You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@trafficserver.apache.org by "Petzel, David" <Da...@disney.com> on 2011/11/18 15:32:11 UTC

Using Multiple/Redundant Backend Origins

Hello,
We are currently in the process of evaluating a new http caching solution and on that list is Apache Traffic Server. We've been reading through the online documentation and have a pretty good feel for the product, however there is one item we are a little clear on. It seems that mapping rules always assume a one-to-one relationship from host to origin.  We are interested in having multiple redundant origins behind a single host. In Varnish they call this a director, and I'm trying to understand what the equivalent in ATS would be.

We do understand that we could use a load balancer VIP as the origin to provide this functionality,  however if possible we're looking to have the "Caching Servers" communicate directly with the origin providers, rather than routing through a VIP.

Could someone point me at the appropriate documentation where this is described?

Thanks

RE: Using Multiple/Redundant Backend Origins

Posted by "Petzel, David" <Da...@disney.com>.
This  looks to be exactly what I was talking about,
Thank You!


-----Original Message-----
From: Eric Balsa [mailto:ericb@apache.org] 
Sent: Friday, November 18, 2011 11:01 AM
To: users@trafficserver.apache.org
Subject: Re: Using Multiple/Redundant Backend Origins

There's also http://svn.apache.org/viewvc/trafficserver/plugins/trunk/balancer/
which is recent code from Y!

Someone would need to take that code & run with it since it's a little bit raw right now. I dont think it does any health checking, etc but just rotates requests among the set of hosts described in the config.

--Eric

On Fri, Nov 18, 2011 at 8:29 AM, Jason Giedymin <ja...@gmail.com> wrote:
> I think most would suggest DNS round robin for more advanced 
> capability I believe some would suggest a custom plug-in On Nov 18, 
> 2011, at 10:27 AM, sridhar basam wrote:
>
>
> On Fri, Nov 18, 2011 at 9:55 AM, Petzel, David 
> <Da...@disney.com>
> wrote:
>>
>> This feels like it's about clustering ATS servers, not the origins 
>> behind it? Our design would like something like this:
>>
>> LOAD BALANCERS à POOL OF ATS SERVERS à POOL OF ORIGINS. In our case 
>> the origin is the application server.
>>
>>
>>
>> So we would have say an Active/Passive LB Pair as the externally 
>> presented host. Behind that VIP would be a pool of 3 or more ATS 
>> servers (for
>> redundancy) and each ATS server would use a POOL of origins.  From 
>> the examples I've seen it appears that there is an assumption by ATS 
>> that the Origin is a VIP, and thus its extracted itself from managing 
>> a pool of hosts for a single origin?
>>
>>
>>
>> Additionally I saw that there are mapping configurations for using 
>> multiple (different) origins. The ATS Server Pool, seems pretty 
>> straightforward, but specifically I'm trying to find how I would 
>> craft a mapping rule such multiple origins for the same mapping.
>
> You could put the origin's behind the LBs too or make the origin an 
> DNS round-robin record.
> I don't know what ATS would do with these records though, ie if it 
> does auto-rotation on lookups and if it fails from unresponsive/dead 
> IPs on a connect.
>   Sridhar
>

RE: Using Multiple/Redundant Backend Origins

Posted by "Petzel, David" <Da...@disney.com>.
OK Cool Thanks. I did note that you mentioned it needed work, I just didn't draw the dots that it meant it wouldn't run under version 3. I thought it just needed some of the items you had listed. In any event, thanks for the clarification and I really do appreciate the responses and feedback I've gotten so far.

-----Original Message-----
From: Leif Hedstrom [mailto:zwoop@apache.org] 
Sent: Monday, November 21, 2011 10:09 PM
To: users@trafficserver.apache.org
Cc: Petzel, David
Subject: Re: Using Multiple/Redundant Backend Origins

On 11/21/11 7:06 PM, Petzel, David wrote:
> I don't know a lick of C so bear with me if this is a dumb question.  I was reading through the docs on plugins and I'm getting the feeling this plugin is for ATS 2x as opposed to 3x (which I downloaded)?
>
> balancer.cc:44:22: error: RemapAPI.h: No such file or directory
>
> I feel like version 3 plugins use ts/remap.h instead?
>
> Again sorry if this is a dumb question.

