fix(kanban): extend stale claim instead of killing live worker (salvage #23071)#23442
Merged
Conversation
Workers running slow models (e.g. kimi-k2.6) can spend longer than DEFAULT_CLAIM_TTL_SECONDS inside a single tool-free LLM call, making no tool calls and therefore not heartbeating. release_stale_claims previously reclaimed these healthy workers, producing the spawn-then-immediately-reclaim loop reported in #23025. When a stale-by-TTL claim's host-local worker PID is still alive, extend the claim (emit a claim_extended event) rather than killing it. enforce_max_runtime / detect_crashed_workers remain the upper bounds for genuinely wedged or dead workers. Reclaim events now also record claim_expires, last_heartbeat_at, worker_pid, and host_local so operators can see why a worker was killed.
Contributor
🔎 Lint report:
|
Closed
13 tasks
steveonjava
added a commit
to steveonjava/hermes-agent
that referenced
this pull request
May 24, 2026
`detect_crashed_workers` calls `_pid_alive` on every `running` task whose claim is held by this host. The check can transiently return False for a freshly-spawned worker (fork → /proc-visibility lag, or reap-race between SIGCHLD and parent reaping). When a second dispatcher ticks inside that window it reclaims the task and spawns a duplicate worker. Add `DEFAULT_CRASH_GRACE_SECONDS = 30` and an `HERMES_KANBAN_CRASH_GRACE_SECONDS` env-var override. `detect_crashed_workers` skips the liveness check when `time.time() - started_at < grace`. The existing 15-minute claim TTL still reclaims genuinely-crashed workers; grace only suppresses the launch-window false positive. `HERMES_KANBAN_CRASH_GRACE_SECONDS=0` is set on the `kanban_home` fixture in `test_kanban_core_functionality.py` so existing tests that assert immediate reclaim retain pre-fix semantics. Companion to merged PR NousResearch#23442 (`release_stale_claims`, closes NousResearch#23025), which addressed the same multi-dispatcher race in the stale-claim path. Related: NousResearch#20015 (`_pid_alive` false-negative behaviour), NousResearch#22926 (stale-claim auto-cleanup).
kshitijk4poor
pushed a commit
that referenced
this pull request
May 27, 2026
`detect_crashed_workers` calls `_pid_alive` on every `running` task whose claim is held by this host. The check can transiently return False for a freshly-spawned worker (fork → /proc-visibility lag, or reap-race between SIGCHLD and parent reaping). When a second dispatcher ticks inside that window it reclaims the task and spawns a duplicate worker. Add `DEFAULT_CRASH_GRACE_SECONDS = 30` and an `HERMES_KANBAN_CRASH_GRACE_SECONDS` env-var override. `detect_crashed_workers` skips the liveness check when `time.time() - started_at < grace`. The existing 15-minute claim TTL still reclaims genuinely-crashed workers; grace only suppresses the launch-window false positive. `HERMES_KANBAN_CRASH_GRACE_SECONDS=0` is set on the `kanban_home` fixture in `test_kanban_core_functionality.py` so existing tests that assert immediate reclaim retain pre-fix semantics. Companion to merged PR #23442 (`release_stale_claims`, closes #23025), which addressed the same multi-dispatcher race in the stale-claim path. Related: #20015 (`_pid_alive` false-negative behaviour),
mathias3
pushed a commit
to mathias3/hermes-agent
that referenced
this pull request
May 28, 2026
`detect_crashed_workers` calls `_pid_alive` on every `running` task whose claim is held by this host. The check can transiently return False for a freshly-spawned worker (fork → /proc-visibility lag, or reap-race between SIGCHLD and parent reaping). When a second dispatcher ticks inside that window it reclaims the task and spawns a duplicate worker. Add `DEFAULT_CRASH_GRACE_SECONDS = 30` and an `HERMES_KANBAN_CRASH_GRACE_SECONDS` env-var override. `detect_crashed_workers` skips the liveness check when `time.time() - started_at < grace`. The existing 15-minute claim TTL still reclaims genuinely-crashed workers; grace only suppresses the launch-window false positive. `HERMES_KANBAN_CRASH_GRACE_SECONDS=0` is set on the `kanban_home` fixture in `test_kanban_core_functionality.py` so existing tests that assert immediate reclaim retain pre-fix semantics. Companion to merged PR NousResearch#23442 (`release_stale_claims`, closes NousResearch#23025), which addressed the same multi-dispatcher race in the stale-claim path. Related: NousResearch#20015 (`_pid_alive` false-negative behaviour),
