agent-hitch 0.2.6 → 0.2.7
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +14 -1
- package/dist/scripts/canary-buildkit-secrets.js +166 -0
- package/dist/scripts/canary-buildkit-secrets.js.map +1 -0
- package/dist/scripts/canary-collision-domains.js +77 -0
- package/dist/scripts/canary-collision-domains.js.map +1 -0
- package/dist/scripts/canary-docker-image-observation.js +181 -0
- package/dist/scripts/canary-docker-image-observation.js.map +1 -0
- package/dist/scripts/canary-gpu-hardware.js +95 -0
- package/dist/scripts/canary-gpu-hardware.js.map +1 -0
- package/dist/scripts/canary-harbor-load.js +365 -0
- package/dist/scripts/canary-harbor-load.js.map +1 -0
- package/dist/scripts/canary-packaged-harness.js +75 -0
- package/dist/scripts/canary-packaged-harness.js.map +1 -0
- package/dist/scripts/canary-resource-load.js +188 -0
- package/dist/scripts/canary-resource-load.js.map +1 -0
- package/dist/scripts/check-architecture.js +8 -4
- package/dist/scripts/check-architecture.js.map +1 -1
- package/dist/scripts/docker-reaper-crash-child.js +21 -0
- package/dist/scripts/docker-reaper-crash-child.js.map +1 -0
- package/dist/src/adapters/catalog.js +1 -0
- package/dist/src/adapters/catalog.js.map +1 -1
- package/dist/src/adapters/discovery.js +1 -1
- package/dist/src/adapters/discovery.js.map +1 -1
- package/dist/src/adapters/providers/claude.js +6 -0
- package/dist/src/adapters/providers/claude.js.map +1 -1
- package/dist/src/adapters/providers/codex.js +6 -0
- package/dist/src/adapters/providers/codex.js.map +1 -1
- package/dist/src/adapters/providers/deepseek.js +6 -0
- package/dist/src/adapters/providers/deepseek.js.map +1 -1
- package/dist/src/adapters/providers/opencode.js +6 -0
- package/dist/src/adapters/providers/opencode.js.map +1 -1
- package/dist/src/adapters/providers/pi.js +6 -0
- package/dist/src/adapters/providers/pi.js.map +1 -1
- package/dist/src/artifacts/index.js +1 -0
- package/dist/src/artifacts/index.js.map +1 -1
- package/dist/src/backends/harbor/backend.js +53 -103
- package/dist/src/backends/harbor/backend.js.map +1 -1
- package/dist/src/backends/harbor/environment-config.js +92 -0
- package/dist/src/backends/harbor/environment-config.js.map +1 -0
- package/dist/src/backends/harbor/index.js +5 -1
- package/dist/src/backends/harbor/index.js.map +1 -1
- package/dist/src/backends/harbor/model-proxy-config.js +23 -0
- package/dist/src/backends/harbor/model-proxy-config.js.map +1 -0
- package/dist/src/backends/harbor/process.js +177 -0
- package/dist/src/backends/harbor/process.js.map +1 -0
- package/dist/src/backends/harbor/runtime-toolchain.js +30 -0
- package/dist/src/backends/harbor/runtime-toolchain.js.map +1 -0
- package/dist/src/backends/harbor/supervisor.js +74 -0
- package/dist/src/backends/harbor/supervisor.js.map +1 -0
- package/dist/src/backends/harbor/task-resources.js +206 -0
- package/dist/src/backends/harbor/task-resources.js.map +1 -0
- package/dist/src/backends/harbor/tools.js +6 -29
- package/dist/src/backends/harbor/tools.js.map +1 -1
- package/dist/src/backends/index.js +4 -1
- package/dist/src/backends/index.js.map +1 -1
- package/dist/src/cli/commands/daemon.js +135 -8
- package/dist/src/cli/commands/daemon.js.map +1 -1
- package/dist/src/cli/commands/eval.js +214 -4
- package/dist/src/cli/commands/eval.js.map +1 -1
- package/dist/src/cli/commands/images.js +40 -0
- package/dist/src/cli/commands/images.js.map +1 -0
- package/dist/src/cli/commands/run.js +6 -3
- package/dist/src/cli/commands/run.js.map +1 -1
- package/dist/src/cli/commands/runs.js +25 -3
- package/dist/src/cli/commands/runs.js.map +1 -1
- package/dist/src/cli/commands/worker.js +79 -0
- package/dist/src/cli/commands/worker.js.map +1 -0
- package/dist/src/cli/main.js +4 -0
- package/dist/src/cli/main.js.map +1 -1
- package/dist/src/cli/output.js +71 -3
- package/dist/src/cli/output.js.map +1 -1
- package/dist/src/control-plane/build-admission.js +69 -0
- package/dist/src/control-plane/build-admission.js.map +1 -0
- package/dist/src/control-plane/build-records.js +6 -0
- package/dist/src/control-plane/build-records.js.map +1 -0
- package/dist/src/control-plane/collisions.js +40 -0
- package/dist/src/control-plane/collisions.js.map +1 -0
- package/dist/src/control-plane/eval-control-state.js +22 -0
- package/dist/src/control-plane/eval-control-state.js.map +1 -0
- package/dist/src/control-plane/eval-control-work.js +39 -0
- package/dist/src/control-plane/eval-control-work.js.map +1 -0
- package/dist/src/control-plane/eval-image-resolution.js +52 -0
- package/dist/src/control-plane/eval-image-resolution.js.map +1 -0
- package/dist/src/control-plane/eval-image-services.js +30 -0
- package/dist/src/control-plane/eval-image-services.js.map +1 -0
- package/dist/src/control-plane/eval-records.js +252 -0
- package/dist/src/control-plane/eval-records.js.map +1 -0
- package/dist/src/control-plane/eval-recovery.js +181 -0
- package/dist/src/control-plane/eval-recovery.js.map +1 -0
- package/dist/src/control-plane/eval-scheduler.js +447 -0
- package/dist/src/control-plane/eval-scheduler.js.map +1 -0
- package/dist/src/control-plane/index.js +20 -0
- package/dist/src/control-plane/index.js.map +1 -0
- package/dist/src/control-plane/local-worker.js +49 -0
- package/dist/src/control-plane/local-worker.js.map +1 -0
- package/dist/src/control-plane/model-capture-planning.js +12 -0
- package/dist/src/control-plane/model-capture-planning.js.map +1 -0
- package/dist/src/control-plane/remote-result-transport.js +172 -0
- package/dist/src/control-plane/remote-result-transport.js.map +1 -0
- package/dist/src/control-plane/remote-work-coordinator.js +332 -0
- package/dist/src/control-plane/remote-work-coordinator.js.map +1 -0
- package/dist/src/control-plane/remote-work-inputs.js +266 -0
- package/dist/src/control-plane/remote-work-inputs.js.map +1 -0
- package/dist/src/control-plane/remote-work-recovery.js +181 -0
- package/dist/src/control-plane/remote-work-recovery.js.map +1 -0
- package/dist/src/control-plane/remote-worker-artifacts.js +135 -0
- package/dist/src/control-plane/remote-worker-artifacts.js.map +1 -0
- package/dist/src/control-plane/remote-worker-client.js +202 -0
- package/dist/src/control-plane/remote-worker-client.js.map +1 -0
- package/dist/src/control-plane/remote-worker-configuration.js +34 -0
- package/dist/src/control-plane/remote-worker-configuration.js.map +1 -0
- package/dist/src/control-plane/remote-worker-credentials.js +58 -0
- package/dist/src/control-plane/remote-worker-credentials.js.map +1 -0
- package/dist/src/control-plane/remote-worker-protocol.js +510 -0
- package/dist/src/control-plane/remote-worker-protocol.js.map +1 -0
- package/dist/src/control-plane/remote-worker-runner.js +214 -0
- package/dist/src/control-plane/remote-worker-runner.js.map +1 -0
- package/dist/src/control-plane/remote-workers.js +377 -0
- package/dist/src/control-plane/remote-workers.js.map +1 -0
- package/dist/src/control-plane/rerun-scheduler.js +411 -0
- package/dist/src/control-plane/rerun-scheduler.js.map +1 -0
- package/dist/src/control-plane/rerun-submission.js +56 -0
- package/dist/src/control-plane/rerun-submission.js.map +1 -0
- package/dist/src/control-plane/resources.js +182 -0
- package/dist/src/control-plane/resources.js.map +1 -0
- package/dist/src/control-plane/scheduler-eval-entry.js +32 -0
- package/dist/src/control-plane/scheduler-eval-entry.js.map +1 -0
- package/dist/src/control-plane/synthetic-result.js +21 -0
- package/dist/src/control-plane/synthetic-result.js.map +1 -0
- package/dist/src/control-plane/work-admission.js +21 -0
- package/dist/src/control-plane/work-admission.js.map +1 -0
- package/dist/src/control-plane/work-dispatcher.js +229 -0
- package/dist/src/control-plane/work-dispatcher.js.map +1 -0
- package/dist/src/controller-runtime/hash.js +12 -1
- package/dist/src/controller-runtime/hash.js.map +1 -1
- package/dist/src/controller-runtime/index.js +1 -1
- package/dist/src/controller-runtime/index.js.map +1 -1
- package/dist/src/controller-runtime/store.js +15 -6
- package/dist/src/controller-runtime/store.js.map +1 -1
- package/dist/src/daemon/client.js.map +1 -1
- package/dist/src/daemon/event-stream.js +46 -0
- package/dist/src/daemon/event-stream.js.map +1 -0
- package/dist/src/daemon/health.js +66 -0
- package/dist/src/daemon/health.js.map +1 -0
- package/dist/src/daemon/launcher.js +28 -1
- package/dist/src/daemon/launcher.js.map +1 -1
- package/dist/src/daemon/logging.js +23 -0