Correct, as Eric (and I) pointed out, it needs some work. Yahoo still uses the "v2" version, the plugins were open sourced as-is. Without the missing support for health checks etc., this particular plugin is fairly useless, hence, I haven't bothered working on it (yet).


-- leif


Re: Using Multiple/Redundant Backend Origins

Posted by Leif Hedstrom <zw...@apache.org>.
On 11/21/11 7:06 PM, Petzel, David wrote:
> I don't know a lick of C so bear with me if this is a dumb question.  I was reading through the docs on plugins and I'm getting the feeling this plugin is for ATS 2x as opposed to 3x (which I downloaded)?
>
> balancer.cc:44:22: error: RemapAPI.h: No such file or directory
>
> I feel like version 3 plugins use ts/remap.h instead?
>
> Again sorry if this is a dumb question.

Correct, as Eric (and I) pointed out, it needs some work. Yahoo still uses 
the "v2" version, the plugins were open sourced as-is. Without the missing 
support for health checks etc., this particular plugin is fairly useless, 
hence, I haven't bothered working on it (yet).


-- leif


RE: Using Multiple/Redundant Backend Origins

Posted by "Petzel, David" <Da...@disney.com>.
I don't know a lick of C so bear with me if this is a dumb question.  I was reading through the docs on plugins and I'm getting the feeling this plugin is for ATS 2x as opposed to 3x (which I downloaded)?

balancer.cc:44:22: error: RemapAPI.h: No such file or directory

I feel like version 3 plugins use ts/remap.h instead?

Again sorry if this is a dumb question.

-----Original Message-----
From: Leif Hedstrom [mailto:zwoop@apache.org] 
Sent: Friday, November 18, 2011 11:55 AM
To: users@trafficserver.apache.org
Cc: Eric Balsa
Subject: Re: Using Multiple/Redundant Backend Origins

On 11/18/2011 09:00 AM, Eric Balsa wrote:
> There's also 
> http://svn.apache.org/viewvc/trafficserver/plugins/trunk/balancer/
> which is recent code from Y!
>
> Someone would need to take that code&  run with it since it's a little 
> bit raw right now. I dont think it does any health checking, etc but 
> just rotates requests among the set of hosts described in the config.


Yeah, it's a bit worse than that :). The way this plugin worked at Yahoo was in combination with an out-of-bandwidth health checker and balancer. 
So, as it is, it's probably fairly useful. I know the Taobao team has changes to make active health checks in ATS, I'm just not sure how that will integrate with 'remapping' rules. There's something missing here, and we need to figure that out  :).

-- Leif


Re: Using Multiple/Redundant Backend Origins

Posted by Leif Hedstrom <zw...@apache.org>.
On 11/18/2011 09:00 AM, Eric Balsa wrote:
> There's also http://svn.apache.org/viewvc/trafficserver/plugins/trunk/balancer/
> which is recent code from Y!
>
> Someone would need to take that code&  run with it since it's a little
> bit raw right now. I dont think it does any health checking, etc but
> just rotates requests among the set of hosts described in the config.


Yeah, it's a bit worse than that :). The way this plugin worked at Yahoo 
was in combination with an out-of-bandwidth health checker and balancer. 
So, as it is, it's probably fairly useful. I know the Taobao team has 
changes to make active health checks in ATS, I'm just not sure how that 
will integrate with 'remapping' rules. There's something missing here, 
and we need to figure that out  :).

-- Leif


Re: Using Multiple/Redundant Backend Origins

Posted by Eric Balsa <er...@apache.org>.
There's also http://svn.apache.org/viewvc/trafficserver/plugins/trunk/balancer/
which is recent code from Y!

