Diagnosing the problem.
My last post had a plug about the migration of our Wordpress instance to a new server. However, it didn’t go completely smoothly. The site had gone down a few times in the first day after the migration, with Wordpress throwing “Error establishing a database connection.” Sure enough, MySQL had gone down. A simple restart of MySQL would bring the site back up, but what caused the crash in the first place?
A peek into
/var/log/mysql/error.log yielded this:
2016-10-12T21:20:50.588667Z 0 [ERROR] InnoDB: mmap(137428992 bytes) failed; errno 12 2016-10-12T21:20:50.588702Z 0 [ERROR] InnoDB: Cannot allocate memory for the buffer pool 2016-10-12T21:20:50.588728Z 0 [ERROR] InnoDB: Plugin initialization aborted with error Generic error 2016-10-12T21:20:50.588749Z 0 [ERROR] Plugin 'InnoDB' init function returned error. 2016-10-12T21:20:50.588758Z 0 [ERROR] Plugin 'InnoDB' registration as a STORAGE ENGINE failed. 2016-10-12T21:20:50.588767Z 0 [ERROR] Failed to initialize plugins. 2016-10-12T21:20:50.588772Z 0 [ERROR] Aborting
So it looks like an out-of-memory error. This VPS only has 512 MB of RAM, so I wouldn’t be surprised. Clearly, some tuning would be necessary. First, we’ll reduce the size of MySQL’s buffer pool, then shrink Apache’s worker pool, and finally add a swapfile just in case memory pressure remains a problem.
The error that we saw was for the allocation of a buffer pool for InnoDB, one of MySQL’s storage engines. We can see from the log that it’s trying to allocate somewhere around 128 MB using
mmap. This corresponds to the default value of the
innodb_buffer_pool_size configuration option. Let’s go ahead and trim this down to about 20 MB – it’ll reduce MySQL’s performance, but we don’t have much of a choice on a machine this small.
On Ubuntu, I put this option in
innodb_buffer_pool_size = 20M
sudo service mysql restart, and rejoice as MySQL no longer uses 25% of your RAM.
Most of Apache’s memory usage comes from the fact that it preemptively forks worker processes in order to handle requests with low latency. This is handled by the
mpm_prefork module, and if enabled, its config can be found in
/etc/apache2/mods-enabled/mpm_prefork.conf (on Ubuntu, at least).
By default, Apache will create 5 processes at startup, keep a minimum of 5 idle processes at all times, allow up to 10 idle processes before they’re reaped, and spawn up to 256 processes at a time under load. Let’s reduce these to something more sane given the constraints of our system:
<IfModule mpm_prefork_module> StartServers 3 MinSpareServers 3 MaxSpareServers 5 MaxRequestWorkers 25 MaxConnectionsPerChild 0 </IfModule>
Now, sudo service apache2 restart` and you’re done.
Creating a swapfile.
Most VPSes don’t give you a swap partition by default, like you would probably create on a dedicated server or your desktop. We can create one using a file on an existing filesystem, in order to make sure there’s extra virtual memory available in case our tuning doesn’t handle everything perfectly.
First, let’s pre-allocate space in the filesystem. We can do this using the
fallocate command. I made a 2 GB swapfile:
$ sudo fallocate -l 2G /swapfile
Now, give it some sane permissions:
$ sudo chmod 600 /swapfile
Next, format the file so it looks like a swap filesystem:
$ sudo mkswap /swapfile
And finally, tell the OS to use it as swap space:
$ sudo swapon /swapfile
Now, we’ve got swap for the time being, but it won’t persist when we reboot the system. To make it persist, simply add a line to
/swapfile none swap sw 0 0
Congratulations, you’re now the proud owner of some swap space.
Hopefully this post will help anyone suffering stability problems with MySQL and Apache on a small VPS. I’ve adapted the instructions here from several sources, most notably this StackOverflow post, and this article from DigitalOcean.