agent-hitch 0.2.15 → 0.2.17
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 +7 -28
- package/README.zh-CN.md +6 -26
- package/dist/scripts/benchmark-dataset-verification.js +118 -0
- package/dist/scripts/benchmark-dataset-verification.js.map +1 -0
- package/dist/scripts/canary-external-endpoint.js +56 -0
- package/dist/scripts/canary-external-endpoint.js.map +1 -0
- package/dist/scripts/canary-native-training.js +8 -6
- package/dist/scripts/canary-native-training.js.map +1 -1
- package/dist/scripts/canary-remote-training.js +8 -6
- package/dist/scripts/canary-remote-training.js.map +1 -1
- package/dist/scripts/check-architecture.js +1 -1
- package/dist/scripts/check-architecture.js.map +1 -1
- package/dist/src/adapters/providers/codex.js +5 -2
- package/dist/src/adapters/providers/codex.js.map +1 -1
- package/dist/src/adapters/providers/model-call.js +1 -2
- package/dist/src/adapters/providers/model-call.js.map +1 -1
- package/dist/src/adapters/providers/training-tool.js +2 -0
- package/dist/src/adapters/providers/training-tool.js.map +1 -1
- package/dist/src/backends/harbor/artifact-pins.js +42 -0
- package/dist/src/backends/harbor/artifact-pins.js.map +1 -0
- package/dist/src/backends/harbor/backend.js +16 -8
- package/dist/src/backends/harbor/backend.js.map +1 -1
- package/dist/src/backends/harbor/bound-credential-helper.js +24 -0
- package/dist/src/backends/harbor/bound-credential-helper.js.map +1 -0
- package/dist/src/backends/harbor/credential-snapshot.js +71 -0
- package/dist/src/backends/harbor/credential-snapshot.js.map +1 -0
- package/dist/src/backends/harbor/environment-config.js +25 -2
- package/dist/src/backends/harbor/environment-config.js.map +1 -1
- package/dist/src/backends/harbor/environment-files.js +143 -0
- package/dist/src/backends/harbor/environment-files.js.map +1 -0
- package/dist/src/backends/harbor/evaluation-binding.js +160 -0
- package/dist/src/backends/harbor/evaluation-binding.js.map +1 -0
- package/dist/src/backends/harbor/host-credential-helper.js +5 -0
- package/dist/src/backends/harbor/host-credential-helper.js.map +1 -1
- package/dist/src/backends/harbor/index.js +2 -0
- package/dist/src/backends/harbor/index.js.map +1 -1
- package/dist/src/backends/harbor/model-proxy-config.js +3 -18
- package/dist/src/backends/harbor/model-proxy-config.js.map +1 -1
- package/dist/src/backends/harbor/portable-regrade.js +15 -3
- package/dist/src/backends/harbor/portable-regrade.js.map +1 -1
- package/dist/src/backends/harbor/python-runtime.js +71 -0
- package/dist/src/backends/harbor/python-runtime.js.map +1 -0
- package/dist/src/backends/harbor/tools.js +6 -4
- package/dist/src/backends/harbor/tools.js.map +1 -1
- package/dist/src/backends/index.js +3 -0
- package/dist/src/backends/index.js.map +1 -1
- package/dist/src/cli/arguments.js +30 -30
- package/dist/src/cli/arguments.js.map +1 -1
- package/dist/src/cli/commands/capabilities.js +7 -3
- package/dist/src/cli/commands/capabilities.js.map +1 -1
- package/dist/src/cli/commands/daemon.js +0 -17
- package/dist/src/cli/commands/daemon.js.map +1 -1
- package/dist/src/cli/commands/eval-binding.js +32 -0
- package/dist/src/cli/commands/eval-binding.js.map +1 -0
- package/dist/src/cli/commands/eval.js +21 -15
- package/dist/src/cli/commands/eval.js.map +1 -1
- package/dist/src/cli/commands/models.js +32 -45
- package/dist/src/cli/commands/models.js.map +1 -1
- package/dist/src/cli/commands/run.js +0 -35
- package/dist/src/cli/commands/run.js.map +1 -1
- package/dist/src/cli/main.js +0 -4
- package/dist/src/cli/main.js.map +1 -1
- package/dist/src/cli/output.js +5 -16
- package/dist/src/cli/output.js.map +1 -1
- package/dist/src/control-plane/eval-records.js +3 -1
- package/dist/src/control-plane/eval-records.js.map +1 -1
- package/dist/src/control-plane/eval-recovery.js +4 -1
- package/dist/src/control-plane/eval-recovery.js.map +1 -1
- package/dist/src/control-plane/eval-scheduler.js +16 -8
- package/dist/src/control-plane/eval-scheduler.js.map +1 -1
- package/dist/src/control-plane/index.js +0 -2
- package/dist/src/control-plane/index.js.map +1 -1
- package/dist/src/control-plane/model-capture-planning.js +3 -4
- package/dist/src/control-plane/model-capture-planning.js.map +1 -1
- package/dist/src/control-plane/remote-model-routes.js +4 -24
- package/dist/src/control-plane/remote-model-routes.js.map +1 -1
- package/dist/src/control-plane/remote-offer-contract.js +2 -2
- package/dist/src/control-plane/remote-offer-contract.js.map +1 -1
- package/dist/src/control-plane/remote-result-transport.js +7 -8
- package/dist/src/control-plane/remote-result-transport.js.map +1 -1
- package/dist/src/control-plane/remote-work-coordinator.js +2 -2
- package/dist/src/control-plane/remote-work-coordinator.js.map +1 -1
- package/dist/src/control-plane/remote-work-recovery.js +2 -2
- package/dist/src/control-plane/remote-work-recovery.js.map +1 -1
- package/dist/src/control-plane/remote-worker-protocol.js +1 -1
- package/dist/src/control-plane/remote-worker-protocol.js.map +1 -1
- package/dist/src/control-plane/remote-worker-runner.js +21 -6
- package/dist/src/control-plane/remote-worker-runner.js.map +1 -1
- package/dist/src/control-plane/remote-workers.js +2 -2
- package/dist/src/control-plane/remote-workers.js.map +1 -1
- package/dist/src/control-plane/rerun-recovery.js +1 -1
- package/dist/src/control-plane/rerun-recovery.js.map +1 -1
- package/dist/src/control-plane/rerun-scheduler.js +21 -14
- package/dist/src/control-plane/rerun-scheduler.js.map +1 -1
- package/dist/src/control-plane/rerun-source.js +5 -2
- package/dist/src/control-plane/rerun-source.js.map +1 -1
- package/dist/src/control-plane/rerun-submission.js +7 -5
- package/dist/src/control-plane/rerun-submission.js.map +1 -1
- package/dist/src/control-plane/synthetic-result.js +3 -2
- package/dist/src/control-plane/synthetic-result.js.map +1 -1
- package/dist/src/control-plane/worker-model-relay.js +1 -1
- package/dist/src/control-plane/worker-model-relay.js.map +1 -1
- package/dist/src/controller-runtime/hash.js +6 -0
- package/dist/src/controller-runtime/hash.js.map +1 -1
- package/dist/src/daemon/client.js +12 -35
- package/dist/src/daemon/client.js.map +1 -1
- package/dist/src/daemon/evaluation-binding.js +33 -0
- package/dist/src/daemon/evaluation-binding.js.map +1 -0
- package/dist/src/daemon/evaluation-submission.js +27 -0
- package/dist/src/daemon/evaluation-submission.js.map +1 -0
- package/dist/src/daemon/index.js +1 -0
- package/dist/src/daemon/index.js.map +1 -1
- package/dist/src/daemon/json-stream.js +98 -0
- package/dist/src/daemon/json-stream.js.map +1 -0
- package/dist/src/daemon/scheduler.js +1 -4
- package/dist/src/daemon/scheduler.js.map +1 -1
- package/dist/src/daemon/server.js +14 -54
- package/dist/src/daemon/server.js.map +1 -1
- package/dist/src/domain/endpoints.js +2 -0
- package/dist/src/domain/endpoints.js.map +1 -0
- package/dist/src/domain/evaluation-binding.js +5 -0
- package/dist/src/domain/evaluation-binding.js.map +1 -0
- package/dist/src/domain/index.js +2 -0
- package/dist/src/domain/index.js.map +1 -1
- package/dist/src/domain/runs.js.map +1 -1
- package/dist/src/evals/benchmark-adapter-manifest.js +95 -32
- package/dist/src/evals/benchmark-adapter-manifest.js.map +1 -1
- package/dist/src/evals/benchmark-verification.js +106 -0
- package/dist/src/evals/benchmark-verification.js.map +1 -0
- package/dist/src/evals/collect-only-rerun.js +3 -2
- package/dist/src/evals/collect-only-rerun.js.map +1 -1
- package/dist/src/evals/directory.js +9 -1
- package/dist/src/evals/directory.js.map +1 -1
- package/dist/src/evals/endpoint-attempts.js +112 -0
- package/dist/src/evals/endpoint-attempts.js.map +1 -0
- package/dist/src/evals/endpoint-usage.js +8 -0
- package/dist/src/evals/endpoint-usage.js.map +1 -0
- package/dist/src/evals/environment-binding.js +9 -0
- package/dist/src/evals/environment-binding.js.map +1 -0
- package/dist/src/evals/eval-artifact-planning.js +10 -1
- package/dist/src/evals/eval-artifact-planning.js.map +1 -1
- package/dist/src/evals/eval-finalization.js +2 -3
- package/dist/src/evals/eval-finalization.js.map +1 -1
- package/dist/src/evals/eval-logical-plan.js +2 -0
- package/dist/src/evals/eval-logical-plan.js.map +1 -1
- package/dist/src/evals/eval-result-builder.js +4 -0
- package/dist/src/evals/eval-result-builder.js.map +1 -1
- package/dist/src/evals/eval-state.js +22 -0
- package/dist/src/evals/eval-state.js.map +1 -0
- package/dist/src/evals/execution-plan.js +6 -2
- package/dist/src/evals/execution-plan.js.map +1 -1
- package/dist/src/evals/index.js +5 -1
- package/dist/src/evals/index.js.map +1 -1
- package/dist/src/evals/infrastructure-retry.js +2 -0
- package/dist/src/evals/infrastructure-retry.js.map +1 -1
- package/dist/src/evals/model-capture-runtime.js +9 -13
- package/dist/src/evals/model-capture-runtime.js.map +1 -1
- package/dist/src/evals/native-phase-evidence.js +8 -2
- package/dist/src/evals/native-phase-evidence.js.map +1 -1
- package/dist/src/evals/planned-execution.js +7 -2
- package/dist/src/evals/planned-execution.js.map +1 -1
- package/dist/src/evals/planned-progress-publisher.js +18 -5
- package/dist/src/evals/planned-progress-publisher.js.map +1 -1
- package/dist/src/evals/planned-retry-execution.js +1 -0
- package/dist/src/evals/planned-retry-execution.js.map +1 -1
- package/dist/src/evals/preparation-rerun.js +0 -1
- package/dist/src/evals/preparation-rerun.js.map +1 -1
- package/dist/src/evals/progress.js +92 -7
- package/dist/src/evals/progress.js.map +1 -1
- package/dist/src/evals/records.js +12 -2
- package/dist/src/evals/records.js.map +1 -1
- package/dist/src/evals/recovery.js +2 -0
- package/dist/src/evals/recovery.js.map +1 -1
- package/dist/src/evals/remote-rerun-execution.js +1 -3
- package/dist/src/evals/remote-rerun-execution.js.map +1 -1
- package/dist/src/evals/remote-verifier-rerun.js +1 -1
- package/dist/src/evals/remote-verifier-rerun.js.map +1 -1
- package/dist/src/evals/request.js +14 -45
- package/dist/src/evals/request.js.map +1 -1
- package/dist/src/evals/rerun-observation.js +31 -0
- package/dist/src/evals/rerun-observation.js.map +1 -0
- package/dist/src/evals/rerun.js +23 -28
- package/dist/src/evals/rerun.js.map +1 -1
- package/dist/src/evals/service.js +45 -42
- package/dist/src/evals/service.js.map +1 -1
- package/dist/src/evals/shared-runtime.js +125 -0
- package/dist/src/evals/shared-runtime.js.map +1 -0
- package/dist/src/evals/trial-import.js +10 -2
- package/dist/src/evals/trial-import.js.map +1 -1
- package/dist/src/evals/trial-publication-recovery.js +1 -1
- package/dist/src/evals/trial-publication-recovery.js.map +1 -1