Bryce-huang
pushed a commit
to wbkunlun/hermes-agent
that referenced
this pull request
May 29, 2026
`detect_crashed_workers` calls `_pid_alive` on every `running` task whose claim is held by this host. The check can transiently return False for a freshly-spawned worker (fork → /proc-visibility lag, or reap-race between SIGCHLD and parent reaping). When a second dispatcher ticks inside that window it reclaims the task and spawns a duplicate worker. Add `DEFAULT_CRASH_GRACE_SECONDS = 30` and an `HERMES_KANBAN_CRASH_GRACE_SECONDS` env-var override. `detect_crashed_workers` skips the liveness check when `time.time() - started_at < grace`. The existing 15-minute claim TTL still reclaims genuinely-crashed workers; grace only suppresses the launch-window false positive. `HERMES_KANBAN_CRASH_GRACE_SECONDS=0` is set on the `kanban_home` fixture in `test_kanban_core_functionality.py` so existing tests that assert immediate reclaim retain pre-fix semantics. Companion to merged PR NousResearch#23442 (`release_stale_claims`, closes NousResearch#23025), which addressed the same multi-dispatcher race in the stale-claim path. Related: NousResearch#20015 (`_pid_alive` false-negative behaviour), #AI commit#
mosaiq-systems
pushed a commit
to mosaiq-systems/hermes-agent
that referenced
this pull request
May 29, 2026
`detect_crashed_workers` calls `_pid_alive` on every `running` task whose claim is held by this host. The check can transiently return False for a freshly-spawned worker (fork → /proc-visibility lag, or reap-race between SIGCHLD and parent reaping). When a second dispatcher ticks inside that window it reclaims the task and spawns a duplicate worker. Add `DEFAULT_CRASH_GRACE_SECONDS = 30` and an `HERMES_KANBAN_CRASH_GRACE_SECONDS` env-var override. `detect_crashed_workers` skips the liveness check when `time.time() - started_at < grace`. The existing 15-minute claim TTL still reclaims genuinely-crashed workers; grace only suppresses the launch-window false positive. `HERMES_KANBAN_CRASH_GRACE_SECONDS=0` is set on the `kanban_home` fixture in `test_kanban_core_functionality.py` so existing tests that assert immediate reclaim retain pre-fix semantics. Companion to merged PR NousResearch#23442 (`release_stale_claims`, closes NousResearch#23025), which addressed the same multi-dispatcher race in the stale-claim path. Related: NousResearch#20015 (`_pid_alive` false-negative behaviour),
KKT-OPT
pushed a commit
to KKT-OPT/hermes-agent
that referenced
this pull request
May 31, 2026
`detect_crashed_workers` calls `_pid_alive` on every `running` task whose claim is held by this host. The check can transiently return False for a freshly-spawned worker (fork → /proc-visibility lag, or reap-race between SIGCHLD and parent reaping). When a second dispatcher ticks inside that window it reclaims the task and spawns a duplicate worker. Add `DEFAULT_CRASH_GRACE_SECONDS = 30` and an `HERMES_KANBAN_CRASH_GRACE_SECONDS` env-var override. `detect_crashed_workers` skips the liveness check when `time.time() - started_at < grace`. The existing 15-minute claim TTL still reclaims genuinely-crashed workers; grace only suppresses the launch-window false positive. `HERMES_KANBAN_CRASH_GRACE_SECONDS=0` is set on the `kanban_home` fixture in `test_kanban_core_functionality.py` so existing tests that assert immediate reclaim retain pre-fix semantics. Companion to merged PR NousResearch#23442 (`release_stale_claims`, closes NousResearch#23025), which addressed the same multi-dispatcher race in the stale-claim path. Related: NousResearch#20015 (`_pid_alive` false-negative behaviour),
gweeteve
pushed a commit
to gweeteve/hermes-agent
that referenced
this pull request
Jun 2, 2026
