|Version 2 (modified by simonmar, 5 years ago) (diff)|
Using Amazon EC2 for building GHC
- you are cycle-challenged. e.g. I (simonmar) use this when I'm working at home on my laptop, so I can use my laptop for development while having a validate going on EC2 in the background.
- you want to validate on a different platform
Right now it's suitable for doing GHC validate on x86/Linux, but we could expand this in the future to include x86_64, more flavours of Linux, and OpenSolaris, all of which will be useful for testing.
It takes a bit of setting up, but once set up you can:
- issue one command to start up an EC2 machine ready to build GHC. A script is provided for this below; it usually takes a couple of minutes to run.
- issue one command to push patches to your machine.
- issue one command to validate.
- issue another command to shut it down (don't forget this, you're paying by the hour!)
The way I have this set up, EC2 costs about $0.06 per hour (I use "spot instances" to make it cheaper). I use a "Medium High-CPU" instance, which is equivalent to a fast dual-core machine; the ordinary bog-standard instance took 1 hour 40 mins to validate when I tried it, whereas the medium high-CPU instance completes validate in under 30 minutes.
Getting set up
Set up access to EC2 using the Amazon instructions: get as far as being able to fire up an instance and SSH to it.
Using the ElasticFox Firefox plugin is a good way to keep an eye on things from Firefox, it's a more responsive alternative to using the web-based management tool.
You'll need to install the Amazon EC2 command-line tools, which require Java to be installed (unfortunately). Install the up-to-date versions from Amazon, not the ones from your Linux distro (even the ones in Ubuntu 9.10 are too outdated). Follow the instructions, and check that things are working by running ec2-describe-instances.
Add to your .bash_profile, or equivalent (replacing keys as appropriate):
export EC2_PRIVATE_KEY=~/.ec2/<your-private-key-file> export EC2_CERT=~/.ec2/<your-certificate-file> # export EC2_URL=https://ec2.eu-west-1.amazonaws.com # export EC2_URL=https://ec2.us-west-1.amazonaws.com export EC2_URL=https://ec2.us-east-1.amazonaws.com
Now, take the attached scripts attachment:start-ec2-instance, attachment:validate, and attachment:prep_instance and save them in your home directory in the same place you saved your key-pair file. Edit start-ec2-instance to point to your key-pair file.
In your ~/.ssh/config file, add this:
# This entry is automatically tweaked by start-ec2-instance Host ec2 Hostname ec2-XXX.compute-1.amazonaws.com User ubuntu IdentityFile XXX.pem
the idea behind this is that start-ec2-instance is going to fix this entry to point to your EC2 instance when it starts up, so that no matter what hostname you are given, you can connect to it with ssh ec2.
Notice that the start-ec2-instance script is configured to use a particular AMI: ami-916c81f8. This is an EBS-backed image with 32-bit Ubuntu Karmic, and tools needed to build GHC installed. The ubuntu user has a full set of GHC HEAD repositories, and it mounts an ephemeral disk to use for builds.
This AMI is in the US-East region for now, because that's where I found the Ubuntu EBS-backed AMI that I used as a base. Ubuntu plan to make EBS-backed images of future releases, so this should get easier.
Why EBS-backed? This means the image is backed by a persistent disk, so I can update it easily, e.g. to bring the repos up to date, or add more tools.
To use EC2, just run start-ec2-instance, wait until it finishes, and it will then tell you that your instance is available and what commands you can run. I normally use GNU screen to manage my session, so the script will copy up the file ~/.screenrc if it exists.
The script is set up to create a spot instance request for a c1.medium instance, and then wait for it to be fulfilled. This is cheaper than just creating an ordinary instance - the price varies depending on demand, but it seems to hover around $0.06 for c1.medium at the moment, and the script uses a maximum of $0.08; this is less than half the price of an on-demand (instead of spot) instance, but the script is easy to modify to use an on-demand instance if you want.
Don't forget to "ssh ec2 sudo halt" when you've finished.
- validate should contact you when it finishes, One way I'd like to do this is to use twidge to send a tweet, which could then be forwarded via SMS to my phone.
- We should make more AMIs available (x86_64, different flavours of Linux, OpenSolaris).
- Make it more suitable for development: e.g. add instructions for using a persistent EBS volume for storage.