- package/dist/src/daemon/logging.js.map +1 -0
- package/dist/src/daemon/scheduler.js +26 -5
- package/dist/src/daemon/scheduler.js.map +1 -1
- package/dist/src/daemon/server.js +255 -98
- package/dist/src/daemon/server.js.map +1 -1
- package/dist/src/daemon/telemetry.js +163 -0
- package/dist/src/daemon/telemetry.js.map +1 -0
- package/dist/src/daemon/worker-routes.js +219 -0
- package/dist/src/daemon/worker-routes.js.map +1 -0
- package/dist/src/domain/bundles.js +2 -0
- package/dist/src/domain/bundles.js.map +1 -0
- package/dist/src/domain/control-plane.js +2 -0
- package/dist/src/domain/control-plane.js.map +1 -0
- package/dist/src/domain/execution-evidence.js +2 -0
- package/dist/src/domain/execution-evidence.js.map +1 -0
- package/dist/src/domain/execution-plan.js +2 -0
- package/dist/src/domain/execution-plan.js.map +1 -0
- package/dist/src/domain/images.js +2 -0
- package/dist/src/domain/images.js.map +1 -0
- package/dist/src/domain/index.js +9 -0
- package/dist/src/domain/index.js.map +1 -1
- package/dist/src/domain/interactions.js +2 -0
- package/dist/src/domain/interactions.js.map +1 -0
- package/dist/src/domain/providers.js +2 -0
- package/dist/src/domain/providers.js.map +1 -0
- package/dist/src/domain/resources.js +2 -0
- package/dist/src/domain/resources.js.map +1 -0
- package/dist/src/domain/workers.js +2 -0
- package/dist/src/domain/workers.js.map +1 -0
- package/dist/src/evals/collect-only-rerun.js +182 -0
- package/dist/src/evals/collect-only-rerun.js.map +1 -0
- package/dist/src/evals/directory.js +30 -0
- package/dist/src/evals/directory.js.map +1 -0
- package/dist/src/evals/docker-engine-stats.js +89 -0
- package/dist/src/evals/docker-engine-stats.js.map +1 -0
- package/dist/src/evals/docker-ownership.js +51 -0
- package/dist/src/evals/docker-ownership.js.map +1 -0
- package/dist/src/evals/docker-reaper.js +144 -0
- package/dist/src/evals/docker-reaper.js.map +1 -0
- package/dist/src/evals/docker-resource-observer.js +207 -0
- package/dist/src/evals/docker-resource-observer.js.map +1 -0
- package/dist/src/evals/environment-image-planning.js +164 -0
- package/dist/src/evals/environment-image-planning.js.map +1 -0
- package/dist/src/evals/eval-artifact-planning.js +56 -0
- package/dist/src/evals/eval-artifact-planning.js.map +1 -0
- package/dist/src/evals/eval-finalization.js +10 -0
- package/dist/src/evals/eval-finalization.js.map +1 -0
- package/dist/src/evals/eval-lifecycle-events.js +16 -0
- package/dist/src/evals/eval-lifecycle-events.js.map +1 -0
- package/dist/src/evals/events.js +53 -5
- package/dist/src/evals/events.js.map +1 -1
- package/dist/src/evals/execution-evidence.js +116 -0
- package/dist/src/evals/execution-evidence.js.map +1 -0
- package/dist/src/evals/execution-leases.js +342 -0
- package/dist/src/evals/execution-leases.js.map +1 -0
- package/dist/src/evals/execution-plan-artifacts.js +72 -0
- package/dist/src/evals/execution-plan-artifacts.js.map +1 -0
- package/dist/src/evals/execution-plan-images.js +69 -0
- package/dist/src/evals/execution-plan-images.js.map +1 -0
- package/dist/src/evals/execution-plan-resources.js +174 -0
- package/dist/src/evals/execution-plan-resources.js.map +1 -0
- package/dist/src/evals/execution-plan.js +415 -0
- package/dist/src/evals/execution-plan.js.map +1 -0
- package/dist/src/evals/harbor-artifact-builder-image.js +99 -0
- package/dist/src/evals/harbor-artifact-builder-image.js.map +1 -0
- package/dist/src/evals/harbor-artifact-builder.js +354 -0
- package/dist/src/evals/harbor-artifact-builder.js.map +1 -0
- package/dist/src/evals/harbor-bridge-error.js +18 -3
- package/dist/src/evals/harbor-bridge-error.js.map +1 -1
- package/dist/src/evals/harbor-phase-timings.js +60 -0
- package/dist/src/evals/harbor-phase-timings.js.map +1 -0
- package/dist/src/evals/index.js +25 -2
- package/dist/src/evals/index.js.map +1 -1
- package/dist/src/evals/infrastructure-retry.js +100 -63
- package/dist/src/evals/infrastructure-retry.js.map +1 -1
- package/dist/src/evals/interaction-capture-import.js +49 -0
- package/dist/src/evals/interaction-capture-import.js.map +1 -0
- package/dist/src/evals/local-docker-provider.js +372 -0
- package/dist/src/evals/local-docker-provider.js.map +1 -0
- package/dist/src/evals/local-eval-planning.js +80 -0
- package/dist/src/evals/local-eval-planning.js.map +1 -0
- package/dist/src/evals/model-capture-plan.js +22 -0
- package/dist/src/evals/model-capture-plan.js.map +1 -0
- package/dist/src/evals/model-capture-runtime.js +114 -0
- package/dist/src/evals/model-capture-runtime.js.map +1 -0
- package/dist/src/evals/model-proxy-runtime-state.js +53 -0
- package/dist/src/evals/model-proxy-runtime-state.js.map +1 -0
- package/dist/src/evals/planned-execution-support.js +54 -0
- package/dist/src/evals/planned-execution-support.js.map +1 -0
- package/dist/src/evals/planned-execution.js +461 -0
- package/dist/src/evals/planned-execution.js.map +1 -0
- package/dist/src/evals/planned-remote-execution.js +40 -0
- package/dist/src/evals/planned-remote-execution.js.map +1 -0
- package/dist/src/evals/planned-retry-lifecycle.js +186 -0
- package/dist/src/evals/planned-retry-lifecycle.js.map +1 -0
- package/dist/src/evals/prepared-harness.js +76 -0
- package/dist/src/evals/prepared-harness.js.map +1 -0
- package/dist/src/evals/progress.js +4 -4
- package/dist/src/evals/progress.js.map +1 -1
- package/dist/src/evals/records.js +20 -3
- package/dist/src/evals/records.js.map +1 -1
- package/dist/src/evals/recovery.js +192 -0
- package/dist/src/evals/recovery.js.map +1 -0
- package/dist/src/evals/remote-infrastructure-retry.js +99 -0
- package/dist/src/evals/remote-infrastructure-retry.js.map +1 -0
- package/dist/src/evals/rerun-inputs.js +89 -0
- package/dist/src/evals/rerun-inputs.js.map +1 -0
- package/dist/src/evals/rerun-slots.js +4 -3
- package/dist/src/evals/rerun-slots.js.map +1 -1
- package/dist/src/evals/rerun-types.js +49 -0
- package/dist/src/evals/rerun-types.js.map +1 -0
- package/dist/src/evals/rerun.js +113 -75
- package/dist/src/evals/rerun.js.map +1 -1
- package/dist/src/evals/result-helpers.js +65 -0
- package/dist/src/evals/result-helpers.js.map +1 -0
- package/dist/src/evals/resume-state.js +36 -0
- package/dist/src/evals/resume-state.js.map +1 -0
- package/dist/src/evals/service-types.js +2 -0
- package/dist/src/evals/service-types.js.map +1 -0
- package/dist/src/evals/service.js +292 -257
- package/dist/src/evals/service.js.map +1 -1
- package/dist/src/evals/task-resources.js +209 -0
- package/dist/src/evals/task-resources.js.map +1 -0
- package/dist/src/evals/trial-environment-evidence.js +87 -0
- package/dist/src/evals/trial-environment-evidence.js.map +1 -0
- package/dist/src/evals/trial-execution-evidence.js +13 -0
- package/dist/src/evals/trial-execution-evidence.js.map +1 -0
- package/dist/src/evals/trial-import-identity.js +31 -0
- package/dist/src/evals/trial-import-identity.js.map +1 -0
- package/dist/src/evals/trial-import.js +143 -145
- package/dist/src/evals/trial-import.js.map +1 -1
- package/dist/src/evals/trial-publication-recovery.js +104 -0
- package/dist/src/evals/trial-publication-recovery.js.map +1 -0
- package/dist/src/evals/trial-publication.js +25 -0
- package/dist/src/evals/trial-publication.js.map +1 -0
- package/dist/src/evals/work-item-artifacts.js +23 -0
- package/dist/src/evals/work-item-artifacts.js.map +1 -0
- package/dist/src/foundation/config.js +17 -0
- package/dist/src/foundation/config.js.map +1 -1
- package/dist/src/foundation/credential-redaction.js +137 -0
- package/dist/src/foundation/credential-redaction.js.map +1 -0
- package/dist/src/foundation/environment-image-references.js +59 -0
- package/dist/src/foundation/environment-image-references.js.map +1 -0
- package/dist/src/foundation/executable.js +45 -5
- package/dist/src/foundation/executable.js.map +1 -1
- package/dist/src/foundation/index.js +5 -2
- package/dist/src/foundation/index.js.map +1 -1
- package/dist/src/foundation/process-identity.js +69 -0
- package/dist/src/foundation/process-identity.js.map +1 -0
- package/dist/src/images/context.js +47 -0
- package/dist/src/images/context.js.map +1 -0
- package/dist/src/images/docker-buildkit.js +137 -0
- package/dist/src/images/docker-buildkit.js.map +1 -0
- package/dist/src/images/dockerfile.js +67 -0
- package/dist/src/images/dockerfile.js.map +1 -0
- package/dist/src/images/gc.js +261 -0
- package/dist/src/images/gc.js.map +1 -0
