HTTP/2 in httpd 2.4.20

Some days before Easter, I reported on the new things in the upcoming 2.4.19 release of Apache httpd. Well, I broke the Windows build in that release and it was scrapped. That is the Apache httpd release process: if you break it, you own both parts!

So, Jim who is the hero doing almost all release management decided to wait until after Easter to tag the next release candidate. Which is supposed to happen today! Hopefully, this time everything works and we can ship a 2.4.20 by the end of the week after the usual testing and voting.

In the meantime, I have not been idle and sneaked in some fixes and a little new stuff that might interest you:

Data is passed more efficiently from the request processing down the main connection. mod_http2 needs to know the frame size of DATA frames before it acutally sends it. Up until now, it first asked the lock protected code that is responsible for data shuffling back and forth how much is available. On transfer it entered the lock again to get the data. With 2.4.20, this is done in one go. Slight performance increase in certain situation.
Slave connections (think of them as kind of, pseudo, internal HTTP/1.1 connections) used to be setup and destroyed for each HTTP/2 request. Now, a set of them is kept for future reuse, saving some effort.
A bug in 2.4.19 was fixed that prevented scoreboard (server-status) updates from happening.
New logging niceties were added, see more below.

Logging HTTP/2

I admit that logging has not been high on my priority list. Probably because I myself do not run statistics on the few servers that I maintain. But a lot of you do and rightly so! So without further ado, here is what possiblities you will have in 2.4.20:

First: the default log format has not been touched. Everything here will not magically confuse your statistics. If you want to make use of new things, you have to add them to a custom log format.
What will however automatically be visible is that the log format %k, will work also in HTTP/2 connections. For h1, it counts how many requests have already been processed on a connection. In h2, where requests run concurrently, it counts how many requests have already been started. And with started I do not mean received. Because received requests (or streams as h2 calls them) are queued. And when a h2 worker becomes available it selects the oldest one with the highest priority to work on.
The environment variable HTTP2, value on|off, flags if HTTP/2 is used on a connection/for a request. You use it as %{HTTP2}e in your custom log formats.
The env var H2_PUSH (also available as deprecated H2PUSH) is also a flag (on|off) which is on only when the PUSH feature of HTTP/2 is available (server and client enable it).
Further, H2_PUSHED is either empty or has the value PUSHED for requests that have been pushed by the server.
H2_PUSHED_ON contains the stream identifier (a 32-bit int) of the request that triggered the push.
H2_STREAM_ID is the stream identifier of the request.
H2_STREAM_TAG is a process wide unique string for a HTTP/2 request, consisting of connection and stream identifier. This is also used when debug logging mod_http2.

And since these are all env variables in httpd you can make use of them in your CGI/PHP/Python or whatever web application.

Hope you enjoy it.

Münster, 04.04.2016,

Stefan Eissing, greenbytes GmbH

Copying and distribution of this file, with or without modification, are permitted in any medium without royalty provided the copyright notice and this notice are preserved. This file is offered as-is, without warranty of any kind. See LICENSE for details.