- package/dist/src/evals/verifier-artifacts.js +1 -0
- package/dist/src/evals/verifier-artifacts.js.map +1 -1
- package/dist/src/evals/verifier-only-rerun.js +1 -1
- package/dist/src/evals/verifier-only-rerun.js.map +1 -1
- package/dist/src/evals/verifier-rerun-result.js +3 -2
- package/dist/src/evals/verifier-rerun-result.js.map +1 -1
- package/dist/src/evals/verifier-score-artifacts.js +9 -1
- package/dist/src/evals/verifier-score-artifacts.js.map +1 -1
- package/dist/src/foundation/config.js +0 -5
- package/dist/src/foundation/config.js.map +1 -1
- package/dist/src/inference/index.js +3 -22
- package/dist/src/inference/index.js.map +1 -1
- package/dist/src/inference/lock.js +44 -130
- package/dist/src/inference/lock.js.map +1 -1
- package/dist/src/model-access/endpoint-contract.js +127 -0
- package/dist/src/model-access/endpoint-contract.js.map +1 -0
- package/dist/src/model-access/endpoint-credentials.js +54 -0
- package/dist/src/model-access/endpoint-credentials.js.map +1 -0
- package/dist/src/model-access/endpoint-probe-stream.js +66 -0
- package/dist/src/model-access/endpoint-probe-stream.js.map +1 -0
- package/dist/src/model-access/endpoint-verification.js +76 -0
- package/dist/src/model-access/endpoint-verification.js.map +1 -0
- package/dist/src/model-access/endpoints.js +105 -0
- package/dist/src/model-access/endpoints.js.map +1 -0
- package/dist/src/model-access/index.js +5 -1
- package/dist/src/model-access/index.js.map +1 -1
- package/dist/src/model-access/proxy.js +57 -14
- package/dist/src/model-access/proxy.js.map +1 -1
- package/dist/src/model-access/remote-model.js +21 -29
- package/dist/src/model-access/remote-model.js.map +1 -1
- package/dist/src/model-access/usage.js +30 -0
- package/dist/src/model-access/usage.js.map +1 -0
- package/dist/src/runs/endpoint-runtime.js +40 -0
- package/dist/src/runs/endpoint-runtime.js.map +1 -0
- package/dist/src/runs/executor.js +28 -42
- package/dist/src/runs/executor.js.map +1 -1
- package/dist/src/runs/local-inference-environment.js +5 -53
- package/dist/src/runs/local-inference-environment.js.map +1 -1
- package/dist/src/runs/manifest.js +1 -0
- package/dist/src/runs/manifest.js.map +1 -1
- package/dist/src/runs/outcome.js +3 -10
- package/dist/src/runs/outcome.js.map +1 -1
- package/dist/src/runs/queued.js +3 -2
- package/dist/src/runs/queued.js.map +1 -1
- package/dist/src/runs/records.js +3 -0
- package/dist/src/runs/records.js.map +1 -1
- package/dist/src/runs/request.js +10 -38
- package/dist/src/runs/request.js.map +1 -1
- package/dist/src/workers/remote-harbor-work-spec.js +8 -10
- package/dist/src/workers/remote-harbor-work-spec.js.map +1 -1
- package/dist/src/workers/remote-harbor-worker.js +3 -0
- package/dist/src/workers/remote-harbor-worker.js.map +1 -1
- package/docs/evidence/eval-preparation-2026-10-09.json +393 -0
- package/docs/evidence/eval-preparation-2026-10-09.md +178 -0
- package/docs/local-model-inference-spec.zh-CN.md +4 -663
- package/docs/resource-storage.zh-CN.md +116 -0
- package/docs/schemas/endpoint-evaluation-attempt.schema.json +29 -0
- package/docs/schemas/eval-artifact-pin.schema.json +14 -0
- package/docs/schemas/eval-execution-plan.schema.json +486 -105
- package/docs/schemas/eval-progress.schema.json +1 -1
- package/docs/schemas/eval-request.schema.json +149 -18
- package/docs/schemas/eval-rerun-submission.schema.json +8 -1
- package/docs/schemas/eval-result.schema.json +310 -56
- package/docs/schemas/evaluation-binding.schema.json +15 -0
- package/docs/schemas/external-endpoint-binding.schema.json +153 -0
- package/docs/schemas/external-endpoint-config.schema.json +93 -0
- package/docs/schemas/inference-execution-evidence.schema.json +253 -3
- package/docs/schemas/inference-lock-v3.schema.json +6 -0
- package/docs/schemas/inference-lock.schema.json +842 -341
- package/docs/schemas/ordered-eval-control.schema.json +7 -1
- package/docs/schemas/ready-endpoint-binding.schema.json +201 -0
- package/docs/schemas/run-manifest.schema.json +6 -0
- package/docs/schemas/run-request.schema.json +122 -20
- package/examples/external-inference/.env.example +16 -0
- package/examples/external-inference/README.md +203 -0
- package/examples/external-inference/accept.mjs +297 -0
- package/examples/external-inference/compose.gateway.yaml +119 -0
- package/examples/external-inference/compose.single.yaml +64 -0
- package/examples/external-inference/compose.tp2.yaml +15 -0
- package/examples/external-inference/deployment.example.json +58 -0
- package/examples/external-inference/endpoint.chat-completions.example.json +13 -0
- package/examples/external-inference/endpoint.responses.example.json +13 -0
- package/integrations/harbor/hitch_build_inputs.py +213 -0
- package/integrations/harbor/hitch_build_process.py +126 -0
- package/integrations/harbor/hitch_compose_preparation.py +274 -0
- package/integrations/harbor/hitch_harbor_agent.py +25 -39
- package/integrations/harbor/hitch_harbor_environment.py +96 -3
- package/integrations/harbor/hitch_image_cache.py +357 -0
- package/integrations/harbor/hitch_python_runtime.py +303 -0
- package/integrations/harbor/hitch_shared_runtime.py +123 -0
- package/integrations/model-call/cli.js +11 -10
- package/integrations/training-tool/cli.js +56 -15
- package/node_modules/smol-toml/LICENSE +24 -0
- package/node_modules/smol-toml/README.md +418 -0
- package/node_modules/smol-toml/dist/date.d.ts +41 -0
- package/node_modules/smol-toml/dist/date.js +127 -0
- package/node_modules/smol-toml/dist/error.d.ts +38 -0
- package/node_modules/smol-toml/dist/error.js +63 -0
- package/node_modules/smol-toml/dist/extract.js +69 -0
- package/node_modules/smol-toml/dist/index.cjs +734 -0
- package/node_modules/smol-toml/dist/index.d.ts +43 -0
- package/node_modules/smol-toml/dist/index.js +33 -0
- package/node_modules/smol-toml/dist/parse.d.ts +36 -0
- package/node_modules/smol-toml/dist/parse.js +149 -0
- package/node_modules/smol-toml/dist/primitive.js +238 -0
- package/node_modules/smol-toml/dist/stringify.d.ts +31 -0
- package/node_modules/smol-toml/dist/stringify.js +181 -0
- package/node_modules/smol-toml/dist/struct.js +179 -0
- package/node_modules/smol-toml/dist/util.d.ts +38 -0
- package/node_modules/smol-toml/dist/util.js +89 -0
- package/node_modules/smol-toml/package.json +68 -0
- package/package.json +15 -4
- package/dist/src/cli/commands/local.js +0 -130
- package/dist/src/cli/commands/local.js.map +0 -1
- package/dist/src/cli/commands/model-node.js +0 -26
- package/dist/src/cli/commands/model-node.js.map +0 -1
- package/dist/src/control-plane/inference-manager.js +0 -357
- package/dist/src/control-plane/inference-manager.js.map +0 -1
- package/dist/src/control-plane/managed-gateway-receipt.js +0 -26
- package/dist/src/control-plane/managed-gateway-receipt.js.map +0 -1
- package/dist/src/control-plane/managed-service-recovery.js +0 -158
- package/dist/src/control-plane/managed-service-recovery.js.map +0 -1
- package/dist/src/daemon/inference-prepare.js +0 -47
- package/dist/src/daemon/inference-prepare.js.map +0 -1
- package/dist/src/inference/daemon-capacity.js +0 -41
- package/dist/src/inference/daemon-capacity.js.map +0 -1
- package/dist/src/inference/device-reservation.js +0 -58
- package/dist/src/inference/device-reservation.js.map +0 -1
- package/dist/src/inference/doctor.js +0 -95
- package/dist/src/inference/doctor.js.map +0 -1
- package/dist/src/inference/generation-recovery.js +0 -83
- package/dist/src/inference/generation-recovery.js.map +0 -1
- package/dist/src/inference/managed-launcher.js +0 -35
- package/dist/src/inference/managed-launcher.js.map +0 -1
- package/dist/src/inference/materialize.js +0 -66
- package/dist/src/inference/materialize.js.map +0 -1
- package/dist/src/inference/model-location.js +0 -34
- package/dist/src/inference/model-location.js.map +0 -1
- package/dist/src/inference/node-client.js +0 -167
- package/dist/src/inference/node-client.js.map +0 -1
- package/dist/src/inference/node-registry.js +0 -79
- package/dist/src/inference/node-registry.js.map +0 -1
- package/dist/src/inference/observation.js +0 -79
- package/dist/src/inference/observation.js.map +0 -1
- package/dist/src/inference/preflight.js +0 -132
- package/dist/src/inference/preflight.js.map +0 -1
- package/dist/src/inference/process-sglang.js +0 -169
- package/dist/src/inference/process-sglang.js.map +0 -1
- package/dist/src/inference/protocol-probe.js +0 -29
- package/dist/src/inference/protocol-probe.js.map +0 -1
- package/dist/src/inference/remote-model-store.js +0 -24
- package/dist/src/inference/remote-model-store.js.map +0 -1
- package/dist/src/inference/runtime-catalog.js +0 -41
- package/dist/src/inference/runtime-catalog.js.map +0 -1
- package/dist/src/inference/runtime-store.js +0 -93
- package/dist/src/inference/runtime-store.js.map +0 -1
- package/dist/src/inference/service-attachment.js +0 -46
- package/dist/src/inference/service-attachment.js.map +0 -1
- package/dist/src/inference/service-inspection.js +0 -38
- package/dist/src/inference/service-inspection.js.map +0 -1
- package/dist/src/inference/sglang.js +0 -279
- package/dist/src/inference/sglang.js.map +0 -1
- package/dist/src/inference/supervisor.js +0 -460
- package/dist/src/inference/supervisor.js.map +0 -1
- package/dist/src/model-access/local-gateway-request.js +0 -139
- package/dist/src/model-access/local-gateway-request.js.map +0 -1
- package/dist/src/model-access/local-gateway.js +0 -323
- package/dist/src/model-access/local-gateway.js.map +0 -1
- package/dist/src/runs/local-inference-run.js +0 -53
- package/dist/src/runs/local-inference-run.js.map +0 -1
|
@@ -1,158 +0,0 @@
|
|
|
1
|
-
import { readdir } from "node:fs/promises";
|
|
2
|
-
import path from "node:path";
|
|
3
|
-
import { parseExecutionLease, parsePhysicalExecution, readExecutionLeases } from "../evals/index.js";
|
|
4
|
-
import { HitchError, readJSON, sha256JSON, statePaths } from "../foundation/index.js";
|
|
5
|
-
import { validateInferenceLockShape } from "../inference/index.js";
|
|
6
|
-
import { remoteModelBinding } from "../model-access/index.js";
|
|
7
|
-
import { parseEvalControl } from "./eval-records.js";
|
|
8
|
-
import { RemoteWorkInputStore } from "./remote-work-inputs.js";
|
|
9
|
-
import { verifyManagedGateway } from "./managed-gateway-receipt.js";
|
|
10
|
-
import { orderedEvalCancelled } from "./ordered-eval-control.js";
|
|
11
|
-
/** Private routes already seal the original gateway credentials before dispatch. */
|
|
12
|
-
export class ManagedServiceRecovery {
|
|
13
|
-
root;
|
|
14
|
-
registry;
|
|
15
|
-
protocol;
|
|
16
|
-
constructor(root, registry, protocol) {
|
|
17
|
-
this.root = root;
|
|
18
|
-
