← Back to team overview

maria-developers team mailing list archive

Re: On the issue of Seconds_behind_master and Parallel Replication

 

Kristian Nielsen <knielsen@xxxxxxxxxxxxxxx> writes:

>  - Should Seconds_Behind_Master be changed as per above in parallel
>    replication (from 10.0 on)?

>  - If so, should the change to Seconds_Behind_Master also be done in the
>    non-parallel case in 10.1? What about 10.0?

So there seems to be agreement that Seconds_Behind_Master should be changed
in 10.1 to only update on commit (also in non-parallel case). This does seem
a better behaviour.

One consequence of this is the following: Suppose the master is idle for 1
hour. The Seconds_Behind_Master is reported as 0 on a slave that is up to
date. Suppose now the master executes a transaction. While that new
transaction is being applied, the Seconds_Behind_Master will be reported as
3600 (one hour), then go back to zero.

This is arguably the correct value - the slave is indeed in the state of the
master one hour ago while executing that transaction. But I still wonder if
this will not just cause another kind of confusion with users (or existing
monitoring tools), as they will interpret it as the slave being far behind
the master. While in reality it is only one transaction behind.

 - Kristian.


Follow ups

References