GitLab

aggregates in a portal experiment, whenever the set of resources
changes. The merged manifest does not replace the manifest at the
aggregate, instead this is a portal specific item that can be retrieved
on experiment nodes with `geni-get portalmanifest`.

Makes for smoother updates, since it takes so long.

buses on those routes. Also turn on floating busids in this mode.

the exit code.

route is which on the map.

endpoints page. Only admins see this at the moment.

does not bind to a conflicting port (typically capserver).

on mobile endpoints. Also make it more apparent when an aggregate is
offline (unreachable), and when an aggregate has been removed from
the experiment (off route) but is not reachable to finished the op.

offline, so that the web interface can pick up on that fact.
Technically, nothing has really changed, not until the aggregate comes
online and the daemon notices it can retry (well, it will notice when
I add that code).

had upgraded to fix a different problem.

  # In April 2021 I did an optimization pass to try to reduce the number of
  # REST API calls made (since they take a minimum of 2s each). Til they have
  # more real world testing, with different firmware versions, make this
  # optional. Set this to zero to revert to the "old ways" which have been
  # working for several years. The biggest difference was avoiding enabling
  # and disabling one port per call. For big LANs, this is a huge improvement.
  #
  # Quick numbers for 38 node LAN (38 ports, 1 VLAN):
  #  Old snmpit -t:  82 seconds, calls:  5 getinfo ( 4 cached),  40 other.
  #  New snmpit -t:  12 seconds, calls:  5 getinfo ( 4 cached),   3 other.
  #  Old snmpit -r: 155 seconds, calls:  7 getinfo ( 6 cached),  77 other.
  #  New snmpit -r:  12 seconds, calls:  7 getinfo ( 6 cached),   3 other.
  # Mostly these are savings of individual calls to enable/disable ports.
  #
  # With multiplexed links (tagged ports):
  #  Old ...

for long periods of time.

offline, the experiment goes into a canceled state, and termination is
periodically retried. But this leaves the routes in use cause that DB
state is not cleared until the instance is deleted. So, for now lets
delete that state so other users can proceed.

The problem is that when the bus comes online, the slice at the bus is
still there (endpoints cannot self terminate), so the portal has to
terminate that slice before it an add the bus to whatever experiment is
currently using that route. In other words, lets kick the can down the
road for a few more days.

Also allow unreserved routes to be used (originally, routes had to be
reserved to be used, but we are relaxing that restriction, at least for
now.

day the experiment starts.