PXE booting failures from tftp on gMAAS

[Mark Shuttleworth]

Am seeing a failure of PXE booting on the Garage MAAS.

In pserv.log I see:

2014-04-04 20:17:15+0100 [TFTP (UDP)] Datagram received from ('192.168.9.1', 46517): <RRQDatagram(filename=pxelinux.0, mode=netascii)>
2014-04-04 20:17:15+0100 [TFTP (UDP)] Datagram received from ('192.168.9.1', 46517): <RRQDatagram(filename=pxelinux.0, mode=netascii)>
2014-04-04 20:17:15+0100 [-] RemoteOriginReadSession starting on 39677
2014-04-04 20:17:15+0100 [-] RemoteOriginReadSession starting on 39677
2014-04-04 20:17:15+0100 [-] Starting protocol <tftp.bootstrap.RemoteOriginReadSession instance at 0x7f146e451290>
2014-04-04 20:17:15+0100 [-] Starting protocol <tftp.bootstrap.RemoteOriginReadSession instance at 0x7f146e451290>
2014-04-04 20:17:21+0100 [-] Session timed out, last wait was 1 seconds long
2014-04-04 20:17:21+0100 [-] Session timed out, last wait was 1 seconds long
2014-04-04 20:17:21+0100 [-] (UDP Port 39677 Closed)
2014-04-04 20:17:21+0100 [-] (UDP Port 39677 Closed)
2014-04-04 20:17:21+0100 [-] Stopping protocol <tftp.bootstrap.RemoteOriginReadSession instance at 0x7f146e451290>
2014-04-04 20:17:21+0100 [-] Stopping protocol <tftp.bootstrap.RemoteOriginReadSession instance at 0x7f146e451290>

TFTP requests to the MAAS server just hang silently.

[Gavin Panella (allenap)]

Someone else recently had a similar problem - can't remember who - but it came down to a routing issue, where outgoing packets from the cluster were going to the wrong network.

[Mark Shuttleworth (sabdfl)]

Other traffic between the cluster and MAAS was unafected, why would TFTP
be different? Also, TFTP was unable to serve files to localhost
connections. I think this is a bug in our TFTPd not a network issue.

Mark