Skip to content

fix: enforce provider cooldown and preserve transport failures - #351

Merged
Moskize91 merged 1 commit into
mainfrom
fix/provider-cooldown-first-failure
Oct 29, 2026
Merged

Moskize91 merged 1 commit into
mainfrom
fix/provider-cooldown-first-failure

Conversation

@Moskize91

@Moskize91 Moskize91 commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

Queued provider calls could bypass a sibling's rate-limit cooldown, and a producer failure could replace an earlier concurrent task failure. Publish cooldowns before releasing capacity, recheck extended cooldowns after queueing, and propagate the execution scope's first failure.

Errored SSE readers now propagate their transport error without misclassifying cancel()'s stored stream error as a cleanup timeout. Add createJsonRequest so embedding hosts can reuse the same full-response retry, deadline, cancellation and retained-capacity policy. JSON parsing and business validation failures do not trigger transport replay.

Validation: 44 runtime regression tests; full workspace regression (228 files, 1537 passed, 4 skipped); workspace lint, typecheck and format checks; workspace build; standalone package installation smoke. The private host's 16 LLM/embedding adapter tests also pass under Node 22 using the current runtime build.

@Moskize91
Moskize91 merged commit bfd8ead into main Oct 29, 2026
22 of 23 checks passed
@Moskize91
Moskize91 deleted the fix/provider-cooldown-first-failure branch October 29, 2026 04:52
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