Someone would need to take that code & run with it since it's a little
bit raw right now. I dont think it does any health checking, etc but
just rotates requests among the set of hosts described in the config.

--Eric

On Fri, Nov 18, 2011 at 8:29 AM, Jason Giedymin
<ja...@gmail.com> wrote:
> I think most would suggest DNS round robin
> for more advanced capability I believe some would suggest a custom plug-in
> On Nov 18, 2011, at 10:27 AM, sridhar basam wrote:
>
>
> On Fri, Nov 18, 2011 at 9:55 AM, Petzel, David <Da...@disney.com>
> wrote:
>>
>> This feels like it’s about clustering ATS servers, not the origins behind
>> it? Our design would like something like this:
>>
>> LOAD BALANCERS à POOL OF ATS SERVERS à POOL OF ORIGINS. In our case the
>> origin is the application server.
>>
>>
>>
>> So we would have say an Active/Passive LB Pair as the externally presented
>> host. Behind that VIP would be a pool of 3 or more ATS servers (for
>> redundancy) and each ATS server would use a POOL of origins.  From the
>> examples I’ve seen it appears that there is an assumption by ATS that the
>> Origin is a VIP, and thus its extracted itself from managing a pool of hosts
>> for a single origin?
>>
>>
>>
>> Additionally I saw that there are mapping configurations for using
>> multiple (different) origins. The ATS Server Pool, seems pretty
>> straightforward, but specifically I’m trying to find how I would craft a
>> mapping rule such multiple origins for the same mapping.
>
> You could put the origin's behind the LBs too or make the origin an DNS
> round-robin record.
> I don't know what ATS would do with these records though, ie if it does
> auto-rotation on lookups and if it fails from unresponsive/dead IPs on a
> connect.
>   Sridhar
>

Re: Using Multiple/Redundant Backend Origins

Posted by Jason Giedymin <ja...@gmail.com>.
I think most would suggest DNS round robin

for more advanced capability I believe some would suggest a custom plug-in

On Nov 18, 2011, at 10:27 AM, sridhar basam wrote:

> 
> 
> On Fri, Nov 18, 2011 at 9:55 AM, Petzel, David <Da...@disney.com> wrote:
> This feels like it’s about clustering ATS servers, not the origins behind it? Our design would like something like this:
> 
> LOAD BALANCERS à POOL OF ATS SERVERS à POOL OF ORIGINS. In our case the origin is the application server.
> 
>  
> 
> So we would have say an Active/Passive LB Pair as the externally presented host. Behind that VIP would be a pool of 3 or more ATS servers (for redundancy) and each ATS server would use a POOL of origins.  From the examples I’ve seen it appears that there is an assumption by ATS that the Origin is a VIP, and thus its extracted itself from managing a pool of hosts for a single origin?
> 
>  
> 
> Additionally I saw that there are mapping configurations for using multiple (different) origins. The ATS Server Pool, seems pretty straightforward, but specifically I’m trying to find how I would craft a mapping rule such multiple origins for the same mapping.
> 
> 
> You could put the origin's behind the LBs too or make the origin an DNS round-robin record. 
> 
> I don't know what ATS would do with these records though, ie if it does auto-rotation on lookups and if it fails from unresponsive/dead IPs on a connect.
> 
>   Sridhar 


Re: Using Multiple/Redundant Backend Origins

Posted by sridhar basam <sr...@basam.org>.
On Fri, Nov 18, 2011 at 9:55 AM, Petzel, David <Da...@disney.com>wrote:

