Graph Flow Control using Galera 3 wsrep_flow_control_paused_ns

[Kenny Gryp]

The current Flow Control graphs are broken in some scenarios.

Clustercheck does a show global status, which resets the wsrep_flow_control_paused status variable. I have found a case where flow control would almost always be 0 on a cacti graph, but the system has massive flow control issues.

Reason for that is that clustercheck was performed on the database 8 times per second. While that's not optimal, doing it once per minute (which is very very low frequency) can also easily distort the flow control graphs.

Galera 3 has a counter wsrep_flow_control_paused_ns which can be used.

[Kenny Gryp (gryp)]

This is not documented yet, see https://github.com/codership/galera/issues/48

[Roman Vynar (roman-vynar)]

To consider:
> wsrep_flow_control_sent, wsrep_flow_control_recv DERIVE in galera 3, GAUGE in galera 2
> wsrep_flow_control_paused is always GAUGE
> wsrep_flow_control_paused_ns is always DERIVE but only exists in galera 3

[Roman Vynar (roman-vynar)]

wsrep_flow_control_paused in v3 is how much % of the time flow control was active since 'flush status' or total server uptime (in general, useless for v3) . wsrep_flow_control_paused_ns is fine.

[Kenny Gryp (gryp)]

coming from:
- https://bugs.launchpad.net/galera/+bug/1180792
- https://bugs.launchpad.net/percona-xtradb-cluster/+bug/1338523