- package/dist/src/images/index.js +10 -0
- package/dist/src/images/index.js.map +1 -0
- package/dist/src/images/manifest.js +109 -0
- package/dist/src/images/manifest.js.map +1 -0
- package/dist/src/images/ownership.js +5 -0
- package/dist/src/images/ownership.js.map +1 -0
- package/dist/src/images/records.js +32 -0
- package/dist/src/images/records.js.map +1 -0
- package/dist/src/images/registry.js +154 -0
- package/dist/src/images/registry.js.map +1 -0
- package/dist/src/images/service.js +224 -0
- package/dist/src/images/service.js.map +1 -0
- package/dist/src/model-access/capture.js +255 -0
- package/dist/src/model-access/capture.js.map +1 -0
- package/dist/src/model-access/index.js +5 -0
- package/dist/src/model-access/index.js.map +1 -0
- package/dist/src/model-access/policy.js +49 -0
- package/dist/src/model-access/policy.js.map +1 -0
- package/dist/src/model-access/proxy.js +336 -0
- package/dist/src/model-access/proxy.js.map +1 -0
- package/dist/src/model-access/records.js +167 -0
- package/dist/src/model-access/records.js.map +1 -0
- package/dist/src/runs/bundle.js +450 -0
- package/dist/src/runs/bundle.js.map +1 -0
- package/dist/src/runs/events.js +12 -1
- package/dist/src/runs/events.js.map +1 -1
- package/dist/src/runs/executor.js +24 -17
- package/dist/src/runs/executor.js.map +1 -1
- package/dist/src/runs/index.js +3 -1
- package/dist/src/runs/index.js.map +1 -1
- package/dist/src/runs/manifest.js +5 -5
- package/dist/src/runs/manifest.js.map +1 -1
- package/dist/src/runs/request.js +6 -1
- package/dist/src/runs/request.js.map +1 -1
- package/dist/src/runs/training-candidate.js +161 -0
- package/dist/src/runs/training-candidate.js.map +1 -0
- package/dist/src/trajectories/provider-capture.js +15 -24
- package/dist/src/trajectories/provider-capture.js.map +1 -1
- package/dist/src/trajectories/providers/deepseek.js +3 -3
- package/dist/src/trajectories/providers/deepseek.js.map +1 -1
- package/dist/src/workers/index.js +3 -0
- package/dist/src/workers/index.js.map +1 -0
- package/dist/src/workers/remote-harbor-work-spec.js +97 -0
- package/dist/src/workers/remote-harbor-work-spec.js.map +1 -0
- package/dist/src/workers/remote-harbor-worker.js +220 -0
- package/dist/src/workers/remote-harbor-worker.js.map +1 -0
- package/docs/schemas/environment-build-index.schema.json +13 -0
- package/docs/schemas/environment-build-record.schema.json +25 -0
- package/docs/schemas/environment-image-manifest.schema.json +62 -0
- package/docs/schemas/eval-control.schema.json +36 -0
- package/docs/schemas/eval-execution-plan.schema.json +216 -0
- package/docs/schemas/eval-rerun-result.schema.json +53 -0
- package/docs/schemas/eval-rerun-state.schema.json +61 -0
- package/docs/schemas/eval-rerun-submission.schema.json +58 -0
- package/docs/schemas/eval-result.schema.json +34 -1
- package/docs/schemas/eval-submission-input.schema.json +14 -0
- package/docs/schemas/eval-submission.schema.json +68 -0
- package/docs/schemas/eval-trial-publication.schema.json +39 -0
- package/docs/schemas/execution-evidence.schema.json +96 -0
- package/docs/schemas/execution-lease.schema.json +50 -0
- package/docs/schemas/execution-provider-status.schema.json +65 -0
- package/docs/schemas/execution-worker.schema.json +53 -0
- package/docs/schemas/harbor-process-exit.schema.json +14 -0
- package/docs/schemas/interaction-capture-ref.schema.json +45 -0
- package/docs/schemas/local-provider-execution.schema.json +33 -0
- package/docs/schemas/model-interaction.schema.json +41 -0
- package/docs/schemas/remote-result-envelope.schema.json +32 -0
- package/docs/schemas/remote-tree-envelope.schema.json +25 -0
- package/docs/schemas/remote-work-offer.schema.json +58 -0
- package/docs/schemas/remote-work-receipt.schema.json +50 -0
- package/docs/schemas/remote-worker-artifact.schema.json +17 -0
- package/docs/schemas/remote-worker-event.schema.json +18 -0
- package/docs/schemas/remote-worker-heartbeat.schema.json +37 -0
- package/docs/schemas/remote-worker-registration.schema.json +52 -0
- package/docs/schemas/result-bundle-index.schema.json +133 -0
- package/docs/schemas/run-environment-images.schema.json +39 -0
- package/docs/schemas/training-data-candidate.schema.json +46 -0
- package/integrations/harbor/hitch_harbor_agent.py +171 -539
- package/integrations/harbor/hitch_harbor_environment.py +331 -0
- package/integrations/harbor/hitch_harbor_task_resources.py +275 -0
- package/integrations/harbor/hitch_harbor_verifier.py +23 -0
- package/package.json +14 -1
|
@@ -0,0 +1,181 @@
|
|
|
1
|
+
import { readdir } from "node:fs/promises";
|
|
2
|
+
import path from "node:path";
|
|
3
|
+
import { SCHEMA_VERSION, atomicWriteJSON, credentialValuesFromEnv, readJSON, safeDiagnosticMessage, withFileLock } from "../foundation/index.js";
|
|
4
|
+
import { EvalEventSink, parseEvalExecutionPlan, readExecutionLeases, recoverLocalDockerEvalLeases } from "../evals/index.js";
|
|
5
|
+
import { idempotencyIndexPath, isTerminalControl, parseEvalControl, parseEvalSubmission, terminalControlState } from "./eval-records.js";
|
|
6
|
+
export async function recoverPersistedEvals(input) {
|
|
7
|
+
const queue = [];
|
|
8
|
+
const entries = await readdir(input.evalsRoot, { withFileTypes: true });
|
|
9
|
+
for (const item of entries) {
|
|
10
|
+
if (!item.isDirectory() || !/^eval_[a-f0-9]{32}$/.test(item.name))
|
|
11
|
+
continue;
|
|
12
|
+
const evalId = item.name;
|
|
13
|
+
const directory = path.join(input.evalsRoot, evalId);
|
|
14
|
+
const submissionValue = await readJSON(path.join(directory, "submission.json"), null);
|
|
15
|
+
const controlValue = await readJSON(path.join(directory, "control.json"), null);
|
|
16
|
+
if (!submissionValue || !controlValue)
|
|
17
|
+
continue;
|
|
18
|
+
const submission = await parseEvalSubmission(submissionValue, evalId);
|
|
19
|
+
const credentialValues = credentialValuesFromEnv(submission.request.pass_env, input.credentialEnv ?? process.env);
|
|
20
|
+
const control = parseEvalControl(controlValue);
|
|
21
|
+
if (await repairIdempotencyIndex(input.root, submission)) {
|
|
22
|
+
const sink = new EvalEventSink(directory, evalId, input.onEvent);
|
|
23
|
+
await sink.open();
|
|
24
|
+
sink.emit({ type: "eval.idempotency-index.recovered", idempotency_key_hash: submission.idempotency_key_hash });
|
|
25
|
+
await sink.close();
|
|
26
|
+
}
|
|
27
|
+
const result = await readJSON(path.join(directory, "result.json"), null);
|
|
28
|
+
if (result) {
|
|
29
|
+
if (!isTerminalControl(control.state))
|
|
30
|
+
await updateControl(directory, (current) => ({ ...withoutAllocation(current), state: terminalControlState(result.status) }));
|
|
31
|
+
continue;
|
|
32
|
+
}
|
|
33
|
+
if (control.state === "queued") {
|
|
34
|
+
queue.push({ evalId, request: submission.request, ...(submission.execution ? { execution: submission.execution } : {}), directory, resumeExisting: false });
|
|
35
|
+
continue;
|
|
36
|
+
}
|
|
37
|
+
if (isTerminalControl(control.state))
|
|
38
|
+
continue;
|
|
39
|
+
const sink = new EvalEventSink(directory, evalId, input.onEvent);
|
|
40
|
+
await sink.open();
|
|
41
|
+
const leases = await readExecutionLeases(directory);
|
|
42
|
+
const localLeases = leases.filter((lease) => lease.provider === "local-docker");
|
|
43
|
+
const providerLeases = leases.filter((lease) => lease.provider !== "local-docker");
|
|
44
|
+
const recoveries = [await recoverLocalDockerEvalLeases({
|
|
45
|
+
root: input.root,
|
|
46
|
+
evalId,
|
|
47
|
+
evalDirectory: directory,
|
|
48
|
+
leases: localLeases,
|
|
49
|
+
env: input.credentialEnv ?? process.env,
|
|
50
|
+
cancelRequested: control.state === "cancelling",
|
|
51
|
+
emit: (event) => sink.emit(event),
|
|
52
|
+
})];
|
|
53
|
+
if (providerLeases.length > 0)
|
|
54
|
+
recoveries.push(input.recoverProviderLeases
|
|
55
|
+
? await input.recoverProviderLeases({
|
|
56
|
+
evalId, evalDirectory: directory, leases: providerLeases,
|
|
57
|
+
cancelRequested: control.state === "cancelling", emit: (event) => sink.emit(event),
|
|
58
|
+
})
|
|
59
|
+
: { status: "ambiguous", recovered_lease_ids: [], code: "execution_state_ambiguous", message: "no recovery provider is available for active remote leases" });
|
|
60
|
+
const recovery = combinedRecovery(recoveries);
|
|
61
|
+
if (control.state === "cancelling") {
|
|
62
|
+
const now = new Date().toISOString();
|
|
63
|
+
await writeSyntheticResult(directory, evalId, submission.request, control, "cancelled", "cancelled", "eval cancellation was recovered after daemon restart", now);
