Config validation: Crash (segfault) when running in Docker about roadrunner HOT 11 CLOSED

Ok, I managed to recreate this issue by corrupting the config file. Can you make sure that http.workers.pool section presented? Can you post your config file sample here?

I have to add more validations for the config to track scenarios when config sections are omitted.

from roadrunner.

Ah, that makes sense. I was providing the minimal config that didn't give a validation error, assuming it would fill in default values as necessary. Here's what I had:

rpc:
  enable: false
http:
  enable: true
  address: 0.0.0.0:8000
  workers:
    command: php rr-entrypoint.php
    relay: pipes
  uploads:
    forbid: []
static:
  enable: true
  dir: public
  forbid:
    - .php

from roadrunner.

That explains :) We have slipped couple validations but I will keep this ticket open until your config given a proper error. We do not have default values for pool options, they always are custom.

Try:

rpc:
  enable: false
http:
  enable: true
  address: 0.0.0.0:8000
  workers:
    command: php rr-entrypoint.php
    relay: pipes
    pool:
       numWorkers: 4
       allocateTimeout: 600000000
       destroyTimeout:  600000000  
  uploads:
      forbid: []
static:
  enable: true
  dir: public
  forbid:
    - .php

from roadrunner.

You can also write configs in json btw.

from roadrunner.

Will be included into next release #13

from roadrunner.

Adding the http.workers.pool section appears to have fixed the crash-on-startup, but an immediately-dying worker (I made a typo and did require_once 'vendor/autoloaad.php';) just put this in the output, which is not terribly useful for debugging:

rr_1            | time="2018-06-19T19:15:56Z" level=error msg="[http]: unable to connect to worker: unexpected response, header is missing: signal: killed"

I was only able to figure out the actual source of the issue by trying to run the worker command myself in a shell; at no point was I able to actually find the PHP error itself when served by rr. Though to be fair, that could be due to bad php.ini settings - I haven't poked at it a whole lot.

Having additional documentation on the config would be super valuable here. For example, I have absolutely no idea how the RPC section works or what it's supposed to accomplish.

from roadrunner.

That's weird, this has been properly tested:

I would have to take a look at this docker container and build our own version later.

from roadrunner.

@Firehed would you mind creating a ticket for that?

Having additional documentation on the config would be super valuable here. For example, I have absolutely no idea how the RPC section works or what it's supposed to accomplish.

from roadrunner.

Will do!

What are your log settings if you run php -i? The base image I started off with has fairly dumb defaults, which could be part of the issue.

If it's relevant, my application's internal logger goes to STDOUT (which is what Docker generally wants), and the front-controller script I'm trying rr with looks like this:

// autoload, framework setup, etc...
$relay = new Spiral\Goridge\StreamRelay(STDIN, STDOUT);
$psr7 = new Spiral\RoadRunner\PSR7Client(new Spiral\RoadRunner\Worker($relay));

while ($req = $psr7->acceptRequest()) {
    try {
        $response = $dispatcher->setRequest($req)->dispatch();
        $psr7->respond($response);
    } catch (\Throwable $e) {
        $psr7->getWorker()->error((string)$e);
    }
}

from roadrunner.

Try setting ini_set('display_errors', 'stderr'); in your script, rr will handle error buffer in this case and forward it to parent docker container with proper error management before that (like displaying in the log).

In some cases (fatal errors on bootload stage) php dies even before stderr can be set, running script manually php script.php actually very solid solution to check if worker bootloads (script must be awaiting for stdin input).

from roadrunner.

Config validation issue has been resolved in 1.0.2

from roadrunner.