this.registry = registry;
|
|
19
|
-
this.protocol = protocol;
|
|
20
|
-
}
|
|
21
|
-
async select(records) {
|
|
22
|
-
const groups = new Map();
|
|
23
|
-
const evals = await readdir(statePaths(this.root).evals, { withFileTypes: true }).catch(error => {
|
|
24
|
-
if (error.code === "ENOENT")
|
|
25
|
-
return [];
|
|
26
|
-
throw error;
|
|
27
|
-
});
|
|
28
|
-
const inputs = new RemoteWorkInputStore(this.root);
|
|
29
|
-
for (const directory of evals) {
|
|
30
|
-
if (!directory.isDirectory() || !/^eval_[a-f0-9]{32}$/.test(directory.name))
|
|
31
|
-
continue;
|
|
32
|
-
const evalDirectory = path.join(statePaths(this.root).evals, directory.name);
|
|
33
|
-
for (const lease of await readExecutionLeases(evalDirectory)) {
|
|
34
|
-
if (lease.provider === "local-docker" || !["offered", "accepted", "running"].includes(lease.state) || Date.parse(lease.expires_at) <= Date.now())
|
|
35
|
-
continue;
|
|
36
|
-
const offer = await this.protocol.findOfferForLease(lease.worker_id, lease.lease_id);
|
|
37
|
-
if (!offer || offer.state !== "accepted")
|
|
38
|
-
continue;
|
|
39
|
-
const worker = await this.registry.get(offer.worker_id);
|
|
40
|
-
if (!worker || worker.revoked_at || worker.generation !== offer.generation)
|
|
41
|
-
continue;
|
|
42
|
-
const ref = offer.inputs?.find(item => item.kind === "work-spec");
|
|
43
|
-
if (!ref)
|
|
44
|
-
continue;
|
|
45
|
-
const spec = await readJSON((await inputs.verify(ref)).path);
|
|
46
|
-
if (spec.model_binding?.kind !== "managed-inference")
|
|
47
|
-
continue;
|
|
48
|
-
const target = await this.protocol.modelRoutes.recoveryTarget(offer);
|
|
49
|
-
if (!target || target.kind !== "managed-inference" || target.binding.kind !== "managed-node"
|
|
50
|
-
|| spec.schema_version !== "2" || sha256JSON(spec.work) !== sha256JSON(offer.work)
|
|
51
|
-
|| sha256JSON(spec.model_binding) !== sha256JSON(remoteModelBinding(target)))
|
|
52
|
-
throw ambiguous("managed route differs from its sealed work input");
|
|
53
|
-
const url = new URL(target.binding.base_url), match = url.pathname.match(/^\/runs\/(run_[a-f0-9]{32})\/v1\/$/);
|
|
54
|
-
if (!match || url.protocol !== "http:" || url.hostname !== "127.0.0.1" || !url.port || url.username || url.password || url.search || url.hash
|
|
55
|
-
|| !/^[a-f0-9]{64}$/.test(target.credential))
|
|
56
|
-
throw ambiguous("original private gateway address or credential is invalid");
|
|
57
|
-
const ownerId = match[1];
|
|
58
|
-
const physical = spec.physical_execution ? parsePhysicalExecution(spec.physical_execution) : undefined;
|
|
59
|
-
if (physical?.kind === "verifier-only")
|
|
60
|
-
throw ambiguous("verifier work cannot recover a model gateway");
|
|
61
|
-
const rerunId = physical && physical.kind !== "physical-infrastructure-retry" ? physical.rerun_id : undefined;
|
|
62
|
-
if (ownerId !== `run_${(rerunId ?? lease.eval_id).slice((rerunId ? "rerun_" : "eval_").length)}`)
|
|
63
|
-
throw ambiguous("gateway owner differs from its original eval or rerun");
|
|
64
|
-
const cacheScopeOwner = rerunId ? `${lease.eval_id}:${rerunId}` : lease.eval_id;
|
|
65
|
-
const evidenceOwner = { kind: "eval", eval_id: lease.eval_id, ...(rerunId ? { rerun_id: rerunId } : {}) };
|
|
66
|
-
if (await this.cancelled(evidenceOwner))
|
|
67
|
-
continue;
|
|
68
|
-
const candidates = records.filter(record => record.inference_id === target.binding.inference_id && record.lease_owner_ids.includes(ownerId)
|
|
69
|
-
&& sha256JSON(record.model_node ?? null) === sha256JSON(target.binding.model_node));
|
|
70
|
-
if (candidates.length !== 1 || candidates[0].state !== "ready")
|
|
71
|
-
throw ambiguous("accepted model work has no unique original ready service");
|
|
72
|
-
const record = candidates[0];
|
|
73
|
-
await verifyManagedGateway(this.root, { serviceId: record.service_id, epoch: record.epoch, inferenceId: record.inference_id,
|
|
74
|
-
isolationKey: record.isolation_key, ownerId, cacheScopeOwner, evidenceOwner, binding: target.binding, credential: target.credential });
|
|
75
|
-
const scopeDirectory = rerunId ? path.join(evalDirectory, "reruns", rerunId) : evalDirectory;
|
|
76
|
-
const lock = validateInferenceLockShape(await readJSON(path.join(scopeDirectory, "inference", "lock.json")));
|
|
77
|
-
const execution = await readJSON(path.join(scopeDirectory, "inference", "execution.json"));
|
|
78
|
-
if (record.inference_id !== lock.inference_id || lock.model_id !== target.modelId
|
|
79
|
-
|| lock.protocol.api !== target.binding.api || lock.generation.max_output_tokens !== target.maxOutputTokens
|
|
80
|
-
|| sha256JSON(lock.model_node ?? null) !== sha256JSON(record.model_node)
|
|
81
|
-
|| record.isolation_key !== sha256JSON({ inference_id: lock.inference_id, cache_scope_owner: cacheScopeOwner })
|
|
82
|
-
|| execution.schema_version !== "2" || execution.run_id !== ownerId || execution.eval_id !== lease.eval_id || execution.rerun_id !== rerunId
|
|
83
|
-
|| sha256JSON(execution.service) !== sha256JSON({ service_id: record.service_id, epoch: record.epoch, isolation_key: record.isolation_key })) {
|
|
84
|
-
throw ambiguous("original inference execution evidence differs from the active gateway");
|
|
85
|
-
}
|
|
86
|
-
const key = `${record.service_id}:${ownerId}`;
|
|
87
|
-
const prior = groups.get(key);
|
|
88
|
-
if (prior) {
|
|
89
|
-
if (sha256JSON(prior.plan.target) !== sha256JSON(target))
|
|
90
|
-
throw ambiguous("one model owner has conflicting private gateways");
|
|
91
|
-
prior.seeds.push(offer);
|
|
92
|
-
}
|
|
93
|
-
else {
|
|
94
|
-
const seeds = [offer];
|
|
95
|
-
const plan = { record, ownerId, lock, cacheScopeOwner, evidenceOwner, target, port: Number(url.port),
|
|
96
|
-
state: () => this.state(seeds, target, evidenceOwner) };
|
|
97
|
-
groups.set(key, { plan, seeds });
|
|
98
|
-
}
|
|
99
|
-
}
|
|
100
|
-
}
|
|
101
|
-
const plans = [];
|
|
102
|
-
for (const { plan } of groups.values())
|
|
103
|
-
if (await plan.state() !== "finished")
|
|
104
|
-
plans.push(plan);
|
|
105
|
-
return plans;
|
|
106
|
-
}
|
|
107
|
-
async cancelled(owner) {
|
|
108
|
-
if (await orderedEvalCancelled(this.root, owner.eval_id, owner.rerun_id))
|
|
109
|
-
return true;
|
|
110
|
-
const directory = path.join(statePaths(this.root).evals, owner.eval_id);
|
|
111
|
-
if (owner.rerun_id) {
|
|
112
|
-
const state = await readJSON(path.join(directory, "reruns", owner.rerun_id, "state.json"));
|
|
113
|
-
if (state.schema_version !== "1" || state.eval_id !== owner.eval_id || state.rerun_id !== owner.rerun_id
|
|
114
|
-
|| !["queued", "running", "completed", "failed", "cancelled"].includes(String(state.status)))
|
|
115
|
-
throw ambiguous("rerun state identity changed");
|
|
116
|
-
const marker = await readJSON(path.join(directory, "reruns", owner.rerun_id, "cancellation.json"), null);
|
|
117
|
-
if (marker && (marker.schema_version !== "1" || marker.eval_id !== owner.eval_id || marker.rerun_id !== owner.rerun_id))
|
|
118
|
-
throw ambiguous("rerun cancellation identity changed");
|
|
119
|
-
return !!marker || ["completed", "failed", "cancelled"].includes(String(state.status));
|
|
120
|
-
}
|
|
121
|
-
const control = parseEvalControl(await readJSON(path.join(directory, "control.json")));
|
|
122
|
-
if (control.eval_id !== owner.eval_id)
|
|
123
|
-
throw ambiguous("eval control identity changed");
|
|
124
|
-
return !!control.cancel_requested_at || ["cancelling", "cancelled", "failed", "succeeded"].includes(control.state);
|
|
125
|
-
}
|
|
126
|
-
async state(seeds, target, owner) {
|
|
127
|
-
if (await this.cancelled(owner))
|
|
128
|
-
return "finished";
|
|
129
|
-
let waiting = false;
|
|
130
|
-
for (const seed of seeds) {
|
|
131
|
-
const offer = await this.protocol.getOffer(seed.worker_id, seed.offer_id);
|
|
132
|
-
if (!offer)
|
|
133
|
-
throw ambiguous("accepted model offer disappeared during recovery");
|
|
134
|
-
if (sha256JSON(identity(offer)) !== sha256JSON(identity(seed)))
|
|
135
|
-
throw ambiguous("accepted model offer identity changed");
|
|
136
|
-
if (offer.state !== "accepted")
|
|
137
|
-
continue;
|
|
138
|
-
const lease = parseExecutionLease(await readJSON(path.join(statePaths(this.root).evals, owner.eval_id, "leases", `${offer.lease.lease_id}.json`)));
|
|
139
|
-
if (lease.epoch !== offer.lease.epoch || !["offered", "accepted", "running"].includes(lease.state) || Date.parse(lease.expires_at) <= Date.now())
|
|
140
|
-
continue;
|
|
141
|
-
const worker = await this.registry.get(offer.worker_id);
|
|
142
|
-
if (!worker || worker.revoked_at || worker.generation !== offer.generation)
|
|
143
|
-
continue;
|
|
144
|
-
await this.protocol.modelRoutes.authorize(offer);
|
|
145
|
-
if (sha256JSON(await this.protocol.modelRoutes.recoveryTarget(offer)) !== sha256JSON(target))
|
|
146
|
-
throw ambiguous("private model route changed during recovery");
|
|
147
|
-
if (worker.worker.status === "ready")
|
|
148
|
-
return "active";
|
|
149
|
-
waiting = true;
|
|
150
|
-
}
|
|
151
|
-
return waiting ? "waiting" : "finished";
|
|
152
|
-
}
|
|
153
|
-
}
|
|
154
|
-
function identity(offer) {
|
|
155
|
-
return { worker: offer.worker_id, generation: offer.generation, offer: offer.offer_id, nonce: offer.nonce, lease: offer.lease, work: offer.work, inputs: offer.inputs };
|
|
156
|
-
}
|
|
157
|
-
function ambiguous(message) { return new HitchError(message, { code: "inference_recovery_ambiguous", exitCode: 12 }); }
|
|
158
|
-
