In one of our Kafka clusters we noticed that fetch sessions are being evicted and lots of clients print `FETCH_SESSION_ID_NOT_FOUND` log messages. We tried to increase max.incremental.fetch.session.cache.slots from 1k to 10k in the brokers but the slots were immediately used up again and slots were evicted as well.
We assume it's caused by one or more clients that leak sessions but we have no clue which one it might be. Is there a good way to figure it out - ideally something like listing the fetch sessions by client.id, sasl username and/or ip address? Optionally, can we limit the number of fetch sessions either per client.id, IP address or SASL user?
It's alarming that a single client can cause these issues (even though they are not critical) for the whole cluster.
Best,
Martin.
We assume it's caused by one or more clients that leak sessions but we have no clue which one it might be. Is there a good way to figure it out - ideally something like listing the fetch sessions by client.id, sasl username and/or ip address? Optionally, can we limit the number of fetch sessions either per client.id, IP address or SASL user?
It's alarming that a single client can cause these issues (even though they are not critical) for the whole cluster.
Best,
Martin.
Comments
Post a Comment