It sounds like one or more of your nodes are under duress. The way to
make this better is to lighten the insert load unfortunately. Check
the logs of the corresponding Cassandra nodes and see if there are
messages being dropped. Decrease the batch size and limit the number
of threads until you get to a point where things can hold up.
Also running 'nodetool cfstats' against the particular node will give
you a picture of what is going on internally. If the above is true,
you should see backups on the pending phase of MESSAGE-SERIALIZER.
No comments:
Post a Comment