perfSONAR User Q&A and Other Discussion

[Mark Feit <>]

Murilo Vetter writes:

 

# pscheduler task --context '{ "schema": 1, "contexts": [ [ { "context": "linuxnns", "data": { "namespace": "ns-ens224" } } ], [ { "context": "linuxnns", "data": { "namespace": "ns-ens256" } } ] ] }' throughput --source 192.168.0.39 --dest 192.168.0.45 --source-node 192.168.0.43 --dest-node 192.168.0.43
Submitting task...
Unable to post task: Error while tasking 192.168.0.43: Unable to post task to 192.168.0.43: ERRO: Task already exists. All participants must be on separate systems.

 

Would it have a way to overcome this error, forcing to do the test? Maybe in a future implementation? Or it is not suppose to?

I tried with 2 different IP on the same host with no namespace configured and it showed the same problem.

 

pScheduler does not allow running of multi-participant tests (throughput and simplestream) where the same server acts as more than one of the participants.  Part of it has to do with the way the API’s resource model is organized and part of it has to do with avoiding running resource-intensive tests on the same server at the same time and getting distorted results.  Any of the single-participant tests (rtt, trace, latency) will work because there’s no second pScheduler involved and they do everything necessary to exercise the namespace changes.  For those, you’d specify a source, a source node and a destination.

 

I do my testing of this on pairs of VMs with a private segment between them.  There’s a utility script in the sources that does the alternate namespace setup and can be customized for your environment:  https://github.com/perfsonar/pscheduler/blob/master/scripts/create-netns

 

--Mark