perfSONAR User Q&A and Other Discussion

[Mark Feit <>]

Andrew Lake writes:

Aside: did you ever look at using influxdb as a backend for perfsonar? 
Perhaps it wasn't around in the early days. However it's highly tuned 
for time-series data, and it compiles down to a standalone binary. 

Yes, it comes up periodically. It falls into the bucket of some people like it, others have had bad experiences with it. A big feature of 4.0 is the ability to write “archiver” plug-ins to ship the data different places, so someone could write a plugin to ship it to influxdb (or maybe even use the rabbitmq archiver we already provide) to do so. 

To expand on that a bit and give a quick peek at what’s coming in the future:

pScheduler includes archiver plugins for Esmond, RabbitMQ, HTTP GET/POST, Syslog and a couple of others you’re not likely to ever use because they were written to test the infrastructure (one sends results to the bit bucket and the other fails periodically).  Which archivers you use and how they’re configured can be set up on a per-task basis; you can also instruct the system to send every result through a specific archiver if you want.  Some of this isn’t available to meshconfig because of limitations in that program, but I imagine we’ll be working to remedy that in the future.

Archiver plugins are dead easy to write with a good understanding of how to get data into your archive (or non-archive; we don’t actually care where you send it) and what the inputs and outputs are to/from the parts of the plugin.  Most of the time, you can get off to a good start by modifying an existing one.

There’s a feature I’m planning to enable in the 4.0.1 release that will let you apply transforms to the JSON result coming out of the system before the archiver sees it.  That will obviate the need to write custom plugins for anything the existing archivers can talk to if all you need to do to make it work is filter the JSON into something else.

--Mark