You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In tests using http.request I had problems exceeding 1k req/s. Below that, the response time is around 1ms, when it gets to 1k, my response time goes to 500ms to 1s for each request. Has anyone had this kind of problem? I didn't find anything about it here.
The text was updated successfully, but these errors were encountered:
All the points mentioned were often analyzed in detail. Both via monitoring tool like Netdata and commands like strace, mpstat and etc. There is no resource exhaustion, ringbuffer overflow or filedescriptors overflow. Sysctl is fine tuned.
I use luasocket inside a lua script called on my PowerDNS server. I validated in many ways and everything pointed to the http.request function I use luasocket to. When I have a stream above 1k of reqs/s, I have horrible degradation in response time with http.request.
Could you help me in a way to be able to debug this better inside luasocket?
It may be an effect of the Nagle's algorithm that seems not to be disabled in http.lua.
Something like h.try(c:setoption('tcp-nodelay', true) in http.lua (function _M.open) may help in this very specific case but may not be desirable for normal HTTP usage.
In tests using http.request I had problems exceeding 1k req/s. Below that, the response time is around 1ms, when it gets to 1k, my response time goes to 500ms to 1s for each request. Has anyone had this kind of problem? I didn't find anything about it here.
The text was updated successfully, but these errors were encountered: