Closed jalvz closed 5 years ago
So far, I failed setting ELASTIC_APM_API_BUFFER_SIZE
to max value
If I start several processes, some appear to succeed completely, and others to fail completely
The only way around that I can see is with distributed load testing:
Very high level, the orchestrator process would spin some machines to perform a benchmark, all with same parameters and benchmark_id
. Then that same process can make the right aggregation query to check for regressions.
Dashboards should work the same as they use averages for everything.
@graphaelli what do you think?
Distributing the load testing makes sense - can we generate enough load from a single host with multiple processes for now or are you envisioning coordination among multiple hosts?
When I wrote that I was thinking multiple hosts, but yeah, starting with one seems more sensible.
This seem to be solved on the newest CI setup. Will keep an eye on things, will close for now
(aka, trigger a full queue without breaking the ES link)