0

I have mariaDB Galera 10.2 Cluster running with 3 nodes; in google cloud VMs with 8 cpu ; 15Gigs Ram / node; My slave_thread setting = 24 ; currently there is no much write happening and local_receive_q_avg= 0.019481 but my Cert_deps_distance= 54 in two nodes and 0 in one node ? What does that mean ?

1

wsrep-cert-deps-distance

Most likely the 0 on one node was the only node being written to. So there are limited galera status changes there as it does not apply in the same was as the other nodes do.

  • So what should I do to make things equal ? BTW the flow_control_pause and sent = 0 always; so it means that everything is committing fine, isn't it ? Please let me know; thanks for the update as well , @danblack – SAGAR Nair Oct 15 '18 at 3:22
  • Unless you have the cluster balanced to have all writes on all nodes (has some down sides), don't make the objective to be equal. 0 flow control pause is good. wsrep_local_cert_failures will tell you if there are failures. Ignore crappy metrics like local_receive_q_avg that average over the entire uptime of the server. Graph many of these metrics (ideas) – danblack Oct 15 '18 at 4:01
  • @danbkack, thank you, we have haproxy loadbalancer infront of it, We don't have that much writes instead we are read hungry; so all writes have been forwarded to all nodes otherwise we will write it to one node and read from all; – SAGAR Nair Oct 15 '18 at 4:46
  • I would like to know how to find the current TPS (Transactions per Second); how to get a added value output with a query if TPS = (Com_commit + Com_rollback) / second ? I think tps is a sort of performance variable that qps. ? I know that qps can be found out by Questions / uptime. whats your view on that; Could u post a query to find TPS in a mysql / MariaDB server ? – SAGAR Nair Oct 15 '18 at 4:49
  • TPS is a good way to go. Averaging over uptime becomes grossly useless for beyond small values of uptime as behaviour change change over time. Most monitoring will show figures captured 1-5 minutes or less. The munin graphs are the ones I look at (which is why I wrote them). Keeping values separate is helpful for looking at them however stacking them in graphs is a way to get local vs remote view and still have a total. – danblack Oct 15 '18 at 5:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.