This ticket is for tracking final deployment steps for WCQS production. This does not include exposing the service to a general public, only steps to ensure up to date WCQS, with fully working Streaming Updater.
Start date: Jan 11th 2022
prereqs:
- https://gerrit.wikimedia.org/r/c/operations/cookbooks/+/745629 - merged
- topic <dc>.mediainfo-streaming-updater.mutation has a retention of 30 days (@Ottomata - can you help with that?)
Dump: https://dumps.wikimedia.your.org/other/wikibase/commonswiki/20220109/commons-20220109-mediainfo.ttl.gz
savepoint(eqiad): swift://rdf-streaming-updater-eqiad.thanos-swift/commons/savepoints/bootstrap_20220109
savepoint(codfw): swift://rdf-streaming-updater-codfw.thanos-swift/commons/savepoints/bootstrap_20220109
Start time (for eventgate flink consumers in Kafka): 2022-01-09T19:00:04Z
Start time of the flink pipeline (for streaming updater consumers on Blazegraphs hosts): TBD
consumer group for flink pipeline: wcqs_streaming_updater
Steps:
- clear the journals on all instances
- Reload WCQS instances with the newest dump (can be done in parallel), streaming updater consumers should be turned off
- wcqs1001
- wcqs1002
- wcqs1003
- wcqs2001
- wcqs2002
- wcqs2003
- Source rev_map.csv from hdfs (automatically generated weekly by airflow)
- set offsets for recent changes events topics based on timestamps for the dump (for both steps see https://wikitech.wikimedia.org/wiki/Wikidata_Query_Service/Streaming_Updater#First_run_(bootstrap) )
- deploy streaming updater consumers to eqiad and codfw (merge puppet patch for streaming updater role)
After the process it should take few hours at most for all the instances to catch up with the lag (dashboard)