|
|
64
|
+
await updateControl(directory, (current) => ({ ...withoutAllocation(current), state: "cancelled", error: { code: "cancelled", message: "eval cancellation was recovered after daemon restart" } }));
|
|
65
|
+
sink.emit({ type: "eval.recovered", status: "cancelled", code: "cancelled" });
|
|
66
|
+
await sink.close();
|
|
67
|
+
continue;
|
|
68
|
+
}
|
|
69
|
+
if (recovery.status === "ambiguous") {
|
|
70
|
+
const now = new Date().toISOString();
|
|
71
|
+
const code = recovery.code || "execution_state_ambiguous";
|
|
72
|
+
const message = safeDiagnosticMessage(recovery.message || "daemon restarted while eval execution state was ambiguous", credentialValues);
|
|
73
|
+
await writeSyntheticResult(directory, evalId, submission.request, control, "failed", code, message, now);
|
|
74
|
+
await updateControl(directory, (current) => ({ ...withoutAllocation(current), state: "failed", error: { code, message } }));
|
|
75
|
+
sink.emit({ type: "eval.recovered", status: "failed", code });
|
|
76
|
+
await sink.close();
|
|
77
|
+
continue;
|
|
78
|
+
}
|
|
79
|
+
const files = await resumableFiles(directory);
|
|
80
|
+
if (files === "inconsistent") {
|
|
81
|
+
const now = new Date().toISOString();
|
|
82
|
+
const message = "daemon restarted with incomplete persisted eval planning state";
|
|
83
|
+
await writeSyntheticResult(directory, evalId, submission.request, control, "failed", "execution_state_ambiguous", message, now);
|
|
84
|
+
await updateControl(directory, (current) => ({ ...withoutAllocation(current), state: "failed", error: { code: "execution_state_ambiguous", message } }));
|
|
85
|
+
sink.emit({ type: "eval.recovered", status: "failed", code: "execution_state_ambiguous" });
|
|
86
|
+
}
|
|
87
|
+
else {
|
|
88
|
+
queue.push({ evalId, request: submission.request, ...(submission.execution ? { execution: submission.execution } : {}), directory, resumeExisting: files === "complete" });
|
|
89
|
+
sink.emit({ type: "eval.recovered", status: "queued", recovery: files === "complete" ? "resume" : "restart-before-execution" });
|
|
90
|
+
}
|
|
91
|
+
await sink.close();
|
|
92
|
+
}
|
|
93
|
+
return queue;
|
|
94
|
+
}
|
|
95
|
+
async function repairIdempotencyIndex(root, submission) {
|
|
96
|
+
const keyHash = submission.idempotency_key_hash;
|
|
97
|
+
if (!keyHash)
|
|
98
|
+
return false;
|
|
99
|
+
return withFileLock(path.join(root, "locks", "eval-idempotency"), keyHash, async () => {
|
|
100
|
+
const file = idempotencyIndexPath(root, keyHash);
|
|
101
|
+
const expected = {
|
|
102
|
+
schema_version: "1",
|
|
103
|
+
eval_id: submission.eval_id,
|
|
104
|
+
submission_digest: submission.submission_digest,
|
|
105
|
+
};
|
|
106
|
+
const existing = await readJSON(file, null);
|
|
107
|
+
if (existing === null) {
|
|
108
|
+
await atomicWriteJSON(file, expected);
|
|
109
|
+
return true;
|
|
110
|
+
}
|
|
111
|
+
if (Object.keys(existing).length !== Object.keys(expected).length
|
|
112
|
+
|| existing.schema_version !== expected.schema_version
|
|
113
|
+
|| existing.eval_id !== expected.eval_id
|
|
114
|
+
|| existing.submission_digest !== expected.submission_digest) {
|
|
115
|
+
throw new TypeError(`eval idempotency index conflicts with recovered submission: ${submission.eval_id}`);
|
|
116
|
+
}
|
|
117
|
+
return false;
|
|
118
|
+
}, { timeoutCode: "idempotency_locked", timeoutExitCode: 12 });
|
|
119
|
+
}
|
|
120
|
+
function combinedRecovery(results) {
|
|
121
|
+
const failure = results.find((result) => result.status === "ambiguous");
|
|
122
|
+
const recovered = results.flatMap((result) => result.recovered_lease_ids);
|
|
123
|
+
return failure
|
|
124
|
+
? { status: "ambiguous", recovered_lease_ids: recovered, ...(failure.code ? { code: failure.code } : {}), ...(failure.message ? { message: failure.message } : {}) }
|
|
125
|
+
: { status: "resumable", recovered_lease_ids: recovered };
|
|
126
|
+
}
|
|
127
|
+
async function resumableFiles(directory) {
|
|
128
|
+
const [plan, executionPlan, progress, resolution] = await Promise.all([
|
|
129
|
+
readJSON(path.join(directory, "plan.json"), null),
|
|
130
|
+
readJSON(path.join(directory, "execution-plan.json"), null),
|
|
131
|
+
readJSON(path.join(directory, "progress.json"), null),
|
|
132
|
+
readJSON(path.join(directory, "resolution.json"), null),
|
|
133
|
+
]);
|
|
134
|
+
const values = [plan, executionPlan, progress, resolution];
|
|
135
|
+
if (values.every((value) => value === null))
|
|
136
|
+
return "absent";
|
|
137
|
+
if (values.some((value) => value === null) || plan?.attempt_execution !== "harbor-task-slots-v1")
|
|
138
|
+
return "inconsistent";
|
|
139
|
+
try {
|
|
140
|
+
const parsed = parseEvalExecutionPlan(executionPlan);
|
|
141
|
+
return parsed.membership === "known"
|
|
142
|
+
&& parsed.work_items.every((item) => item.task_ids.length === 1 && item.slots.length === 1 && item.logical_attempt !== null)
|
|
143
|
+
? "complete"
|
|
144
|
+
: "inconsistent";
|
|
145
|
+
}
|
|
146
|
+
catch {
|
|
147
|
+
return "inconsistent";
|
|
148
|
+
}
|
|
149
|
+
}
|
|
150
|
+
async function writeSyntheticResult(directory, evalId, request, control, status, code, message, completedAt) {
|
|
151
|
+
if (await readJSON(path.join(directory, "result.json"), null))
|
|
152
|
+
return;
|
|
153
|
+
await atomicWriteJSON(path.join(directory, "result.json"), {
|
|
154
|
+
schema_version: SCHEMA_VERSION,
|
|
155
|
+
eval_id: evalId,
|
|
156
|
+
status,
|
|
157
|
+
exit_code: status === "cancelled" ? 9 : 12,
|
|
158
|
+
error: { code, message },
|
|
159
|
+
benchmark_id: request.benchmark_id,
|
|
160
|
+
benchmark_revision: request.benchmark_revision,
|
|
161
|
+
trials: [],
|
|
162
|
+
started_at: control.created_at,
|
|
163
|
+
completed_at: completedAt,
|
|
164
|
+
});
|
|
165
|
+
}
|
|
166
|
+
async function updateControl(directory, update) {
|
|
167
|
+
const current = parseEvalControl(await readJSON(path.join(directory, "control.json")));
|
|
168
|
+
await atomicWriteJSON(path.join(directory, "control.json"), parseEvalControl({
|
|
169
|
+
...update(current),
|
|
170
|
+
schema_version: "1",
|
|
171
|
+
eval_id: current.eval_id,
|
|
172
|
+
generation: current.generation + 1,
|
|
173
|
+
created_at: current.created_at,
|
|
174
|
+
updated_at: new Date().toISOString(),
|
|
175
|
+
}));
|
|
176
|
+
}
|
|
177
|
+
function withoutAllocation(control) {
|
|
178
|
+
const { allocation_id: _allocationId, ...rest } = control;
|
|
179
|
+
return rest;
|
|
180
|
+
}
|
|
181
|
+
//# sourceMappingURL=eval-recovery.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"eval-recovery.js","sourceRoot":"","sources":["../../../src/control-plane/eval-recovery.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAC3C,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,cAAc,EAAE,eAAe,EAAE,uBAAuB,EAAE,QAAQ,EAAE,qBAAqB,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AACjJ,OAAO,EAAE,aAAa,EAAE,sBAAsB,EAAE,mBAAmB,EAAE,4BAA4B,EAAE,MAAM,mBAAmB,CAAC;AAE7H,OAAO,EAAE,oBAAoB,EAAE,iBAAiB,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAAM,mBAAmB,CAAC;AAUzI,MAAM,CAAC,KAAK,UAAU,qBAAqB,CAAC,KAY3C;IACC,MAAM,KAAK,GAAyB,EAAE,CAAC;IACvC,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,KAAK,CAAC,SAAS,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IACxE,KAAK,MAAM,IAAI,IAAI,OAAO,EAAE,CAAC;QAC3B,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,qBAAqB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;YAAE,SAAS;QAC5E,MAAM,MAAM,GAAG,IAAI,CAAC,IAAc,CAAC;QACnC,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;QACrD,MAAM,eAAe,GAAG,MAAM,QAAQ,CAAiB,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,iBAAiB,CAAC,EAAE,IAAI,CAAC,CAAC;QACtG,MAAM,YAAY,GAAG,MAAM,QAAQ,CAAiB,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE,IAAI,CAAC,CAAC;QAChG,IAAI,CAAC,eAAe,IAAI,CAAC,YAAY;YAAE,SAAS;QAChD,MAAM,UAAU,GAAG,MAAM,mBAAmB,CAAC,eAAe,EAAE,MAAM,CAAC,CAAC;QACtE,MAAM,gBAAgB,GAAG,uBAAuB,CAAC,UAAU,CAAC,OAAO,CAAC,QAAQ,EAAE,KAAK,CAAC,aAAa,IAAI,OAAO,CAAC,GAAG,CAAC,CAAC;QAClH,MAAM,OAAO,GAAG,gBAAgB,CAAC,YAAY,CAAC,CAAC;QAC/C,IAAI,MAAM,sBAAsB,CAAC,KAAK,CAAC,IAAI,EAAE,UAAU,CAAC,EAAE,CAAC;YACzD,MAAM,IAAI,GAAG,IAAI,aAAa,CAAC,SAAS,EAAE,MAAM,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;YACjE,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC;YAClB,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,kCAAkC,EAAE,oBAAoB,EAAE,UAAU,CAAC,oBAAoB,EAAE,CAAC,CAAC;YAC/G,MAAM,IAAI,CAAC,KAAK,EAAE,CAAC;QACrB,CAAC;QACD,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,aAAa,CAAC,EAAE,IAAI,CAAC,CAAC;QACzG,IAAI,MAAM,EAAE,CAAC;YACX,IAAI,CAAC,iBAAiB,CAAC,OAAO,CAAC,KAAK,CAAC;gBAAE,MAAM,aAAa,CAAC,SAAS,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,iBAAiB,CAAC,OAAO,CAAC,EAAE,KAAK,EAAE,oBAAoB,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC;YACpK,SAAS;QACX,CAAC;QACD,IAAI,OAAO,CAAC,KAAK,KAAK,QAAQ,EAAE,CAAC;YAC/B,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,EAAE,UAAU,CAAC,OAAO,EAAE,GAAG,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,UAAU,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,SAAS,EAAE,cAAc,EAAE,KAAK,EAAE,CAAC,CAAC;YAC5J,SAAS;QACX,CAAC;QACD,IAAI,iBAAiB,CAAC,OAAO,CAAC,KAAK,CAAC;YAAE,SAAS;QAC/C,MAAM,IAAI,GAAG,IAAI,aAAa,CAAC,SAAS,EAAE,MAAM,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;QACjE,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC;QAClB,MAAM,MAAM,GAAG,MAAM,mBAAmB,CAAC,SAAS,CAAC,CAAC;QACpD,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,cAAc,CAAC,CAAC;QAChF,MAAM,cAAc,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,cAAc,CAAC,CAAC;QACnF,MAAM,UAAU,GAA8B,CAAC,MAAM,4BAA4B,CAAC;gBAChF,IAAI,EAAE,KAAK,CAAC,IAAI;gBAChB,MAAM;gBACN,aAAa,EAAE,SAAS;gBACxB,MAAM,EAAE,WAAW;gBACnB,GAAG,EAAE,KAAK,CAAC,aAAa,IAAI,OAAO,CAAC,GAAG;gBACvC,eAAe,EAAE,OAAO,CAAC,KAAK,KAAK,YAAY;gBAC/C,IAAI,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC;aAClC,CAAC,CAAC,CAAC;QACJ,IAAI,cAAc,CAAC,MAAM,GAAG,CAAC;YAAE,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,qBAAqB;gBACxE,CAAC,CAAC,MAAM,KAAK,CAAC,qBAAqB,CAAC;oBAClC,MAAM,EAAE,aAAa,EAAE,SAAS,EAAE,MAAM,EAAE,cAAc;oBACxD,eAAe,EAAE,OAAO,CAAC,KAAK,KAAK,YAAY,EAAE,IAAI,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC;iBACnF,CAAC;gBACF,CAAC,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,mBAAmB,EAAE,EAAE,EAAE,IAAI,EAAE,2BAA2B,EAAE,OAAO,EAAE,4DAA4D,EAAE,CAAC,CAAC;QAChK,MAAM,QAAQ,GAAG,gBAAgB,CAAC,UAAU,CAAC,CAAC;QAC9C,IAAI,OAAO,CAAC,KAAK,KAAK,YAAY,EAAE,CAAC;YACnC,MAAM,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;YACrC,MAAM,oBAAoB,CAAC,SAAS,EAAE,MAAM,EAAE,UAAU,CAAC,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,WAAW,EAAE,sDAAsD,EAAE,GAAG,CAAC,CAAC;YAClK,MAAM,aAAa,CAAC,SAAS,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,iBAAiB,CAAC,OAAO,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,sDAAsD,EAAE,EAAE,CAAC,CAAC,CAAC;YACpM,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,MAAM,EAAE,WAAW,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC,CAAC;YAC9E,MAAM,IAAI,CAAC,KAAK,EAAE,CAAC;YACnB,SAAS;QACX,CAAC;QACD,IAAI,QAAQ,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YACpC,MAAM,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;YACrC,MAAM,IAAI,GAAG,QAAQ,CAAC,IAAI,IAAI,2BAA2B,CAAC;YAC1D,MAAM,OAAO,GAAG,qBAAqB,CAAC,QAAQ,CAAC,OAAO,IAAI,2DAA2D,EAAE,gBAAgB,CAAC,CAAC;YACzI,MAAM,oBAAoB,CAAC,SAAS,EAAE,MAAM,EAAE,UAAU,CAAC,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;YACzG,MAAM,aAAa,CAAC,SAAS,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,iBAAiB,CAAC,OAAO,CAAC,EAAE,KAAK,EAAE,QAAQ,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC;YAC5H,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC;YAC9D,MAAM,IAAI,CAAC,KAAK,EAAE,CAAC;YACnB,SAAS;QACX,CAAC;QACD,MAAM,KAAK,GAAG,MAAM,cAAc,CAAC,SAAS,CAAC,CAAC;QAC9C,IAAI,KAAK,KAAK,cAAc,EAAE,CAAC;YAC7B,MAAM,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;YACrC,MAAM,OAAO,GAAG,gEAAgE,CAAC;YACjF,MAAM,oBAAoB,CAAC,SAAS,EAAE,MAAM,EAAE,UAAU,CAAC,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,2BAA2B,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;YAChI,MAAM,aAAa,CAAC,SAAS,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,iBAAiB,CAAC,OAAO,CAAC,EAAE,KAAK,EAAE,QAAQ,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,2BAA2B,EAAE,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC;YACzJ,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,2BAA2B,EAAE,CAAC,CAAC;QAC7F,CAAC;aAAM,CAAC;YACN,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,EAAE,UAAU,CAAC,OAAO,EAAE,GAAG,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,UAAU,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,SAAS,EAAE,cAAc,EAAE,KAAK,KAAK,UAAU,EAAE,CAAC,CAAC;YAC3K,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,MAAM,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,KAAK,UAAU,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,0BAA0B,EAAE,CAAC,CAAC;QAClI,CAAC;QACD,MAAM,IAAI,CAAC,KAAK,EAAE,CAAC;IACrB,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,KAAK,UAAU,sBAAsB,CAAC,IAAY,EAAE,UAA2D;IAC7G,MAAM,OAAO,GAAG,UAAU,CAAC,oBAAoB,CAAC;IAChD,IAAI,CAAC,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3B,OAAO,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,kBAAkB,CAAC,EAAE,OAAO,EAAE,KAAK,IAAI,EAAE;QACpF,MAAM,IAAI,GAAG,oBAAoB,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QACjD,MAAM,QAAQ,GAAG;YACf,cAAc,EAAE,GAAG;YACnB,OAAO,EAAE,UAAU,CAAC,OAAO;YAC3B,iBAAiB,EAAE,UAAU,CAAC,iBAAiB;SAChD,CAAC;QACF,MAAM,QAAQ,GAAG,MAAM,QAAQ,CAAiC,IAAI,EAAE,IAAI,CAAC,CAAC;QAC5E,IAAI,QAAQ,KAAK,IAAI,EAAE,CAAC;YACtB,MAAM,eAAe,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;YACtC,OAAO,IAAI,CAAC;QACd,CAAC;QACD,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM;eAC5D,QAAQ,CAAC,cAAc,KAAK,QAAQ,CAAC,cAAc;eACnD,QAAQ,CAAC,OAAO,KAAK,QAAQ,CAAC,OAAO;eACrC,QAAQ,CAAC,iBAAiB,KAAK,QAAQ,CAAC,iBAAiB,EAAE,CAAC;YAC/D,MAAM,IAAI,SAAS,CAAC,+DAA+D,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC;QAC3G,CAAC;QACD,OAAO,KAAK,CAAC;IACf,CAAC,EAAE,EAAE,WAAW,EAAE,oBAAoB,EAAE,eAAe,EAAE,EAAE,EAAE,CAAC,CAAC;AACjE,CAAC;AAED,SAAS,gBAAgB,CAAC,OAAkC;IAC1D,MAAM,OAAO,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,MAAM,KAAK,WAAW,CAAC,CAAC;IACxE,MAAM,SAAS,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,mBAAmB,CAAC,CAAC;IAC1E,OAAO,OAAO;QACZ,CAAC,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,mBAAmB,EAAE,SAAS,EAAE,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,GAAG,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE;QACpK,CAAC,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,mBAAmB,EAAE,SAAS,EAAE,CAAC;AAC9D,CAAC;AAED,KAAK,UAAU,cAAc,CAAC,SAAiB;IAC7C,MAAM,CAAC,IAAI,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;QACpE,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,WAAW,CAAC,EAAE,IAAI,CAAC;QACjF,QAAQ,CAAiB,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,qBAAqB,CAAC,EAAE,IAAI,CAAC;QAC3E,QAAQ,CAAiB,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,eAAe,CAAC,EAAE,IAAI,CAAC;QACrE,QAAQ,CAAiB,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,iBAAiB,CAAC,EAAE,IAAI,CAAC;KACxE,CAAC,CAAC;IACH,MAAM,MAAM,GAAG,CAAC,IAAI,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,CAAC,CAAC;IAC3D,IAAI,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,IAAI,CAAC;QAAE,OAAO,QAAQ,CAAC;IAC7D,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,IAAI,CAAC,IAAI,IAAI,EAAE,iBAAiB,KAAK,sBAAsB;QAAE,OAAO,cAAc,CAAC;IACxH,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,sBAAsB,CAAC,aAAa,CAAC,CAAC;QACrD,OAAO,MAAM,CAAC,UAAU,KAAK,OAAO;eAC/B,MAAM,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,IAAI,IAAI,CAAC,eAAe,KAAK,IAAI,CAAC;YAC5H,CAAC,CAAC,UAAU;YACZ,CAAC,CAAC,cAAc,CAAC;IACrB,CAAC;IAAC,MAAM,CAAC;QAAC,OAAO,cAAc,CAAC;IAAC,CAAC;AACpC,CAAC;AAED,KAAK,UAAU,oBAAoB,CACjC,SAAiB,EACjB,MAAc,EACd,OAAoB,EACpB,OAAsB,EACtB,MAA8B,EAC9B,IAAY,EACZ,OAAe,EACf,WAAmB;IAEnB,IAAI,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,aAAa,CAAC,EAAE,IAAI,CAAC;QAAE,OAAO;IACtE,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,aAAa,CAAC,EAAE;QACzD,cAAc,EAAE,cAAc;QAC9B,OAAO,EAAE,MAAM;QACf,MAAM;QACN,SAAS,EAAE,MAAM,KAAK,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE;QAC1C,KAAK,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE;QACxB,YAAY,EAAE,OAAO,CAAC,YAAY;QAClC,kBAAkB,EAAE,OAAO,CAAC,kBAAkB;QAC9C,MAAM,EAAE,EAAE;QACV,UAAU,EAAE,OAAO,CAAC,UAAU;QAC9B,YAAY,EAAE,WAAW;KAC1B,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,aAAa,CAAC,SAAiB,EAAE,MAAiD;IAC/F,MAAM,OAAO,GAAG,gBAAgB,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC,CAAC,CAAC,CAAC;IACvF,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE,gBAAgB,CAAC;QAC3E,GAAG,MAAM,CAAC,OAAO,CAAC;QAClB,cAAc,EAAE,GAAG;QACnB,OAAO,EAAE,OAAO,CAAC,OAAO;QACxB,UAAU,EAAE,OAAO,CAAC,UAAU,GAAG,CAAC;QAClC,UAAU,EAAE,OAAO,CAAC,UAAU;QAC9B,UAAU,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KACrC,CAAC,CAAC,CAAC;AACN,CAAC;AAED,SAAS,iBAAiB,CAAC,OAAsB;IAC/C,MAAM,EAAE,aAAa,EAAE,aAAa,EAAE,GAAG,IAAI,EAAE,GAAG,OAAO,CAAC;IAC1D,OAAO,IAAI,CAAC;AACd,CAAC"}
