Network boot from MAAS sometimes fails at "grub>" prompt

[Rod Smith]

On SOME (but not all) boots via MAAS, GRUB hangs at the "grub>" prompt. This happens about 10% or 20% of the time on affected servers (at least ostwald and meitner, two Supermicro servers).

The MAAS rackd.log file shows that the node has requested and received GRUB:

2020-11-17 10:08:47 provisioningserver.rackdservices.tftp: [info] bootx64.efi requested by 10.1.10.82
2020-11-17 10:08:47 provisioningserver.rackdservices.tftp: [info] bootx64.efi requested by 10.1.10.82
2020-11-17 10:08:50 provisioningserver.rackdservices.tftp: [info] grubx64.efi requested by 10.1.10.82

On a failed boot, the process then stops; the node does not request grub.cfg, as happens normally. Watching the console, I see several notices that read "error: Couldn't send network packet." (See attached screen shot.)

At the grub> prompt, net_ls_addr shows the expected IP address, and net_ls_routes shows a routing table; however, net_bootps results in an error message stating "can't find command `net_bootps`" (see second attached screen shot.)

Once the system is hung, typing "exit" at the "grub>" prompt causes the server to try the next boot option, which usually works (booting via another network interface, in the case of our servers).

As noted, this problem occurs on a minority of boots. It can affect reboots after deployment, and if it occurs during deployment, it can prevent deployment because the server will hang at the "grub>" prompt.

[Julian Andres Klode (juliank)]

Hello!

This seems to be a duplicate of https://launchpad.net/bugs/1900668 - could you check that? I have some list of things in there I'm looking for, mostly running with debug=all and trying current SRU in proposed with has a tftp fix (probably unrelated).

Thanks!

[Julian Andres Klode (juliank)]

The proper command name is net_bootp, not net_bootps

[Julian Andres Klode (juliank)]

Sorry for the many comments, but net_dhcp is a more modern replacement for net_bootp too AFAIUI.

[Rod Smith (rodsmith)]

Ah, I thought the fact that "net_bootps" returned a different error meant this was a different bug; but if that's a typo, then perhaps not. Also, typing "normal" (as in comment #5 there) doesn't work. Still, it does look like it's likely a duplicate.

[Rod Smith (rodsmith)]

It turns out that "normal" does work, although I could swear it didn't work earlier. Perhaps I mis-typed the command. In any event, I'm marking this bug report as a duplicate now.