Kandev: ACP session/new stream-not-connected failure followed by delivery-checkpoint-missing on resume #85
Assignees
dreampiler (dreampiler)
Clear assignees
dreampiler (dreampiler)
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: dreampiler/kandev#85
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Ops (Codex) observation 2026-10-14 14:15 KST: three AIWS sessions (09258d6e/81576ff8, ab93eca0/28467289, 09258d6e/6e207469) failed at start with 'session/new failed: agent.session.new request failed: agent stream not connected while waiting for response' and later on resume with 'agent delivery checkpoint missing'; all FAILED. Determine causality from logs/code, recommend resume behaviour for checkpoint-less sessions (resume vs fresh_start vs explicit failure), state whether AIWS watchdog b34bf7d8 needs an item, and propose a fix with R1 upstream comparison. Kandev task 30da0ef6 under root 8e3eb6b7.
PR-1 merged: dreampiler/kandev#311 (head fe5ba6fc39, squash 2ba5d50f7c). Removed the isResumedSession exclusion in canReplayFreshBootstrapEvents so a resume with no committed native conversation replays the retained generation-1 startup events as a fresh bootstrap. Narrow lifecycle tests, go vet, go build, golangci-lint pass. Remaining per controller: PR-2 (admission-deadline transport hardening, after PR-1) and PR-3/c258. Follow-up: keep PR-2/PR-3 on this task plan checklist.
KANDEV-FU-1014-ACP-STREAM (Kandev task 30da0ef6). Causal verdict confirmed: the startup session/new failure and the later resume checkpoint-missing are one causal chain. Merged: PR #311 2ba5d50f7c (resume admit, checkpoint-less fresh_start), PR #320 9f005901b3 (bounded durable admission-insert retry, 3x/200ms), PR #328 62191089a6 (teardown-time startup SetModel failure classified as transport, not application_failed). Remaining: PR-3 reduces the misclassification but does not remove the startup-step/teardown race; start_model.go is upstream-same -> upstream transplant candidate (W091 handoff); the operational checkpoint recovery stays with existing owners/AIWS.
KANDEV-FU-1014-ACP-STREAM final: causal verdict confirmed (startup session/new failure and resume checkpoint-missing are one chain); resume recommendation fresh_start; b34 monitoring item present. Merged: PR #311 2ba5d50f7c (resume admit), PR #320 9f005901b3 (durable admission bounded retry), PR #328 62191089a6 (SetModel transport misclassification), PR #331 344e29335f2 (execution cleanup ownership: reaper completed-claim + no-runtime stop settle). Installed 94e2cfbf minimal hotfix candidate for the two cleanup surfaces prepared (branch hotfix/94-reaper-noruntime-stop, head 4b445ee02f, task_operations.go +84/-19, hooks passed) - install/apply is the operations approval boundary. Remaining: #328 upstream prep via 93c6b4d0 -> W-091 (condition #331 merged recorded, not started here); unresolved_durable_work/dynamic_route_pending residue belongs to 44e94a1b/44e scope.