Skip to main content

​
Checklist

  • The API is reachable only through something that authenticates callers.
  • Only /parse, /jobs/, and /batch/ are routed to ingest.
  • ingest.maxUploadSize and ingest.scratchSize fit your largest request.
  • scaling.maxWorkers and your node capacity match peak load.
  • Workers run on their own nodes, and every component runs on amd64 nodes.
  • orchestrator.workLeaseSeconds is longer than your slowest unit takes.
  • Results are collected within orchestrator.jobRetentionSeconds, or read from the output volume.
  • The output storage expires old results, if they should not be kept forever.

​
Exposing the API

The chart creates a ClusterIP Service, fineparser-ingest, on port 8080, and no Ingress. In-cluster clients can call http://fineparser-ingest.fineparser.svc.cluster.local:8080. To expose it outside the cluster, put your own Ingress, Gateway, or load balancer in front of it:
  • Authenticate callers. Ingest has no authentication. Anyone who reaches it can submit documents and download results.
  • Route only /parse and the /jobs/ and /batch/ prefixes.
  • Allow large bodies, at least ingest.maxUploadSize.
  • Allow long requests. Set timeouts long enough for your largest upload to finish.

​
Sizing uploads

A request needs about twice its size in ingest scratch space.

​
Results and retention

A finished job’s status is kept for orchestrator.jobRetentionSeconds (86400). "infinity" keeps it forever. Results on the output volume are never deleted by FineParser and stay downloadable after the status expires.