Skip to content

[codex] Expose deploy config mechanics#61

Closed
xu16601526267 wants to merge 1 commit into
developfrom
codex/product-mechanism-fixes
Closed

[codex] Expose deploy config mechanics#61
xu16601526267 wants to merge 1 commit into
developfrom
codex/product-mechanism-fixes

Conversation

@xu16601526267
Copy link
Copy Markdown
Contributor

Summary

This PR fixes three product-mechanism issues that made AIMA deploy troubleshooting hard to trust:

  • separates deploy.dry_run resolved catalog/default config from the effective config after CheckFit runtime adjustments
  • filters LLM-only CLI flags before emitting runtime command args for non-LLM service wrappers such as image/ASR/TTS services
  • exposes service/catalog context in status and diagnostics so UI/CLI context splits and stale overlays are visible

Details

deploy.dry_run now keeps the legacy config field as the effective deployment config for compatibility, while also adding resolved_config, effective_config, fit_adjustments, resolved_provenance, and effective_provenance. This prevents UI/agent callers from confusing catalog defaults with resource-driven automatic adjustments.

Runtime flag generation now accepts catalog model type and engine context. LLM knobs like max_model_len, gpu_memory_utilization, and mem_fraction_static are preserved for LLM/vLLM/SGLang-style commands, but are not passed into service wrappers that do not expose those flags.

system.status, system.diagnostics, and catalog status now include service context such as data_dir, overlay_dir, catalog load time, latest overlay mtime, and a reload hint when overlays changed after the process loaded the catalog.

Validation

Passed:

  • go test ./internal/knowledge
  • go test ./internal/runtime -run "TestConfigToFlags" -count=1
  • go test ./cmd/aima -run "TestResolveDeploymentKeepsResolvedAndEffectiveConfigSeparate|TestServiceContextReportsStaleOverlayHint" -count=1
  • git diff --check

Known local Windows environment failures when running go test ./internal/knowledge ./internal/runtime ./cmd/aima:

  • TestBuildRunArgs_ModelFileVolume expects Unix /mnt/... Docker volume formatting but the test ran under Windows paths
  • several native stale-port tests fail under this local Windows process/port behavior
  • TestResolveLocalModelPathNoPullUsesSymlinkMirrorSource cannot create symlinks without elevated Windows privileges

@xu16601526267
Copy link
Copy Markdown
Contributor Author

Superseded by the split draft PRs: #63 for dry-run resolved/effective config, #64 for engine-aware config flag filtering, and #65 for service/catalog context diagnostics. Closing this aggregate PR to keep review boundaries clean.

@xu16601526267
Copy link
Copy Markdown
Contributor Author

Closing in favor of split draft PRs #63, #64, and #65.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants