perfSONAR User Q&A and Other Discussion

[Mark Feit <>]

"Hitch, A Ryan (53629) CIV USN NIWC PACIFIC CA (USA)"    (via perfsonar-user 
Mailing List) writes:

    I have a test grid with 10 probes that has been running latencybg and 
throughput tests for a few weeks and I see an increasing number of 
non-starter tasks. These seem to be resulting from the following error in 
pscheduler.log, for example:
    Sep 11 16:44:49 s00-perf journal: runner ERROR   2208189: Failed to post 
run for result: duplicate key value violates unique constraint "run_uuid_key"
    Sep 11 16:44:49 s00-perf jounrla: runner ERROR   DETAIL: Key 
(uuid)=(95bd7ecf-e106-4cd6-8ab7-14a096843966) already exists.

That falls under the heading of "shouldn't happen" or at least "should only 
happen on exceedingly-rare occasions."  Yours would be the first report of a 
collision in almost 5 years of developing and running pScheduler.  That fact 
that it happened at all is odd; the fact that it's happening with any 
regularity is really, really strange.

UUIDs for things in the database get there two ways:  one is when a row is 
inserted; the other is when a run is sent over by another pScheduler instance 
as part of a two-participant test like throughput.  In both cases, the UUIDs 
are generated randomly using a function in the pgcrypto extension which, I 
assume, depends on the kernel for randomness.

Could you outline any ways in which the system deviates from a stock 
installation, specifically anything that might have an effect on random 
number generation by programs not running as root?

--Mark