Seen on recent build from stable/helium branch.
The NPE then breaks pipeline, which stops additional BGP processing.
When ODL comes up after restart, bundle activation is such that this bug does not affect basic BGP test result.

Here are snippets from karaf.log. I have not checked whether the message was parsed correctly, neither I have checked whether BGP speaker is even allowed to send such an update message.

2014-12-08 10:37:41,779 | TRACE | oupCloseable-4-4 | BGPByteToMessageDecoder

2014-12-08 10:37:41,788 | DEBUG | oupCloseable-4-4 | BGPUpdateMessageParser | 234 - org.opendaylight.bgpcep.bgp-parser-impl - 0.3.3.SNAPSHOT | BGP Update message was parsed Update [_pathAttributes=PathAttributes [augmentation=[PathAttributes2 [_mpUnreachNlri=MpUnreachNlri [_afi=class org.opendaylight.yang.gen.v1.urn.opendaylight.params.xml.ns.yang.bgp.linkstate.rev131125.LinkstateAddressFamily, _safi=class org.opendaylight.yang.gen.v1.urn.opendaylight.params.xml.ns.yang.bgp.linkstate.rev131125.LinkstateSubsequentAddressFamily, augmentation=[]]]]], augmentation=[]].

2014-12-08 10:37:41,790 | WARN | oupCloseable-4-4 | DefaultChannelPipeline | 184 - 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.
java.lang.NullPointerException
at org.opendaylight.protocol.bgp.linkstate.LinkstateAdjRIBsIn.removeRoutes(LinkstateAdjRIBsIn.java:216)[258:org.opendaylight.bgpcep.bgp-linkstate:0.3.3.SNAPSHOT]
at org.opendaylight.protocol.bgp.rib.impl.RIBImpl.updateTables(RIBImpl.java:209)[237:org.opendaylight.bgpcep.bgp-rib-impl:0.3.3.SNAPSHOT]
...