Re: drop experimental from http2 for 2.4.next?

[Jacob Champion]

On 04/17/2017 01:24 AM, Stefan Eissing wrote:

As much as I like to work on it, I am certainly not able to do
that by myself. So, yes, I welcome getting rid of experimental.


Sounds good to me.

I agree with Nick that we should probably switch to RTC at the same 
time, for backports to 2.4.x.


--Jacob

Re: Apache leaves shared memory segments

[Reindl Harald]

and also "-k stop" don't remove them

read HTTPD_PID < "$PROFILE_ROOT/logs/httpd.pid"
$HTTPD_BINARY -f "$PROFILE_ROOT/httpd.conf" -k stop
exit 0

[builduser@testserver:/rpmbuild/PHP-PGO]$ ipcs -m | wc -l
12
[builduser@testserver:/rpmbuild/PHP-PGO]$ ./profile.sh
[builduser@testserver:/rpmbuild/PHP-PGO]$ ipcs -m | wc -l
13
[builduser@testserver:/rpmbuild/PHP-PGO]$ ./profile.sh
[builduser@testserver:/rpmbuild/PHP-PGO]$ ipcs -m | wc -l
14

[builduser@testserver:/rpmbuild/PHP-PGO]$ ipcs -m
-- Shared Memory Segments 
keyshmid  owner  perms  bytes  nattch status
0x661403ad 0  root   6001000   2
0x151450bd 32769  root   6001000   6
0x0121019a 65538  builduser  6002560
0xc121019a 98307  builduser  6002560
0x8621019a 131076 builduser  6002560
0xdf21019a 163845 builduser  6002560
0x1b210199 196614 builduser  6002560
0x99210199 229383 builduser  6002560
0xe2210199 262152 builduser  6002560
0x5d210199 294921 builduser  6002560


Am 17.04.2017 um 13:28 schrieb Reindl Harald:

https://bz.apache.org/bugzilla/show_bug.cgi?id=7838
that still happens with 2.4.25

"killall httpd 2> /dev/null" in a script starting a temporary httpd for 
php-pgo-profiling since it's a SIGTERM should not leave them and finally 
fail after enough runs to allocate shm segment for auth_digest until you 
reboot


  killall httpd 2> /dev/null
  sleep 2
  rm -f "$PROFILE_ROOT/php.ini"
  rm -f "$PROFILE_ROOT/httpd.conf"
  rm -f "$PROFILE_ROOT/logs/sess_"*
  rm -f "$PROFILE_ROOT/logs/authdigest"*
  rm -f "$PROFILE_ROOT/logs/httpd.pid"
  rm -f "$PROFILE_ROOT/logs/modsec"*

the 256 bytes are pretty clear from "AuthDigestShmemSize 256"

$HTTPD_BINARY -X -f "$PROFILE_ROOT/httpd.conf" to make sure there is 
only one httpd process should not take longer than the 2 seconds sleep 
given that it handles 1500 serial requests within 10 seconds before


[builduser@testserver:/rpmbuild/PHP-PGO]$ ipcs -m

-- Shared Memory Segments 
keyshmid  owner  perms  bytes  nattch status
0x661403ad 0  root   6001000   2
0x151450bd 32769  root   6001000   6
0x0121019a 65538  builduser  6002560
0xc121019a 98307  builduser  6002560
0x8621019a 131076 builduser  6002560

Apache leaves shared memory segments

[Reindl Harald]

https://bz.apache.org/bugzilla/show_bug.cgi?id=7838
that still happens with 2.4.25

"killall httpd 2> /dev/null" in a script starting a temporary httpd for 
php-pgo-profiling since it's a SIGTERM should not leave them and finally 
fail after enough runs to allocate shm segment for auth_digest until you 
reboot


 killall httpd 2> /dev/null
 sleep 2
 rm -f "$PROFILE_ROOT/php.ini"
 rm -f "$PROFILE_ROOT/httpd.conf"
 rm -f "$PROFILE_ROOT/logs/sess_"*
 rm -f "$PROFILE_ROOT/logs/authdigest"*
 rm -f "$PROFILE_ROOT/logs/httpd.pid"
 rm -f "$PROFILE_ROOT/logs/modsec"*

the 256 bytes are pretty clear from "AuthDigestShmemSize 256"

$HTTPD_BINARY -X -f "$PROFILE_ROOT/httpd.conf" to make sure there is 
only one httpd process should not take longer than the 2 seconds sleep 
given that it handles 1500 serial requests within 10 seconds before


[builduser@testserver:/rpmbuild/PHP-PGO]$ ipcs -m

-- Shared Memory Segments 
keyshmid  owner  perms  bytes  nattch status
0x661403ad 0  root   6001000   2
0x151450bd 32769  root   6001000   6
0x0121019a 65538  builduser  6002560
0xc121019a 98307  builduser  6002560
0x8621019a 131076 builduser  6002560

Re: drop experimental from http2 for 2.4.next?

[Stefan Eissing]

These modules, they grow up so fast...

For the project, it would be good to drop that "experimental" and 
treat HTTP/2 as an integral part of httpd. Not only for political
posturing (which is important), but also for very technical reasons.

Looking at https://w3techs.com/technologies/details/ce-http2/all/all
one can see that HTTP/2 is used by 13% of all sites, which is almost
double from 1 year ago. Firefox telemetry reports HTTP/2.0 now 
on 35% of all responses received.

What needs to be done? From what I saw in the last two years, these 
are key areas to improve:

  1. separation of semantics and serialisation
  2. connections with >1 requests simultaneously

mod_http need to spin off a mod_http1 with the parts that read
and write headers, handle chunked encoding in requests
and responses. etc.

mpm needs facilities for processing slave connections and assign
its resources to slave/master connections in fair and performant
ways.

As much as I like to work on it, I am certainly not able to do
that by myself. So, yes, I welcome getting rid of experimental.

Cheers,

-Stefan


> Am 16.04.2017 um 15:00 schrieb Jim Jagielski :
> 
> Agreed. The "tag" is being used to keep it out of distros
> as well as to continue the FUD that httpd doesn't "really"
> support http/2.
> 
>> On Apr 15, 2017, at 5:02 PM, Eric Covener  wrote:
>> 
>> Hi everyone, shall we drop experimental from mod_http2 for 2.4.next?
>> 
>> We could drop it and keep CTR.
>> 
>> -- 
>> Eric Covener
>> cove...@gmail.com
>