|
|
@@ -0,0 +1,447 @@
|
|
|
1
|
+
import { mkdir, rm } from "node:fs/promises";
|
|
2
|
+
import path from "node:path";
|
|
3
|
+
import { HitchError, atomicWriteJSON, credentialValuesFromEnv, ensureDir, hitchRootId, readJSON, safeDiagnosticMessage, sha256Bytes, sha256JSON, statePaths, withFileLock } from "../foundation/index.js";
|
|
4
|
+
import { newEvalId, parseEvalExecutionPlan, readExecutionLeases, reapOwnedDockerResources, resolveLocalDatasetTaskIds, runEval, validateEvalId } from "../evals/index.js";
|
|
5
|
+
import { scaleResources } from "./resources.js";
|
|
6
|
+
import { CollisionLockManager } from "./collisions.js";
|
|
7
|
+
import { assertExecutionPolicySupported, defaultEvalExecutionPolicy, idempotencyIndexPath, isTerminalControl, normalizeEvalSubmissionInput, parseEvalControl, reconcileIdempotencyKeys, terminalControlState, validateIdempotencyKey } from "./eval-records.js";
|
|
8
|
+
import { WorkItemDispatcher } from "./work-dispatcher.js";
|
|
9
|
+
import { workItemAdmission } from "./work-admission.js";
|
|
10
|
+
import { localProviderStatusSnapshot, localWorkerSnapshot } from "./local-worker.js";
|
|
11
|
+
import { recoverPersistedEvals } from "./eval-recovery.js";
|
|
12
|
+
import { applyEvalPhase, applyEvalWorkItem, settleEvalWorkItems } from "./eval-control-work.js";
|
|
13
|
+
import { EvalImageServices } from "./eval-image-services.js";
|
|
14
|
+
import { writeSyntheticEvalResult } from "./synthetic-result.js";
|
|
15
|
+
import { RemoteWorkCoordinator } from "./remote-work-coordinator.js";
|
|
16
|
+
import { schedulerCapturePlan, schedulerQueuedEval } from "./scheduler-eval-entry.js";
|
|
17
|
+
import { emitPersistedEvalEvent, updateEvalControl } from "./eval-control-state.js";
|
|
18
|
+
export class EvalScheduler {
|
|
19
|
+
root;
|
|
20
|
+
evalsRoot;
|
|
21
|
+
resources;
|
|
22
|
+
trialResources;
|
|
23
|
+
workerId;
|
|
24
|
+
provider;
|
|
25
|
+
collisionDomainId;
|
|
26
|
+
executor;
|
|
27
|
+
onEvent;
|
|
28
|
+
unsubscribe;
|
|
29
|
+
unsubscribeCollisions;
|
|
30
|
+
collisions;
|
|
31
|
+
workItems;
|
|
32
|
+
dockerResourceReaper;
|
|
33
|
+
environmentImages;
|
|
34
|
+
remoteWork;
|
|
35
|
+
credentialEnv;
|
|
36
|
+
queue = [];
|
|
37
|
+
active = new Map();
|
|
38
|
+
completions = new Map();
|
|
39
|
+
accepting = true;
|
|
40
|
+
draining = false;
|
|
41
|
+
constructor({ root, resources, trialResources, executor = runEval, onEvent = () => { }, collisions = new CollisionLockManager(), workerId, provider = "local-docker", collisionDomainId, dockerResourceReaper = reapOwnedDockerResources, environmentImageResolver, environmentImageBuilder, remoteWorkers, remoteWorkerProtocol, credentialEnv = process.env }) {
|
|
42
|
+
this.root = root;
|
|
43
|
+
this.evalsRoot = statePaths(root).evals;
|
|
44
|
+
this.resources = resources;
|
|
45
|
+
this.trialResources = trialResources;
|
|
46
|
+
const rootHash = hitchRootId(root);
|
|
47
|
+
this.workerId = workerId || `worker_${rootHash}`;
|
|
48
|
+
this.provider = provider;
|
|
49
|
+
this.collisionDomainId = collisionDomainId || `local-docker-root:${rootHash}`;
|
|
50
|
+
this.executor = executor;
|
|
51
|
+
this.onEvent = onEvent;
|
|
52
|
+
this.collisions = collisions;
|
|
53
|
+
this.workItems = new WorkItemDispatcher({ resources, collisions });
|
|
54
|
+
this.dockerResourceReaper = dockerResourceReaper;
|
|
55
|
+
this.credentialEnv = credentialEnv;
|
|
56
|
+
this.environmentImages = new EvalImageServices({ root, provider, resources, onEvent: this.onEvent, ...(environmentImageResolver ? { resolver: environmentImageResolver } : {}), ...(environmentImageBuilder ? { builder: environmentImageBuilder } : {}) });
|
|
57
|
+
this.remoteWork = remoteWorkers && remoteWorkerProtocol
|
|
58
|
+
? new RemoteWorkCoordinator({ root, registry: remoteWorkers, protocol: remoteWorkerProtocol, collisions })
|
|
59
|
+
: undefined;
|
|
60
|
+
this.unsubscribe = resources.subscribe(() => this.scheduleDrain());
|
|
61
|
+
this.unsubscribeCollisions = collisions.subscribe(() => this.scheduleDrain());
|
|
62
|
+
}
|
|
63
|
+
async initialize() {
|
|
64
|
+
await ensureDir(this.evalsRoot);
|
|
65
|
+
await this.recoverInterruptedEvals();
|
|
66
|
+
if (this.provider === "local-docker") {
|
|
67
|
+
try {
|
|
68
|
+
const report = await this.dockerResourceReaper({ root: this.root });
|
|
69
|
+
this.onEvent({ type: "docker.reaper.completed", root_id: report.root_id, scanned: report.scanned, deleted: report.deleted.length, issues: report.issues.length });
|
|
70
|
+
}
|
|
71
|
+
catch (error) {
|
|
72
|
+
this.onEvent({ type: "docker.reaper.failed", code: error.code || "docker_reaper_failed" });
|
|
73
|
+
}
|
|
74
|
+
}
|
|
75
|
+
this.scheduleDrain();
|
|
76
|
+
}
|
|
77
|
+
async submit(input, options = {}) {
|
|
78
|
+
if (!this.accepting)
|
|
79
|
+
throw new HitchError("daemon is shutting down", { code: "daemon_shutting_down", exitCode: 12 });
|
|
80
|
+
const normalized = await normalizeEvalSubmissionInput(input, { provider: this.provider, trialResources: this.trialResources });
|
|
81
|
+
const remote = normalized.execution.provider !== this.provider;
|
|
82
|
+
assertExecutionPolicySupported(normalized.execution, remote ? normalized.execution.provider : this.provider);
|
|
83
|
+
if (remote && !this.remoteWork)
|
|
84
|
+
throw new HitchError(`execution provider is unavailable: ${normalized.execution.provider}`, { code: "execution_provider_unavailable", exitCode: 10 });
|
|
85
|
+
if (remote && normalized.execution.build.mode !== "backend")
|
|
86
|
+
throw new HitchError("remote worker image prebuild is not available yet", { code: "remote_build_mode_unsupported", exitCode: 10 });
|
|
87
|
+
if (remote && await resolveLocalDatasetTaskIds(normalized.request.dataset) === null)
|
|
88
|
+
throw new HitchError("remote workers require enumerable task membership", { code: "remote_opaque_membership_unsupported", exitCode: 10 });
|
|
89
|
+
const modelCapturePlan = await schedulerCapturePlan({ request: normalized.request, execution: normalized.execution, localProvider: this.provider, localStatus: this.providerSnapshot(), ...(this.remoteWork ? { remoteWork: this.remoteWork } : {}) });
|
|
90
|
+
const idempotencyKey = reconcileIdempotencyKeys(normalized.idempotencyKey, options.idempotencyKey);
|
|
91
|
+
const canFit = remote
|
|
92
|
+
? await this.remoteWork.canEverFit(normalized.execution.provider, normalized.execution.resources.default_trial)
|
|
93
|
+
: this.resources.canEverFit(normalized.execution.resources.default_trial);
|
|
94
|
+
if (!canFit) {
|
|
95
|
+
throw new HitchError("one eval trial exceeds the daemon resource capacity", {
|
|
96
|
+
code: "resource_request_unsatisfiable",
|
|
97
|
+
exitCode: 10,
|
|
98
|
+
});
|
|
99
|
+
}
|
|
100
|
+
const submissionDigest = sha256JSON({ request: normalized.request, execution: normalized.execution });
|
|
101
|
+
if (idempotencyKey !== undefined)
|
|
102
|
+
validateIdempotencyKey(idempotencyKey);
|
|
103
|
+
if (idempotencyKey) {
|
|
104
|
+
const keyHash = sha256Bytes(idempotencyKey);
|
|
105
|
+
return withFileLock(path.join(this.root, "locks", "eval-idempotency"), keyHash, async () => {
|
|
106
|
+
const indexPath = idempotencyIndexPath(this.root, keyHash);
|
|
107
|
+
const existing = await readJSON(indexPath, null);
|
|
108
|
+
if (existing) {
|
|
109
|
+
if (existing.submission_digest !== submissionDigest || typeof existing.eval_id !== "string") {
|
|
110
|
+
throw new HitchError("idempotency key was already used for a different eval request", {
|
|
111
|
+
code: "idempotency_conflict",
|
|
112
|
+
exitCode: 2,
|
|
113
|
+
});
|
|
114
|
+
}
|
|
115
|
+
return validateEvalId(existing.eval_id);
|
|
116
|
+
}
|
|
117
|
+
const entry = await this.persistSubmission(normalized.request, normalized.execution, modelCapturePlan, submissionDigest, keyHash);
|
|
118
|
+
try {
|
|
119
|
+
await atomicWriteJSON(indexPath, { schema_version: "1", eval_id: entry.evalId, submission_digest: submissionDigest });
|
|
120
|
+
}
|
|
121
|
+
catch (error) {
|
|
122
|
+
await rm(entry.directory, { recursive: true, force: true });
|
|
123
|
+
throw error;
|
|
124
|
+
}
|
|
125
|
+
return this.enqueue(entry);
|
|
126
|
+
}, { timeoutCode: "idempotency_locked", timeoutExitCode: 12 });
|
|
127
|
+
}
|
|
128
|
+
return this.enqueue(await this.persistSubmission(normalized.request, normalized.execution, modelCapturePlan, submissionDigest));
|
|
129
|
+
}
|
|
