Mailing-List: contact users-help@trafficserver.apache.org; run by ezmlm
Precedence: bulk
Reply-To: users@trafficserver.apache.org
Received-SPF: pass (nike.apache.org: domain of jdavidlists@gmail.com
 designates 209.85.223.177 as permitted sender)
MIME-Version: 1.0
Sender: jdavidlists@gmail.com
In-Reply-To: 
 <CAMujjx2zb0OHABXBbwcCoxw4OiU3eVNXwZRo9pjge0SLqE8SDw@mail.gmail.com>
References: 
 <CABXB=RQgR2=CFgwB+kH9_sCbbfhz0C0+97kYiJn7RAkX-mytDg@mail.gmail.com>
	<3D01BE33-22BE-46EF-A070-701B5920A08A@apache.org>
	<4461ADD8-10CA-48DE-AC25-A443AD2B854A@apache.org>
	<CAMujjx2zb0OHABXBbwcCoxw4OiU3eVNXwZRo9pjge0SLqE8SDw@mail.gmail.com>
Date: Fri, 27 Feb 2015 12:01:01 -0500
Message-ID: 
 <CABXB=RTCuvHcNfCEVc2UWZMPq-fW3dGrmQNRXi9V7jc6NmgqKQ@mail.gmail.com>
Subject: Re: Migrating from squid
From: J David <j.david.lists@gmail.com>
To: users@trafficserver.apache.org
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

On Fri, Feb 27, 2015 at 10:50 AM, Leif Hedstrom <zwoop@apache.org> wrote:
> Dealing with external squid helpers is a bit wonky, but you probably coul=
d implement something in a plugin that does it. The fact that you are doing=
 so much weirdness (MySql, Memcached) makes it particularly tough, I=E2=80=
=99m not sure how Squid deals with that?

Squid prestarts a (configurable, large) number of external rewriters
sufficient to handle the #requests * delay product.

On Fri, Feb 27, 2015 at 11:10 AM, Faysal Banna <degreane@gmail.com> wrote:
> I have been doing this using nothing but lua injecting/retrieving  data f=
rom
> mysql, mongodb , sqlite ...

Can the lua approach hold resources open between requests, preferably
in some kind of managed resource pool, like the APR offers in Apache?
Forking a process from lua for every incoming request will lead to a
*tremendous* amount of context-switching overhead at high request
rates.  And that's assuming that we pare the process being exec'd down
to some sort of thin client, as the current rewriter takes several
seconds to spin up and assimilate initial data.  And that kind of
split is definitely reasonable & doable.

This *might* work if we can get all the info about the request at the
Lua level and dispatch all the various types of responses (backend
selection for existing URL, 301+new URL, 4XX/5XX error) from there.
But to get really good performance, the best approach would probably
be to maintain a persistent pool of open sockets to the external
logic.  Is that possible?

Thanks!