On 12/23/2013 11:50 AM, Adam Thompson wrote: > On 13-12-22 06:44 AM, Stuart Henderson wrote: >>> I'm seeing lots of "nsd[11026]: error: sendto failed: No route to host" >> You may need to raise net.inet.udp.sendspace > Done. Raised from ~9000ish (default) to 41600. Errors still occurring > periodically, no perceptible change in rate. > If that was the problem in the first place, wouldn't the error be > different (ENOBUFS instead of EHOSTUNREACH)? > > If it's of any interest, the error often occurs in bursts (n>=4 within > syslog's "last message repeated /n/ times" window). > You might get a hint from netstat statistics by capturing them at intervals and seeing if there's anything unique about periods with errors.
The fact that the problem is intermittent strongly suggests a cacheing problem somewhere. "No route to host", of course, suggests that the arp cache is not working for you for some reason, routing tables are messed up, or that dns lookups are returning nonsense or unusable addresses for destination names. If the program will log errors (including at minimum the destination address) or if you could add syslog or local disk file logs of errors including destination name, IP, size, and timestamp, this might help to go further. Errors for local destinations suggest arp problems. Errors for remote destinations suggest dns or some form of route flapping. If your network has any sort of dynamic routing (for instance, multiple outgoing interfaces with failover) that's easy to misconfigure. Geoff Steckel
