vxData production DB restore + cluster storage gotchas
— Postgres restore-from-S3 runbook, the vxData production deployment
contract, and the cluster-wide traps: topolvm-no-replication, ArgoCD
self-heal, the Argo 128KB ARGO_TEMPLATE limit, nginx worker_processes,
imagePullSecrets, registry mirrors vs pull-through caches, and bulk rclone
migrations.
Coder workspace template — where the template lives,
the gpu_enabled MIG toggle, and the ArgoCD exclusion that makes template
changes a manual Coder update.
Keel auto-redeploy — how Keel is deployed
(polling, no webhooks) and the annotations that wire a Deployment to a
mutable tag.
vxData
vxData SDK + schema model — resource/payload
model, the current SDK surface, the child_identifiers tree-navigation
gotcha, and the apps/vxdata-api local dev loop.
vxData jobs ingestion patterns — writing an
ingestion job in apps/vxdata-jobs: discovery, mapping, stable identifiers,
linking, plus the polars and vxAnnotate data traps.
vxAnnotate ↔ vxData sync — ID mapping, why mask
decoding belongs in vxa-sdk, and which job template to copy.
PHI-safe data handling — what deliberately holds
PHI, the missing guardrails, and the PHI-safe profiling recipe. Read before
profiling or exporting anything.
Apps and packaging
Dashboard app — local dev and deploy —
mock-mode dev, the @virdx/ui-components build prerequisite, metrics
sources, the Containerfile COPY-allowlist trap, and the image-tag + Keel
deploy model.
Dashboard histo tile serving — the
dashboard-api/histo router (large_image TIFF + OME-Zarr streaming) and
the read-only storage hostPath mounts it needs.
Conda packaging and releases — the actual
*-package.yml / *-release.yml flow, multi-output recipes for in-repo path
dependencies, and the ABI/lockfile pitfalls.
