On Wed, 29 Nov 2023, Dmitry Karpov via curl-library wrote:
Is it something known or a not previously noticed regression in 8.4.0?
When I looked into this now, I noticed a regression in multi_getsock() which
makes libcurl re-calculate the waiting periods during rate-limiting way too
often. That's wasteful and in my tests it made the limited speed differ more
from the desired value than what it should otherwise do.
This said: I could not see a 20% deviation in my tests. I saw a much smaller
one, that seems to be mostly gone with my fix. So I can't say for sure if this
really is the problem you see or if there is an additional one.
Anyway, my proposed first fix is here:
https://github.com/curl/curl/pull/12430
If you apply that patch, then doing debug outputs in the
Curl_pgrsLimitWaitTime() function is much easier and way less "floody". If you
keep seeing this problem we can then double-check the calculations there to
check that they seem reasonable. They do for me.
--
/ daniel.haxx.se
| Commercial curl support up to 24x7 is available!
| Private help, bug fixes, support, ports, new features
| https://curl.se/support.html
--
Unsubscribe: https://lists.haxx.se/mailman/listinfo/curl-library
Etiquette: https://curl.se/mail/etiquette.html
