GitLab

pubsubd. The main point is that instead of being able to run the SSL
pubsubd at the Mothership only, any site can be a Portal and needs to
run it. So for example my elabinelab is a real Portal, which is very
handy for testing.

(Along the way, adding an option to "predict" to compute the minimum free
node count over a type period.)

Closes #373.

(I had a disk image containing unmodifiable binary software that would
overwrite dhcpcd's sane copy of /etc/resolv.conf, at a nondeterministic
point in time, with something completely bogus.  That screwed up
startcmdstatus reports; this helps out with that case (in combination
with other custom scripting that returns /etc/resolv.conf to sanity).

Note though that we only retry infinitely once runstartup has
successfully gone to the background; up til then, we're limited to about
a minute's worth of retries.  Likewise, we don't retry forever if
runstartup itself experiences an error.  We only retry forever if we
actually have a status to send.

was really inefficient. At about 1200 nodes each successive node was
taking over a second to process. Much better now.

Also note that this commit disables all Jacks when the number of nodes
is greater then 200. Temporary.

If we support provenance but not deltas, then we do not use the
newer create-versioned-image when creating images from Xen vnodes.
However, we had a bug in that path where we would then not pass the
imageid argument to the old script, resulting in us spewing the image
out to stdout which got put in the logfile.

events that need to be transformed into portal events. This cuts the
number of events received by 50 percent, which keeps pubsub from backing
up on sending the events back. Helpful for really big experiments
(1000+ nodes).

If the TBScriptLock caller provides a debug message, it will be stored
in a file, and other blocked TBScriptLock callers will get (possibly
slightly racy) info about who holds the lock.

Then, use this in libvnode_xen to get some info about long calls to xl
(create|halt|reboot|etc).

Also enable lockdebug in libvnode_xen for now.

This is really only needed on the mothership.

cluster is not reachable or its BUSY. Just mark it canceled so the
apt_daemon will pick it up.

the extend modal, tell the user they cannot extend their experiment
until the cluster is back online, try again later. This is a stopgap, we
probably need a better way to handle transient failure in contacting
clusters when doing extensions.

an experiment from the cluster (say, cause its offline).

1. Reservation system now groks experiment lockdown and swappable. When
   swapping in, lockdown and swappable mean the expected end of the
   experiment is never.

2. Reservation library now handles changes to lockdowm, swappable, and
   autoswap (timeout). editexp now hands these changes off to a new
   script called manage_expsettings, which can be called by hand since
   we might need to force a change (I am not changing the classic UI, if
   a change is not allowed by the res system, we have to do it by hand).

3. Minor fixes to reservation library.