You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
AKO-147 Bound Slurm inspection and honor terminal sentinel before accounting queries #147
Fresh Montage run montage-clean-real-20261011-v1 has a physical mProject1 terminal sentinel with exit_code=0 and runtime about479s while API still reports running well afterwards. Slurm inspection reads sentinel then unconditionally runs sacct, squeue and scontrol without bounded remote deadlines; inspectActivities waits for every worker before consuming any result. SSH inspection warnings are intermittent and must not be confused with failed jobs or trigger duplicates. Diagnose the active transport/accounting latency; honor trustworthy terminal sentinels before querying accounting, bound read-only inspection requests, return transient warnings without treating timeouts as terminal, and test that unavailable/hung accounting cannot hide a completed sentinel or block all other handles. Keep artifact/output validation, metrics and exact attempt identity. Do not deploy by restarting live activity submissions without a verified safe point. Local main commits only, no push.
Read-only Slurm inspection is bounded and terminal sentinel/manifests take precedence over unavailable accounting; provider tests pass. Located through AKO issue-number commits 026ff18 and verified integrated in published main 4aecc4c via PR #175. Full Go suite was rerun on main successfully for this audit; PR Go/docs CI is green. Closure covers this implementation/investigation scope only; remaining scientific forecast gates are tracked separately.
Activity
ovvesley commentedon Oct 14, 2026
Read-only Slurm inspection is bounded and terminal sentinel/manifests take precedence over unavailable accounting; provider tests pass. Located through AKO issue-number commits 026ff18 and verified integrated in published main 4aecc4c via PR #175. Full Go suite was rerun on main successfully for this audit; PR Go/docs CI is green. Closure covers this implementation/investigation scope only; remaining scientific forecast gates are tracked separately.