Mailing List Archive

Bugs and Cherokee State
Hi,

I'm glad to read that someone (Stefan de Konink), want to give some love to the "stable" branch of Cherokee. I was on the point of changing to something else (probably Nginx). And I'm sure I'm not the only one. The list was mostly dead, the community seems to collapse quickly. This is bad signs for an open source project.

Here are the issues that bugs me:

1. Recently my client upgrade Firefox from 3.6 to version 10 ESR. Result: all pages of their site served under HTTPS are a lot slower (at least 3x slower).

After a few tests the problem really seem to come from the combination of Firefox 10+, SSL/TLS and Cherokee (1.2.101). I tried many sites that can be view in HTTPS and HTTP with Firefox 10 and Firefox 3.6 and most sites served by Cherokee were more slow only with Firefox 10 under HTTPS. I didn't see any significant differences with site served by other servers.

The slowness really come from the server, not the browser. If I look in Firebug on the network panel, my typical page under HTTPS and FF10 take around 1.15 sec. to load, most of this time is waiting. Same page with FF3.6 under HTTPS took 200 ms.


2. The Timeout settings are not respected under SSL/TLS. To whatever value I set the timeout setting
(via
general > network settings > timeout
or
vserver > behavior > restrictions > timeout),
the server always timeout after 15 sec.

It's with cherokee 1.2.101 on Ubuntu 10.04 with uwsgi (-t 600) and over SSL.

I ask about this one on the list before and I only got 1 answer of someone having the same problem.


3. The Google Chrome bug.

I can open issues in the tracker if needed.


One last think, I really think that Alvaro didn't dealt with his transition really well. He didn't give any news for a long time and after that it say that it will work on a new Cherokee. That sounds to most Cherokee users like "I will not fix your bugs". But the problem is that for most people the most important thing in web server is that it work and work now. After struggling with many SSL related bugs (and some not SSL related) hearing that from Alvaro make me feel that I bet on the wrong horse.

Etienne

_______________________________________________
Cherokee mailing list
Cherokee@lists.octality.com
http://lists.octality.com/listinfo/cherokee
Re: Bugs and Cherokee State [ In reply to ]
Hi Etienne,

> The slowness really come from the server, not the browser. If I look in
> Firebug on the network panel, my typical page under HTTPS and FF10 take
> around 1.15 sec. to load, most of this time is waiting. Same page with
> FF3.6 under HTTPS took 200 ms.

Could you check if the same things happens using the *latest* stable
github version? There are quite some stability changes in so I hope it is
actually fixed. If not, please let us know.

Stefan
_______________________________________________
Cherokee mailing list
Cherokee@lists.octality.com
http://lists.octality.com/listinfo/cherokee
Re: Bugs and Cherokee State [ In reply to ]
On Mon, 11 Jun 2012, Etienne Desautels wrote:

> 1. Recently my client upgrade Firefox from 3.6 to version 10 ESR.
> Result: all pages of their site served under HTTPS are a lot slower (at
> least 3x slower).

I have tested in Chrome with and without false-start
(--disble-ssl-false-start) and it becomes clear that the first request in
the keep-alive request takes about one second 'extra', the one second
sounds like a programmed timeout to me.

Thus although we have 'fixed' the POST breakage, which is obviously a step
in the right direction, figuring out where this one second comes from (a
delay induced by the webserver, or a anticipation from the client on the
webserver not doing something yet) sounds like a good thing more people
can help with using CHEROKEE_TRACE=all.

I have asked Etienne to log a bug in the Google Bugtracker.


Stefan

_______________________________________________
Cherokee mailing list
Cherokee@lists.octality.com
http://lists.octality.com/listinfo/cherokee