Mailing-List: contact users-help@trafficserver.apache.org; run by ezmlm
Precedence: bulk
Reply-To: users@trafficserver.apache.org
Received-SPF: pass (athena.apache.org: domain of baconandtuna@gmail.com
 designates 74.125.82.67 as permitted sender)
MIME-Version: 1.0
In-Reply-To: <5350220C.5010608@thelounge.net>
References: 
 <CAHa6MXSHzs-RZLzZ=Ss77J897T3oS=vgwVFuCFZFcEAzuQ43=A@mail.gmail.com>
	<5350220C.5010608@thelounge.net>
Date: Thu, 17 Apr 2014 12:58:42 -0700
Message-ID: 
 <CAHa6MXRPGNdCG9OK4Kv2bEyCshE+gU+wmos84o4YCTn0ZoBAqw@mail.gmail.com>
Subject: Re: ATS x datacenter(s)
From: Bacon Tuna <baconandtuna@gmail.com>
To: users@trafficserver.apache.org
Content-Type: multipart/alternative; boundary=047d7b87442cbdee1904f7427544

--047d7b87442cbdee1904f7427544
Content-Type: text/plain; charset=UTF-8

I'm actually not concerned about the backend availability as I am about the
gateways. If the gateways die nothing works. Here's a little more info in a
scenario format:

Client requests object from ATS.
Request is handled by geographic load balancer and directed to a cluster in
once data center 'A'.
The cluster fetches the object and caches it.
Client makes request for object again.
Request is handled by geo load balancer and directed to cluster in data
center 'B'.
Client requests object purged from cache.
Purge request is handled by geo load balancer and directed to cluster in
data center 'A'
Cluster invalidates object.

At this point the cache across 'A' and 'B' is inconsistent. Further
requests would either retrieve a "stale" object or fetch a new object and
cache it. Both objects would persist.

Do folks just preach "eventual consistency", or handle this via some other
means?


On Thu, Apr 17, 2014 at 11:48 AM, Reindl Harald <h.reindl@thelounge.net>wrote:

>
>
> Am 17.04.2014 20:41, schrieb Bacon Tuna:
> > I haven't had luck finding white papers or documentation discussing
> strategies for running ATS clusters across
> > multiple data centers or availability zones to support high availability?
> >
> > Any chance someone on here can point me in the right direction?
> >
> > Do people load balance multiple clusters with separate caches, or do
> they employ some mechanism to maintain
> > consistency across regions?
>
> i would let caches alone, it get's filled alone and has no imprtant data
>
> if you only have one backend and some ATS for load-balancing you are
> completly fine - your spread your load and take a lot of away from
> the backend server possible resulting in you have no need for more
> backend servers - the high ability itself of the backend can by
> something like VMware HA or whatever doing failover
>
> if you need more than one backendserver you should take much
> more care to keep sessions between the backends synchronous
> to avoid troubles having one request of the same client on
> two different backend servers and log the client out
>
>
>

--047d7b87442cbdee1904f7427544
Content-Type: text/html; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr">I&#39;m actually not concerned about the backend availabil=
ity as I am about the gateways. If the gateways die nothing works. Here&#39=
;s a little more info in a scenario format:<div><br></div><div>Client reque=
sts object from ATS.=C2=A0</div>
<div>Request is handled by geographic load balancer and directed to a clust=
er in once data center &#39;A&#39;.</div>
<div>The cluster fetches the object and caches it.</div><div>Client makes r=
equest for object again.</div><div>Request is handled by geo load balancer =
and directed to cluster in data center &#39;B&#39;.</div><div>Client reques=
ts object purged from cache.</div>
<div>Purge request is handled by geo load balancer and directed to cluster =
in data center &#39;A&#39;</div><div>Cluster invalidates object.</div><div>=
<br></div><div>At this point the cache across &#39;A&#39; and &#39;B&#39; i=
s inconsistent. Further requests would either retrieve a &quot;stale&quot; =
object or fetch a new object and cache it. Both objects would persist.<br>
</div><div><br></div><div>Do folks just preach &quot;eventual consistency&q=
uot;, or handle this via some other means?</div><div><br></div><div><br></d=
iv></div><div class=3D"gmail_extra"><br><br><div class=3D"gmail_quote">On T=
hu, Apr 17, 2014 at 11:48 AM, Reindl Harald <span dir=3D"ltr">&lt;<a href=
=3D"mailto:h.reindl@thelounge.net" target=3D"_blank">h.reindl@thelounge.net=
</a>&gt;</span> wrote:<br>
<blockquote class=3D"gmail_quote" style=3D"margin:0 0 0 .8ex;border-left:1p=
x #ccc solid;padding-left:1ex"><br>
<br>
Am 17.04.2014 20:41, schrieb Bacon Tuna:<br>
<div><div class=3D"h5">&gt; I haven&#39;t had luck finding white papers or =
documentation discussing strategies for running ATS clusters across<br>
&gt; multiple data centers or availability zones to support high availabili=
ty?<br>
&gt;<br>
&gt; Any chance someone on here can point me in the right direction?<br>
&gt;<br>
&gt; Do people load balance multiple clusters with separate caches, or do t=
hey employ some mechanism to maintain<br>
&gt; consistency across regions?<br>
<br>
</div></div>i would let caches alone, it get&#39;s filled alone and has no =
imprtant data<br>
<br>
if you only have one backend and some ATS for load-balancing you are<br>
completly fine - your spread your load and take a lot of away from<br>
the backend server possible resulting in you have no need for more<br>
backend servers - the high ability itself of the backend can by<br>
something like VMware HA or whatever doing failover<br>
<br>
if you need more than one backendserver you should take much<br>
more care to keep sessions between the backends synchronous<br>
to avoid troubles having one request of the same client on<br>
two different backend servers and log the client out<br>
<br>
<br>
</blockquote></div><br></div>

--047d7b87442cbdee1904f7427544--