Repository navigation
fix(backend): eval runner opts out of output-style and memory injection - #13414
Merged
diegosouzapw merged 3 commits intoSep 15, 2026
Merged
Conversation
The 'does not duplicate custom Jina specialty models' test asserted
catalog IDs prefixed with 'jina-ai/' (the connection/provider ID), but
the catalog builder uses the provider alias 'jina' — which was
introduced in v3.8.36. The test above it in the same file ('does not
duplicate imported Jina specialty models') already asserts 'jina/' and
passes; this second test was simply never updated.
Fix the two assert.equal strings from 'jina-ai/' to 'jina/' to match
the current product behavior.
Fixes diegosouzapw#13313 (Jina portion — pt-BR compression integration tests need
separate investigation)
…iegosouzapw#13308) The VACUUM INTO path in createManagedDbBackup() — reached from the health-check-repair flow — never called the retention helper, so db_backups/ grew without bound (observed: 10+ GB across 18 snapshots in 8 days). The sibling path (backupDbFile → backup.ts) already calls cleanupDbBackups() after each snapshot; the health-check path simply forgot to do the same. Fix: after a successful VACUUM INTO in createManagedDbBackup(), call pruneBackupDirectory() from backupRetention.ts using the same env-var precedence (DB_BACKUP_MAX_FILES / DB_BACKUP_RETENTION_DAYS) as the manual/scheduled backup path. Pruning is wrapped in a try/catch so a best-effort failure never obscures the backup result. Regression test: db-backup-healthcheck-prune-13308.test.ts seeds a backup directory with MAX_DB_BACKUPS + 5 families and asserts that pruneBackupDirectory removes exactly the overflow.
…on (diegosouzapw#13139) The eval runner sent cases down the ordinary chat path, so every case picked up output-style persona messages and memory context/tools. There was no way to opt out, so evals measured injected context as much as the model. Added x-omniroute-compression: off and x-omniroute-no-memory: true headers to the eval runner's request, matching the documented opt-out mechanism used by self-managed clients. Added 3 regression tests verifying the headers are set correctly. Fixes diegosouzapw#13139
diegosouzapw
merged commit Sep 15, 2026
76c928d
into
diegosouzapw:release/v3.8.51
8 of 16 checks passed
muhamadgalihsaputra
pushed a commit
to niyatna/NiyatnaRoute
that referenced
this pull request
Sep 27, 2026
…on (diegosouzapw#13414) The eval runner sends `x-omniroute-compression: off` and `x-omniroute-no-memory: true`, so cases measure the model rather than injected output styles or retrieved memory (diegosouzapw#13139). Both headers are the existing per-request opt-outs honored by `chatCore` (`open-sse/handlers/chatCore/headers.ts`). Validated in one consolidated batch of this series (37 PRs boarded together on `release/v3.8.51`): `typecheck:core`, `check:open-sse-typecheck` and `check:dashboard-typecheck` clean; ESLint clean on every changed file; file-size, complexity, cognitive-complexity, changelog-integrity, docs-counts, docs-sync and migration-numbering gates green (only the pre-existing `open-sse/utils/stream.ts` file-size red remains, inherited from the base); 3,743 focused `node:test` cases plus 34 vitest cases green. Thanks @KooshaPari!
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The eval runner now opts out of output-style and memory injection so that eval cases measure the model, not injected context.
Problem
executeEvalCase()sent cases down the ordinary chat path with onlyContent-Typeand optionallyAuthorizationheaders. Three injections applied:memory_*tools appended to the requestThese are documented opt-outs via
x-omniroute-compression: offandx-omniroute-no-memory: true, but the eval runner set neither.Effect: pass rates varied wildly depending on output-style selection and API key presence, measuring injected context rather than model capability.
Fix
Added both opt-out headers to the eval runner's request:
x-omniroute-compression: off— disables output-style persona injectionx-omniroute-no-memory: true— disables memory context and memory tools injectionTest results
Fixes #13139