fix (WMS): reserve the post-processing margin before advertising the slot - #8746
Draft
aldbr wants to merge 1 commit into
Draft
fix (WMS): reserve the post-processing margin before advertising the slot#8746aldbr wants to merge 1 commit into
aldbr wants to merge 1 commit into
Conversation
…slot The JobAgent advertised the whole of the remaining slot to the matcher and published it in /LocalSite/CPUTimeLeft, but the payload never got all of it: the JobWrapper still has to upload the outputs and the logs once the payload stops, and the Watchdog was quietly holding StopMargin back at the far end for that. Nothing upstream knew. A job whose declared CPUTime just fits the slot is matched into it and then killed before it finishes, and an elastic payload sizing itself against /LocalSite/CPUTimeLeft over-commits by the same margin. Either way the work done so far is lost, never having been uploaded. Take the margin off once, in JobAgent.initialize(), where the slot enters the agent. Every later use of it -- the filling-mode check, the CPU work advertised to the matcher, /LocalSite/CPUTimeLeft and so the payload itself -- then works from what a payload may actually consume, without any of them knowing a reserve exists. The Watchdog no longer deducts it, the budget it reads being already net, and stops the payload when that budget is spent. PushJobAgent is left alone: it post-processes on the server, so its remote allocation runs the payload and nothing else. Restore the graceful stop, on that same budget. A payload killed at the end of its budget loses whatever it had produced but not written; an event generator stopped mid-event loses the whole file. The JDL options for this are still parsed, but the code behind them went with MJF in 1968c85 and nothing has sent the signal since. The slot budget is a better clock than MJF was: it is the one __checkTimeLeft already counts down, so the warning and the hard stop cannot disagree. StopSigRegex also matches the command line again, as it did before 0e67f78, so the application is signalled rather than the shells and wrappers sharing its process tree -- and only within this JobWrapper's own descendants, so other jobs on the node are untouched however well they match. Both thresholds are now in CPU work rather than seconds: StopSigStartWork and StopSigFinishWork replace StopSigStartSeconds and StopSigFinishSeconds. How much work a payload has done, and how much it needs to stop, are questions about computation, and the same computation costs more wall clock on a slower node -- so one figure means the same thing across a heterogeneous fleet where a figure in seconds does not. StopSigFinishWork has no default, leaving the graceful stop off until a value is given; StopSigStartWork defaults to it, so a payload is never stopped having done less work than stopping it costs. All the options may now be set in the CS as well as per job, since a JDL is fixed at submission and cannot reach work already in the system. Unchanged for any job that does not ask: the signal is sent only where StopSigRegex and StopSigFinishWork are both set, and only once. Deleted along the way are wallClockCheckSeconds and wallClockCheckCount, which throttled reading MJF's files from disk and have nothing left to throttle. Documented in a new Allocated time management page: where the budget comes from and in which units, who may spend which part of it, and how to measure the values rather than guess them. BEGINRELEASENOTES *WorkloadManagement FIX: the JobAgent reserves the JobWrapper's StopMargin before advertising the slot to the matcher and publishing it in /LocalSite/CPUTimeLeft, so that jobs are neither matched nor sized against time the post-processing will need. The Watchdog no longer deducts it a second time. NEW: the Watchdog can again signal a payload to wind down before its budget runs out (StopSigRegex, StopSigNumber, StopSigStartWork, StopSigFinishWork), now configurable in Systems/WorkloadManagement/<INSTANCE>/JobWrapper as well as per job in the JDL. CHANGE: StopSigStartSeconds and StopSigFinishSeconds are replaced by StopSigStartWork and StopSigFinishWork, expressed in CPU work. CHANGE: new "Allocated time management" page in the Administrator Guide. ENDRELEASENOTES Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MXoFjUCdmqT3oP6FEwc8K7
aldbr
marked this pull request as draft
August 25, 2026 13:30
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
1 participant
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.
Closes #8745
Should be tested with #8528
BEGINRELEASENOTES
*WorkloadManagement
FIX: the JobAgent reserves the JobWrapper's StopMargin before advertising the slot to the matcher and publishing it in /LocalSite/CPUTimeLeft, so that jobs are neither matched nor sized against time the post-processing will need. The Watchdog no longer deducts it a second time.
NEW: the Watchdog can again signal a payload to wind down before its budget runs out (StopSigRegex, StopSigNumber, StopSigStartWork, StopSigFinishWork), now configurable in Systems/WorkloadManagement//JobWrapper as well as per job in the JDL. CHANGE: StopSigStartSeconds and StopSigFinishSeconds are replaced by StopSigStartWork and StopSigFinishWork, expressed in CPU work.
CHANGE: new "Allocated time management" page in the Administrator Guide.
ENDRELEASENOTES