Hi guys,

 

I was wondering if anyone could suggest me a way to cluster Pound. 

As it is becoming a central point of failure in my organisation I am looking
for ideas and feedback from your various experiences . 

I already checked several solutions like LVS / piranha solution for RH, as
well as DNS round robin (why not ?). 

 

Which would be  the best in terms of maintenance and reliability , as It is
aimed at a production network ? 

 

Thanks guys .

 

Yann

> I was wondering if anyone could suggest me a way to cluster Pound. [...]

I'm pretty sure this has been discussed in the archives.  Take a look there.

Personally I've got a hot-backup system based on OpenBSD using carp. 
It's pretty simple, but it ignores sticky sessions, which I'm not using 
anyway.

Regards,[...]

Carbonell, Yann wrote:[...]
Yann,

I would have thought HA-Linux or Keepalived for clustering Pound would 
make more sense.
If you want to load balance pound instances then LVS + HA-Linux or 
Keepalived.

Regards,
Malcolm.

Thanks for the input  Dave,

I'll look into the mail archive, to be honest I quickly checked it and I
might have missed the information.

Thanks

Yann.

www.banctec.com[...]

Hi,

But would the LVS server become a single point of failure too ? Let's say if
I put it in front of the Pound HA ?

Thanks
[...]

Carbonell, Yann napsal(a):[...]

We are using keepalived to switch between two pound servers. The 
keepalived is configured to transfer the IP address and fire up pound on 
the secondary machine, should the primary go off-line. It works 
smoothly, though we haven't actually had a real disaster yet (fingers 
crossed :) ), just tests. We use it on Linux.
[...]

we run pound setup with linux-ha active/passive for failover. use  
rsync to synchronize configs.

drawback to this is that when pound server dies connection state  
isn't replicated. Not such a huge issue as we use pound mainly as a  
transparent proxy and we use another pair of servers running IPVS +  
linux-ha for load balancing.

if your interested i can post the haresorces and the resorce  files  
to the list tho they aren't necessarily pretty.

On Jun 4, 2007, at 12:35 PM, Carbonell, Yann wrote:
[...]

We are currently using spread [1] and wackamole [2] to facilitate a HA 
pound environment and it is working quite well.  It is also very easy to 
setup.  We are pushing about 200Mbit/sec and 2500 req/sec through our 
load balancers, and have had no issues with regards to failover.  We 
don't use persistent sessions, so this solution does not address that.

There are a couple of things to keep in mind:

1) A failover will only be triggered if the heartbeat/ping between the 2 
servers fails.   You will probably want to write some sort of monitoring 
script that initiates a failover if the pound service is not responding 
but the server is still online.  This can easily be accomplished via 
something like monit[3]

2) `echo 1 > /proc/sys/net/ipv4/ip_nonlocal_bind` is your friend

Barry

[1] http://www.spread.org/
[2] http://www.backhand.org/wackamole/
[3] http://www.tildeslash.com/monit/


Carbonell, Yann wrote:[...]

> We are using keepalived to switch between two pound servers. The 
keepalived is configured to transfer the IP address and fire up pound on 
the secondary machine

What is the difference between keepalive and Linux HA ? They both transfer
the floating IP from one server to another in case of failure right .
Why keepalive more than Linux HA ? 


Thanks

Yann



-----Original Message-----
From: Michal Taborsky - Internet Mall [mailto:michal.taborsky(at)mall.cz] 
Sent: June 4, 2007 4:03 PM
To: pound(at)apsis.ch
Subject: Re: [Pound Mailing List] Pound Cluter ideas ?

Carbonell, Yann napsal(a):[...]

We are using keepalived to switch between two pound servers. The 
keepalived is configured to transfer the IP address and fire up pound on 
the secondary machine, should the primary go off-line. It works 
smoothly, though we haven't actually had a real disaster yet (fingers 
crossed :) ), just tests. We use it on Linux.
[...]

hi,

On 6/5/07, Carbonell, Yann <Yann.Carbonell(at)banctec.ca> wrote:[...]

we use wackamole and spread to ensure that the IPs we have in DNS are
always available if a pound load balancer goes down.

let me know if you want more detail.

cheers
justin

We use LVS in Direct Routing mode. In general, this works well, but one
thing to be aware of is the time between service checks is Direct Routing
mode - if one Pound node goes down between checks it's possible that a user
might get a timeout or 503 etc.

Cheers,
D

On 6/5/07, justin randell <justin.randell(at)gmail.com> wrote:[...]

Hi,

I'm also looking to setup an HA Pound cluster, possibly with wackamole, and
any pointers about configuration would be welcome. :)
[...]

Ok thanks for the input, 

But as I see it I would have to setup  another level of LVS-HA server on
front of two Pound proxy. Dod I get it right ?

Thanks

Yann.
[...]

Thanks  barry  ,


I checked the Spread/wackamole websites and it sounds very interesting, I
didn't know about those tools. But first I have to check if the product we
use for our web services requires session persistent , but I doubt it.
[...]

I am already using Monit and I get you point regarding the service restart. 
I will read more about Spread to understand the concdept and might contact
you later if I have further questions as I don't understand how Monit will
help if you have several backends . 
Do you have a sort of diagram that would illustrate the implementation ?.
Perhaps  I could find it on the wackamole site.

Yann
[...]

Carbonell, Yann wrote:[...][...]

Monit would be used to monitor the pound process or the responsiveness 
of port 80 on the LB itself, not the backends.  If pound were to stop 
responding, and would not restart, it would trigger an IP failover to 
your 2nd LB.  traffic routing to the backends is something pound takes 
care of.
[...]

Not really, but it's pretty simple.  We have n + 1 load balancers in an 
active-active configuration and we have many backends behind them. 
Using wackamole allows you to do n + 1 rather than having simply pairs 
of devices. Depending on the datacenter it can be 50-100 backends.

Barry
[...]
>>> Hi guys,
>>>
>>>
>>>
>>> I was wondering if anyone could suggest me a way to cluster Pound.
>>>
>>> As it is becoming a central point of failure in my organisation I
am[...]
>>> for ideas and feedback from your various experiences .
>>>
>>> I already checked several solutions like LVS / piranha solution
for RH,[...]
>>> well as DNS round robin (why not ?).
>>>
>>>
>>>
>>> Which would be  the best in terms of maintenance and reliability ,
as It[...]
>>> aimed at a production network ?
>>>
>>>
>>>
>>> Thanks guys .
>>>
>>>
>>>
>>> Yann
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>[...][...]