130
|
+
async persistSubmission(normalized, execution, modelCapturePlan, submissionDigest, keyHash) {
|
|
131
|
+
const evalId = newEvalId();
|
|
132
|
+
const directory = path.join(this.evalsRoot, evalId);
|
|
133
|
+
await mkdir(directory, { mode: 0o700 });
|
|
134
|
+
try {
|
|
135
|
+
const now = new Date().toISOString();
|
|
136
|
+
const submission = {
|
|
137
|
+
schema_version: "1",
|
|
138
|
+
eval_id: evalId,
|
|
139
|
+
request: normalized,
|
|
140
|
+
execution,
|
|
141
|
+
submission_digest: submissionDigest,
|
|
142
|
+
...(keyHash ? { idempotency_key_hash: keyHash } : {}),
|
|
143
|
+
submitted_at: now,
|
|
144
|
+
};
|
|
145
|
+
const control = {
|
|
146
|
+
schema_version: "1",
|
|
147
|
+
eval_id: evalId,
|
|
148
|
+
generation: 0,
|
|
149
|
+
state: "queued",
|
|
150
|
+
requested_parallelism: execution.max_parallelism,
|
|
151
|
+
admitted_parallelism: 0,
|
|
152
|
+
active_leases: [],
|
|
153
|
+
queued_work_items: [],
|
|
154
|
+
terminal_work_items: [],
|
|
155
|
+
created_at: now,
|
|
156
|
+
updated_at: now,
|
|
157
|
+
};
|
|
158
|
+
await atomicWriteJSON(path.join(directory, "request.json"), normalized);
|
|
159
|
+
await atomicWriteJSON(path.join(directory, "submission.json"), submission);
|
|
160
|
+
await atomicWriteJSON(path.join(directory, "control.json"), control);
|
|
161
|
+
await this.emitPersisted(directory, evalId, { type: "eval.queued", requested_parallelism: execution.max_parallelism, model_capture: modelCapturePlan });
|
|
162
|
+
return this.queuedEval(evalId, normalized, execution, modelCapturePlan, directory);
|
|
163
|
+
}
|
|
164
|
+
catch (error) {
|
|
165
|
+
await rm(directory, { recursive: true, force: true });
|
|
166
|
+
throw error;
|
|
167
|
+
}
|
|
168
|
+
}
|
|
169
|
+
enqueue(entry) {
|
|
170
|
+
this.queue.push(entry);
|
|
171
|
+
this.scheduleDrain();
|
|
172
|
+
return entry.evalId;
|
|
173
|
+
}
|
|
174
|
+
async status(evalIdValue) {
|
|
175
|
+
const evalId = validateEvalId(evalIdValue);
|
|
176
|
+
const directory = path.join(this.evalsRoot, evalId);
|
|
177
|
+
const request = await readJSON(path.join(directory, "request.json"), null);
|
|
178
|
+
const submission = await readJSON(path.join(directory, "submission.json"), null);
|
|
179
|
+
const controlValue = await readJSON(path.join(directory, "control.json"), null);
|
|
180
|
+
if (!request || !controlValue)
|
|
181
|
+
return null;
|
|
182
|
+
const control = parseEvalControl(controlValue);
|
|
183
|
+
const persistedResult = await readJSON(path.join(directory, "result.json"), null);
|
|
184
|
+
const result = isTerminalControl(control.state) ? persistedResult : null;
|
|
185
|
+
const progress = await readJSON(path.join(directory, "progress.json"), null);
|
|
186
|
+
const active = this.active.get(evalId);
|
|
187
|
+
const workItems = this.workItems.evalSnapshot(evalId);
|
|
188
|
+
const admitted = active?.fineGrained ? workItems?.active ?? 0 : control.admitted_parallelism;
|
|
189
|
+
return {
|
|
190
|
+
request,
|
|
191
|
+
execution: submission?.execution ?? null,
|
|
192
|
+
control,
|
|
193
|
+
progress,
|
|
194
|
+
result,
|
|
195
|
+
leases: await readExecutionLeases(directory),
|
|
196
|
+
effective_parallelism: {
|
|
197
|
+
requested: control.requested_parallelism,
|
|
198
|
+
admitted,
|
|
199
|
+
running: active ? admitted : 0,
|
|
200
|
+
},
|
|
201
|
+
};
|
|
202
|
+
}
|
|
203
|
+
async cancel(evalIdValue) {
|
|
204
|
+
const evalId = validateEvalId(evalIdValue);
|
|
205
|
+
const queuedIndex = this.queue.findIndex((entry) => entry.evalId === evalId);
|
|
206
|
+
if (queuedIndex >= 0) {
|
|
207
|
+
const [entry] = this.queue.splice(queuedIndex, 1);
|
|
208
|
+
if (!entry)
|
|
209
|
+
return "not_cancellable";
|
|
210
|
+
const now = new Date().toISOString();
|
|
211
|
+
await this.emitPersisted(entry.directory, evalId, { type: "eval.cancel.requested", phase: "queued" });
|
|
212
|
+
await writeSyntheticEvalResult({ directory: entry.directory, evalId: entry.evalId, request: entry.request, status: "cancelled", code: "cancelled", message: "eval cancelled before launch", completedAt: now });
|
|
213
|
+
await this.updateControl(entry.directory, (control) => ({
|
|
214
|
+
...control,
|
|
215
|
+
state: "cancelled",
|
|
216
|
+
admitted_parallelism: 0,
|
|
217
|
+
cancel_requested_at: now,
|
|
218
|
+
}));
|
|
219
|
+
await this.emitPersisted(entry.directory, evalId, { type: "eval.cancelled", phase: "queued" });
|
|
220
|
+
return "accepted";
|
|
221
|
+
}
|
|
222
|
+
const active = this.active.get(evalId);
|
|
223
|
+
if (active) {
|
|
224
|
+
const directory = path.join(this.evalsRoot, evalId);
|
|
225
|
+
const now = new Date().toISOString();
|
|
226
|
+
await this.emitPersisted(directory, evalId, { type: "eval.cancel.requested", phase: "running" });
|
|
227
|
+
const control = await this.updateControl(directory, (control) => isTerminalControl(control.state) ? control : ({
|
|
228
|
+
...control,
|
|
229
|
+
state: "cancelling",
|
|
230
|
+
cancel_requested_at: control.cancel_requested_at || now,
|
|
231
|
+
}));
|
|
232
|
+
if (isTerminalControl(control.state))
|
|
233
|
+
return "terminal";
|
|
234
|
+
active.controller.abort();
|
|
235
|
+
return "accepted";
|
|
236
|
+
}
|
|
237
|
+
const status = await this.status(evalId).catch(() => null);
|
|
238
|
+
if (!status)
|
|
239
|
+
return "not_found";
|
|
240
|
+
return isTerminalControl(status.control.state) || status.result ? "terminal" : "not_cancellable";
|
|
241
|
+
}
|
|
242
|
+
snapshot() {
|
|
243
|
+
return {
|
|
244
|
+
queued: this.queue.length,
|
|
245
|
+
running: this.active.size,
|
|
246
|
+
accepting: this.accepting,
|
|
247
|
+
active: [...this.active.entries()].map(([evalId, active]) => ({
|
|
248
|
+
eval_id: evalId,
|
|
249
|
+
requested_parallelism: active.requestedParallelism,
|
|
250
|
+
admitted_parallelism: active.fineGrained ? this.workItems.evalSnapshot(evalId)?.active ?? 0 : active.admittedParallelism,
|
|
251
|
+
...(active.lease ? { allocation_id: active.lease.allocation.allocation_id } : {}),
|
|
252
|
+
collision_keys: active.collisions?.keys ?? [],
|
|
253
|
+
scheduling: active.fineGrained ? "work-item-drr-v1" : "coarse-eval-v1",
|
|
254
|
+
})),
|
|
255
|
+
collisions: this.collisions.snapshot(),
|
|
256
|
+
work_items: this.workItems.snapshot(),
|
|
257
|
+
};
|
|
258
|
+
}
|
|
259
|
+
workerSnapshot() {
|
|
260
|
+
return localWorkerSnapshot({ workerId: this.workerId, provider: this.provider, collisionDomainId: this.collisionDomainId, accepting: this.accepting, resources: this.resources });
|
|
261
|
+
}
|
|
262
|
+
providerSnapshot() {
|
|
263
|
+
return localProviderStatusSnapshot({ workerId: this.workerId, provider: this.provider, collisionDomainId: this.collisionDomainId, accepting: this.accepting, resources: this.resources });
|
|
264
|
+
}
|
|
265
|
+
async shutdown() {
|
|
266
|
+
if (!this.accepting && this.completions.size === 0)
|
|
267
|
+
return;
|
|
268
|
+
this.accepting = false;
|
|
269
|
+
for (const entry of [...this.queue])
|
|
270
|
+
await this.cancel(entry.evalId);
|
|
271
|
+
for (const [evalId] of this.active)
|
|
272
|
+
await this.cancel(evalId);
|
|
273
|
+
await Promise.all([...this.completions.values()]);
|
|
274
|
+
this.workItems.close();
|
|
275
|
+
this.environmentImages.close();
|
|
276
|
+
this.unsubscribe();
|
|
277
|
+
this.unsubscribeCollisions();
|
|
278
|
+
}
|
|
279
|
+
scheduleDrain() {
|
|
280
|
+
if (this.draining)
|
|
281
|
+
return;
|
|
282
|
+
queueMicrotask(() => this.drain());
|
|
283
|
+
}
|
|
284
|
+
drain() {
|
|
285
|
+
if (this.draining || !this.accepting)
|
|
286
|
+
return;
|
|
287
|
+
this.draining = true;
|
|
288
|
+
try {
|
|
289
|
+
for (;;) {
|
|
290
|
+
const selected = this.selectRunnable();
|
|
291
|
+
if (!selected)
|
|
292
|
+
break;
|
|
293
|
+
const { index, parallelism, lease, collisions, fineGrained } = selected;
|
|
294
|
+
const [entry] = this.queue.splice(index, 1);
|
|
295
|
+
if (!entry) {
|
|
296
|
+
lease?.release();
|
|
297
|
+
collisions?.release();
|
|
298
|
+
continue;
|
|
299
|
+
}
|
|
300
|
+
this.start(entry, parallelism, fineGrained, lease, collisions);
|
|
301
|
+
}
|
|
302
|
+
}
|
|
303
|
+
finally {
|
|
304
|
+
this.draining = false;
|
|
305
|
+
}
|
|
306
|
+
}
|
|
307
|
+
selectRunnable() {
|
|
308
|
+
for (let index = 0; index < this.queue.length; index += 1) {
|
|
309
|
+
const entry = this.queue[index];
|
|
310
|
+
if (entry.fineGrained)
|
|
311
|
+
return { index, parallelism: entry.execution.max_parallelism, fineGrained: true };
