pg_up != 1 doesn't work when I stopping PostgreSQL

[moovs]

pg_up != 1 alert doesn't work when I stopping PostgreSQL, I think it's happens because when PostgreSQL is down postgres_exporter can't connect to DB.

is this the expected behavior for this exporter and this metric or I did something wrong?

thanks!

[alexey-gavrilov-flant]

I have the same problem. this is because the exporter starts to return results longer than the "Scrape Duration".
Moreover, this happens cyclically every 30s.

[moovs]

ok, thank you! Do you have any idea how can this be fix? @alexey-gavrilov-flant

[alexey-gavrilov-flant]

I see two ways to fix it

1. changing quantity here to 1 https://github.com/wrouesnel/postgres_exporter/blob/master/cmd/postgres_exporter/postgres_exporter.go#L960
2. or here the increasing duration https://github.com/wrouesnel/postgres_exporter/blob/master/cmd/postgres_exporter/postgres_exporter.go#L970 https://github.com/wrouesnel/postgres_exporter/blob/master/cmd/postgres_exporter/postgres_exporter.go#L977
3. but the most correct way. to make the variable from which the last written state will be read. and the connector will work in a separate thread and update the value as new data is received, blocking the variable for a minimum amount of time.

[moovs]

I see two ways to fix it

1. changing quantity here to 1 https://github.com/wrouesnel/postgres_exporter/blob/master/cmd/postgres_exporter/postgres_exporter.go#L960
2. or here the increasing duration https://github.com/wrouesnel/postgres_exporter/blob/master/cmd/postgres_exporter/postgres_exporter.go#L970 https://github.com/wrouesnel/postgres_exporter/blob/master/cmd/postgres_exporter/postgres_exporter.go#L977
3. but the most correct way. to make the variable from which the last written state will be read. and the connector will work in a separate thread and update the value as new data is received, blocking the variable for a minimum amount of time.

or use absent() function in alert condition:
absent(pg_up{instance="192.168.0.0:9187"})
for me it's the most easy way)

[frenkye]

You can't be serious suggesting using absent for this. Just testing v0.8 and when i stop cluster, the exporter takes a lot of time to respond.

There is only one solution which is right, the exporter should respond with metric pg_up 0 when cluster is down. That can be even with only pg_up the only metrics.