Skip to main content

Operator

Presentation​

The operator does 2 things :

  • monitor 0 or more buckets to check for changes inside (taken from Vstore config directly)
    • if any changes is detected, the Vstore deployment is restarted
  • provide a REST API that tells you whether Vstore is currently ready to serve a given version of its model
    • you provide the version number as a pathparam/queryparam
    • the api return :
      • NotReady : if none of the pod are in the desired version
      • InProgress : If at least one of the pod is in the desired version but at least one isn't
      • Ready : if all the pod are in the desired version

operator API​

The current API is to be used inside a kubernetes cluster and as such does not use TLS.

Authentication​

You need to provide your Vstore token the same was as if you were querying Vstore itself, as a bearer token in a Authorization header.

Without this token, or an invalid token you will receive a http 401 error.

Get Vstore readiness for a given model version​

# Request
GET /vstore/model/status?version=<version_number>

# Response
{"status":"<status>"}

The version_number you provide must match the version returned by the GET /v0/models to consider the Vstore that answered as Ready.

The three possible status are :

  • NotReady : if 0 replica have the desired version
  • InProgress : if 1 or more replica, but no all of them, have teh desired version
  • Ready : if all the replicas have the desired version

Note: when a Vstore pod is unable to answer (eg: restarting) it is considered as not ready