//# sourceMappingURL=managed-service-recovery.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"managed-service-recovery.js","sourceRoot":"","sources":["../../../src/control-plane/managed-service-recovery.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAC3C,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,mBAAmB,EAAE,sBAAsB,EAAE,mBAAmB,EAAE,MAAM,mBAAmB,CAAC;AACrG,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACtF,OAAO,EAAE,0BAA0B,EAAE,MAAM,uBAAuB,CAAC;AACnE,OAAO,EAAE,kBAAkB,EAAE,MAAM,0BAA0B,CAAC;AAE9D,OAAO,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAGrD,OAAO,EAAE,oBAAoB,EAAE,MAAM,yBAAyB,CAAC;AAC/D,OAAO,EAAE,oBAAoB,EAAE,MAAM,8BAA8B,CAAC;AACpE,OAAO,EAAE,oBAAoB,EAAE,MAAM,2BAA2B,CAAC;AAejE,oFAAoF;AACpF,MAAM,OAAO,sBAAsB;IACZ;IAAuB;IAAyC;IAArF,YAAqB,IAAY,EAAW,QAA8B,EAAW,QAA8B;QAA9F,SAAI,GAAJ,IAAI,CAAQ;QAAW,aAAQ,GAAR,QAAQ,CAAsB;QAAW,aAAQ,GAAR,QAAQ,CAAsB;IAAG,CAAC;IAEvH,KAAK,CAAC,MAAM,CAAC,OAAmC;QAC9C,MAAM,MAAM,GAAG,IAAI,GAAG,EAA4E,CAAC;QACnG,MAAM,KAAK,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,EAAE;YAC9F,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,EAAE,CAAC;YAAC,MAAM,KAAK,CAAC;QACjF,CAAC,CAAC,CAAC;QACH,MAAM,MAAM,GAAG,IAAI,oBAAoB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACnD,KAAK,MAAM,SAAS,IAAI,KAAK,EAAE,CAAC;YAC9B,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,IAAI,CAAC,qBAAqB,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;gBAAE,SAAS;YACtF,MAAM,aAAa,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,SAAS,CAAC,IAAI,CAAC,CAAC;YAC7E,KAAK,MAAM,KAAK,IAAI,MAAM,mBAAmB,CAAC,aAAa,CAAC,EAAE,CAAC;gBAC7D,IAAI,KAAK,CAAC,QAAQ,KAAK,cAAc,IAAI,CAAC,CAAC,SAAS,EAAE,UAAU,EAAE,SAAS,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,UAAU,CAAC,IAAI,IAAI,CAAC,GAAG,EAAE;oBAAE,SAAS;gBAC3J,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,iBAAiB,CAAC,KAAK,CAAC,SAAS,EAAE,KAAK,CAAC,QAAQ,CAAC,CAAC;gBACrF,IAAI,CAAC,KAAK,IAAI,KAAK,CAAC,KAAK,KAAK,UAAU;oBAAE,SAAS;gBACnD,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;gBACxD,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,UAAU,IAAI,MAAM,CAAC,UAAU,KAAK,KAAK,CAAC,UAAU;oBAAE,SAAS;gBACrF,MAAM,GAAG,GAAG,KAAK,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,WAAW,CAAC,CAAC;gBAAC,IAAI,CAAC,GAAG;oBAAE,SAAS;gBACtF,MAAM,IAAI,GAAG,MAAM,QAAQ,CAA0B,CAAC,MAAM,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;gBACtF,IAAK,IAAI,CAAC,aAAgD,EAAE,IAAI,KAAK,mBAAmB;oBAAE,SAAS;gBACnG,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC;gBACrE,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,IAAI,KAAK,mBAAmB,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,KAAK,cAAc;uBACvF,IAAI,CAAC,cAAc,KAAK,GAAG,IAAI,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC;uBAC/E,UAAU,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,UAAU,CAAC,kBAAkB,CAAC,MAAM,CAAC,CAAC;oBAAE,MAAM,SAAS,CAAC,kDAAkD,CAAC,CAAC;gBACpJ,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,KAAK,GAAG,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,oCAAoC,CAAC,CAAC;gBAC/G,IAAI,CAAC,KAAK,IAAI,GAAG,CAAC,QAAQ,KAAK,OAAO,IAAI,GAAG,CAAC,QAAQ,KAAK,WAAW,IAAI,CAAC,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,QAAQ,IAAI,GAAG,CAAC,QAAQ,IAAI,GAAG,CAAC,MAAM,IAAI,GAAG,CAAC,IAAI;uBACxI,CAAC,gBAAgB,CAAC,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC;oBAAE,MAAM,SAAS,CAAC,2DAA2D,CAAC,CAAC;gBAC7H,MAAM,OAAO,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC;gBAC1B,MAAM,QAAQ,GAAG,IAAI,CAAC,kBAAkB,CAAC,CAAC,CAAC,sBAAsB,CAAC,IAAI,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;gBACvG,IAAI,QAAQ,EAAE,IAAI,KAAK,eAAe;oBAAE,MAAM,SAAS,CAAC,8CAA8C,CAAC,CAAC;gBACxG,MAAM,OAAO,GAAG,QAAQ,IAAI,QAAQ,CAAC,IAAI,KAAK,+BAA+B,CAAC,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC;gBAC9G,IAAI,OAAO,KAAK,OAAO,CAAC,OAAO,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,EAAE;oBAAE,MAAM,SAAS,CAAC,uDAAuD,CAAC,CAAC;gBAC3K,MAAM,eAAe,GAAG,OAAO,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,OAAO,IAAI,OAAO,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC;gBAChF,MAAM,aAAa,GAAG,EAAE,IAAI,EAAE,MAAe,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC;gBACnH,IAAI,MAAM,IAAI,CAAC,SAAS,CAAC,aAAa,CAAC;oBAAE,SAAS;gBAClD,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,YAAY,KAAK,MAAM,CAAC,OAAO,CAAC,YAAY,IAAI,MAAM,CAAC,eAAe,CAAC,QAAQ,CAAC,OAAO,CAAC;uBACtI,UAAU,CAAC,MAAM,CAAC,UAAU,IAAI,IAAI,CAAC,KAAK,UAAU,CAAC,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,CAAC;gBACtF,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC,IAAI,UAAU,CAAC,CAAC,CAAE,CAAC,KAAK,KAAK,OAAO;oBAAE,MAAM,SAAS,CAAC,0DAA0D,CAAC,CAAC;gBAC7I,MAAM,MAAM,GAAG,UAAU,CAAC,CAAC,CAAE,CAAC;gBAC9B,MAAM,oBAAoB,CAAC,IAAI,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,WAAW,EAAE,MAAM,CAAC,YAAY;oBACzH,YAAY,EAAE,MAAM,CAAC,aAAa,EAAE,OAAO,EAAE,eAAe,EAAE,aAAa,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU,EAAE,CAAC,CAAC;gBACzI,MAAM,cAAc,GAAG,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,QAAQ,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC;gBAC7F,MAAM,IAAI,GAAG,0BAA0B,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,EAAE,WAAW,CAAC,CAAC,CAAC,CAAC;gBAC7G,MAAM,SAAS,GAAG,MAAM,QAAQ,CAA0B,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,EAAE,gBAAgB,CAAC,CAAC,CAAC;gBACpH,IAAI,MAAM,CAAC,YAAY,KAAK,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,QAAQ,KAAK,MAAM,CAAC,OAAO;uBAC5E,IAAI,CAAC,QAAQ,CAAC,GAAG,KAAK,MAAM,CAAC,OAAO,CAAC,GAAG,IAAI,IAAI,CAAC,UAAU,CAAC,iBAAiB,KAAK,MAAM,CAAC,eAAe;uBACxG,UAAU,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,KAAK,UAAU,CAAC,MAAM,CAAC,UAAU,CAAC;uBACrE,MAAM,CAAC,aAAa,KAAK,UAAU,CAAC,EAAE,YAAY,EAAE,IAAI,CAAC,YAAY,EAAE,iBAAiB,EAAE,eAAe,EAAE,CAAC;uBAC5G,SAAS,CAAC,cAAc,KAAK,GAAG,IAAI,SAAS,CAAC,MAAM,KAAK,OAAO,IAAI,SAAS,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO,IAAI,SAAS,CAAC,QAAQ,KAAK,OAAO;uBACzI,UAAU,CAAC,SAAS,CAAC,OAAO,CAAC,KAAK,UAAU,CAAC,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,aAAa,EAAE,MAAM,CAAC,aAAa,EAAE,CAAC,EAAE,CAAC;oBAC/I,MAAM,SAAS,CAAC,uEAAuE,CAAC,CAAC;gBAC3F,CAAC;gBACD,MAAM,GAAG,GAAG,GAAG,MAAM,CAAC,UAAU,IAAI,OAAO,EAAE,CAAC;gBAC9C,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;gBAC9B,IAAI,KAAK,EAAE,CAAC;oBACV,IAAI,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,UAAU,CAAC,MAAM,CAAC;wBAAE,MAAM,SAAS,CAAC,kDAAkD,CAAC,CAAC;oBAC9H,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;gBAC1B,CAAC;qBAAM,CAAC;oBACN,MAAM,KAAK,GAAG,CAAC,KAAK,CAAC,CAAC;oBACtB,MAAM,IAAI,GAA+B,EAAE,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,eAAe,EAAE,aAAa,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC;wBAC9H,KAAK,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,MAAM,EAAE,aAAa,CAAC,EAAE,CAAC;oBAC1D,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC;gBACnC,CAAC;YACH,CAAC;QACH,CAAC;QACD,MAAM,KAAK,GAAiC,EAAE,CAAC;QAC/C,KAAK,MAAM,EAAE,IAAI,EAAE,IAAI,MAAM,CAAC,MAAM,EAAE;YAAE,IAAI,MAAM,IAAI,CAAC,KAAK,EAAE,KAAK,UAAU;gBAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAChG,OAAO,KAAK,CAAC;IACf,CAAC;IAEO,KAAK,CAAC,SAAS,CAAC,KAAkD;QACxE,IAAI,MAAM,oBAAoB,CAAC,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,QAAQ,CAAC;YAAE,OAAO,IAAI,CAAC;QACtF,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;QACxE,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;YACnB,MAAM,KAAK,GAAG,MAAM,QAAQ,CAA0B,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,YAAY,CAAC,CAAC,CAAC;YACpH,IAAI,KAAK,CAAC,cAAc,KAAK,GAAG,IAAI,KAAK,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO,IAAI,KAAK,CAAC,QAAQ,KAAK,KAAK,CAAC,QAAQ;mBACnG,CAAC,CAAC,QAAQ,EAAE,SAAS,EAAE,WAAW,EAAE,QAAQ,EAAE,WAAW,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;gBAAE,MAAM,SAAS,CAAC,8BAA8B,CAAC,CAAC;YAChJ,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,mBAAmB,CAAC,EAAE,IAAI,CAAC,CAAC;YACzI,IAAI,MAAM,IAAI,CAAC,MAAM,CAAC,cAAc,KAAK,GAAG,IAAI,MAAM,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO,IAAI,MAAM,CAAC,QAAQ,KAAK,KAAK,CAAC,QAAQ,CAAC;gBAAE,MAAM,SAAS,CAAC,qCAAqC,CAAC,CAAC;YAChL,OAAO,CAAC,CAAC,MAAM,IAAI,CAAC,WAAW,EAAE,QAAQ,EAAE,WAAW,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;QACzF,CAAC;QACD,MAAM,OAAO,GAAG,gBAAgB,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC,CAAC,CAAC,CAAC;QACvF,IAAI,OAAO,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO;YAAE,MAAM,SAAS,CAAC,+BAA+B,CAAC,CAAC;QACxF,OAAO,CAAC,CAAC,OAAO,CAAC,mBAAmB,IAAI,CAAC,YAAY,EAAE,WAAW,EAAE,QAAQ,EAAE,WAAW,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;IACrH,CAAC;IAEO,KAAK,CAAC,KAAK,CAAC,KAA0B,EAAE,MAAc,EAAE,KAAkD;QAChH,IAAI,MAAM,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC;YAAE,OAAO,UAAU,CAAC;QACnD,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC1E,IAAI,CAAC,KAAK;gBAAE,MAAM,SAAS,CAAC,kDAAkD,CAAC,CAAC;YAChF,IAAI,UAAU,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,KAAK,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;gBAAE,MAAM,SAAS,CAAC,uCAAuC,CAAC,CAAC;YACzH,IAAI,KAAK,CAAC,KAAK,KAAK,UAAU;gBAAE,SAAS;YACzC,MAAM,KAAK,GAAG,mBAAmB,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,GAAG,KAAK,CAAC,KAAK,CAAC,QAAQ,OAAO,CAAC,CAAC,CAAC,CAAC;YACnJ,IAAI,KAAK,CAAC,KAAK,KAAK,KAAK,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC,SAAS,EAAE,UAAU,EAAE,SAAS,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,UAAU,CAAC,IAAI,IAAI,CAAC,GAAG,EAAE;gBAAE,SAAS;YAC3J,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;YACxD,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,UAAU,IAAI,MAAM,CAAC,UAAU,KAAK,KAAK,CAAC,UAAU;gBAAE,SAAS;YACrF,MAAM,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YACjD,IAAI,UAAU,CAAC,MAAM,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC,KAAK,UAAU,CAAC,MAAM,CAAC;gBAAE,MAAM,SAAS,CAAC,6CAA6C,CAAC,CAAC;YAC7J,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,OAAO;gBAAE,OAAO,QAAQ,CAAC;YACtD,OAAO,GAAG,IAAI,CAAC;QACjB,CAAC;QACD,OAAO,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,CAAC;IAC1C,CAAC;CACF;AACD,SAAS,QAAQ,CAAC,KAAwB;IACxC,OAAO,EAAE,MAAM,EAAE,KAAK,CAAC,SAAS,EAAE,UAAU,EAAE,KAAK,CAAC,UAAU,EAAE,KAAK,EAAE,KAAK,CAAC,QAAQ,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC;AAC1K,CAAC;AACD,SAAS,SAAS,CAAC,OAAe,IAAI,OAAO,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,8BAA8B,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC"}
|
|
@@ -1,47 +0,0 @@
|
|
|
1
|
-
import { HitchError, invalidInput } from "../foundation/index.js";
|
|
2
|
-
import { readBodyJSON } from "./http.js";
|
|
3
|
-
import { parseModelNodeBinding } from "../domain/index.js";
|
|
4
|
-
export async function prepareInference(request, response, manager) {
|
|
5
|
-
const value = await readBodyJSON(request);
|
|
6
|
-
if (!value || typeof value !== "object" || Array.isArray(value))
|
|
7
|
-
throw invalidInput("prepare requires an object");
|
|
8
|
-
const body = value;
|
|
9
|
-
if (Object.keys(body).some((key) => !["model", "device", "profile", "offline", "inference_id", "model_node"].includes(key))
|
|
10
|
-
|| typeof body.model !== "string" || !body.model.startsWith("local/")
|
|
11
|
-
|| !["auto", "cpu", "cuda", "metal"].includes(String(body.device))
|
|
12
|
-
|| !["baseline", "throughput"].includes(String(body.profile)) || typeof body.offline !== "boolean") {
|
|
13
|
-
throw invalidInput("invalid local inference preparation selection");
|
|
14
|
-
}
|
|
15
|
-
if (body.model_node !== undefined)
|
|
16
|
-
body.model_node = parseModelNodeBinding(body.model_node);
|
|
17
|
-
if (body.inference_id !== undefined && (typeof body.inference_id !== "string" || !/^sha256:[a-f0-9]{64}$/.test(body.inference_id)))
|
|
18
|
-
throw invalidInput("invalid inference lock digest");
|
|
19
|
-
const controller = new AbortController();
|
|
20
|
-
const abort = () => { if (!response.writableEnded)
|
|
21
|
-
controller.abort(); };
|
|
22
|
-
request.once("aborted", abort);
|
|
23
|
-
response.once("close", abort);
|
|
24
|
-
// OCI download plus cold model startup can exceed a regular HTTP header timeout.
|
|
25
|
-
// Stream progress immediately; the final result/error is still structured JSON.
