The Short: We're hitting max connections and seeing very poor performance from MySQL during peak business hours and cannot determine if it is a load problem or a configuration problem.

The Long: To begin, here’s our server specs hosted on Digital Ocean:

Cores: 20


total used free shared buff/cache available Mem: 62G 48G 4.7G 216M 10G 14G

MySQL WorkBench Dashboard Snapshot:

MySQL Workbench Dashboard

We typically have over 200 entries in processlist, almost all of which are "Opening tables", "closing tables", and, slightly less frequent, "cleaning up". The cleaning up entries have COMMAND killed and NULL INFO (SELECT * FROM INFORMATION_SCHEMA.PROCESSLIST). The only queries we really see taking very long are some INSERTS into tables with heavy use, and UPDATE commands with the primary key in the WHERE for tables with heavy us. Their state is “update” in the processlist.

MySQL .cnf settings:


We have 800+ databases on our single live server, and many thousands of users using an app that makes frequent server requests with lots of individual record updates or inserts that need to sync to multiple users.

The Question:

Given the .cnf settings and our server specs, in relation to the MySQL WorkBench dashboard snapshot, should we be experiencing the kind of slowdown/connections we're seeing with this configuration? We constantly hit our Max Connections and get SQLSTATE[HY000] [2002] Resource temporarily unavailable errors in PHP from PDO.

Does anyone have any suggestions to improve the performance? Different config settings? We know long term solutions, like putting large clients on separate servers, among others—but for the short term, is there any way to optimize our server? Our app and website are barely usable during the rush hours.

Additional Stat:

Linux 4.0.4-301.fc22.x86_64 04/25/2016 _x86_64_ (20 CPU)

avg-cpu: %user %nice %system %iowait %steal %idle

8.51 0.01 0.67 0.28 0.35 90.18

Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn vda
269.78 1681.90 2180.52 37679232421 48849706700

  • What about another system metrics? What is CPU load and utilization? If there is lot of IO WAIT, your disk subsystem is bottle neck. – Ondra Sniper Flidr Apr 25 '16 at 23:04
  • See the updated io stat at the bottom of my post. – Pete_1 Apr 25 '16 at 23:20

My money is on "configuration" – but probably in the opposite way you anticipate... I suspect you've configured it too much.

It looks to me as if you've been "tuning" your server's performance right into the ground.

If you have been using a "tuning" script – stop doing that.

For example... table_definition_cache=6000 and table_open_cache=6000 ... what is the justification for this?

Most of the parameters you have configured away from their defaults seem like guesswork.

Unless you benchmarked each of these customizations, individually, before and after you made them, individually... well, you probably see where I'm going... they're probably unjustified, and you are likely to be better off without them. Bigger isn't always better. Bigger can be harder to juggle.

I would expect good things if you'd simply comment out almost all of these settings except for innodb_* and let the server run with more parameters at their default values.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.