0

We have an "cache" (javax.cache.Cache) implementation that is a wrapper of Hazelcast's IMap. We use a composite Object key.

We upgraded from version 3.12.5 to 5.1.1. When I deploy the system on a local Windows machine, all works well. But when I deploy the system into an Kubernetes environment, the map just "does not work". Values do not get persisted into the map (after a put operation). An Hazelcast cluster does get formed so it does not seem to be an auto discovery issue. I also have another K8S env in which it does work properly.

I enabled Hazelcast's diagnostic mode and it does not seem to show me anything useful. I do not get any error or warn messages from the com.hazelcast.* package. The same issue happened also when I tried version 4.x.

I am trying to explore ways which will help to the realise what is the issue here. Thank you.

Ido Gal
  • 528
  • 10
  • 26
  • 1
    How do you configure the client? Can the client read the values it is writing? Are you sure your client connects to the cluster you are monitoring? – František Hartman Apr 25 '22 at 11:26
  • I work only with Hazelcast members. No clients. – Ido Gal Apr 25 '22 at 12:08
  • I see, so what does _Values do not get persisted into the map_ mean exactly? That members don't see the values they write? That other members don't see them? – František Hartman Apr 25 '22 at 12:14
  • I work with a single member (pod, node) in order to reduce the complexity of the problem. Yes, a member does not see the values that it wrote. Same happens in a configuration of 2 or 2 nodes/pods/members. – Ido Gal Apr 25 '22 at 12:15
  • Well, it looks like a misconfiguration on your application side, share your config, code how you create & access the cache etc. – František Hartman Apr 25 '22 at 13:47
  • After some (quite a lot...) tinkering, I noticed that the issue occurs when I set the map's max-idle value to the allowed max value, 2147483647. Still occurs if I use 2147483646, but stops if I use 214748364. – Ido Gal Apr 26 '22 at 16:45

2 Answers2

0

Turns out it is a bug. Hazelcast recommends to use the value of 0 instead.

Ido Gal
  • 528
  • 10
  • 26
0

I had the same problem but i was migrating from hazelcast 3.11.1 to 5.1.2, I found the IMap in "com.hazelcast.map" not in "com.hazelcast.core"

HICHAM ASBIKA
  • 1
  • 1