- Role discovery sucks right now — broham = weak, too much setup, ick
- Make broham a webservice in front of pluggable backend?
- Get people’s blessing that cluster_service_discovery is a good impl.
- Documentation: moar plz.
- Videocast: a visual walkthrough would help adoption
- Bootstrap: bless a minimal bootstrap method
- Should we use packages or scripts?
- ChefHadoopNfsMaster: smoother instantiation of a standalone (starter) cluster.
- chef server walkthrough could use work
- EBS volume creation: create, attach, format, populate databag
- Make sure
- Local cluster: separate out cloud code to work with local code
- Rackspace & other clouds: Abstract out EC2-specific codes
- JClouds
- Poolparty
- The scripts are insufficently DRY — bring some of this into poolparty trunk?
- Complain if the validation.pem (etc) is missing
- bugs:
- this thing with the [defunct] processes
- Tuning for generic node — is EC2-specific tuning necessary?
- Admin tasks
- Make namenode formatting safe
- recipe to do namenode metadata backup?
- EBS attachment = still shaky
- Secondary Namenode on non-master: Right now it’s really hard to do something like ‘make one slave also be the 2NN’. This should be easy
- Patched version: this is still hard
- If I set a hostname (‘gibbon.infochimps.com’) on a
- runit script flapping
- Need scripts
- cluster discovery integration?
- runit script flapping
- Newer point version
- cassandra shouldn’t start before the /etc/cassandra is present
- Tuning for cluster size
- Need scripts
- cluster discovery
- Monitoring, etc:
- Scribe
- Ganglia
- Elephant Friends
- Hive
- Mahout
- Dynamic DNS
- Whirr: integrate &or play nice with Whirr
- Opscode cookbooks acceptance: Clean up and document, metadata.rb, etc so that we can put the cookbooks in for chef adoption.
- Performance qualification: run some chimpmark tasks across clusters and report on completion times