Architectural Overview for OpenStack Compute
Live Notes may be taken for this topic at: http://etherpad.openstack.org/Architecture and http://etherpad.openstack.org/nova-archdoc
“Small” components, loosely coupled
- Queue based (currently AMQP/RabbitMQ)
- Flexible schema for datastore (currently Redis)
- LDAP (allows for integration with MS Active Directory via translucent proxy)
Workers & Web hooks (be of the web)
- Asynchronous everything (don't block)
- Components (queue, datastore, http endpoints, ...) should scale independently and allow visibility into internal state (for the pretty charts/operations)
Development goals
Testing & Continuous Integration
- Fakes (allows development on a laptop)
- Adaptable (goal is to make integration with existing resources at organization easier)
Queue
Each worker/agent listens on a general topic, and a subtopic for that node. Example would be "compute" & "compute:hostname"
Messages in the queue are currently Topic, Method, Arguments - which maps to a method in the python class for the worker
- exposed via method calls
rpc.cast to broadcast the message and not wait for a response
rpc.call to send a message and wait for the response
Datastore
- Pre-Austin, data is stored in Redis 2.0 (RC)
- Do the work on write - make reads FAST
- maintain indexes / lists of common subsets
- use pools (SETs in redis) that are drained for IPs instead of tracking what is allocated
Delta
- Scheduler does not exist (instances are distributed via the queue to the first worker that consumes the message)
- Object store in Nova is a naive stub which would be replaced with Cloud Files in Production (a simple object store that mimics Cloud Files might be good for development)
- Tornado should be phased out for WSGI-based web framework
Networking
Currently, there are three strategies for networking, implemented by different managers:
FlatManager -- ip addresses are grabbed from a network and injected into the image on launch. All instances are attached to the same manually configured bridge.
- FlatDHCPManager -- ip addresses are grabbed from a network, and a single bridge is created for all instances. A dhcp server is started to pass out addresses
VlanManager -- each project gets its own vlan, bridge and network. A dhcpserver is started for each vlan, and all instances are bridged into that vlan.
The implementation of creating bridges, vlans, dhcpservers, and firewall rules is done by the driver linux_net. This layer of abstraction is so that we can at some point support configuring hardware switches etc. using the same managers.
For more discussion of network architecture, see Networking.