(sorry for late response.. I caught you on Slack as well, but responding to
Yes we have recently added a stage in our continuous delivery pipeline to
catch performance regressions.
This stage right now is not using routing-perf-release, it is currently
using a command-line (golang-based) http load generator and we are
currently measuring and emitting requests / sec per run. The stage also
will fail if the results of the run are below a specific threshold.
We also have a datadog dashboard with all the routing related metrics we
are getting from the firehose.
Important thing to note is that this is just the beginning of our
performance analysis related work on gorouter,
Shannon (Routing PM) is heading up our next initiative which focuses on
gorouter performance and metrics.
Part of this work will likely include enhancing our performance automation
test suite and more info on metrics.
I'll let Shannon chime in if he had additional info to add.
On Wed, Apr 6, 2016 at 5:47 AM, Michal Tekel <
That helps a lot. Looks like (from slack chat and pivotal tracker) that
the routingteam already has a job set-up which runs perf. tests each hour
and posts results to datadog, which is something very similar to what we
were looking to do. I got in touch with them on slack...
On 5 April 2016 at 17:39, Guillaume Berche <bercheg(a)gmail.com> wrote:
I understand the routing team is working on automated performance tests
that aims at detecting performance regressions on the gorouter. We have not
yet set up execution of those at Orange yet. More in the pointers below.
The routing team was considering to open sourcing the routing ci repo, I'm
not sure of the current status now.
https://www.pivotaltracker.com/n/projects/1358110 with search:
label:"gorouter performance" includedone:true includes current perf
regression and effort for automated perf testing and determining max
request rate and concurrent requests
Hope this helps,
On Mon, Apr 4, 2016 at 4:08 PM, Michal Tekel <
I wanted to check if you know about a good way to spot performance
regressions on CF and if anyone is doing it (at least partially) already.
We are mainly interested in discovering user-facing degradation. That is,
app operations (deploy, scale, delete, manage routes) and routing layer
operations (latency per request, total throughput, SSL termination
capacity) - esp. in connection with (heavier) routing services usage.
We did a brief investigation about the best way to automate this kind of
regression testing, but didn't find any complete solution. We had a look at
[CF PATs](1) tool and made a [container](2) it can be run from. But that
only gives us metrics on app operations and no way of spotting regressions
or automating the tests.
We were wondering if perhaps someone already does any kind of
performance measurements as a part of their CF build/deploy
pipelines/process and what their experience with tracking the results is...