• Bug
    • Status: Confirmation
    • Medium
    • Resolution: Unresolved
    • None
    • None
    • None
    • None


      While trying to limit the RAM of each PXC pod in order to be able to provide one MySQL cluster for each customer we noticed two things:

      1. In case we limit the RAM using regular resource limits, an OOM kill disrupts the cluster replication (this being in the middle of an active workload) and a lot of grief ensues;
      2. In case we don't limit the memory at all, it appears the pod allocates as many RAM as it deems necessary. Here, the problem is that once the workload is done, the memory never gets deallocated...

      To trigger the memory spike, we noticed huge INSERT INTO queries usually do the trick. Think mysqldump files, since we mainly notice the RAM allocation while moving our clients from elsewhere to PXC running on our cluster.

      As our business revolves around solving the first problem (successfully limiting the RAM), we posted the original message to the forum. In any case, considering the fact the CR example is being used to test this memory management, we don't know how to proceed.

      So the problem is that the CR in the example don't limit the RAM at all, if anyone set the resource/limits using the value of 1GB of RAM they'll likely incur in the pod being killed by an OOM kill depending on the workload and if this happens, the cluster becomes very unstable.


        Issue Links



              lalit.choudhary Lalit Choudhary
              davi Davi S Evangelista
              0 Vote for this issue
              8 Start watching this issue



                Time Tracking

                  Original Estimate - Not Specified
                  Not Specified
                  Remaining Estimate - Not Specified
                  Not Specified
                  Time Spent - 2 days, 1 hour
                  2d 1h

                  Smart Checklist