Thanks, Gabor! We switched back to the hashmap state backend for now. We'll troubleshoot if we go back to rocksdb. Jad
On Wed, Nov 6, 2024 at 6:51 AM Gabor Somogyi <gabor.g.somo...@gmail.com> wrote: > Hi Jad, > > There is no low hanging fruit here if you really want to find this out. > Such case the memory manager tries to allocate and deallocate the total > memory which is prepared for. > When not all the memory is available then it's not going to be successful > and you see the mentioned exception. > > I would suggest the following: > * Presume this is already fulfilled but double check that Java 8u72+ used > (as the exception message writes) > * Create a custom Flink image with additional log entries > in UnsafeMemoryBudget > * Remote debug with a breakpoint in UnsafeMemoryBudget > > All in all one must find out why the available memory is not equals to the > total memory in the memory manager. > > G > > > On Tue, Oct 29, 2024 at 6:03 PM Jad Naous <j...@grepr.ai> wrote: > >> Hi Flink Community, >> I'd really appreciate your help. We're trying to switch from using the >> heap state backend to rocksdb, and have been encountering a warning "Not >> all slot managed memory is freed at TaskSlot..." when the pipeline >> restarts. Any pointers to troubleshoot this issue? >> Many thanks! >> Jad. >> >
