Using pcc-mock to try and push 2x50000 tunnels, connections went down.
Even after closing pcc-mock and waiting (low CPU utilization on ODL), 31185 tunnels were left present in pcep-topology.

Expected result for fixed ODL: When queue gets full, pcep-topology should retry after some time, or do some kind of blocking insert. Input from PCEP session would be slowed down or blocked, and this state would be recognized to suspend usual deadtimer event consequences.

Segment of karaf.log attached, most of exceptions happened two times, for each PCEP session once. Highlights:

2014-10-09 16:17:53,140 | ERROR | oupCloseable-4-2 | DOMDataCommitCoordinatorImpl | 157 - org.opendaylight.controller.sal-broker-impl - 1.1.0.Helium | The commit executor's queue is full - submit task was rejected.
(2 times)

2014-10-09 16:17:53,160 | ERROR | oupCloseable-4-2 | TopologyNodeState | 261 - org.opendaylight.bgpcep.pcep-topology-provider - 0.3.1.Helium | Unexpected transaction failure in node
(2 times)

2014-10-09 16:17:53,167 | ERROR | oupCloseable-4-2 | AbstractTopologySessionListener | 261 - org.opendaylight.bgpcep.pcep-topology-provider - 0.3.1.Helium | Failed to update internal state for session
(2 times)

2014-10-09 16:17:53,175 | INFO | oupCloseable-4-2 | PCEPSessionImpl | 251 - org.opendaylight.bgpcep.pcep-impl - 0.3.1.Helium | Closing PCEP session:
(2 times)

2014-10-09 16:17:53,191 | WARN | oupCloseable-4-3 | DefaultChannelPipeline | 126 - io.netty.common - 4.0.23.Final | An exceptionCaught() event was fired, and it reached at the tail of the pipeline. It usually means the last handler in the pipeline did not handle the exception.
(roughly thousand times in full log)