.0.

Local TS wrapper test results: 95 passed / 95.

Co-authored-by: Manoj Prabhakar Paidiparthy <mpaidiparthy@microsoft.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant