Lexemes were not imported on wdqs1009 and when the streaming-updater-consumer started consuming lexemes from the flink ouput streams the triple store did not match what was expected causing diffs to generate a lot of inconsistencies.
The full flink output is still retained in kafka (1 month retention) and we could try to recover lexemes without having to do a full import again.
- add an option to the streaming-updater-consumer to ingest only lexemes
- import the lexemes from /srv/wdqs/lex-munged/
- start a manual streaming-updater-consumer re-reading the whole mutation stream filtered on lexeme from offset 12080891
- stop it when it catches up with the normal consumer