corosync/TODO

Application Interface Specification TODO list

Generic Items
-------------
* EVT, DLOCK, MSG APIs functionality need to be developed.
* Error checking on parameters could use improvement.
* Allow AIS Executive to configure cluster name.
* Compliance testing of return values would be helpful.
* Support B.01.01 version of spec (currently support A.01.01 version).
* Consider implementing SOCK_SEQPACKET for the AF_UNIX family of sockets on
  Linux.  This would save an extra system call every time an operation must
  be done from the API.
* There are lots of TODO's in the code that need attention.

Group Messaging Interface
-------------------------
* Very important: implement full EVS semantics when holes occur in
  delivery messages after a configuration change but before the new
  configuration is delivered.
* Very important: block new messages from being multicast until recovery
  of each service has completed after a configuration change.  This could be 
  done with a "plug" in the token which "stops" any GMI_PRIO_MED or GMI_PRIO_LOW
  messages from being multicast until all members of the configuration have 
  unplugged the token.  Then queued messages in MED or LOW priority can be
  sent, ensuring correct partition operation.
* Implement error creation config file to test GMI since all my lossy
  hardware has been fixed.
* Add secrecy/authentication to group messaging interface.
* Add support for multiple rings with gateway ring to ring for added scalability
  in LANs.
* Add support for multiple rings with gateway tuned to long haul networks for
  added scalability in WANs.  Look at spread.org as a design.
* Add support for low delivery-time delay FIFO messages.
* Add support for SAFE ordering.
* Add support for encryption/authentication using Helix.  nonce will start
  at zero and increment for every message sent or rotation on the ring.  Group
  key produced using group key generation protocol.
  
Cluster Membership
------------------
* Make timeout on SaClmClusterNodeGet work.  Currently the timeout is 5
  seconds, but the spec requires the timeout to be specified in the API call.

Availability Management Framework
---------------------------------
* Very Important: Implement configuration change support.  This includes partitions.
* Currently the executive can record and manage only one component service
  instance per component.  As a result, one component cannot act as standby/active
  for two other components in the system (AIS Spec page 74 Figure 16. Example of
  n+1 redundancy model).
  - Fix to follow spec.
* If a user of the AMF library doesn't respond with saAmfResponse, the state
  of the application will never change.  Fix by adding timeouts to readiness state
  and ha state changes to force saAmfResponse state changes triggered.
* Implement resource proxy functions. (currently dummy functions)
* Implement pending operations. (currently dummy functions)
* Implement NWAY and NWAYACTIVE redundancy models.

Checkpointing
-------------
* Very Important: Implement configuration change support.  This includes partitions.
* Implement expiration times on checkpoints.