@valhallasw I told @Fae to reopen that ticket, so if anyone's, it is my fault for reopening that. I apologize (I was actually monitoring that).

BTW, this will have to wait, because there is another process crashing labsdb hosts, and I have first to investigate that.

As we discussed by email, I am still waiting for a reason to unthrottle the account, and what measures have been taken to avoid another OOM.

To clarify, I am not asking why you need more than one connection. I know and can understand that. I am asking what has been done to limit the concurrency (e.g. I have added a limit to no more than 5 concurrent queries at the same time/I have limited the query execution to 300 seconds).

I thought that was covered by explaining on Phab how use of '-once' makes multiple jobs impossible for the BLP report. As multiple jobs are impossible, there is no reason to expect that this type of job could create a job backlog. All the other Faebot jobs run just a couple of times a day and cannot pose any sort of risk for a future outage, while the BLP report was running every 5 minutes as it was a popular vandalism prevention tool.

No alternative has been suggested as a helpful change from my side apart from this one that I made in November 2015, though presumably work has gone on in the background better to identify and manage unexpectedly large job queue backlogs.

So, how many concurrent connections do you need?

Probably 5 connections is sufficient. Some of my reports rely on more that one database link at the same time to make the SQL, though most are one at a time even when multiple queries are made for the report.

I'll keep an eye on error logs in case some reports drop out due to clashes with each other, though I could probably fix that by jiggling with crontab.