When using multiple slaves, each with their own perforce workspace, polling appears to be broken causing endless rebuilds.

One slave does the polling, finding changes:

P4 Task: establishing connection.
... server: perforce.somewhere.com:1666
... node: A
P4: Polling with cstat: //

/... +
... p4 change -o 113712 +
... found change: 113712
... p4 change -o 113686 +
... found change: 113686

Triggering a build on another host:

P4 Task: establishing connection.
... server: perforce.somewhere.com:1666
... node: B

Which already has all the changes!

It seems that the p4-plugin is relying only on the perforce servers knowledge of what the client has to decide if a new build is needed, which breaks if multiple clients are polling/building.

Maybe it would be better to save the last changelist id on the master and only build if a new changelist is found on when querying the server?