Re: [I] [client] Avoid repeated synchronous schema fetches in ClientSchemaGetter [fluss]

["fxbing (via GitHub)" <gi...@apache.org>]

fxbing commented on issue #3336:
URL: https://github.com/apache/fluss/issues/3336#issuecomment-5267283911

   We observed production impact consistent with this issue in a Fluss-to-Fluss CDC workload while consuming records encoded with historical schema IDs.
   
   On a synchronous cache miss, `ClientSchemaGetter#getSchema` fetched the schema through `Admin#getTableSchema`, but did not populate `schemasById`. Repeated decoding of records with the same schema ID could therefore repeatedly hit the metadata path.
   
   During the incident, ZooKeeper cached-read traffic, source TabletServer network traffic and request queues, and CDC lag increased together, while TabletServer CPU and disk were not saturated. After rolling restarts and backlog catch-up, the metadata-read amplification disappeared and the read rate fell below its pre-incident baseline.
   
   Restart is only a temporary mitigation: the amplification can recur when an uncached schema ID is encountered.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: issues-unsubscribe@fluss.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org