503 Errors

alexgreen

Hi, I've been randomly receiving 503 Service Unavailable errors from nginx, when accessing something via php. I don't have anything custom on the php service, and the requests are fast (not a long running process that might time out). Regular text files/etc are fine, so it seems to be coming from php. I do have 6 different php versions active across 2 stacks, and quite a few sites on the apache service of each.

Indigo

That's an odd one. Next time it happens, could you attempt to access the site in question using the port number specified in its Apache site config? That bypasses the nginx reverse proxy, and should help determine whether it's the rp, or the Apache origin site at fault.

Unfortunately I fear this will be difficult for me to reproduce. Any chance you can narrow it down? eg does the 503 ever happen with just one of the two stacks active? Are you accessing the sites in an unusual way (eg from a script vs from a browser) or anything else you can think of? Is your Mac running low on RAM at the time or anything in that vein?

I have many stacks running simultaneously all day long with almost all the PHPs spread amongst them and it all works fine for me — not sure if that's of any comfort though 😆

alexgreen

Thanks for your quick reply - I'm on an M3 Pro Macbook with plenty of ram, and I've tried with both stacks but same issue. It's really random - sometimes I can develop for hours without an issue, other times every request, or every second request gets a 503. This could be to a simple php file that just outputs phpinfo() or to api endpoints of an app or index.php files.

I've been able to replicate the error via the apache port as well.

alexgreen

Also I get the error when accessed via a browser or curl. At the moment it's usually every second request. I don't have any other requests happening at the same time.

Indigo

Hmm, your last statement is certainly valuable to know. We can rule out an nginx rp config, which is good.

Sorry to hear it's happening so frequently. That must be quite frustrating.

Are you able to use Nginx instead of Apache and if so, see if it has the same issue? I'm assuming likely not due to .htaccess requirements or something, but worth asking.

Since it's happening a lot to you, and you've seen it even with pages just doing a simple phpinfo(), I'm wondering if you might be able to replicate it without your specific projects in the mix. Would you have time to create a new empty stack containing just an apache server and a bunch of PHPs and see if you can replicate it with a stack you could share with me?

alexgreen

Hi, I'm getting the same issue when run through nginx instead of apache, except instead of a 503 I get a 502. This is on a new stack for just the test.

I'm not actually getting the issue with simple phpinfo() files (probably wasn't before, not sure now).

I'm thinking it's not actually an issue with IndigoStack, but the Mac.

I've done more testing and I've found that the error is caused either by connecting to postgres or gettext. Accessing a php file with just echo gettext("hello"); will cause the error every time, postgres just sometimes. I found this: https://github.com/Homebrew/homebrew-core/issues/137431

I tried the suggestions of adding:

env['PGGSSENCMODE'] = disable
env[OBJC_DISABLE_INITIALIZE_FORK_SAFETY] = YES
env[LC_ALL] = C

To etc/php-fpm.d/www.conf, to no avail.

Indigo

Hey, thanks for your patience waiting for me to get back to this. To be honest I was kinda busy and saw "I'm thinking it's not actually an issue with IndigoStack", sighed a huge sigh of relief and thought "I'll come back to that later" haha

Thanks also for such deep research on it, and for providing such a foolproof replication of the issue. Turns out you were very close indeed with your attempted fix. I don't use Postgres (yet) but the following fixed the gettext error for me (ref this discussion):

env[LC_ALL]=en_US.UTF-8
env[LC]=en_US.UTF-8
env[LANG]=en_US.UTF-8
env[LC_MESSAGES]=en_US.UTF-8
env[PGGSSENCMODE]=disable

I put that at the end of an etc/php-fpm.d/www.conf override (I think you've mastered Indigo overrides, but let me know if not), and the gettext error went away immediately. I've added the env[PGGSSENCMODE]=disable in the hopes it will fix the Postgres side of the issue, but as mentioned I can't vouch for that.

If you have time, I'd really value your input on whether this fixes the issue for you. If it does, I'll roll it into the next Indigo release.

Note, configuring en_US.UTF-8 is a bit of a hack, as not everybody wants en_US nor UTF-8, so I am unsure how universally acceptable it is to set them; I've done a lot of i18n over the years but the frameworks I've used for it (Drupal & Laravel) handle it all at a higher level than this php-fpm configuration. Do you think en_US.UTF-8 will be ok to be rolled into Indigo's config?

alexgreen

That fixed it! For both gettext as well as for postgres. Thanks so much, I'm happy to be using Indigo again and not having to worry about docker. And yes I think en_US.UTF-8 will be fine as a good default that could be overriden if needed.

Indigo

Thanks for your confirmation, much appreciated. And very glad it fixed it, that was a very strange error indeed.

I've just added this config into all the PHP builds, which should show up in the next Indigo release.

mrpercival

I am having the same issue, but im not sure where I find the file you are referring to, could you help me out a bit?

mrpercival

I found a www conf in the .indigo directory, but any time I update it, it gets over written the next time the server loads

mrpercival

OK never mind, I figured it out,

Indigo

Hey @mrpercival — glad you worked it out! I think I need to release the next version of Indigo already! 🙂