GitLab

feature on nodes that are prereserved to the same project as the
experiment, and add a 0.2 weight desire to the vtop file. Rob says it is
okay to always add this desire, so that is what I did.

experiment in the project, but without a prereserve setting, then set the
pre-reserve (as per Mike's request a few months back). But also need to
check that if the node is allocated to the project, and it does have a
pre-reserve, that pre-reserve must actually match the project and have the
right reservation name.

on the APT cluster.

A couple of Clemson machines have 47 disks each!

proof of concept, the NS converter path now uses ns2rspec piped into the
newly added rspec2genlib, which becomes the source code for NS based
experiments that are being converted to Portal experiments.

recently and currently active users and projects (these lists are portal
specific!). The second two tabs do dynamic search of the database via ajax
calls, to populate the lists. This is cause listing all users statically on
the page would be insane (tablesorter does not perform well on such giant
lists).

Plus other little related tweaks.

Apparently, specifying "--unit==1" to grub did not work because grub
didn't think it had a unit 1 (aka, COM2), so it rejected the "serial"
command. Output still worked due to redirection (I think), but it was
expecting input from the keyboard.

It looks like the number of valid serial ports is compiled into grub and
who knows how it was built for Ubuntu. Anyway, by specifying "--port=0x2F8"
in addition, that will override the unit test and all is well. Strange.

Also enhanced the Linux slicefix to deal with fixing /etc/default/grub so
that those wacky console settings will remain in effect even if grub is
re-configured (which some packages do).

genilib).

Need to rethink. We still rely on bootinfo sent events on the Moonshots
(e.g., when going into PXEWAIT).

Since the Moonshot cluster will soon have x86 nodes and those nodes do
call bootinfo (unlike the ARM nodes), make sure that bootinfo does not
send PXEBOOTING/BOOTING events (since dhcpd already has) by starting it
with the new -E option.

Formalize by adding a BOOTINFO_EVENTS defs var that defaults to 1.
This ensures that a nobootinfo version of the dhcpd.conf template is
installed and that bootinfo is invoked with the -E option.
Only defs-cloudlab-utah should set it to zero.

Use on Cloudlab Utah where we configure dhcpd to send the events
(needed for the uboot'ed ARM cartridges).

This was rendered obsolete by the addition of the bs-agent 1+ years ago.

This will be used by the power command to tell it to try to power on a
machine that fails to "cycle". ipmitool (or IPMI) seems to fail by default
if you try to cycle a powered-off node.