My team is in pre-production and were chasing some odd behaviors that we 
believe are contributing to stability issues with our application and our
Cassandra cluster.
First some environment information:
Akka version - 2.4.11
Cassandra Version - 3.3 - 3 node Cassandra in 2 DC
Now based on my understanding of Akka persistence in a health system I'd
expect my writes to far outpace my reads in the journal since the
persistent actor should only execute a journal read on creation to see if
there is any journal to playback. After my actor receives RecoverCompleted
I'd expect no more reads and for writes to dominate my journal
transactions. Based on our datadog graphs we're seeing our reads being
constant and often outpacing our writes. I also encountered similar
behavior when I did a simple test with the DynamoDB journal plugin. I
provisioned 15 for both reads and writes per second. When I ran my simple
developer test I looked at my cloudwatch graphs and saw my reads exceeded
provisioned readers but the writes were within expected parameters. So am I
missing something in the Akka infrastructure that is constantly reading
from the journal table?
Regards,
Richard
Richard Ney 's gravatar image asked Jan 11 2017 at 10:10 in Akka-User by Richard Ney

0 Answers