Docs for this project are now generated in Sphinx, and will shortly be hosted on the project page.
an amazon/eucalyptus/rackspace cloud clone in python, amqp, tornado, ...
[X] generate cloud cert on run if it doesn't already exist (used for bundling)
[X] api calls are validated and ran as a user
[X] X509 certs for users
[x] add concept of admin
[X] Deliver creds to user (novarc, x509, ...)
[X] users exist and have ec2 credentials
[X] user can create and destroy keypairs
[x] euca-run-instances requests nodes to spawn instances
[ ] can only launch images that user owns or is public
[x] keypair is added when running instance
[ ] nodes have a quota on # of instances
[X] can allocate and attach public IPs
[X] network state persists
[x] each user gets a subnet and vlan for their instances
[x] node downloads proper image from S3 (verify image via content-md5)
[x] instances can access their user-data, meta-data
[X] hard code all instances for a user into "default" security group: deny except 22, 80, 443, and 1194
[x] instances state from multiple nodes overwrite each other (update_state code)
[x] euca-get-console-output works
[x] euca-terminate-instances works
[X] euca-run-instances uses userdata, instance size, image, keypair, ... (all api params)
[X] euca-describe-instances works
[x] euca-describe-instances only returns instances I have permissions to
[x] can launch from many different images
[x] ignore kernel/ramdisk from user, hardcode for now
[x] can launch different sizes
[ ] implement REBOOT in the cloud controller
[ ] add more local IPs to cloudpipe
[ ] unblock ping for own subnet
[X] NAT to public internet works from instances
[ ] access to other users instances only works on "default" protocols
[X] terminate should send to only the approriate node
[X] BUG: running -n N+1 instances when you have N results in only N instances launched
- seems to be an issue with multiprocess.Process
the _launch call doesn't occur when two Processes are running at the same time
INFO:root:Done create image for: i-286573
DEBUG:root:Arrived in _launch, thanks to callback on deferred. <- only happens first time
[X] BUG: launching multiple instances show the incorrect IP in describe-instance during while pending
[x] describe-instances doesn't show public ips
[ ] When instances are shutdown or terminated, clean them up (detach IP and volume)
[x] euca-upload-bundle: buckets have owners and are private (only accessible by owner and admin)
[x] euca-register: registration works and decrypts image with cloud's cert
[x] euca-describe-images: returns only images that user has access to (public or theirs)
[x] images are owned by user, and private by default
[X] can create volumes
[X] can destroy volumes
[X] can attach volumes
[ ] detach on destroy?
[X] s3server's register decryption should be done in bkg:
[ ] build debs - perhaps use git-buildpackage?
[ ] remove eucalyptus specific terminology in favor of amazon (emi -> ami, ?)
[ ] documentation/SOPs for backup, updating, ?
[ ] add license headers - apache license
[ ] rewrite code such as partition2disk that is too close to eucalyptus
[ ] review code for internal (nasa) info
[ ] init.d scripts & location for configuration files
[ ] Logging clean-up: system should (default?) to using syslog
[ ] verify user is allowed to execute commands - for each API method!
[ ] when instances are terminated, IP addresses are reclaimed
[ ] node/node_worker is bad name for instances node as storage is a node too
[ ] describe instances should be returned ordered by ami_launch_index
[ ] multiprocess the cloud for x509 generation
[ ] bin/users.py uses command line flags
[ ] more space for instance ids
[ ] get Dean to update switch configuration
[ ] BUG: cloud using boto to communicate with OSS means that if OSS throws a
500 in response to the error it will lock up the cloud by doing time.sleep
and retrying for about a minute (5 times, the delay doubling each time)
[X] Port existing users, images from euca
[X] Port cloudpipe and dashboard, install them
[ ] Remove secgroups from dashboard UI
[ ] Configure instance sizes
[X] dd warns blocksize of 512 is slow in partition2disk.convert
[ ] tiny (5x overcommitted CPU) shouldn't be on same nodes as regular sizes
[ ] lvm instead of file based disk images?
[ ] decrypt in python is slow!
[ ] users have quotas
[ ] fix fingerprint generation on creation of keypair
[ ] api to modify private/public (image attributes) works
[ ] proper security groups
[ ] projects / groups
[ ] RBAC - roles based control
[ ] on cloud launch, it should broadcast to nodes to report their current state
[ ] throttling for reporting state from node/storage/...
(report back at least every minute, at most once a second, only when things change)
[ ] support for ephemeral and swap on disk image generation
[ ] windows support
[ ] out-of-band console - example for XEN: http://nostarch.com/xen.htm and
http://book.xen.prgmr.com/mediawiki/index.php/Prgmr_menu