[I got this reported privately and the reporter wants to keep its employer's name out of this. I'm filing it here so that I won't forget about this issue.]
The above commit introduced a pretty big performance regression when
dealing with large number of requests going through an HTTP proxy + ssl.
I¹ve measured around a 4.5x increase in time spent in the attached
application vs reverting this change. We noticed the problem when we
upgraded from 7.25 to 7.36 recently and it had a pretty negative impact on
some application logic we have. With the test application you will need
to change the proxy to your own local one of course.
Without ssl the attached application is very fast for me.. see the
following timings all based off of 7.37.0
time with https:
time with http:
time with https and cb3e6dfa3511e75334f70f53370ce3f1d711c904 reverted:
time with http and cb3e6dfa3511e75334f70f53370ce3f1d711c904 reverted:
(I'll attach the test application later after editing out sensitive names)