GitLab

Fedora 8.

NS file is wrong. Well, worse then that, the query was wrong and it
was setting all the resource records for that experiment to the new NS
file. Highly annoying, probably no way to recover the lost record of
NS files.

It doesn't look promising that I will be back to it this week, sigh...

  endexp <pid>/<eid>
rather than
  endexp <pid> <eid>
which is no longer supported

tables.

Remove the tunnels table since that stuff is now in the
lan_member_attributes table.

Add node_id to the lan_members table. While its nice having that as an
attribute, it makes joins more difficult and slower.

nightly, and make repairs for the common things I see happening.  The
current version repairs most of the errors that have crept in since
the Epoch, and then rebuilds the summary stats table (user_stats,
group_stats, project_stats). My intent is to add a "daily_stats" table
as well since scanning the testbed_stats and experiment_resources
table to generate range data is getting pretty slow as more records
enter the system. We can also use a daily_stats table to generate
graphs on the fly as Jay requested a few weeks ago.

Not done yet, hope to return to it later this week.

script.

(Rob,Mike,Jay) have overflowed an unsigned int.

Some parts very "hacky", lots of XXX / TODOs.
BARELY tested!!!! Don't use on the production system yet.
Need to verify that things didn't break, especially with the automanager,
the bgmon.pl "outage detection stuff", etc...

See code's todos and also the file
flexlab-canonicalization-misc-notes.txe

At this point:
 - mangerclient & automanagerclient did not change.
 - The manager, when a probe request is received, looks at the DB table
   "tool_spec" to retrieve the proper wrapperpath, formal parameter list, etc..
 - The manager assembles a message and sends it out to the appropriate
   path probers, which operate on the request as they have been
   (no major internal changes to the path probers).
 - The path probers call the toolwrapper to run the tests
 - The toolwrappers return a canonical result for that type of metric.
 - The path probers send the result and the actual parameters used by the
   tool to the data collector.
 - The data collector uses the toolname and actual parameters to get an
   idx number. This idx is stored in the "measured_by" field of this
   measurement's entry in the "pair_data" table.

arguments and result parsing handled by separate application wrappers.
Currently fping and iperf supported.
Not fully functional at this point: user interface needs extending
to support the generalization, as well as getting the results back
to "opsrecv".

Also merged some of Mike's changes to allow for "private" managers
to use during testing.

Result was redundant info returned.

now, this is keyed off nodetype.  Lots of hardcoded constants and config
stuff moved to attributes in the db.  You can now set per-PLC and
per-slice attributes, so you can (for instance) use different auth info
whenever you want.  Experiments can use preexisting slices if somebody
sets up the db before swapin.  Also, we no longer have to rely on
slices.xml to sync up nodes/sites with PLC... can use xmlrpc instead.

Lots of code cleanup, improved some abstractions, etc.

previously would get interpreted as a (non-existent) '1' command line
argument to the tmcc binary.

table is going to get dropped.

sure this matters, having two sets of stats numbers has proven to be a
disaster.  I should just get rid of them or better, regen them with a
periodic script from the experiment stats.

does the loopback device properly inside vservers. Also remove the
tmcd pem files from inside the vserver so that forced to go through
the proxy.

resource record. This has really messed up the counts for anything
with plab nodes (there are too few jail node experiments to worry
about) so need to come up with something to fix the resource records.

creating a whole disk image with the "Easy" form.

which I needed for the CMU emulab.

created during the range.

PLABSUPPORT=0.

experiment "start" (initial creation with swapin). Been this way for a
few months, but the fix to the DB was easy:

	update experiment_resources as r
	left join testbed_stats as s on s.rsrcidx=r.idx
	 set swapin_time=UNIX_TIMESTAMP(tstamp)
        where action='start' and exitcode=0 and swapin_time=0;

dropped out of the tally. Also break out swapmods (active vs inactive)
and preload vs starts.

- latency measurements under 2 second intervals now operate in "continuous"
  mode: a single "ping" process runs per destination, versus the "normal"
  approach of forking a new fping process for each measurement instance.

- Note, that I found a bug which is not fixed in this checkin: if a latency
  test is running in continuous mode and another latency measurement request
  is received with a shorter interval, the "continuous" ping process is not
  re-initialized with the new, shorter interval.