`detect_crashed_workers` calls `_pid_alive` on every `running` task whose claim is held by this host. The check can transiently return False for a freshly-spawned worker (fork → /proc-visibility lag, or reap-race between SIGCHLD and parent reaping). When a second dispatcher ticks inside that window it reclaims the task and spawns a duplicate worker. Add `DEFAULT_CRASH_GRACE_SECONDS = 30` and an `HERMES_KANBAN_CRASH_GRACE_SECONDS` env-var override. `detect_crashed_workers` skips the liveness check when `time.time() - started_at < grace`. The existing 15-minute claim TTL still reclaims genuinely-crashed workers; grace only suppresses the launch-window false positive. `HERMES_KANBAN_CRASH_GRACE_SECONDS=0` is set on the `kanban_home` fixture in `test_kanban_core_functionality.py` so existing tests that assert immediate reclaim retain pre-fix semantics. Companion to merged PR NousResearch#23442 (`release_stale_claims`, closes NousResearch#23025), which addressed the same multi-dispatcher race in the stale-claim path. Related: NousResearch#20015 (`_pid_alive` false-negative behaviour),
alt-glitch
pushed a commit
that referenced
this pull request
Jun 14, 2026
`detect_crashed_workers` calls `_pid_alive` on every `running` task whose claim is held by this host. The check can transiently return False for a freshly-spawned worker (fork → /proc-visibility lag, or reap-race between SIGCHLD and parent reaping). When a second dispatcher ticks inside that window it reclaims the task and spawns a duplicate worker. Add `DEFAULT_CRASH_GRACE_SECONDS = 30` and an `HERMES_KANBAN_CRASH_GRACE_SECONDS` env-var override. `detect_crashed_workers` skips the liveness check when `time.time() - started_at < grace`. The existing 15-minute claim TTL still reclaims genuinely-crashed workers; grace only suppresses the launch-window false positive. `HERMES_KANBAN_CRASH_GRACE_SECONDS=0` is set on the `kanban_home` fixture in `test_kanban_core_functionality.py` so existing tests that assert immediate reclaim retain pre-fix semantics. Companion to merged PR #23442 (`release_stale_claims`, closes #23025), which addressed the same multi-dispatcher race in the stale-claim path. Related: #20015 (`_pid_alive` false-negative behaviour),
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
2 participants
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
Stops the kanban dispatcher from killing healthy workers that are slow. Workers running slow models (kimi-k2.6 was the reported case) can spend longer than the 15-min
DEFAULT_CLAIM_TTL_SECONDSinside a single tool-free LLM call — they make no tool calls, so they don't heartbeat, so the dispatcher used to mark the claim stale and SIGTERM the worker mid-flight. The respawned worker hit the same trap, producing the empty-workspace stall loop reported in #23025.How
release_stale_claims()now checks if the worker's host-local PID is alive before reclaiming. If alive: extend the claim by anotherDEFAULT_CLAIM_TTL_SECONDSand emit aclaim_extendedevent. If dead (or non-host-local): reclaim as before.Upper bounds are unchanged:
enforce_max_runtimestill hard-caps task runtime per themax_runtime_secondscolumn (catches genuinely-stuck-but-PID-alive workers — deadlocks, infinite loops).detect_crashed_workersstill reaps workers whose PID has vanished.The host-local check (
lock.startswith(host_prefix)from_claimer_id().split(":", 1)[0]) means we only trust_pid_alivewhen the lock was set by THIS host. Cross-host claims (rare; happens if you migrate the kanban DB between machines) fall through to the normal reclaim path because we can't safely interpret a PID number from a different host.Changes
hermes_cli/kanban_db.py::release_stale_claims: add the live-PID extension branch with a CAS-guarded UPDATE, run-row sync, andclaim_extendedevent. The reclaim path's payload is also enriched withclaim_expires,last_heartbeat_at,worker_pid,host_local, andnowso operators can tell fromtask_eventswhether a kill was timing-driven or a genuinely-dead worker.tests/hermes_cli/test_kanban_db.py: 2 new tests (test_stale_claim_with_live_pid_extends_instead_of_reclaiming+test_stale_claim_reclaim_event_records_diagnostic_payload) and the existingtest_stale_claim_reclaimedflipped to_pid_alive=Falseso it exercises the still-correct dead-PID reclaim path.Validation
tests/hermes_cli/test_kanban_db.py(stale-claim/reclaim subset)tests/hermes_cli/test_kanban_db.py + test_kanban_core_functionality.py + tests/tools/test_kanban_tools.pyCloses #23025 via salvage. Salvage of #23071. Original commit by @konsisumer cherry-picked with authorship preserved (re-attributed during salvage from
der@konsi.orgto the GitHub-noreply form for release-notes credit). AUTHOR_MAP entry added.