|
|
26
|
-
response.writeHead(200, { "content-type": "application/x-ndjson", "cache-control": "no-store" });
|
|
27
|
-
const emit = (event) => { if (!response.destroyed)
|
|
28
|
-
response.write(`${JSON.stringify(event)}\n`); };
|
|
29
|
-
emit({ type: "inference.preparing", message: "Preparing and validating local inference runtime" });
|
|
30
|
-
const heartbeat = setInterval(() => emit({ type: "inference.progress" }), 15_000);
|
|
31
|
-
heartbeat.unref();
|
|
32
|
-
try {
|
|
33
|
-
const prepared = await manager.prepare(body, controller.signal, emit);
|
|
34
|
-
emit({ type: "inference.prepared", result: prepared });
|
|
35
|
-
}
|
|
36
|
-
catch (error) {
|
|
37
|
-
emit({ type: "inference.prepare.failed", error: { message: error.message,
|
|
38
|
-
code: error instanceof HitchError ? error.code : "inference_route_unavailable", exit_code: error instanceof HitchError ? error.exitCode : 12 } });
|
|
39
|
-
}
|
|
40
|
-
finally {
|
|
41
|
-
clearInterval(heartbeat);
|
|
42
|
-
request.removeListener("aborted", abort);
|
|
43
|
-
response.removeListener("close", abort);
|
|
44
|
-
response.end();
|
|
45
|
-
}
|
|
46
|
-
}
|
|
47
|
-
//# sourceMappingURL=inference-prepare.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"inference-prepare.js","sourceRoot":"","sources":["../../../src/daemon/inference-prepare.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAClE,OAAO,EAAE,YAAY,EAAE,MAAM,WAAW,CAAC;AACzC,OAAO,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAE3D,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,OAAwB,EAAE,QAAwB,EAAE,OAA8B;IACvH,MAAM,KAAK,GAAG,MAAM,YAAY,CAAC,OAAO,CAAC,CAAC;IAC1C,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,MAAM,YAAY,CAAC,4BAA4B,CAAC,CAAC;IAClH,MAAM,IAAI,GAAG,KAAgC,CAAC;IAC9C,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,SAAS,EAAE,cAAc,EAAE,YAAY,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;WACtH,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,QAAQ,CAAC;WAClE,CAAC,CAAC,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;WAC/D,CAAC,CAAC,UAAU,EAAE,YAAY,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,IAAI,OAAO,IAAI,CAAC,OAAO,KAAK,SAAS,EAAE,CAAC;QACrG,MAAM,YAAY,CAAC,+CAA+C,CAAC,CAAC;IACtE,CAAC;IACD,IAAI,IAAI,CAAC,UAAU,KAAK,SAAS;QAAE,IAAI,CAAC,UAAU,GAAG,qBAAqB,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;IAC5F,IAAI,IAAI,CAAC,YAAY,KAAK,SAAS,IAAI,CAAC,OAAO,IAAI,CAAC,YAAY,KAAK,QAAQ,IAAI,CAAC,uBAAuB,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAAE,MAAM,YAAY,CAAC,+BAA+B,CAAC,CAAC;IACxL,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,MAAM,KAAK,GAAG,GAAG,EAAE,GAAG,IAAI,CAAC,QAAQ,CAAC,aAAa;QAAE,UAAU,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC;IACzE,OAAO,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;IAC/B,QAAQ,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;IAC9B,iFAAiF;IACjF,gFAAgF;IAChF,QAAQ,CAAC,SAAS,CAAC,GAAG,EAAE,EAAE,cAAc,EAAE,sBAAsB,EAAE,eAAe,EAAE,UAAU,EAAE,CAAC,CAAC;IACjG,MAAM,IAAI,GAAG,CAAC,KAA8B,EAAE,EAAE,GAAG,IAAI,CAAC,QAAQ,CAAC,SAAS;QAAE,QAAQ,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;IAC5H,IAAI,CAAC,EAAE,IAAI,EAAE,qBAAqB,EAAE,OAAO,EAAE,kDAAkD,EAAE,CAAC,CAAC;IACnG,MAAM,SAAS,GAAG,WAAW,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,oBAAoB,EAAE,CAAC,EAAE,MAAM,CAAC,CAAC;IAClF,SAAS,CAAC,KAAK,EAAE,CAAC;IAClB,IAAI,CAAC;QACH,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,OAAO,CAAC,IAA4C,EAAE,UAAU,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;QAC9G,IAAI,CAAC,EAAE,IAAI,EAAE,oBAAoB,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC,CAAC;IACzD,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,CAAC,EAAE,IAAI,EAAE,0BAA0B,EAAE,KAAK,EAAE,EAAE,OAAO,EAAG,KAAe,CAAC,OAAO;gBACjF,IAAI,EAAE,KAAK,YAAY,UAAU,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,6BAA6B,EAAE,SAAS,EAAE,KAAK,YAAY,UAAU,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC;IACtJ,CAAC;YAAS,CAAC;QACT,aAAa,CAAC,SAAS,CAAC,CAAC;QACzB,OAAO,CAAC,cAAc,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;QACzC,QAAQ,CAAC,cAAc,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QACxC,QAAQ,CAAC,GAAG,EAAE,CAAC;IACjB,CAAC;AACH,CAAC"}
|
|
@@ -1,41 +0,0 @@
|
|
|
1
|
-
import { statfs } from "node:fs/promises";
|
|
2
|
-
import path from "node:path";
|
|
3
|
-
import { runCommand } from "../foundation/index.js";
|
|
4
|
-
/** Supply optional resource dimensions for the automatically started daemon.
|
|
5
|
-
* Explicit user limits, including zero, always take precedence. */
|
|
6
|
-
export async function localInferenceDaemonEnvironment(root, env = process.env, options = {}) {
|
|
7
|
-
const result = { ...env };
|
|
8
|
-
if (result.HITCH_CAPACITY_EPHEMERAL_DISK_MIB === undefined) {
|
|
9
|
-
const bytes = await (options.freeDiskBytes ?? availableDiskBytes)(root);
|
|
10
|
-
result.HITCH_CAPACITY_EPHEMERAL_DISK_MIB = String(Math.floor(bytes / 1024 ** 2));
|
|
11
|
-
}
|
|
12
|
-
if (result.HITCH_CAPACITY_GPUS === undefined) {
|
|
13
|
-
try {
|
|
14
|
-
const invoke = options.run ?? ((executable, args) => runCommand(executable, args, {
|
|
15
|
-
env, timeoutMs: 5_000, failureCode: "gpu_detection_failed",
|
|
16
|
-
}));
|
|
17
|
-
const observed = await invoke(env.HITCH_NVIDIA_SMI_PATH || "nvidia-smi", ["-L"]);
|
|
18
|
-
const count = observed.stdout.split(/\r?\n/).filter((line) => /^GPU \d+:/.test(line)).length;
|
|
19
|
-
if (count > 0)
|
|
20
|
-
result.HITCH_CAPACITY_GPUS = String(count);
|
|
21
|
-
}
|
|
22
|
-
catch { /* CPU-only daemon capacity remains valid. */ }
|
|
23
|
-
}
|
|
24
|
-
return result;
|
|
25
|
-
}
|
|
26
|
-
async function availableDiskBytes(directory) {
|
|
27
|
-
let current = path.resolve(directory);
|
|
28
|
-
for (;;) {
|
|
29
|
-
try {
|
|
30
|
-
const info = await statfs(current, { bigint: true });
|
|
31
|
-
return Number(info.bavail * info.bsize);
|
|
32
|
-
}
|
|
33
|
-
catch (error) {
|
|
34
|
-
const parent = path.dirname(current);
|
|
35
|
-
if (error.code !== "ENOENT" || parent === current)
|
|
36
|
-
throw error;
|
|
37
|
-
current = parent;
|
|
38
|
-
}
|
|
39
|
-
}
|
|
40
|
-
}
|
|
41
|
-
//# sourceMappingURL=daemon-capacity.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"daemon-capacity.js","sourceRoot":"","sources":["../../../src/inference/daemon-capacity.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AAC1C,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAGpD;mEACmE;AACnE,MAAM,CAAC,KAAK,UAAU,+BAA+B,CAAC,IAAY,EAAE,MAAyB,OAAO,CAAC,GAAG,EAAE,UAGtG,EAAE;IACJ,MAAM,MAAM,GAAG,EAAE,GAAG,GAAG,EAAE,CAAC;IAC1B,IAAI,MAAM,CAAC,iCAAiC,KAAK,SAAS,EAAE,CAAC;QAC3D,MAAM,KAAK,GAAG,MAAM,CAAC,OAAO,CAAC,aAAa,IAAI,kBAAkB,CAAC,CAAC,IAAI,CAAC,CAAC;QACxE,MAAM,CAAC,iCAAiC,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC;IACnF,CAAC;IACD,IAAI,MAAM,CAAC,mBAAmB,KAAK,SAAS,EAAE,CAAC;QAC7C,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,IAAI,CAAC,CAAC,UAAkB,EAAE,IAAc,EAAE,EAAE,CAAC,UAAU,CAAC,UAAU,EAAE,IAAI,EAAE;gBAClG,GAAG,EAAE,SAAS,EAAE,KAAK,EAAE,WAAW,EAAE,sBAAsB;aAC3D,CAAC,CAAC,CAAC;YACJ,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,GAAG,CAAC,qBAAqB,IAAI,YAAY,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC;YACjF,MAAM,KAAK,GAAG,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC;YAC7F,IAAI,KAAK,GAAG,CAAC;gBAAE,MAAM,CAAC,mBAAmB,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;QAC5D,CAAC;QAAC,MAAM,CAAC,CAAC,6CAA6C,CAAC,CAAC;IAC3D,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,KAAK,UAAU,kBAAkB,CAAC,SAAiB;IACjD,IAAI,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;IACtC,SAAS,CAAC;QACR,IAAI,CAAC;YACH,MAAM,IAAI,GAAG,MAAM,MAAM,CAAC,OAAO,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC;YACrD,OAAO,MAAM,CAAC,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;QAC1C,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;YACrC,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ,IAAI,MAAM,KAAK,OAAO;gBAAE,MAAM,KAAK,CAAC;YAC1F,OAAO,GAAG,MAAM,CAAC;QACnB,CAAC;IACH,CAAC;AACH,CAAC"}
|
|
@@ -1,58 +0,0 @@
|
|
|
1
|
-
import { mkdir, readdir, rm } from "node:fs/promises";
|
|
2
|
-
import { homedir } from "node:os";
|
|
3
|
-
import path from "node:path";
|
|
4
|
-
import { HitchError, atomicWriteJSON, hitchRootId, readJSON, sha256JSON, withFileLock } from "../foundation/index.js";
|
|
5
|
-
export const defaultDeviceReservationDirectory = () => path.join(homedir(), ".cache", "agent-hitch", "inference-devices");
|
|
6
|
-
export async function reservedInferenceDevices(directory = defaultDeviceReservationDirectory()) {
|
|
7
|
-
let files;
|
|
8
|
-
try {
|
|
9
|
-
files = await readdir(directory);
|
|
10
|
-
}
|
|
11
|
-
catch (error) {
|
|
12
|
-
if (error.code === "ENOENT")
|
|
13
|
-
return [];
|
|
14
|
-
throw error;
|
|
15
|
-
}
|
|
16
|
-
const devices = [];
|
|
17
|
-
for (const name of files.filter((name) => /^[a-f0-9]{64}\.json$/.test(name))) {
|
|
18
|
-
const record = await readJSON(path.join(directory, name), null);
|
|
19
|
-
if (record?.gpu_uuid)
|
|
20
|
-
devices.push(record.gpu_uuid);
|
|
21
|
-
}
|
|
22
|
-
return devices;
|
|
23
|
-
}
|
|
24
|
-
/** Survives daemon death. Only confirmed container cleanup releases a reservation;
|
|
25
|
-
* a dead PID alone is never evidence that a GPU is free. Shared across Hitch roots. */
|
|
26
|
-
export async function reserveInferenceDevice(directory, root, serviceId, gpuUuid) {
|
|
27
|
-
await mkdir(directory, { recursive: true, mode: 0o700 });
|
|
28
|
-
const key = sha256JSON(gpuUuid).slice(7);
|
|
29
|
-
await withFileLock(directory, key, async () => {
|
|
30
|
-
const file = path.join(directory, `${key}.json`);
|
|
31
|
-
const existing = await readJSON(file, null);
|
|
32
|
-
if (existing)
|
|
33
|
-
throw new HitchError(`GPU ${gpuUuid} is reserved by another inference service; stop or recover its owning Hitch daemon first`, {
|
|
34
|
-
code: "inference_device_in_use", exitCode: 12,
|
|
35
|
-
});
|
|
36
|
-
await atomicWriteJSON(file, { root_id: hitchRootId(root), service_id: serviceId, gpu_uuid: gpuUuid });
|
|
37
|
-
});
|
|
38
|
-
}
|
|
39
|
-
export async function releaseInferenceDevice(directory, root, serviceId) {
|
|
40
|
-
let files;
|
|
41
|
-
try {
|
|
42
|
-
files = await readdir(directory);
|
|
43
|
-
}
|
|
44
|
-
catch (error) {
|
|
45
|
-
if (error.code === "ENOENT")
|
|
46
|
-
return;
|
|
47
|
-
throw error;
|
|
48
|
-
}
|
|
49
|
-
for (const name of files.filter((name) => /^[a-f0-9]{64}\.json$/.test(name))) {
|
|
50
|
-
await withFileLock(directory, name.slice(0, -5), async () => {
|
|
51
|
-
const file = path.join(directory, name);
|
|
52
|
-
const record = await readJSON(file, null);
|
|
53
|
-
