WIP plan please fix:
- [] Install & test elasticsearch 6.5.4 to relforge
-- [] copy the .ltrstore index from production to verify compatibility before upgrading
- [] ES6 upgrade of the search cluster in codfw
-- [] stop mjolnir daemon
-- [] disable apifeatureusage logstash output to codfw
-- [] ship a mw-config patch to stop writing to codfw (note the time)
-- [] disable mirroring of translate ($wgTranslateClustersAndMirrors)
-- [] bring down all search clusters in codfw
-- [] ship new plugins to codfw machines
-- [] install elasticsearch6 and the plugins
-- [] test on psi
--- [] bring a single master up and verify plugins load and config is OK
--- [] bring other nodes up to start recovery and make sure that indices are loading properly
-- [] redo previous steps on omega & chi
-- [] restart mjolnir
-- [] ship a mw-config patch to resume sending writes to codfw
-- [] run forceSearchIndex.php to index lost updates to codfw
-- [] refresh TTMServer index in codfw
-- [] resume apifeatureusage logstash output to codfw
- [] merge MW es6 branch to master on Elastica & CirrusSearch
-- [] deploy a config patch to switch to codfw based on train version
[] Wait for the train to roll forward, if something's wrong block the train and revert merge commits from es6 to master (Elastica/CirrusSearch)- [] Upgrade deployment-prep to ES6
[] Make sure the train is up and running (it must not be rolled back without backporting ES6) on the new ES6 version on all wikis- [] Wait for the train to roll forward, if something's wrong block the train and revert merge commits from es6 to master (Elastica/CirrusSearch)
[] ES6 upgrade search cluster in eqiad (- [] Make sure the train is up and running (it must not be rolled back without backporting ES6) on the new ES6 version on all wikis
- [] ES6 upgrade of the search cluster in eqiad
-- [] follow same steps used for codfw)
-- [] switch traffic back to eqiad