Also note, that it’s always a good idea (and strongly recommended) to
stick
to the latest version especially if you’re using new or/and experimental
modules like SPDY.
The “stable” branch receives only critical bugfixes, and misses many
other.
Thank you very much for the comment. I’d appreciate it a lot.
From server production perspective, The ‘mainline’ branch is not really
useful. Not only the SPDY code changed but everything changed or can
change. This gives a lot of extra time to test everything through and
through.
So it is strongly recommended to use stable code as far as possible in
any
production environment.
And about experimental modules like SPDY. Yes you are right its
experimental. But for web servers you have no choice now days.
You can’t optimize a site to the max without SPDY.
Coming back to 200ms.
200ms extra, unnecessary delay is deadly for a web server.
All the way in a world dominated by internet search engines where every
ms
counts.
It’s like a heart attack.
I do not understand that this problem is not referred as critical bug.
It
should even be a blocker.
Your changelog writes about: ‘now the “tcp_nodelay” directive works with
SPDY connections’. This is a major loss of function because the existing
1.6. tcp_nodelay won’t work as it should be.
Understood. Apart from the fact what SPDY can mean for someone specific.
There is a flaw in and which prevents “tcp_nodelay” in Nginx 1.6. to
function correctly.
Understood. Apart from the fact what SPDY can mean for someone specific.
There is a flaw in and which prevents “tcp_nodelay” in Nginx 1.6. to
function correctly.
How to fix that.
There are several options:
you can backport 1.7 diff to 1.6;
you can use 1.7 in production (this is what we actually recommend
to do; e.g. nginx-plus based on 1.7 branch currently).
We will discuss merging this code to 1.6 for the next release but we
don’t have a schedule for 1.6.3 yet.