Alex Rousskov

There are several options for configuring HttpStateData::inBuf capacity: 0. read_ahead_gap: This is dead wrong, for many reasons (partially detailed in the PR description). This is what Squid is using today....

> We have many other similar buffers which need to be fixed for example in ssl/bio.cc, tunnel.cc and icap code Yes, but I think that is a secondary concern because...

TLDR: @yadij, my [question](https://github.com/squid-cache/squid/pull/736#issuecomment-705014452) is about HttpStateData::inBuf capacity. It is not clear to me what you mean by "server I/O buffer", but it feels like you are talking about something...

> [Christos](https://github.com/squid-cache/squid/pull/736#issuecomment-706508832): It would be interesting to see test programs to simulate various cases. For example to check if and how linux kernel decreases the read buffer sizes. My sense...

I do not plan on making more changes and have removed the Draft/WIP designation from this PR. This complex PR deserves another pair of eyes, so I will keep it...

> the size of inBuf is a sub-set of the message bytes received and not yet sent to client. The above statement is technically true but is largely irrelevant because...

This PR may fix bug [5170](https://bugs.squid-cache.org/show_bug.cgi?id=5170) as well, although these changes were not focused on or driven by that specific problem (which was reported a year after the last PR...

For the record, I disagree with the first three of the above four arguments. For item 4, more discussion/specifics are needed for me to finalize my opinion. Let's discuss on...

I removed `S-waiting-for-QA` because, I assume, you will merge this PR when it is approved regardless of whether somebody volunteers to perform more ICAP tests than you are willing to...

I polished PR title and description to follow Squid Project conventions. This adjustment should remove the M-failed-description label.