if (record?.root_id === hitchRootId(root) && record.service_id === serviceId)
|
|
54
|
-
await rm(file);
|
|
55
|
-
});
|
|
56
|
-
}
|
|
57
|
-
}
|
|
58
|
-
//# sourceMappingURL=device-reservation.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"device-reservation.js","sourceRoot":"","sources":["../../../src/inference/device-reservation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE,EAAE,MAAM,kBAAkB,CAAC;AACtD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,UAAU,EAAE,eAAe,EAAE,WAAW,EAAE,QAAQ,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAGtH,MAAM,CAAC,MAAM,iCAAiC,GAAG,GAAG,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,EAAE,QAAQ,EAAE,aAAa,EAAE,mBAAmB,CAAC,CAAC;AAE1H,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAAC,SAAS,GAAG,iCAAiC,EAAE;IAC5F,IAAI,KAAe,CAAC;IACpB,IAAI,CAAC;QAAC,KAAK,GAAG,MAAM,OAAO,CAAC,SAAS,CAAC,CAAC;IAAC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACvD,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,EAAE,CAAC;QAClE,MAAM,KAAK,CAAC;IACd,CAAC;IACD,MAAM,OAAO,GAAa,EAAE,CAAC;IAC7B,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,sBAAsB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;QAC7E,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAqB,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC,CAAC;QACpF,IAAI,MAAM,EAAE,QAAQ;YAAE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IACtD,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;uFACuF;AACvF,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAAC,SAAiB,EAAE,IAAY,EAAE,SAAiB,EAAE,OAAe;IAC9G,MAAM,KAAK,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC;IACzD,MAAM,GAAG,GAAG,UAAU,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IACzC,MAAM,YAAY,CAAC,SAAS,EAAE,GAAG,EAAE,KAAK,IAAI,EAAE;QAC5C,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,GAAG,GAAG,OAAO,CAAC,CAAC;QACjD,MAAM,QAAQ,GAAG,MAAM,QAAQ,CAAqB,IAAI,EAAE,IAAI,CAAC,CAAC;QAChE,IAAI,QAAQ;YAAE,MAAM,IAAI,UAAU,CAAC,OAAO,OAAO,0FAA0F,EAAE;gBAC3I,IAAI,EAAE,yBAAyB,EAAE,QAAQ,EAAE,EAAE;aAC9C,CAAC,CAAC;QACH,MAAM,eAAe,CAAC,IAAI,EAAE,EAAE,OAAO,EAAE,WAAW,CAAC,IAAI,CAAC,EAAE,UAAU,EAAE,SAAS,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC;IACxG,CAAC,CAAC,CAAC;AACL,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAAC,SAAiB,EAAE,IAAY,EAAE,SAAiB;IAC7F,IAAI,KAAe,CAAC;IACpB,IAAI,CAAC;QAAC,KAAK,GAAG,MAAM,OAAO,CAAC,SAAS,CAAC,CAAC;IAAC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACvD,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO;QAC/D,MAAM,KAAK,CAAC;IACd,CAAC;IACD,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,sBAAsB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;QAC7E,MAAM,YAAY,CAAC,SAAS,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,IAAI,EAAE;YAC1D,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YACxC,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAqB,IAAI,EAAE,IAAI,CAAC,CAAC;YAC9D,IAAI,MAAM,EAAE,OAAO,KAAK,WAAW,CAAC,IAAI,CAAC,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS;gBAAE,MAAM,EAAE,CAAC,IAAI,CAAC,CAAC;QAC/F,CAAC,CAAC,CAAC;IACL,CAAC;AACH,CAAC"}
|
|
@@ -1,95 +0,0 @@
|
|
|
1
|
-
import { readFile } from "node:fs/promises";
|
|
2
|
-
import { HitchError, runCommand } from "../foundation/index.js";
|
|
3
|
-
/** Static eligibility only. Loading, CUDA visibility and the protocol are verified by prepare. */
|
|
4
|
-
export async function doctorLocalInference(backend, options = {}) {
|
|
5
|
-
const platform = options.platform ?? process.platform;
|
|
6
|
-
const architecture = options.architecture ?? process.arch;
|
|
7
|
-
const env = options.env ?? process.env;
|
|
8
|
-
const invoke = options.run ?? ((executable, args) => runCommand(executable, args, {
|
|
9
|
-
env, timeoutMs: 10_000, failureCode: "inference_runtime_unavailable", failureExitCode: 3,
|
|
10
|
-
}));
|
|
11
|
-
const checks = {};
|
|
12
|
-
checks.platform = platform === "linux" && architecture === "x64"
|
|
13
|
-
? pass("linux/amd64") : fail(`P0 requires linux/amd64; detected ${platform}/${architecture}`);
|
|
14
|
-
if (backend === "metal") {
|
|
15
|
-
checks.catalog = fail("Metal/MLX is a P1 backend and has no P0 runtime catalog entry");
|
|
16
|
-
return result(backend, checks);
|
|
17
|
-
}
|
|
18
|
-
try {
|
|
19
|
-
const docker = options.dockerExecutable || env.HITCH_DOCKER_PATH || "docker";
|
|
20
|
-
const server = await invoke(docker, ["info", "--format", "{{json .}}"]);
|
|
21
|
-
const info = JSON.parse(server.stdout);
|
|
22
|
-
checks.docker = info.OSType === "linux" && ["x86_64", "amd64"].includes(String(info.Architecture))
|
|
23
|
-
? pass(`Docker ${String(info.ServerVersion)}`) : fail("Docker Engine must run linux/amd64");
|
|
24
|
-
const context = await invoke(docker, ["context", "inspect", "--format", "{{json .Endpoints.docker.Host}}"]);
|
|
25
|
-
const endpoint = env.DOCKER_CONTEXT ? JSON.parse(context.stdout) : env.DOCKER_HOST || JSON.parse(context.stdout);
|
|
26
|
-
checks.docker_endpoint = typeof endpoint === "string" && endpoint.startsWith("unix:///")
|
|
27
|
-
? pass("local Unix socket") : fail("managed inference requires a local Docker Unix socket");
|
|
28
|
-
}
|
|
29
|
-
catch (error) {
|
|
30
|
-
checks.docker = fail(`Docker is unavailable: ${error.message}`);
|
|
31
|
-
}
|
|
32
|
-
if (backend === "cpu") {
|
|
33
|
-
let info = "";
|
|
34
|
-
try {
|
|
35
|
-
info = await (options.readCpuInfo ?? (() => readFile("/proc/cpuinfo", "utf8")))();
|
|
36
|
-
}
|
|
37
|
-
catch { }
|
|
38
|
-
const flags = new Set(info.toLowerCase().split(/[^a-z0-9_]+/));
|
|
39
|
-
const missing = ["amx_tile", "amx_int8", "amx_bf16"].filter((flag) => !flags.has(flag));
|
|
40
|
-
checks.amx = missing.length ? fail(`missing CPU flags: ${missing.join(", ")}`) : pass("AMX tile/int8/bf16");
|
|
41
|
-
return result(backend, checks);
|
|
42
|
-
}
|
|
43
|
-
let gpu;
|
|
44
|
-
try {
|
|
45
|
-
const observed = await invoke(options.nvidiaSmiExecutable || env.HITCH_NVIDIA_SMI_PATH || "nvidia-smi", [
|
|
46
|
-
"--query-gpu=uuid,name,memory.total,memory.free,driver_version,compute_cap", "--format=csv,noheader,nounits",
|
|
47
|
-
]);
|
|
48
|
-
const candidates = observed.stdout.trim().split(/\r?\n/).map(parseGpu);
|
|
49
|
-
// The pinned image uses CUDA 13.0. Require the native driver floor; do not
|
|
50
|
-
// assume forward-compat packages work on an arbitrary device/driver pair.
|
|
51
|
-
const eligible = candidates.filter((entry) => Number(entry.driver_version.split(".")[0]) >= 580
|
|
52
|
-
&& [8.0, 8.6, 8.9, 9.0, 10.0, 10.3, 12.0].includes(Number(entry.compute_capability))
|
|
53
|
-
&& entry.free_memory_mib >= (options.requiredMemoryMiB ?? 0)
|
|
54
|
-
&& !options.excludedDeviceUuids?.includes(entry.uuid)
|
|
55
|
-
&& (!options.deviceConstraint || entry.uuid === options.deviceConstraint));
|
|
56
|
-
gpu = eligible[0];
|
|
57
|
-
checks.cuda = gpu ? pass(`${gpu.name} (${gpu.uuid}), ${gpu.free_memory_mib} MiB free, driver ${gpu.driver_version}`)
|
|
58
|
-
: fail(`no eligible CUDA 13 GPU: require driver >=580, supported SM, ${options.requiredMemoryMiB ?? 0} MiB free${options.deviceConstraint ? ` on ${options.deviceConstraint}` : ""}`);
|
|
59
|
-
}
|
|
60
|
-
catch (error) {
|
|
61
|
-
checks.cuda = fail(`NVIDIA CUDA device is unavailable: ${error.message}`);
|
|
62
|
-
}
|
|
63
|
-
return { ...result(backend, checks), ...(gpu ? { gpu } : {}) };
|
|
64
|
-
}
|
|
65
|
-
export async function resolveLocalInferenceDevice(requested, options = {}) {
|
|
66
|
-
for (const backend of requested === "auto" ? ["cuda", "cpu"] : [requested]) {
|
|
67
|
-
const doctor = await doctorLocalInference(backend, options);
|
|
68
|
-
if (doctor.ready)
|
|
69
|
-
return { backend, doctor };
|
|
70
|
-
if (requested !== "auto")
|
|
71
|
-
throw unavailable(doctor);
|
|
72
|
-
}
|
|
73
|
-
throw new HitchError("no eligible local inference device is available; run hitch local doctor for diagnostics", {
|
|
74
|
-
code: "inference_device_unsupported", exitCode: 3,
|
|
75
|
-
});
|
|
76
|
-
}
|
|
77
|
-
function parseGpu(line) {
|
|
78
|
-
const [uuid, name, total, free, driver, sm] = line.split(",").map((part) => part.trim());
|
|
79
|
-
if (!uuid?.startsWith("GPU-") || !name || !driver || !sm || !Number.isFinite(Number(sm))
|
|
80
|
-
|| !Number.isFinite(Number(total)) || Number(total) <= 0 || !Number.isFinite(Number(free)) || Number(free) < 0) {
|
|
81
|
-
throw new TypeError("nvidia-smi returned an invalid GPU record");
|
|
82
|
-
}
|
|
83
|
-
return { uuid, name, memory_mib: Number(total), free_memory_mib: Number(free), driver_version: driver, compute_capability: sm };
|
|
84
|
-
}
|
|
85
|
-
function result(backend, checks) {
|
|
86
|
-
return { schema_version: "1", backend, ready: Object.values(checks).every((check) => check.status === "pass"), checks };
|
|
87
|
-
}
|
|
88
|
-
function pass(message) { return { status: "pass", message }; }
|
|
89
|
-
function fail(message) { return { status: "fail", message }; }
|
|
90
|
-
function unavailable(doctor) {
|
|
91
|
-
return new HitchError(`${doctor.backend} local inference is unavailable: ${Object.values(doctor.checks).filter((c) => c.status === "fail").map((c) => c.message).join("; ")}`, {
|
|
92
|
-
code: "inference_device_unsupported", exitCode: 3,
|
|
93
|
-
});
|
|
94
|
-
}
|
|
95
|
-
