The exported replication lag does not handle all failure modes, and can
report 0 for replicas that are out of sync and incapable of recovery.
A proper replacement for that metric would require a different approach
(see e.g. #1007), but for a lot of folks, simply exporting the age of
the last replay can provide a pretty strong signal for something being
amiss.
I think this solution might be preferable to #977, though the lag
metric needs to be fixed or abandoned eventually.
Signed-off-by: Conrad Hoffmann <ch@bitfehler.net>
Use const definitions to make collector registration consistent.
* Use collector subsystem name consistently.
* Fix up replication metric name unit.
Signed-off-by: SuperQ <superq@gmail.com>
