How can I fix recurring PHP 503 Errors on my Apache-mod_proxy_fcgi-PHP-FPM Server?

Question

i'm having trouble with my php-fpm setup. Maybe you guys can point me in the right direction. First off, everything is working fine. But every now and then, I'll get 503 Errors. Those errors are gone once I reload the website. They only ever appear on php sites and are not isolated to one domain or one framework. I am receiving the 503 errors in PHPmyAdmin, Wordpress and Typo3. Those are the 3 sites I have tested. They are on seperate vhosts and have different php-fpm pools, but they share the same php-fpm master prozess.

The server I am running is Apache 2.4 (MPM-Event Workers), without mod_php or cgi/fastcgi. Instead, I'm using mod_proxy and mod_proxy_fcgi to pass every .php file to my php-fpm process. One thing to note is, the server is not yet in production, so there is little to no traffic at all. Server Hardware is strong, 12 VCores und 32 GB Ram.

My mod_proxy and mod_proxy_fcgi setups are default - I haven't changed anything there.

My vhost config (the proxy part):

<FilesMatch "\.php$">
        SetHandler "proxy:unix:///opt/php-5.6.11/var/run/php5-fpm-mywebsite.sock|fcgi://mywebsite/"
    </FilesMatch>
    <Proxy fcgi://mywebsite/ enablereuse=on retry=0>
</Proxy>

Note: I have had max=10 in the Proxy directive before and it seemed to produce the 503 error more often. Now that I have removed max=10, it seems to occur less. Might just be coincidence though.

My PHP-FPM Pool Config (the relevant parts):

listen = var/run/php5-fpm-mywebsite.sock
listen.owner = mywebsite
listen.group = www-data
listen.mode = 0660
listen.backlog = 65535

user = mywebsite
group = www-data
listen.allowed_clients = 127.0.0.1

pm = ondemand
pm.max_children = 20
pm.process_idle_timeout = 15s

request_terminate_timeout = 300s
rlimit_files = 131072
rlimit_core = unlimited
catch_workers_output = no

My PHP-FPM Config (the relevant parts):

emergency_restart_threshold = 10
emergency_restart_interval = 1m
process_control_timeout = 10
events.mechanism = epoll

My PHP.ini for the PHP-FPM Master Process. Everything not listed here, is either default php settings or should not be relevant:

memory_limit = 400M
upload_max_filesize = 20M
post_max_size = 20M
max_execution_time = 600
max_input_time  = -1
max_input_vars = 10000
suhosin.get.max_vars = 10000
suhosin.post.max_vars = 10000

[Zend]
zend_extension=/opt/php-5.6.11/lib/php/extensions/no-debug-non-zts-20131226/ioncube.so
zend_extension=opcache.so
opcache.revalidate_freq=0
;opcache.validate_timestamps=0
opcache.max_accelerated_files=50000
opcache.memory_consumption=256
opcache.interned_strings_buffer=16
opcache.fast_shutdown=1

[APC]
extension=apcu.so
apc.enabled=1
apc.shm_segments = 1
apc.shm_size=256M
apc.ttl=7200
apc.user_ttl=7200
apc.gc_ttl=3600
apc.stat=1
apc.enable_cli=0
apc.file_update_protection=2
apc.max_file_size=2M
apc.include_once_override=0
apc.mmap_file_mask=/tmp/apc.XXXXXX
apc.cache_by_default=1
apc.use_request_time=1
apc.slam_defense=0
apc.stat_ctime=0
apc.canonicalize=1
apc.write_lock=1
apc.report_autofilter=0
apc.rfc1867=0
apc.rfc1867_prefix =upload_
apc.rfc1867_name=APC_UPLOAD_PROGRESS
apc.rfc1867_freq=0
apc.rfc1867_ttl=3600
apc.lazy_classes=0
apc.lazy_functions=0

extension=memcache.so
extension=memcached.so

Note: Memcached has 1 GB of memory allocated to it.

Apache Error Log

The actual error message from the apache error.log. The error message if always the same. (I have enabled verbose proxy logging):

