fix: evict pooled connection when a request does not complete (8.0.x)

[gjtorikian]

Backport to the 8.0 series (targets a v8.0.2 patch release via release-please on this branch).

What changed

• lib/workos/base_client.rb: wrap Net::HTTP#request in begin/ensure and evict the cached keep-alive connection unless the request completed. Any exit other than a returned response — a connection error the rescue below knows about, one it doesn't, or a non-local exit such as an application-level Timeout.timeout or Thread#kill — now removes and closes the pooled socket instead of leaving it mid-stream for the next request on the same thread.
• test/workos/test_base_client.rb: adds a real-socket regression test (test_aborted_request_does_not_leak_its_response_to_the_next_request), StubConnection-based coverage of the evict/keep paths, and a teardown that clears the fiber-local connection cache between tests (this branch predates the Thread.current[] storage change in v9.1.0, so the teardown clears Fiber[:workos_connections]).
• .github/workflows/release-please.yml: trigger release-please on pushes to v8.0.x with target-branch: v8.0.x, so merging this PR opens the 8.0.2 release PR automatically.

Verification

• Full test/workos/test_base_client.rb suite passes across multiple seeds (20 runs, 37 assertions).
• Without the lib change, the new regression test fails with the second request reading the first request's abandoned response.
• standardrb clean on both changed files.

[greptile-apps]

Greptile Summary

This backport makes incomplete HTTP requests evict their pooled sockets, adds regression coverage for exceptional and asynchronous abort paths, and retargets release automation to the v8.0.x maintenance branch.

• Wraps Net::HTTP#request with failure-path connection cleanup.
• Adds unit and real-socket regression coverage plus test cache teardown.
• Configures release-please to create v8.0.x release PRs.

Confidence Score: 4/5

The PR appears safe to merge, with a non-blocking opportunity to limit failure cleanup to the specific timeout-keyed connection that failed.

The incomplete-request socket is reliably removed and the release/test changes align with the backport, but endpoint-wide eviction can unnecessarily discard other healthy pooled connections.

Files Needing Attention: lib/workos/base_client.rb

Important Files Changed

Filename	Overview
lib/workos/base_client.rb	Correctly evicts potentially desynchronized sockets on all incomplete exits, but cleanup also closes healthy timeout-specific sibling connections for the endpoint.
test/workos/test_base_client.rb	Adds focused stub and real-socket coverage for eviction and connection-retention behavior, with cache cleanup between tests.
.github/workflows/release-please.yml	Retargets release-please triggers and PR creation to the v8.0.x maintenance branch.

Sequence Diagram

sequenceDiagram
participant App
participant Client as BaseClient
participant Pool as Connection pool
participant API
App->>Client: execute_request
Client->>Pool: connection_for(base, timeout)
Pool-->>Client: cached Net::HTTP
Client->>API: request
API--xClient: request aborted before response returns
Client->>Pool: evict_connection(base)
App->>Client: next request
Client->>Pool: connection_for(base, timeout)
Pool-->>Client: fresh connection

Loading

Prompt To Fix All With AI

### Issue 1
lib/workos/base_client.rb:153
**Eviction drops healthy sibling connections**
When the same fiber has connections to this endpoint cached with different timeout values, an incomplete request calls `evict_connection(base)`, which closes every timeout-specific entry rather than only the failed socket. Subsequent requests using those healthy entries must reconnect, adding connection and TLS setup latency.
---
For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

_{Reviews (1): Last reviewed commit: "ci: run release-please against the v8.0...." | Re-trigger Greptile}

[greptile-apps]

Eviction drops healthy sibling connections

When the same fiber has connections to this endpoint cached with different timeout values, an incomplete request calls evict_connection(base), which closes every timeout-specific entry rather than only the failed socket. Subsequent requests using those healthy entries must reconnect, adding connection and TLS setup latency.

Knowledge Base Used:Core client infrastructure

Prompt To Fix With AI

This is a comment left during a code review.
Path: lib/workos/base_client.rb
Line: 153
Comment:
**Eviction drops healthy sibling connections**
When the same fiber has connections to this endpoint cached with different timeout values, an incomplete request calls `evict_connection(base)`, which closes every timeout-specific entry rather than only the failed socket. Subsequent requests using those healthy entries must reconnect, adding connection and TLS setup latency.
**Knowledge Base Used:**[Core client infrastructure](https://app.greptile.com/workos/-/custom-context/knowledge-base/workos/workos-ruby/-/docs/core-client.md)---
For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.