Update machine setup docs

[pietroalbini]

r? @Mark-Simulacrum

[Mark-Simulacrum]

I believe the direction on setting up the attached disk are missing (i.e., how to mount, move docker, setup /tmp, etc.)

[Mark-Simulacrum]

When trying to stand up a machine I was also unable to figure out how to view current logs beyond just literally running the command outside of systemd -- is there something I'm missing?

[Mark-Simulacrum]

I also ran into trouble getting the crater image to actually start -- it turns out that if the version of the crater agent and server aren't in sync then it might not start; however, another agent that is already running on an older commit is fine. It feels like we need a version negotiation or something so that at least there's a better error than this: rust-lang/rust#58899 (comment) which is essentially impossible to debug.

[pietroalbini]

I believe the direction on setting up the attached disk are missing (i.e., how to mount, move docker, setup /tmp, etc.)

I'll try to remember how to do that.

When trying to stand up a machine I was also unable to figure out how to view current logs beyond just literally running the command outside of systemd -- is there something I'm missing?

journalctl -u crater-agent should do the trick.

I also ran into trouble getting the crater image to actually start -- it turns out that if the version of the crater agent and server aren't in sync then it might not start; however, another agent that is already running on an older commit is fine. It feels like we need a version negotiation or something so that at least there's a better error than this: rust-lang/rust#58899 (comment) which is essentially impossible to debug.

Yeah, the problem is that a recent PR changed the protocol and I didn't deploy the changes yet. Version negotiation would be awesome indeed.

[Mark-Simulacrum]

journalctl -u crater-agent should do the trick.

Unfortunately, it didn't. On crater-2, running that command ends with Nov 29 01:31:49 ip-172-30-0-241 systemd[1]: Starting Crater agent... which is quite long ago.

[Mark-Simulacrum]

Hm, that does work on a freshly created machine, so I'm not sure what the problem is.