As on operator of WDQS / WCQS, I want to deploy Flink in a way that is robust so that we can deploy WDQS Streaming updater on top of it.
The strategy as discussed in part in T247058 is to use k8s to provide compute resources and Swift as storage. Communication with the teams in charge of k8s and Flink has already started, but our needs have to be formalized, and the concrete strategy for implementation needs to be defined.
The wikidata query service streaming updater is currently deployed to Yarn and is in beta. Current production plans are for the updater to be deployed in the Kubernetes cluster to the staging cluster and to the Eqiad cluster. No plans right now for multi-cluster deployment.
- Review charts
- rdf-streaming-updater namespaces in k8s.
- rdf-streaming-updater accounts in k8s.
- rdf-streaming-updater puppet private tokens.
- rdf-streaming-updater helmfile.d files
-
LVS setupnot needed -
DNS for LVS recordsnot needed -
Discovery DNSnot needed - Monitoring dashboard rdf-streaming-updater: https://grafana.wikimedia.org/d/gCFgfpG7k/flink-session-cluster?orgId=1 and https://grafana.wikimedia.org/d/fdU5Zx-Mk/wdqs-streaming-updater?orgId=1&from=now-12h&to=now
- Integration and Acceptance tests: currently updating query-preview.wikidata.org
AC:
- Flink is deployed on a production k8s cluster
- stream of TTL update is available in Kafka and ready to be consumed