Skip to content

[PR] fix: retain active lanes during conversation recall eviction - #6997

Closed
Afloat16 wants to merge 3 commits into
ranxianglei:masterfrom
Afloat16:2026-11-17_recall-lane-retention
Closed

Afloat16 wants to merge 3 commits into
ranxianglei:masterfrom
Afloat16:2026-11-17_recall-lane-retention

Conversation

@Afloat16

Copy link
Copy Markdown
Contributor

Problem

Global conversation-recall eviction chooses the oldest individual session, so an idle sibling can evict a lane containing a recently learned or recalled session. Record-only traffic also leaves 65 lanes until the next operation.

Change

Compare each lane by its most recent session activity, then enforce the existing 64-lane limit immediately after learning. TTL, unique-boundary matching, and the per-lane session limit stay unchanged.

Verification

  • Base: 301e1a5459a77d5578206927c8062a58381af642 (master, 0.1.215).
  • Three new deterministic tests fail on that base.
  • Recall retention, existing conversation recall, tool-call fingerprint and dsh integration suites: 18 tests passed.
  • npm run typecheck, npm run build, and git diff --check: passed.
  • No kernel, wire, config, release, or dependency changes. Human merge tier applies to conversation identity.

Closes #6995

@ranxianglei ranxianglei changed the title fix: retain active lanes during conversation recall eviction [PR] fix: retain active lanes during conversation recall eviction Nov 17, 2026
ework-agent and others added 2 commits November 17, 2026 11:07
…sation

The fix adds a trailing sweep(now) that enforces the 64-lane limit and
refreshes stats after the insertion. The leading sweep(now) ran the same
TTL + lane-limit + stats work with the same `now` moments earlier, so it
is redundant. Removing it keeps one sweep per learn (the hot path taken on
every attributed dispatch) with no behavior change: the per-lane LRU plus
the trailing sweep reach the identical final state (verified: 18/18
targeted recall/fingerprint/dsh tests pass with and without it).
Merge upstream master 6a641f3, including the architecture module-map and schedule-independent ranxianglei#2929 probe fixes.

Preserve the existing PR head and exact contribution patch. Validation of this refreshed head runs in upstream CI; no local test rerun is claimed.
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

Re-verified live check state before touching anything: the three failing checks in this wake (test (windows-latest, 22, 4), test (ubuntu-latest, 24, 3), test (ubuntu-latest, 22, 3)) were from the older head 3c9d19b37a (run 44008094246). The branch head has since moved to de81b162 ("fix: sync PR #6997 with merged upstream CI fixes", merging upstream master 6a641f37), and the full CI run on that head (44050511981) is all green — 50/50 check runs completed with success/skipped (verified via the live check-runs API for the head sha; the mirror cache was stale at ~3h and cross-checked once).

No merge conflict (mergeable_state: behind — branch is behind master, which is not a blocker). No [ework:ci-held] marker on the head commit. Nothing to fix this run — stopping here.

@ranxianglei

Copy link
Copy Markdown
Owner

This fix is included verbatim (original commits and authorship preserved) in the merged batch PR #7074 — the changes are already on master. Closing to avoid a duplicate merge. 该修复已随批量 PR #7074 原样合入 master,为避免重复合并,关闭此 PR。

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.

Conversation recall evicts an active lane when it contains an older sibling

2 participants