//# sourceMappingURL=doctor.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"doctor.js","sourceRoot":"","sources":["../../../src/inference/doctor.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAE5C,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AA4BhE,kGAAkG;AAClG,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,OAA8B,EAAE,UAAkC,EAAE;IAC7G,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,QAAQ,CAAC;IACtD,MAAM,YAAY,GAAG,OAAO,CAAC,YAAY,IAAI,OAAO,CAAC,IAAI,CAAC;IAC1D,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,IAAI,OAAO,CAAC,GAAG,CAAC;IACvC,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,IAAI,CAAC,CAAC,UAAkB,EAAE,IAAc,EAAE,EAAE,CAAC,UAAU,CAAC,UAAU,EAAE,IAAI,EAAE;QAClG,GAAG,EAAE,SAAS,EAAE,MAAM,EAAE,WAAW,EAAE,+BAA+B,EAAE,eAAe,EAAE,CAAC;KACzF,CAAC,CAAC,CAAC;IACJ,MAAM,MAAM,GAA2C,EAAE,CAAC;IAC1D,MAAM,CAAC,QAAQ,GAAG,QAAQ,KAAK,OAAO,IAAI,YAAY,KAAK,KAAK;QAC9D,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,qCAAqC,QAAQ,IAAI,YAAY,EAAE,CAAC,CAAC;IAChG,IAAI,OAAO,KAAK,OAAO,EAAE,CAAC;QACxB,MAAM,CAAC,OAAO,GAAG,IAAI,CAAC,+DAA+D,CAAC,CAAC;QACvF,OAAO,MAAM,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;IACjC,CAAC;IACD,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,OAAO,CAAC,gBAAgB,IAAI,GAAG,CAAC,iBAAiB,IAAI,QAAQ,CAAC;QAC7E,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,MAAM,EAAE,CAAC,MAAM,EAAE,UAAU,EAAE,YAAY,CAAC,CAAC,CAAC;QACxE,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAA4B,CAAC;QAClE,MAAM,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,KAAK,OAAO,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;YAChG,CAAC,CAAC,IAAI,CAAC,UAAU,MAAM,CAAC,IAAI,CAAC,aAAa,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,oCAAoC,CAAC,CAAC;QAC9F,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,MAAM,EAAE,CAAC,SAAS,EAAE,SAAS,EAAE,UAAU,EAAE,iCAAiC,CAAC,CAAC,CAAC;QAC5G,MAAM,QAAQ,GAAG,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,WAAW,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QACjH,MAAM,CAAC,eAAe,GAAG,OAAO,QAAQ,KAAK,QAAQ,IAAI,QAAQ,CAAC,UAAU,CAAC,UAAU,CAAC;YACtF,CAAC,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,uDAAuD,CAAC,CAAC;IAChG,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QAAC,MAAM,CAAC,MAAM,GAAG,IAAI,CAAC,0BAA2B,KAAe,CAAC,OAAO,EAAE,CAAC,CAAC;IAAC,CAAC;IAC/F,IAAI,OAAO,KAAK,KAAK,EAAE,CAAC;QACtB,IAAI,IAAI,GAAG,EAAE,CAAC;QACd,IAAI,CAAC;YAAC,IAAI,GAAG,MAAM,CAAC,OAAO,CAAC,WAAW,IAAI,CAAC,GAAG,EAAE,CAAC,QAAQ,CAAC,eAAe,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC,CAAA,CAAC;QACnG,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC;QAC/D,MAAM,OAAO,GAAG,CAAC,UAAU,EAAE,UAAU,EAAE,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC;QACxF,MAAM,CAAC,GAAG,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,sBAAsB,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;QAC5G,OAAO,MAAM,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;IACjC,CAAC;IACD,IAAI,GAA+B,CAAC;IACpC,IAAI,CAAC;QACH,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,OAAO,CAAC,mBAAmB,IAAI,GAAG,CAAC,qBAAqB,IAAI,YAAY,EAAE;YACtG,2EAA2E,EAAE,+BAA+B;SAC7G,CAAC,CAAC;QACH,MAAM,UAAU,GAAG,QAAQ,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;QACvE,2EAA2E;QAC3E,0EAA0E;QAC1E,MAAM,QAAQ,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,cAAc,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG;eAC1F,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC;eACjF,KAAK,CAAC,eAAe,IAAI,CAAC,OAAO,CAAC,iBAAiB,IAAI,CAAC,CAAC;eACzD,CAAC,OAAO,CAAC,mBAAmB,EAAE,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC;eAClD,CAAC,CAAC,OAAO,CAAC,gBAAgB,IAAI,KAAK,CAAC,IAAI,KAAK,OAAO,CAAC,gBAAgB,CAAC,CAAC,CAAC;QAC7E,GAAG,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;QAClB,MAAM,CAAC,IAAI,GAAG,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,GAAG,CAAC,IAAI,KAAK,GAAG,CAAC,IAAI,MAAM,GAAG,CAAC,eAAe,qBAAqB,GAAG,CAAC,cAAc,EAAE,CAAC;YAClH,CAAC,CAAC,IAAI,CAAC,gEAAgE,OAAO,CAAC,iBAAiB,IAAI,CAAC,YAAY,OAAO,CAAC,gBAAgB,CAAC,CAAC,CAAC,OAAO,OAAO,CAAC,gBAAgB,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IAC1L,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QAAC,MAAM,CAAC,IAAI,GAAG,IAAI,CAAC,sCAAuC,KAAe,CAAC,OAAO,EAAE,CAAC,CAAC;IAAC,CAAC;IACzG,OAAO,EAAE,GAAG,MAAM,CAAC,OAAO,EAAE,MAAM,CAAC,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC;AACjE,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,2BAA2B,CAAC,SAA+B,EAAE,UAAkC,EAAE;IACrH,KAAK,MAAM,OAAO,IAAI,SAAS,KAAK,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,KAAK,CAAU,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,EAAE,CAAC;QACpF,MAAM,MAAM,GAAG,MAAM,oBAAoB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QAC5D,IAAI,MAAM,CAAC,KAAK;YAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;QAC7C,IAAI,SAAS,KAAK,MAAM;YAAE,MAAM,WAAW,CAAC,MAAM,CAAC,CAAC;IACtD,CAAC;IACD,MAAM,IAAI,UAAU,CAAC,yFAAyF,EAAE;QAC9G,IAAI,EAAE,8BAA8B,EAAE,QAAQ,EAAE,CAAC;KAClD,CAAC,CAAC;AACL,CAAC;AACD,SAAS,QAAQ,CAAC,IAAY;IAC5B,MAAM,CAAC,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;IACzF,IAAI,CAAC,IAAI,EAAE,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,MAAM,IAAI,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;WACnF,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC;QACjH,MAAM,IAAI,SAAS,CAAC,2CAA2C,CAAC,CAAC;IACnE,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,CAAC,KAAK,CAAC,EAAE,eAAe,EAAE,MAAM,CAAC,IAAI,CAAC,EAAE,cAAc,EAAE,MAAM,EAAE,kBAAkB,EAAE,EAAE,EAAE,CAAC;AAClI,CAAC;AACD,SAAS,MAAM,CAAC,OAA8B,EAAE,MAA8C;IAC5F,OAAO,EAAE,cAAc,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,KAAK,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;AAC1H,CAAC;AACD,SAAS,IAAI,CAAC,OAAe,IAA4B,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC;AAC9F,SAAS,IAAI,CAAC,OAAe,IAA4B,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC;AAC9F,SAAS,WAAW,CAAC,MAA+B;IAClD,OAAO,IAAI,UAAU,CAAC,GAAG,MAAM,CAAC,OAAO,oCAAoC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE;QAC7K,IAAI,EAAE,8BAA8B,EAAE,QAAQ,EAAE,CAAC;KAClD,CAAC,CAAC;AACL,CAAC"}
|
|
@@ -1,83 +0,0 @@
|
|
|
1
|
-
import path from "node:path";
|
|
2
|
-
import { HitchError, atomicWriteJSON, hitchRootId, readJSON, sha256JSON, statePaths, withFileLock } from "../foundation/index.js";
|
|
3
|
-
import { parseInferenceServiceHandle, parseInferenceServiceRecord, parseModelNodeBinding } from "./manifest.js";
|
|
4
|
-
import { loadModelNodeClient, observeModelNode } from "./node-registry.js";
|
|
5
|
-
const SHA = /^sha256:[a-f0-9]{64}$/;
|
|
6
|
-
const directory = (root, id) => path.join(statePaths(root).inferenceServices, id);
|
|
7
|
-
function serviceDigest(record) {
|
|
8
|
-
// A delayed startup projection may fill in an initially unknown handle.
|
|
9
|
-
// The sealed node receipt owns that handle and validates every later projection.
|
|
10
|
-
return sha256JSON({ service_id: record.service_id, inference_id: record.inference_id, model_node: record.model_node,
|
|
11
|
-
epoch: record.epoch, isolation_key: record.isolation_key, owner_id: record.owner_id });
|
|
12
|
-
}
|
|
13
|
-
export function validateInferenceGenerationRelease(root, record, current, value) {
|
|
14
|
-
const source = record.model_node;
|
|
15
|
-
const r = value;
|
|
16
|
-
const keys = ["schemaVersion", "kind", "serviceId", "ownerId", "inferenceId", "previousNode", "node", "sourceIdentityDigest", "handle",
|
|
17
|
-
"admissionOnly", "state", "resourcesReleased", "gpuSeconds", "devices", "previousBootDigest", "currentBootDigest"];
|
|
18
|
-
if (!source || source.node_id !== current.node_id || source.generation === current.generation || record.container_id
|
|
19
|
-
|| !r || Object.keys(r).sort().join(",") !== keys.sort().join(",") || r.schemaVersion !== 2 || r.kind !== "inference-generation-release"
|
|
20
|
-
|| r.serviceId !== record.service_id || r.ownerId !== hitchRootId(root) || r.inferenceId !== record.inference_id
|
|
21
|
-
|| sha256JSON(r.previousNode) !== sha256JSON({ nodeId: source.node_id, generation: source.generation })
|
|
22
|
-
|| sha256JSON(r.node) !== sha256JSON({ nodeId: current.node_id, generation: current.generation })
|
|
23
|
-
|| ![r.sourceIdentityDigest, r.previousBootDigest, r.currentBootDigest].every(v => typeof v === "string" && SHA.test(v))
|
|
24
|
-
|| r.previousBootDigest === r.currentBootDigest || r.state !== "stopped" || r.resourcesReleased !== true
|
|
25
|
-
|| typeof r.admissionOnly !== "boolean" || typeof r.gpuSeconds !== "number" || !Number.isFinite(r.gpuSeconds) || r.gpuSeconds < 0
|
|
26
|
-
|| !Array.isArray(r.devices) || r.devices.some(v => typeof v !== "string" || !/^GPU-[a-fA-F0-9-]+$/.test(v)) || new Set(r.devices).size !== r.devices.length
|
|
27
|
-
|| r.admissionOnly && (r.handle !== null || r.gpuSeconds !== 0 || r.devices.length !== 0))
|
|
28
|
-
throw ambiguous();
|
|
29
|
-
if (r.handle !== null) {
|
|
30
|
-
const handle = parseInferenceServiceHandle(r.handle);
|
|
31
|
-
if (handle.kind !== "process" || handle.node_id !== source.node_id || handle.generation !== source.generation
|
|
32
|
-
|| handle.service_id !== record.service_id)
|
|
33
|
-
throw ambiguous();
|
|
34
|
-
}
|
|
35
|
-
if (record.service_handle && sha256JSON(record.service_handle) !== sha256JSON(r.handle))
|
|
36
|
-
throw ambiguous();
|
|
37
|
-
return structuredClone(r);
|
|
38
|
-
}
|
|
39
|
-
/** A confirmed release remains valid if the original node generation is no longer reachable. */
|
|
40
|
-
export async function readInferenceGenerationRelease(root, record) {
|
|
41
|
-
const value = await readJSON(path.join(directory(root, record.service_id), "generation-release.json"), null);
|
|
42
|
-
if (!value)
|
|
43
|
-
return null;
|
|
44
|
-
if (Object.keys(value).sort().join(",") !== "model_node,receipt,receipt_digest,schema_version,service_digest"
|
|
45
|
-
|| value.schema_version !== "2" || value.service_digest !== serviceDigest(record) || value.receipt_digest !== sha256JSON(value.receipt))
|
|
46
|
-
throw ambiguous();
|
|
47
|
-
const node = parseModelNodeBinding(value.model_node);
|
|
48
|
-
validateInferenceGenerationRelease(root, record, node, value.receipt);
|
|
49
|
-
return value;
|
|
50
|
-
}
|
|
51
|
-
/** Explicitly drain an old generation. No service or experiment is rebound to the new generation. */
|
|
52
|
-
export async function reconcileModelNodeService(root, serviceId, binding) {
|
|
53
|
-
if (!/^inference_[a-f0-9]{32}$/.test(serviceId))
|
|
54
|
-
throw ambiguous();
|
|
55
|
-
const current = parseModelNodeBinding(binding);
|
|
56
|
-
return withFileLock(statePaths(root).inferenceOperationLocks, sha256JSON({ kind: "generation-recovery", serviceId }), async () => {
|
|
57
|
-
const stateFile = path.join(directory(root, serviceId), "state.json");
|
|
58
|
-
const record = parseInferenceServiceRecord(await readJSON(stateFile));
|
|
59
|
-
if (record.service_id !== serviceId || !record.model_node || record.model_node.node_id !== current.node_id
|
|
60
|
-
|| record.model_node.generation === current.generation)
|
|
61
|
-
throw ambiguous();
|
|
62
|
-
let stored = await readInferenceGenerationRelease(root, record);
|
|
63
|
-
if (!stored) {
|
|
64
|
-
const client = await loadModelNodeClient(root, current);
|
|
65
|
-
await observeModelNode(client, current);
|
|
66
|
-
const receipt = validateInferenceGenerationRelease(root, record, current, await client.call("inference.recover", {
|
|
67
|
-
serviceId, ownerId: hitchRootId(root), inferenceId: record.inference_id,
|
|
68
|
-
previousNode: { nodeId: record.model_node.node_id, generation: record.model_node.generation }, expectedHandle: record.service_handle ?? null,
|
|
69
|
-
}));
|
|
70
|
-
stored = { schema_version: "2", service_digest: serviceDigest(record), model_node: current, receipt_digest: sha256JSON(receipt), receipt };
|
|
71
|
-
await atomicWriteJSON(path.join(directory(root, serviceId), "generation-release.json"), stored);
|
|
72
|
-
}
|
|
73
|
-
const latest = parseInferenceServiceRecord(await readJSON(stateFile));
|
|
74
|
-
if (serviceDigest(latest) !== serviceDigest(record))
|
|
75
|
-
throw ambiguous();
|
|
76
|
-
