golem15/msd-core
4
0
Fork 0
Code Issues 1 Pull Requests Actions 2 Packages Projects Releases Wiki Activity

fix(#1164): wire external-job config keys + document wave:post choice (#2006)

Browse Source 
Refinements A-D to the external-job capability (PR #1998 follow-up):

A. Document why the contribution registers at execute:wave:post: #1164 asks
   for wave:pre, but execute-phase.md only dispatches wave:post today (wave:pre
   is declared in the loop host contract but not rendered). Wiring wave:pre is
   a core-loop change #1164 puts out of scope; the executor honors the
   runtime_budget classification guidance before running any tagged task.

B. external_job.artifact_dir is now consumed (was declared but unused): the
   adapter resolves it via the canonical capability-config seam and surfaces
   the resolved root in submit output.

C. external_job.submit_timeout_ms / poll_timeout_ms are now read from config
   (were shadowed by env-only reads). Precedence: env > config > registry
   default; non-numeric config values fall back (no guessing, no NaN).

D. CLI surface gains unit coverage: parseFlags, findPlanningDir,
   resolveExternalJobSettings, formatShowReport.

Regenerates capability-registry.cjs from the updated capability.json.
This commit is contained in:
Tom Boucher
2026-07-05 14:04:00 -04:00
committed by GitHub
parent e2bfe4dc67
commit 97730e59a1
7 changed files with 326 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
capabilities/external-job/capability.json
View File

@@ -3,7 +3,7 @@
"role": "feature",
"version": "1.7.0-rc.2",
"title": "Async external-job scheduler adapter",
"description": "Default-off producer of the async external-job manifest (#1164). At execute:wave:post an executor can externalize long-running compute (SLURM first, scheduler-pluggable), commit a .planning/async-jobs/<job>.json manifest, defer SUMMARY.md, and return external_job_waiting. The core loop (#1165) consumes the manifest; this capability is the only thing that writes it.",
"description": "Default-off producer of the async external-job manifest (#1164). At execute:wave:post an executor can externalize long-running compute (SLURM first, scheduler-pluggable), commit a .planning/async-jobs/<job>.json manifest, defer SUMMARY.md, and return external_job_waiting. The core loop (#1165) consumes the manifest; this capability is the only thing that writes it. NOTE on contribution point: #1164 specifies execute:wave:pre, but execute-phase.md only dispatches execute:wave:post today (wave:pre is declared in the loop host contract but not rendered); wiring wave:pre dispatch is a core-loop change #1164 explicitly puts out of scope, so this capability registers at wave:post and the executor honors the runtime_budget classification guidance before running any tagged task. The adapter (scripts/slurm-adapter.cjs) reads external_job.submit_timeout_ms / poll_timeout_ms / artifact_dir through the canonical capability-config seam (env override > config > registry default).",
"tier": "full",
"requires": [],
"engines": {
@@ -35,17 +35,17 @@
"external_job.artifact_dir": {
"type": "string",
"default": "Artifacts/jobs",
"description": "Root for per-job artifact directories (e.g. Artifacts/jobs/<jobid>/). Avoids fixed log paths and hardcoding a cluster/project layout."
"description": "Root for per-job artifact directories (e.g. Artifacts/jobs/<jobid>/). Avoids fixed log paths and hardcoding a cluster/project layout. Surfaced by the adapter (`slurm-adapter.cjs submit` prints the resolved value); override via GSD_EXTERNAL_JOB_ARTIFACT_DIR."
},
"external_job.submit_timeout_ms": {
"type": "number",
"default": 30000,
"description": "Hard timeout (ms) for the scheduler submit subprocess (e.g. sbatch). Bounded per CLAUDE.md unbounded-subprocess policy."
"description": "Hard timeout (ms) for the scheduler submit subprocess (e.g. sbatch). Bounded per CLAUDE.md unbounded-subprocess policy. Read by the adapter (env GSD_SLURM_SUBMIT_TIMEOUT_MS overrides)."
},
"external_job.poll_timeout_ms": {
"type": "number",
"default": 15000,
"description": "Hard timeout (ms) for the scheduler poll subprocess (squeue, with sacct fallback)."
"description": "Hard timeout (ms) for the scheduler poll subprocess (squeue, with sacct fallback). Read by the adapter (env GSD_SLURM_POLL_TIMEOUT_MS overrides)."
}
},
"steps": [],

10
capabilities/external-job/fragments/execute-wave-post.md
View File

@@ -1,4 +1,12 @@
<!-- external-job capability — execute:wave:post fragment, injected into the executor (#1164). -->
<!-- external-job capability — execute:wave:post fragment, injected into the executor (#1164).
Why wave:post, not wave:pre (#1164 refinement A): execute-phase.md only
dispatches execute:wave:post today — wave:pre is declared in the loop host
contract but not rendered. Wiring wave:pre dispatch is a core-loop change
#1164 puts out of scope. The executor therefore honors this classification
guidance BEFORE running any task tagged <runtime_budget>long_compute</runtime_budget>,
whether in the current or a subsequent wave, and externalizes rather than
blocking the turn. -->
## Externalize long-running compute (async external job)