Requirements

• CfnCluster - launches cluster, get from pip
• awscli - for uploading files to S3
  • Comes as part of CfnCluster from pip
• AWS Credentials
• jq - for json parsing in connect.sh

Initial Setup

1. Put your aws credentials in your ~/.aws directory and the CfnCluster config file (./config)
2. Create and download your private key from AWS
3. Create new IAM role:
   1. Attach policies in /policies
   2. Copy policies from http://cfncluster.readthedocs.io/en/latest/iam.html
4. Change the name of the private key from CfnClusterKey to <your_key> in ./config
5. Move mysqlauth.sh.template to mysqlauth.sh and fill in database info
6. Move buckets.sh.template to buckets.sh and fill in s3 info
7. symlink mysqlauth.sh and buckets.sh to /backend/shared

Usage

1. Once set up, just run launch.sh
2. After any changes to the backend run sync.sh

sync.sh

Uploads all off the files in ./backend, ~/.aws/credentials to s3. Uses the sync functionality for incremental changes, so any moved files will also be moved in S3 as opposed to duplicated.

query_db.sh

Usage: ./query_db.sh "<query>" "<args>"

• <args> is passed to mysql
• A universal helper script which sends a query to the database
• Database is determined from mysqlauth.sh

mysqlauth.sh

• File containing database information
• All values must be in quotes to ensure proper parsing in memdata.py (in ./backend/shared)

buckets.sh

• File containing names of s3 buckets
• backend - bucket where files in /backend are stored
• builds - bucket where builds of openmpi are stored (for purposes of reproducing errors)
• logs - bucket where log files from /var/log and $HOME on the cluster are stored (debugging)
• All values must be in quotes to ensure proper parsing in memdata.py (in ./backend/shared)

config:

CfnCluster configuration file. Look over the documentation, it is extremely helpful: http://cfncluster.readthedocs.io/en/latest/

postinstallargs.cfg

• File containing a list of tests to run
• Each line is appended to the previously mentioned CfnCluster config file when a cluster is launched
• To add a new test copy a previous line and follow the format outlined in the header:
  • The source can either be git or a url for a tarball
  • Cores on Head and Cores on Compute are the number of cores on the head node and compute nodes respectively
  • Number of compute is the number of compute nodes to start
  • Test name is the location of the test to be run on the cluster

launch.sh

Used to launch a cluster using the information found in postinstallargs.cfg

Usage: ./launch [cluster-name-base]
If the base name is supplied, each cluster will be named "-", where is the number of the cluster being launched. This number will start at zero and increment by one with each line in postinstallargs.cfg.

connect.sh

Used to connect via ssh to the head node of a selected cluster Usage: ./connect.sh <path-to-keyfile>

Subdirectories

• backend: all of the files put on the cluster are contained here tests to be run are in backend/tests
• data: Contains all graphs and scripts to generate them

Possible Improvements

• Have launch.sh detect rollbacks and retry launch. This would require either
  1. launching without --nowait
  2. having a process that watches the cluster and restarts it if it rolls back This can also enable moving shutdown.sh (backend/shared/shutdown.sh) to outside of the cluster