|
|
312
|
+
const unit = entry.execution.resources.default_trial;
|
|
313
|
+
const parallelism = this.resources.maximumUnits(unit, entry.execution.max_parallelism);
|
|
314
|
+
if (parallelism < 1)
|
|
315
|
+
continue;
|
|
316
|
+
const collisions = this.collisions.tryAcquire(entry.evalId, entry.collisionKeys);
|
|
317
|
+
if (!collisions)
|
|
318
|
+
continue;
|
|
319
|
+
const lease = this.resources.tryAcquire(entry.evalId, "eval", scaleResources(unit, parallelism));
|
|
320
|
+
if (lease)
|
|
321
|
+
return { index, parallelism, fineGrained: false, lease, collisions };
|
|
322
|
+
collisions.release();
|
|
323
|
+
}
|
|
324
|
+
return null;
|
|
325
|
+
}
|
|
326
|
+
start(entry, parallelism, fineGrained, lease, collisions) {
|
|
327
|
+
const controller = new AbortController();
|
|
328
|
+
this.active.set(entry.evalId, { controller, requestedParallelism: entry.execution.max_parallelism, admittedParallelism: parallelism, fineGrained, ...(lease ? { lease } : {}), ...(collisions ? { collisions } : {}) });
|
|
329
|
+
const completion = this.execute(entry, parallelism, fineGrained, lease, controller)
|
|
330
|
+
.catch((error) => this.recordUnexpectedFailure(entry, error))
|
|
331
|
+
.finally(() => {
|
|
332
|
+
this.active.delete(entry.evalId);
|
|
333
|
+
this.completions.delete(entry.evalId);
|
|
334
|
+
lease?.release();
|
|
335
|
+
collisions?.release();
|
|
336
|
+
this.scheduleDrain();
|
|
337
|
+
});
|
|
338
|
+
this.completions.set(entry.evalId, completion);
|
|
339
|
+
}
|
|
340
|
+
async execute(entry, parallelism, fineGrained, lease, controller) {
|
|
341
|
+
const remote = entry.execution.provider !== this.provider;
|
|
342
|
+
const queuedControl = parseEvalControl(await readJSON(path.join(entry.directory, "control.json")));
|
|
343
|
+
await this.emitPersisted(entry.directory, entry.evalId, {
|
|
344
|
+
type: "eval.dispatch.started",
|
|
345
|
+
duration_ms: Math.max(0, Date.now() - Date.parse(queuedControl.created_at)),
|
|
346
|
+
requested_parallelism: entry.execution.max_parallelism,
|
|
347
|
+
admitted_parallelism: fineGrained ? 0 : parallelism,
|
|
348
|
+
});
|
|
349
|
+
await this.updateControl(entry.directory, (control) => ({
|
|
350
|
+
...control,
|
|
351
|
+
state: "planning",
|
|
352
|
+
admitted_parallelism: fineGrained ? 0 : parallelism,
|
|
353
|
+
active_leases: [],
|
|
354
|
+
...(lease ? { allocation_id: lease.allocation.allocation_id } : {}),
|
|
355
|
+
}));
|
|
356
|
+
const result = await this.executor({
|
|
357
|
+
evalId: entry.evalId,
|
|
358
|
+
request: entry.request,
|
|
359
|
+
normalizedRequest: entry.request,
|
|
360
|
+
maxConcurrentOverride: parallelism,
|
|
361
|
+
executionResources: entry.execution.resources.default_trial,
|
|
362
|
+
...(entry.modelCapturePlan ? { modelCapturePlan: entry.modelCapturePlan } : {}),
|
|
363
|
+
executionResourceSource: "submission-default",
|
|
364
|
+
executionStrategy: "local-task-slots-v1",
|
|
365
|
+
environmentBuildMode: entry.execution.build.mode,
|
|
366
|
+
...this.environmentImages.runOptions(),
|
|
367
|
+
executionWorker: {
|
|
368
|
+
workerId: remote ? "worker_remote_pool" : this.workerId,
|
|
369
|
+
provider: entry.execution.provider,
|
|
370
|
+
collisionDomainId: remote ? `remote-pool:${entry.execution.provider}` : this.collisionDomainId,
|
|
371
|
+
...(lease ? { parentAllocationId: lease.allocation.allocation_id } : {}),
|
|
372
|
+
},
|
|
373
|
+
...(fineGrained && !remote ? {
|
|
374
|
+
workItemAdmission: workItemAdmission({ dispatcher: this.workItems, request: entry.request, collisionDomainId: this.collisionDomainId }),
|
|
375
|
+
} : {}),
|
|
376
|
+
...(remote && this.remoteWork ? { remoteWorkExecutor: this.remoteWork.execute } : {}),
|
|
377
|
+
precreated: true,
|
|
378
|
+
resumeExisting: entry.resumeExisting,
|
|
379
|
+
root: this.root,
|
|
380
|
+
env: this.credentialEnv,
|
|
381
|
+
signal: controller.signal,
|
|
382
|
+
onEvent: this.onEvent,
|
|
383
|
+
...(remote ? {} : { dockerResourceReaper: this.dockerResourceReaper }),
|
|
384
|
+
onControlPhase: async (phase, work) => {
|
|
385
|
+
await this.updateControl(entry.directory, (control) => applyEvalPhase(control, phase, work?.queuedWorkItems, work?.terminalWorkItems));
|
|
386
|
+
},
|
|
387
|
+
onWorkItemState: async (workId, leaseId, state) => {
|
|
388
|
+
await this.updateControl(entry.directory, (control) => applyEvalWorkItem(control, workId, leaseId, state));
|
|
389
|
+
},
|
|
390
|
+
});
|
|
391
|
+
if (!await readJSON(path.join(entry.directory, "result.json"), null)) {
|
|
392
|
+
await atomicWriteJSON(path.join(entry.directory, "result.json"), result);
|
|
393
|
+
}
|
|
394
|
+
const state = terminalControlState(result.status);
|
|
395
|
+
await this.updateControl(entry.directory, (control) => {
|
|
396
|
+
const { allocation_id: _allocationId, ...released } = control;
|
|
397
|
+
return {
|
|
398
|
+
...settleEvalWorkItems(released),
|
|
399
|
+
state,
|
|
400
|
+
admitted_parallelism: fineGrained ? 0 : parallelism,
|
|
401
|
+
...(result.error ? { error: result.error } : {}),
|
|
402
|
+
};
|
|
403
|
+
});
|
|
404
|
+
}
|
|
405
|
+
async recordUnexpectedFailure(entry, error) {
|
|
406
|
+
const now = new Date().toISOString();
|
|
407
|
+
const cancelled = this.active.get(entry.evalId)?.controller.signal.aborted === true;
|
|
408
|
+
const code = cancelled ? "cancelled" : "control_plane_error";
|
|
409
|
+
const message = cancelled
|
|
410
|
+
? "eval was cancelled"
|
|
411
|
+
: safeDiagnosticMessage(error, credentialValuesFromEnv(entry.request.pass_env, this.credentialEnv));
|
|
412
|
+
await writeSyntheticEvalResult({ directory: entry.directory, evalId: entry.evalId, request: entry.request, status: cancelled ? "cancelled" : "failed", code, message, completedAt: now });
|
|
413
|
+
await this.updateControl(entry.directory, (control) => {
|
|
414
|
+
const { allocation_id: _allocationId, ...released } = control;
|
|
415
|
+
return { ...settleEvalWorkItems(released), state: cancelled ? "cancelled" : "failed", error: { code, message } };
|
|
416
|
+
});
|
|
417
|
+
this.onEvent({ type: "eval.scheduler.error", eval_id: entry.evalId, code });
|
|
418
|
+
}
|
|
419
|
+
async recoverInterruptedEvals() {
|
|
420
|
+
const remoteWork = this.remoteWork;
|
|
421
|
+
const recovered = await recoverPersistedEvals({
|
|
422
|
+
root: this.root, evalsRoot: this.evalsRoot, onEvent: this.onEvent, credentialEnv: this.credentialEnv,
|
|
423
|
+
...(remoteWork ? { recoverProviderLeases: (input) => remoteWork.recoverEvalLeases(input) } : {}),
|
|
424
|
+
});
|
|
425
|
+
for (const entry of recovered) {
|
|
426
|
+
const execution = entry.execution || defaultEvalExecutionPolicy(entry.request, { provider: this.provider, trialResources: this.trialResources, buildMode: "backend" });
|
|
427
|
+
assertExecutionPolicySupported(execution, execution.provider === this.provider ? this.provider : execution.provider);
|
|
428
|
+
if (execution.provider !== this.provider && !this.remoteWork)
|
|
429
|
+
throw new HitchError(`execution provider is unavailable: ${execution.provider}`, { code: "execution_provider_unavailable", exitCode: 10 });
|
|
430
|
+
const persistedPlan = entry.resumeExisting
|
|
431
|
+
? parseEvalExecutionPlan(await readJSON(path.join(entry.directory, "execution-plan.json")))
|
|
432
|
+
: null;
|
|
433
|
+
const capturePlan = persistedPlan ? persistedPlan.model_capture : await schedulerCapturePlan({ request: entry.request, execution, localProvider: this.provider, localStatus: this.providerSnapshot(), ...(this.remoteWork ? { remoteWork: this.remoteWork } : {}) });
|
|
434
|
+
this.queue.push(await this.queuedEval(entry.evalId, entry.request, execution, capturePlan, entry.directory, entry.resumeExisting));
|
|
435
|
+
}
|
|
436
|
+
}
|
|
437
|
+
async queuedEval(evalId, request, execution, modelCapturePlan, directory, resumeExisting = false) {
|
|
438
|
+
return schedulerQueuedEval({ evalId, request, execution, ...(modelCapturePlan ? { modelCapturePlan } : {}), directory, collisionDomainId: this.collisionDomainId, resumeExisting });
|
|
439
|
+
}
|
|
440
|
+
async updateControl(directory, update) {
|
|
441
|
+
return updateEvalControl(directory, update);
|
|
442
|
+
}
|
|
443
|
+
async emitPersisted(directory, evalId, event) {
|
|
444
|
+
return emitPersistedEvalEvent(directory, evalId, this.onEvent, event);
|
|
445
|
+
}
|
|
446
|
+
}
|
|
447
|
+
//# sourceMappingURL=eval-scheduler.js.map
|