[proxy:debug] [pid 141760:tid 140526898214656] mod_proxy.c(1159): [client myclient] AH01143: Running scheme unix handler (attempt 0), referer: http://mywebsite/website/typo3/install/index.php?TYPO3_INSTALL[type]=cleanup
[proxy_fcgi:debug] [pid 141760:tid 140526898214656] mod_proxy_fcgi.c(879): [client myclient] AH01076: url: fcgi://mywebsite//var/www/html/mywebsite/htdocs/typo3site/website/index.php proxyname: (null) proxyport: 0, referer: http://mywebsite/website/typo3/install/index.php?TYPO3_INSTALL[type]=cleanup
[proxy_fcgi:debug] [pid 141760:tid 140526898214656] mod_proxy_fcgi.c(886): [client myclient] AH01078: serving URL fcgi://mywebsite//var/www/html/mywebsite/htdocs/typo3site/website/index.php, referer: http://mywebsite/website/typo3/install/index.php?TYPO3_INSTALL[type]=cleanup
[proxy:debug] [pid 141760:tid 140526898214656] proxy_util.c(2147): AH00942: FCGI: has acquired connection for (mywebsite)
[proxy:debug] [pid 141760:tid 140526898214656] proxy_util.c(2200): [client myclient] AH00944: connecting fcgi://mywebsite//var/www/html/mywebsite/htdocs/typo3site/website/index.php to mywebsite:8000, referer: http://mywebsite/website/typo3/install/index.php?TYPO3_INSTALL[type]=cleanup
[proxy:debug] [pid 141760:tid 140526898214656] proxy_util.c(2237): [client myclient] AH02545: fcgi: has determined UDS as /opt/php-5.6.11/var/run/php5-fpm-mywebsite.sock, referer: http://mywebsite/website/typo3/install/index.php?TYPO3_INSTALL[type]=cleanup
[proxy:debug] [pid 141760:tid 140526898214656] proxy_util.c(2409): [client myclient] AH00947: connected //var/www/html/mywebsite/htdocs/typo3site/website/index.php to httpd-UDS:0, referer: http://mywebsite/website/typo3/install/index.php?TYPO3_INSTALL[type]=cleanup
[proxy_fcgi:error] [pid 141760:tid 140526898214656] [client myclient] AH01067: Failed to read FastCGI header, referer: http://mywebsite/website/typo3/install/index.php?TYPO3_INSTALL[type]=cleanup
[proxy_fcgi:error] [pid 141760:tid 140526898214656] (104)Connection reset by peer: [client myclient] AH01075: Error dispatching request to : , referer: http://mywebsite/website/typo3/install/index.php?TYPO3_INSTALL[type]=cleanup
[proxy:debug] [pid 141760:tid 140526898214656] proxy_util.c(2162): AH00943: FCGI: has released connection for (mywebsite)

Now my question:

How can I fix recurring PHP 503 Errors on my Webserver?

My thoughts:

• Maybe mod_proxy_fcgi in UDS mode. But isn't it bad to deactive UDS, performance wise? Can I tweak anything there?
• mod_proxy or mod_proxy_fcgi not working correctly with php-fpm or not set up properly?
• APC or ZendOPCache or Memcached messing things up? I wouldn't say its the Memory allocated to them, because there is hardly anything going on on the server and there is plenty of free memory
• Some Problems with the php.ini config?
• Some Problems with the php-fpm config or php-fpm pool config?

I'm not an export on those things, so I'm having a hard time figuring things out. Apache with php-fpm doesn't seem that common either, most google results are nginx based, which doesn't help me much.

Maybe someone here can help me?

Thanks alot!!

Answer 1

Remove the enablereuse=on option on your Proxy line, so that it reads

<Proxy fcgi://mywebsite/ retry=0>

That solved the " Error dispatching request to : , referer... " problem for me.

Answer 2

I had the same exact issue and after trying to understand what was the cause I found that for us the reason was a buggy plugin.

In particular we fixed the issue by disabling this wordpress plugin: https://wordpress.org/plugins/custom-css-js/

We were using php 7 and wordpress 4.5.6.

So for us was not due to a misconfiguration php, apache or any cache system. Neither the problem was a no resource available (CPU/RAM) issue. The problem was due to a buggy plugin.

Was very hard to find which plugins caused the issue. We understood thanks to php debugging options, so I suggest to add these lines below into wp-config.php to debug:

define('WP_DEBUG', true);
define('WP_DEBUG_LOG', true);
define('WP_DEBUG_DISPLAY', false);
@ini_set('display_errors',0);

Best luck!

Answer 3

As the other posts suggested, the enablereuse=on should not be used when PHP-FPM daemon is running via a Unix socket.

The Apache documentation for mod_proxy_fcgi confirms this by saying,

UDS does not currently support connection reuse

UDS means Unix Domain Sockets, which means when you're running PHP-FPM as a socket (.sock), instead of the default TCP port method.

Nathan Zachary goes into great detail about configuring Apache + PHP-FPM as a socket, among other related topics.

I experienced crazy, unpredictable Apache responses immediately after doing a POST with enablereuse=on and UDS. As soon as I removed enablereuse=on and restarted the services the issue went away entirely.

More on symptoms and debugging this issue from the mod_h2 folks.

Answer 4

Is listen = var/run/php5-fpm-mywebsite.sock correct ?

As for virtual host case, you can try these configuration :

• Apache ( in VirtualHost tag )

  ProxyPassMatch ^/(.*\.php(/.*)?)$ fcgi://127.0.0.1:9000/DOCUMENT_ROOT_OF_VHOST/$1
  

• PHP-FPM pool

  listen = 127.0.0.1:9000
  

If you have no concerns about .htaccess , nginx is recommended.

Answer 5

I had an issue similar to this and fixed it by moving the folders out of /var/lib/php/opcache/ and then restarting php-fpm.

Possibly Relevant: A user was spamming a PHP URL last night, causing these errors to repeat in our Apache log.

[pid 925] worker.c(1613): AH00288: scoreboard is full, not at MaxRequestWorkers

I am guessing that it caused some cache corruption in a folder located in /var/lib/php/opcache.

For some reason, resetting opcache via the opcache control panel, nor restarting apache/php-fpm, seems to clear those folders out. You have to do it manually.

But that seems to have worked for me.

Answer 6

We were hitting the default 60 seconds timeout. I added the line ProxyTimeout 300 to the .conf file within the directory /etc/apache2/sites-enabled to make Apache2 wait longer for PHP.

Check how fast you get the HTTP 503 error. If it is within a few seconds, this answer is not for you.

You can find the documentation here: https://httpd.apache.org/docs/2.4/mod/mod_proxy.html#proxytimeout