So the big upgrade from 14.04 to 16.04 came up a little unexpectedly 2 weeks ago. many small package changes, and went from Percona 5.6.33-79.0-log (on 14.04) to 5.6.37-82.2-log (on 16.04) - and performance is really not very good since the upgrade. In fact, it fails to meet our minimum requirements. A mass insert process that, triggers & all included, used to run at ~7 r/s, now barely musters 0.7 r/s, a figure improved (not enough to enable the daily process to complete in under 24h) to 1.5 r/s by setting sync_master & sync_relay_log_info to 0
Has anyone else experienced this performance regression issue upgrading from Trusty to Xenial, or have any ideas where I might look to see what is going on?
You might already be using MySQL 5.6 Performance Schema but just in case here’s a link to a webinar and slides that might help you troubleshoot. [url]Technical Presentations from Percona
The team here also suggested that for systemd you might need also to set LimitNOFILE if you could check that?
I have been looking at performance schema data and found nothing really telling. though I think without setting up a proxy to redirect identical output to one of each, it will be hard to tell, because in a 4-shard environment there is only one shard whose data-set/usage-pattern combination surfaces these problems.
As for systemd - I am looking at the docs for it on dev.mysql.com and there is no info. for MySQL 5.6 - Does Percona 5.6 need systemd configuration, and where would I find information on how to set the relevant options?
Mmm … the only thing I found on dev.mysql.com is this [url]https://dev.mysql.com/doc/refman/5.7/en/using-systemd.html[/url] but I imagine that you might have found that, it’s not specific to performance and is part of the 5.7 documentation not 5.6. I’ve not found anything published just yet on systemd config for Percona 5.6 but I’ll keep looking.