Configuring Phusion Passenger on a 256MB Ubuntu Slice @Slicehost

I installed Phusion Passenger a few weeks ago to get a production environment up and running for my Rails app. It was super easy to install and felt like magic. I checked off the item on my TODO and moved on. After all it didn’t say “YOUR APP WILL BE SLOW AS HELL IF YOU DON’T READ THIS MANUAL!!!” in their documentation. Well…it was too slow…and here’s how I fixed it.

Referencing the Phusion Passenger User Guide I was able to make changes that made my app run about 10 times faster. If you want to do this in 15 minutes, read this article. If you have a few hours, read through the user guide and experiment. My hardware setup is a 256MB Ubuntu VPS at my favorite hosting company – Slicehost.

The following are global settings I added to my httpd.conf:

RailsSpawnMethod smart
PassengerUseGlobalQueue on
PassengerMaxPoolSize 2
PassengerPoolIdleTime 0
PassengerMaxRequests 1000
RailsAutoDetect off
RailsAppSpawnerIdleTime 0
PassengerStatThrottleRate 5

RailsSpawnMethod smart/smart-lv2/conservative
This can speed up the spawn time tremendously (by as much as 90%) if your app is compatible. However, some Ruby on Rails applications and libraries are not compatible with smart spawning. Set your configuration to smart spawning and if your app works, then you are all good. If not, use another spawning method.

PassengerUseGlobalQueue ON/OFF
Turning this option ON tells Passenger to add requests to a global queue when your Rails instances are busy. As soon as an instance becomes available it will grab a request from the queue. If the option is OFF, each process will maintain its own queue. Localized queues can hurt you if an instance is processing a 60-second request and Passenger adds a 1-second request to the instance’s queue. ( Imagine waiting in a 10 person non-express line at Whole Foods when all you want is a bottle of water ). Therefore, it only makes sense to turn this OFF for the 5% performance gain if all of your requests take about the same amount of time.

PassengerMaxPoolSize INTEGER
This simply says how many application instances Passenger is aloud to Spawn. The default is 6 but the user guide recommends 2 for a 256MB slice so that’s what I went with. More could consume a lot of memory and crash my server. You can use 30 if you have 2GB of RAM.

***PassengerPoolIdleTime INTEGER
This is the minimum time an application instance can be idle without Passenger shutting it down. The default value was 300 which means if my app received no traffic for 300 seconds, it could shut down all of my app instances. The result? A visitor would have to wait way too long when viewing my website. I triple starred this because I think this setting alone will greatly improve your app speed. Note there is an obvious memory trade-off because your application instances will always be running.

PassengerMaxRequests INTEGER
This provides a safety net for memory leaks. It tells Passenger to restart an application instance after it receives a certain number of requests. My app is young so I am keeping it at 1000 requests but ideally you should set this to zero meaning no maximum.

RailsAutoDetect ON/OFF
Turn off auto-detection and tell Passenger exactly which directory contains your Rails app. Do this by adding the following line to your VirtualHost configuration block:

<VirtualHost>
    ...
    RailsBaseURI /
    ...
</VirtualHost>

RailsAppSpawnerIdleTime INTEGER
This is a similar setting to PassengerPoolIdleTime but applied to the ApplicationSpawner server. If the ApplicationSpawner server hasn’t done anything for the specified amount of time, it will automatically shut down. I set this at zero so it will never shut down and therefore take 10% of the time to spawn a Rails process as it would with the ApplicationSpawner server shut down. I would imagine this becomes increasingly important as your PassengerMaxPoolSize increases. Note that if your application is at it’s maximum number of instances allocated (PassengerMaxPoolSize), you may want the Spawner to shut down immediately after it spawns an app instance since the Spawner won’t be used much. Set the value to 1 if this is the case. I will probably do this in the future when my website has a more constant load ( another option is to write a script to periodically hit your Rails app and keep it at a constant load ).

PassengerStatThrottleRate INTEGER
Tell Passenger to only check for restart.txt up to once every 5 seconds instead of once per processed request.

Finally, try to turn on PassengerHighPerformance:

<VirtualHost>
    ...
    PassengerHighPerformance on
    ...
</VirtualHost>

Adding that line to your VirtualHost configuration block will improve your performance but make your Rails app incompatible with some Apache modules. If you turn that on, restart apache, and your Rails app works fine, chances are you are in the clear. Just keep this in mind for the future in case you add more Apache modules.

If you enjoyed this post, make sure you subscribe to my RSS feed!