Open bisgaard-itis opened 3 days ago
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 87.9%. Comparing base (
cafbf96
) to head (0922c6b
). Report is 310 commits behind head on master.
Thanks!
This limit will only take place for requests coming via public
network. For the internal (within docker swarm cluster) communication it will not apply.
Q: are you adding a sleep+retry in the client for this code?
there is already the retry strategy in place with exponential backoff. That should also be triggered by a 429
very nice! Q: what is your strategy to determine this value (e.g. why did you start in 25 simultaneous requests?)
When running Werner's "load test" I see in our grafana dashboard that the api-server is at times handling around 50 requests simultaniously. That seems to add quite a performance overhead, so 25 is chosen to try and avoid that heavy load. Not sure this is optimal of course. So we will have to tune it as we go
@YuryHrytsuk and @mrnicegyu11 can I get you guys (codeowners) to accept this one?
What do these changes do?
Related issue/s
5952
How to test
run local deployment with
API_SERVER_INFLIGHTREQ_AMOUNT=1
in.env
. Then runwith
<api-server-port>
=the port exposing the api-server through traefik and<auth>
=valid authorization. Observe theoutput*.txt
files and see that only a single request received a 2xx return code and the others received a 429.Dev-ops checklist