validateInferenceGenerationRelease(root, latest, stored.model_node, stored.receipt);
|
|
77
|
-
await atomicWriteJSON(stateFile, { ...latest, state: "stopped", lease_owner_ids: [], updated_at: new Date().toISOString() });
|
|
78
|
-
return { schema_version: "2", service_id: serviceId, inference_id: record.inference_id, model_node: record.model_node,
|
|
79
|
-
state: "stopped", resources_released: true, gpu_seconds: stored.receipt.gpuSeconds, generation_release: stored };
|
|
80
|
-
});
|
|
81
|
-
}
|
|
82
|
-
function ambiguous() { return new HitchError("prior model-node generation has no matching confirmed service release", { code: "inference_recovery_ambiguous", exitCode: 12 }); }
|
|
83
|
-
//# sourceMappingURL=generation-recovery.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"generation-recovery.js","sourceRoot":"","sources":["../../../src/inference/generation-recovery.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,UAAU,EAAE,eAAe,EAAE,WAAW,EAAE,QAAQ,EAAE,UAAU,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAClI,OAAO,EAAE,2BAA2B,EAAE,2BAA2B,EAAE,qBAAqB,EAAE,MAAM,eAAe,CAAC;AAChH,OAAO,EAAE,mBAAmB,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AAY3E,MAAM,GAAG,GAAG,uBAAuB,CAAC;AACpC,MAAM,SAAS,GAAG,CAAC,IAAY,EAAE,EAAU,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,iBAAiB,EAAE,EAAE,CAAC,CAAC;AAClG,SAAS,aAAa,CAAC,MAAgC;IACrD,wEAAwE;IACxE,iFAAiF;IACjF,OAAO,UAAU,CAAC,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU,EAAE,YAAY,EAAE,MAAM,CAAC,YAAY,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU;QACjH,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,aAAa,EAAE,MAAM,CAAC,aAAa,EAAE,QAAQ,EAAE,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;AAC3F,CAAC;AAED,MAAM,UAAU,kCAAkC,CAAC,IAAY,EAAE,MAAgC,EAAE,OAA2B,EAAE,KAAc;IAC5I,MAAM,MAAM,GAAG,MAAM,CAAC,UAAU,CAAC;IACjC,MAAM,CAAC,GAAG,KAA4C,CAAC;IACvD,MAAM,IAAI,GAAG,CAAC,eAAe,EAAE,MAAM,EAAE,WAAW,EAAE,SAAS,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,EAAE,sBAAsB,EAAE,QAAQ;QACpI,eAAe,EAAE,OAAO,EAAE,mBAAmB,EAAE,YAAY,EAAE,SAAS,EAAE,oBAAoB,EAAE,mBAAmB,CAAC,CAAC;IACrH,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,OAAO,KAAK,OAAO,CAAC,OAAO,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,CAAC,UAAU,IAAI,MAAM,CAAC,YAAY;WAC/G,CAAC,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,IAAI,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,aAAa,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,KAAK,8BAA8B;WACrI,CAAC,CAAC,SAAS,KAAK,MAAM,CAAC,UAAU,IAAI,CAAC,CAAC,OAAO,KAAK,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,WAAW,KAAK,MAAM,CAAC,YAAY;WAC7G,UAAU,CAAC,CAAC,CAAC,YAAY,CAAC,KAAK,UAAU,CAAC,EAAE,MAAM,EAAE,MAAM,CAAC,OAAO,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU,EAAE,CAAC;WACpG,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,UAAU,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,OAAO,EAAE,UAAU,EAAE,OAAO,CAAC,UAAU,EAAE,CAAC;WAC9F,CAAC,CAAC,CAAC,CAAC,oBAAoB,EAAE,CAAC,CAAC,kBAAkB,EAAE,CAAC,CAAC,iBAAiB,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;WACrH,CAAC,CAAC,kBAAkB,KAAK,CAAC,CAAC,iBAAiB,IAAI,CAAC,CAAC,KAAK,KAAK,SAAS,IAAI,CAAC,CAAC,iBAAiB,KAAK,IAAI;WACrG,OAAO,CAAC,CAAC,aAAa,KAAK,SAAS,IAAI,OAAO,CAAC,CAAC,UAAU,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,UAAU,GAAG,CAAC;WAC9H,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,qBAAqB,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,IAAI,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,OAAO,CAAC,MAAM;WACzJ,CAAC,CAAC,aAAa,IAAI,CAAC,CAAC,CAAC,MAAM,KAAK,IAAI,IAAI,CAAC,CAAC,UAAU,KAAK,CAAC,IAAI,CAAC,CAAC,OAAO,CAAC,MAAM,KAAK,CAAC,CAAC;QAAE,MAAM,SAAS,EAAE,CAAC;IAC/G,IAAI,CAAC,CAAC,MAAM,KAAK,IAAI,EAAE,CAAC;QACtB,MAAM,MAAM,GAAG,2BAA2B,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QACrD,IAAI,MAAM,CAAC,IAAI,KAAK,SAAS,IAAI,MAAM,CAAC,OAAO,KAAK,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,UAAU,KAAK,MAAM,CAAC,UAAU;eACxG,MAAM,CAAC,UAAU,KAAK,MAAM,CAAC,UAAU;YAAE,MAAM,SAAS,EAAE,CAAC;IAClE,CAAC;IACD,IAAI,MAAM,CAAC,cAAc,IAAI,UAAU,CAAC,MAAM,CAAC,cAAc,CAAC,KAAK,UAAU,CAAC,CAAC,CAAC,MAAM,CAAC;QAAE,MAAM,SAAS,EAAE,CAAC;IAC3G,OAAO,eAAe,CAAC,CAAC,CAAC,CAAC;AAC5B,CAAC;AAED,gGAAgG;AAChG,MAAM,CAAC,KAAK,UAAU,8BAA8B,CAAC,IAAY,EAAE,MAAgC;IACjG,MAAM,KAAK,GAAG,MAAM,QAAQ,CAAuB,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,MAAM,CAAC,UAAU,CAAC,EAAE,yBAAyB,CAAC,EAAE,IAAI,CAAC,CAAC;IACnI,IAAI,CAAC,KAAK;QAAE,OAAO,IAAI,CAAC;IACxB,IAAI,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,iEAAiE;WACxG,KAAK,CAAC,cAAc,KAAK,GAAG,IAAI,KAAK,CAAC,cAAc,KAAK,aAAa,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,cAAc,KAAK,UAAU,CAAC,KAAK,CAAC,OAAO,CAAC;QAAE,MAAM,SAAS,EAAE,CAAC;IAC7J,MAAM,IAAI,GAAG,qBAAqB,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;IACrD,kCAAkC,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IACtE,OAAO,KAAK,CAAC;AACf,CAAC;AAED,qGAAqG;AACrG,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,IAAY,EAAE,SAAiB,EAAE,OAA2B;IAC1G,IAAI,CAAC,0BAA0B,CAAC,IAAI,CAAC,SAAS,CAAC;QAAE,MAAM,SAAS,EAAE,CAAC;IACnE,MAAM,OAAO,GAAG,qBAAqB,CAAC,OAAO,CAAC,CAAC;IAC/C,OAAO,YAAY,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,uBAAuB,EAAE,UAAU,CAAC,EAAE,IAAI,EAAE,qBAAqB,EAAE,SAAS,EAAE,CAAC,EAAE,KAAK,IAAI,EAAE;QAC/H,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,SAAS,CAAC,EAAE,YAAY,CAAC,CAAC;QACtE,MAAM,MAAM,GAAG,2BAA2B,CAAC,MAAM,QAAQ,CAAC,SAAS,CAAC,CAAC,CAAC;QACtE,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,CAAC,MAAM,CAAC,UAAU,IAAI,MAAM,CAAC,UAAU,CAAC,OAAO,KAAK,OAAO,CAAC,OAAO;eACrG,MAAM,CAAC,UAAU,CAAC,UAAU,KAAK,OAAO,CAAC,UAAU;YAAE,MAAM,SAAS,EAAE,CAAC;QAC5E,IAAI,MAAM,GAAG,MAAM,8BAA8B,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QAChE,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,MAAM,MAAM,GAAG,MAAM,mBAAmB,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;YACxD,MAAM,gBAAgB,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;YACxC,MAAM,OAAO,GAAG,kCAAkC,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,MAAM,CAAC,IAAI,CAAC,mBAAmB,EAAE;gBAC/G,SAAS,EAAE,OAAO,EAAE,WAAW,CAAC,IAAI,CAAC,EAAE,WAAW,EAAE,MAAM,CAAC,YAAY;gBACvE,YAAY,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,CAAC,OAAO,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU,CAAC,UAAU,EAAE,EAAE,cAAc,EAAE,MAAM,CAAC,cAAc,IAAI,IAAI;aAC7I,CAAC,CAAC,CAAC;YACJ,MAAM,GAAG,EAAE,cAAc,EAAE,GAAG,EAAE,cAAc,EAAE,aAAa,CAAC,MAAM,CAAC,EAAE,UAAU,EAAE,OAAO,EAAE,cAAc,EAAE,UAAU,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC;YAC3I,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,SAAS,CAAC,EAAE,yBAAyB,CAAC,EAAE,MAAM,CAAC,CAAC;QAClG,CAAC;QACD,MAAM,MAAM,GAAG,2BAA2B,CAAC,MAAM,QAAQ,CAAC,SAAS,CAAC,CAAC,CAAC;QACtE,IAAI,aAAa,CAAC,MAAM,CAAC,KAAK,aAAa,CAAC,MAAM,CAAC;YAAE,MAAM,SAAS,EAAE,CAAC;QACvE,kCAAkC,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,MAAM,CAAC,OAAO,CAAC,CAAC;QACpF,MAAM,eAAe,CAAC,SAAS,EAAE,EAAE,GAAG,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,eAAe,EAAE,EAAE,EAAE,UAAU,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;QAC7H,OAAO,EAAE,cAAc,EAAE,GAAG,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,CAAC,YAAY,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU;YACnH,KAAK,EAAE,SAAS,EAAE,kBAAkB,EAAE,IAAI,EAAE,WAAW,EAAE,MAAM,CAAC,OAAO,CAAC,UAAU,EAAE,kBAAkB,EAAE,MAAM,EAAE,CAAC;IACrH,CAAC,CAAC,CAAC;AACL,CAAC;AACD,SAAS,SAAS,KAAiB,OAAO,IAAI,UAAU,CAAC,uEAAuE,EAAE,EAAE,IAAI,EAAE,8BAA8B,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC"}
|
|
@@ -1,35 +0,0 @@
|
|
|
1
|
-
import { loadModelNodeClient } from "./node-registry.js";
|
|
2
|
-
import { ProcessSGLangLauncher } from "./process-sglang.js";
|
|
3
|
-
import { DockerSGLangLauncher } from "./sglang.js";
|
|
4
|
-
import { readInferenceGenerationRelease } from "./generation-recovery.js";
|
|
5
|
-
/** Dispatch by the frozen model location; the Harbor provider is independent. */
|
|
6
|
-
export class ManagedSGLangLauncher {
|
|
7
|
-
docker;
|
|
8
|
-
constructor(docker = new DockerSGLangLauncher()) {
|
|
9
|
-
this.docker = docker;
|
|
10
|
-
}
|
|
11
|
-
async start(input) {
|
|
12
|
-
if (!input.lock.model_node)
|
|
13
|
-
return this.docker.start(input);
|
|
14
|
-
return new ProcessSGLangLauncher(await loadModelNodeClient(input.root, input.lock.model_node)).start(input);
|
|
15
|
-
}
|
|
16
|
-
async attach(input) {
|
|
17
|
-
if (!input.lock.model_node || await readInferenceGenerationRelease(input.root, input.record)) {
|
|
18
|
-
throw new TypeError("only an unreleased model-node process supports attachment");
|
|
19
|
-
}
|
|
20
|
-
return new ProcessSGLangLauncher(await loadModelNodeClient(input.root, input.lock.model_node)).attach(input);
|
|
21
|
-
}
|
|
22
|
-
async stopOrphan(root, record) {
|
|
23
|
-
if (!record.model_node)
|
|
24
|
-
return this.docker.stopOrphan?.(root, record) ?? "ambiguous";
|
|
25
|
-
try {
|
|
26
|
-
if (await readInferenceGenerationRelease(root, record))
|
|
27
|
-
return "stopped";
|
|
28
|
-
return await new ProcessSGLangLauncher(await loadModelNodeClient(root, record.model_node)).stopOrphan(root, record);
|
|
29
|
-
}
|
|
30
|
-
catch {
|
|
31
|
-
return "ambiguous";
|
|
32
|
-
}
|
|
33
|
-
}
|
|
34
|
-
}
|
|
35
|
-
//# sourceMappingURL=managed-launcher.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"managed-launcher.js","sourceRoot":"","sources":["../../../src/inference/managed-launcher.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AACzD,OAAO,EAAE,qBAAqB,EAAE,MAAM,qBAAqB,CAAC;AAC5D,OAAO,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAEnD,OAAO,EAAE,8BAA8B,EAAE,MAAM,0BAA0B,CAAC;AAE1E,iFAAiF;AACjF,MAAM,OAAO,qBAAqB;IACH;IAA7B,YAA6B,SAAyB,IAAI,oBAAoB,EAAE;QAAnD,WAAM,GAAN,MAAM,CAA6C;IAAG,CAAC;IACpF,KAAK,CAAC,KAAK,CAAC,KAAwB;QAClC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU;YAAE,OAAO,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;QAC5D,OAAO,IAAI,qBAAqB,CAAC,MAAM,mBAAmB,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IAC9G,CAAC;IACD,KAAK,CAAC,MAAM,CAAC,KAAwB;QACnC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,IAAI,MAAM,8BAA8B,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC;YAC7F,MAAM,IAAI,SAAS,CAAC,2DAA2D,CAAC,CAAC;QACnF,CAAC;QACD,OAAO,IAAI,qBAAqB,CAAC,MAAM,mBAAmB,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAC/G,CAAC;IACD,KAAK,CAAC,UAAU,CAAC,IAAY,EAAE,MAAgC;QAC7D,IAAI,CAAC,MAAM,CAAC,UAAU;YAAE,OAAO,IAAI,CAAC,MAAM,CAAC,UAAU,EAAE,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,WAAW,CAAC;QACrF,IAAI,CAAC;YACH,IAAI,MAAM,8BAA8B,CAAC,IAAI,EAAE,MAAM,CAAC;gBAAE,OAAO,SAAS,CAAC;YACzE,OAAO,MAAM,IAAI,qBAAqB,CAAC,MAAM,mBAAmB,CAAC,IAAI,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACtH,CAAC;QACD,MAAM,CAAC;YAAC,OAAO,WAAW,CAAC;QAAC,CAAC;IAC/B,CAAC;CACF"}
|