mysql.replication.seconds_behind_master remains 0 when slave is stopped or broken

[dcrosta]

If you stop a MySQL slave (or if its replication becomes broken due to corruption on the master or an un-replicate-able event in the binlog), the field Seconds_Behind_Master becomes NULL. According to metric explorer, the value remains "0" for this metric, even though that's (very likely) not true.

I would propose that either:

1. The metric is set to something like -1 (an impossible value in normal circumstances), so that users can create metric alerts, or
2. The (old-style) check be given an option to send events when replication is broken. Perhaps this is best done after Move mysql to checks.d #391.
3. The metric stops reporting, though I don't like this because you can't set up metric alerts about a metric that has stopped reporting (yet?)

If you'd like to do option 1 or 3, I'm happy to do the work and send a pull request. Option 2 probably requires more time than I could commit to.

[olidb2]

Another possibility would be to have the metric count seconds after the
last reported value. Is that information we can get from the DB, or would
that require us to keep state in the agent?

On Wed, Mar 27, 2013 at 9:50 AM, Dan Crosta notifications@github.comwrote:

If you stop a MySQL slave (or if its replication becomes broken due to
corruption on the master or an un-replicate-able event in the binlog), the
field Seconds_Behind_Master becomes NULL. According to metric explorer,
the value remains "0" for this metric, even though that's (very likely) not
true.

[image: replication]https://f.cloud.github.com/assets/35122/308319/30318d9c-96e5-11e2-845e-680d5d71f917.png

I would propose that either:

1. The metric is set to something like -1 (an impossible value in
   normal circumstances), so that users can create metric alerts, or
2. The (old-style) check be given an option to send events when
   replication is broken. Perhaps this is best done after Move mysql to checks.d #391Move mysql to checks.d #391
   .
3. The metric stops reporting, though I don't like this because you
   can't set up metric alerts about a metric that has stopped reporting (yet?)

If you'd like to do option 1 or 3, I'm happy to do the work and send a
pull request. Option 2 probably requires more time than I could commit to.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/425
.

[dcrosta]

If the replication is broken for any reason it seems that Seconds_Behind_Master becomes NULL, so this would be something the agent would have to compute.

[alq666]

Released with 4.0.1

[alq666]

With the metric mysql.slave_running