Hello, Just for the Storm's users community information: since today this morning (CET), I am running my pre-production Storm 2.6.0 cluster (10+ topologies relying on storm-kafka, storm-hdfs, etc) using Java 21 runtime (more precisely using OpenJ9 21.0.1 milestone 1 on Linux x64 on Redhat).
Java 21 is used for Nimbus, Nimbus UI, all Supervisor and Logview processes, and for Zookeepers, and of course to execute the topologies code. I had no need to fix anything. One (maybe) important detail: all topologies' code was build with JDK 17, and it was previously running with IBM Semeru 17.0.8.1 (alias OpenJ9). I will keep this setup on the preproduction cluster for say 2 months before No/NoGo for production use, carefully monitoring all stats (topologies latencies, capacities, CPU/Memory/IOPS usage on Supervisor nodes, etc) - so far so good but it's too early to conclude. Note: I won't try compiling our topologies with JDK21 before a while, but it looks like Storm code can be compiled with JDK21 thanks to Richard Zowalla recent attention to make it possible! If others have similar experience to share, they are welcome to let us know ! Thanks, Alexandre
