Add support to -LEAVING error in GETJOB and other commands

[mariano]

The GETJOB command, or other commands, may return a -LEAVING error instead of blocking. This error should be considered by the client library as a request to connect to a different node, since the node it is connected to is not able to serve the request since it is leaving the cluster. Nodes in this state have a very high priority number published via HELLO, so will be unlikely to be picked at the next connection attempt. See #25 regarding the latter

[Revisor]

I think this should be a responsibility of the Manager, and I would do it like this:

1. Execute a command
2. If any command returns the -LEAVING error, switch to a better node (switchNodeIfNeeded())
3. Execute the command again, on the new node
4. If the command returns the -LEAVING error, switch to a better node...

Until

• we connect to a non-leaving node. In that case continue program execution
• or we run out of nodes to connect to (switchNodeIfNeeded() throws a ConnectionException). In that case we have no other solution than to give up.

[mariano]

Totally +1

On Monday, 7 March 2016, Revisor notifications@github.com wrote:

I think this should be a responsibility of the Manager, and I would do it
like this:

1. Execute a command
2. If any command returns the -LEAVING error, switch to a better node (
   switchNodeIfNeeded())
3. Execute the command again, on the new node
4. If the command returns the -LEAVING error, switch to a better
   node...

Until

• we connect to a non-leaving node. In that case continue program
  execution
• or we run out of nodes to connect to (switchNodeIfNeeded() throws a
  ConnectionException). In that case we have no other solution than to
  give up.

—
Reply to this email directly or view it on GitHub
#28 (comment).