Watch metadata algorithm waits for 1 second for metadata to be synced. If 1 second is not enough to metadata sync, then retry helper gets invoked. This retry helper will wait until for existing watcher to become in-sync which has its own timeout.
DDL service manager closes the retry helper after a timeout.
Before version 6.6.2, the timeout was 0.5 seconds.
In version 6.6.2, the timeout was increased to 5 seconds.
In the attached CBSE (using version 6.6.0), the metadata file size is already large. So, to be on safer side, the timeout should be increased to 30 seconds (which is default in version 7.0 onwards).
Note that such timeout does not exist in case of query client.
Note that the workaround in this case is disruptive as the node with large metadata file size needs to be removed from the cluster.
|For Gerrit Dashboard: MB-50231|
|167173,2||MB-50231: Increase timeout for watcher to go live to 30 Sec||mad-hatter||indexing||Status: MERGED||+2||+1|