I'm running a microservice in a openshift environment using hazelcast 4.1.1 and the 2.2.1 kubernetes discovery plugin. I have configured hazelcast in embedded mode and I'm running 4 instances of that service. When I scale down the application from 4 to 3 pods the whole migration does not finish and my application throws exception due to WrongTargetException all the time (after one minute).
I analyzed the diagnostic file and I believe that the error comes from the index calculation. If I disable all my indices on my maps, everthing works like a charm. I think this might be related to https://github.com/hazelcast/hazelcast/issues/18079
It seems that the deserialization of my objects are called for each index separatly. Since we have configured a custom (de-)serializer which also applies some compression (LZ4) the migration takes ages.
Can somebody confirm my assumtions? Or are there any other known issues with index calculation and migration?
question from:
https://stackoverflow.com/questions/65860208/slow-hazelcast-migration-when-using-index 与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…