> This feels like it’s about clustering ATS servers, not the origins behind
> it? Our design would like something like this:****
>
> LOAD BALANCERS à POOL OF ATS SERVERS à POOL OF ORIGINS. In our case the
> origin is the application server.****
>
> ** **
>
> So we would have say an Active/Passive LB Pair as the externally presented
> host. Behind that VIP would be a pool of 3 or more ATS servers (for
> redundancy) and each ATS server would use a POOL of origins.  From the
> examples I’ve seen it appears that there is an assumption by ATS that the
> Origin is a VIP, and thus its extracted itself from managing a pool of
> hosts for a single origin?****
>
> ** **
>
> Additionally I saw that there are mapping configurations for using
> multiple (different) origins. The ATS Server Pool, seems pretty
> straightforward, but specifically I’m trying to find how I would craft a
> mapping rule such multiple origins for the same mapping.
>

You could put the origin's behind the LBs too or make the origin an DNS
round-robin record.

I don't know what ATS would do with these records though, ie if it does
auto-rotation on lookups and if it fails from unresponsive/dead IPs on a
connect.

  Sridhar

RE: Using Multiple/Redundant Backend Origins

Posted by "Petzel, David" <Da...@disney.com>.
This feels like it's about clustering ATS servers, not the origins behind it? Our design would like something like this:
LOAD BALANCERS --> POOL OF ATS SERVERS --> POOL OF ORIGINS. In our case the origin is the application server.

So we would have say an Active/Passive LB Pair as the externally presented host. Behind that VIP would be a pool of 3 or more ATS servers (for redundancy) and each ATS server would use a POOL of origins.  From the examples I've seen it appears that there is an assumption by ATS that the Origin is a VIP, and thus its extracted itself from managing a pool of hosts for a single origin?

Additionally I saw that there are mapping configurations for using multiple (different) origins. The ATS Server Pool, seems pretty straightforward, but specifically I'm trying to find how I would craft a mapping rule such multiple origins for the same mapping.

From: Jason Giedymin [mailto:jason.giedymin@gmail.com]
Sent: Friday, November 18, 2011 9:45 AM
To: users@trafficserver.apache.org
Subject: Re: Using Multiple/Redundant Backend Origins

I believe what your looking for is clustering ATS which serve multiple origins (in round robin/ etc) manner.

https://cwiki.apache.org/confluence/display/TS/Clustering

There are still some kinks in it but its getting better.

On Nov 18, 2011, at 9:32 AM, Petzel, David wrote:


Hello,
We are currently in the process of evaluating a new http caching solution and on that list is Apache Traffic Server. We've been reading through the online documentation and have a pretty good feel for the product, however there is one item we are a little clear on. It seems that mapping rules always assume a one-to-one relationship from host to origin.  We are interested in having multiple redundant origins behind a single host. In Varnish they call this a director, and I'm trying to understand what the equivalent in ATS would be.

We do understand that we could use a load balancer VIP as the origin to provide this functionality,  however if possible we're looking to have the "Caching Servers" communicate directly with the origin providers, rather than routing through a VIP.

Could someone point me at the appropriate documentation where this is described?

Thanks


Re: Using Multiple/Redundant Backend Origins

Posted by Jason Giedymin <ja...@gmail.com>.
I believe what your looking for is clustering ATS which serve multiple origins (in round robin/ etc) manner. 

https://cwiki.apache.org/confluence/display/TS/Clustering

There are still some kinks in it but its getting better.

On Nov 18, 2011, at 9:32 AM, Petzel, David wrote:

> Hello,
> We are currently in the process of evaluating a new http caching solution and on that list is Apache Traffic Server. We’ve been reading through the online documentation and have a pretty good feel for the product, however there is one item we are a little clear on. It seems that mapping rules always assume a one-to-one relationship from host to origin.  We are interested in having multiple redundant origins behind a single host. In Varnish they call this a director, and I’m trying to understand what the equivalent in ATS would be.
>  
> We do understand that we could use a load balancer VIP as the origin to provide this functionality,  however if possible we’re looking to have the “Caching Servers” communicate directly with the origin providers, rather than routing through a VIP.
>  
> Could someone point me at the appropriate documentation where this is described?
>  
> Thanks