My client got ROS pushback error.
I suggested them to tune the MergeOutInterval and CintainerPerProjectionLimit parameter and run do_tm_tack ('mergeout').
They are concerned that this issue has been caused by a few number of CPU cores.
Is there a possibility that TM Merge Out processing will be delayed or not performed due to a small number of CPU cores?
Vertica version: 9.1.1
Hardware specs:
4 cores is a bit light, and it isn't the number that come in an E5-2699, so I'm assuming this is a cloud/VM scenario.
But I think it probably isn't just the number of cores. Look things in this order:
1. Is the TM running (and not keeping up) or is the TM waiting for something to do?
2. If it is running, how many threads, and is it achieving 1GB / min?
3. The TM resource pool settings... how much concurrency and how much memory?
Basically, 4 cores isn't that many, but the setup should still benefit from the regular TM tuning guide.
Maybe check if TM has been disabled or stopped first just in case TM never have the chance to run. Then as Chuck pointed out, look at TM pool setup, memory and concurrency. If that looks normal and TM suppose to run, I would either look at tuple_mover_operations to see if any mergeout happened but failed, that should also give you idea if it did happen or tried but failed.
Already have an account? Login
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.
77 4th Avenue
Waltham, MA 02451 USA
