[j-nsp] ldp flaps after bgp reconfiguration ?

[Alexandre Snarskii]

Hi!

Seeing a bit strange behaviour on one of our routers: 

That's M20, JUNOS 6.4R3.4, connected to one of major internet 
exchanges. Router configured for peer with 224 neighbors. 

Every morining our homebrewed script used to check as-sets of our 
peers in order to reconfigure input as-path-group's, then upload
changes to router, and then commit those changes (all at once). 

Not every day, but for some times per month ldp got reset rigth
after that commit: 

Jun 17 07:18:08 lo0.ProblemRouter mgd[66302]: UI_COMMIT: User 'updater'
performed commit: no comment
Jun 17 07:19:05 lo0.PeerRouter rpd[2617]: RPD_LDP_NBRDOWN: LDP neighbor
<ProblemRouterIP> (ge-1/1/0.0) is down
Jun 17 07:19:05 lo0.PeerRouter rpd[2617]: RPD_LDP_SESSIONDOWN: LDP
session <ProblemRouterIP> is down, reason: all adjacencies down
Jun 17 07:19:05 lo0.PeerRouter rpd[2617]: RPD_LDP_NBRUP: LDP neighbor
<ProblemRouterIP> (ge-1/1/0.0) is up
Jun 17 07:19:15 lo0.ProblemRouter rpd[2706]: RPD_LDP_SESSIONDOWN: LDP
session <PeerRouterIP> is down, reason: connection reset
Jun 17 07:19:33 lo0.ProblemRouter rpd[2706]: RPD_LDP_SESSIONUP: LDP session
<PeerRouterIP> is up
 
My only assumption is that rpd may get too busy recalculating policies
to get ldp time to send/process keepalives....

Any comments or suggestions on that problem ? 
Should we modify logic of our script to perform one commit per 
modified as-path-group ? Or should we just upgrade junos to more 
modern version ?