perfSONAR User Q&A and Other Discussion

[Mark Feit <>]

Michael Thomas writes:

 

I recently set up a perfsonar server and configured it to run throughput
tests with two other sites in our group.  The iperf3 tests from my site
to the remote locations works fine, but all incoming tests fail…

…

Error from perfsonar.ligo-wa.caltech.edu:
   iperf3 output has neither 'sum_sent' nor 'sum' field, and one of them
is required

 

This part’s a bug.  The plugin should have detected the error before looking for real data in the output.  I’ve opened a ticket on that and will fix it in 5.0:  https://github.com/perfsonar/pscheduler/issues/1270.

Diagnostics from perfsonar.ligo-wa.caltech.edu:
   /usr/bin/iperf3 -p 5201 -B 198.129.208.245 -c 208.69.128.92 -t 10
--json --rsa -public-key-path /var/run/pscheduler-server/runner/tmp/tmp_j91baqu/tmpumiughjb/public-key
--username <username redacted>

 

Unrelated note:  There’s no need to redact those; they’re only valid for one measurement.

   {
         "error":        "test authorization failed"
   }

How can I further debug this authorization failure?

 

The fault is at the far end (208.69.128.92, ligo-la).  I did a telnet to port 5201 on that host and got an answer, which is strange because there was no throughput test scheduled when I did it.  I followed it up with an attempt to run iperf3 against it and got a similar authorization failure.  Odds are very, very good there’s a stray iperf3 process in server mode on that host.  If you’d like an additional set of eyes while you poke around on that system, drop me a line off-list and we’ll set something up.

 

The production code will attempt to hunt down and terminate processes that appear to be orphaned iperf3 servers, but it isn’t perfect.  The latest version of iperf3 has some switches that will let us prevent the server process from running past the scheduled end time of the measurement and there’s already code in the pipeline for 5.0 to take advantage of it.

 

--Mark