Mailing List Archive

wackamole www tests?
Hi

As far as I can see, wackamole currently communicates via Spread, and takes
over IP's if it can not talk to the other server.

However, I would find it more useful if it could send HTTP GET request to
the other server and wait for a specific response to determine if the web
server is actually up. (since the server may be up but Apache could have
died).

Also, when I un-plugged the Ethernet cable from the server, neither of the 2
servers I was testing with seemed to notice.

Regards
Anthony
wackamole www tests? [ In reply to ]
On Tuesday, October 2, 2001, at 01:01 PM, Anthony Walker wrote:
> As far as I can see, wackamole currently communicates via Spread, and
> takes
> over IP's if it can not talk to the other server.

Bingo.

> However, I would find it more useful if it could send HTTP GET request
> to
> the other server and wait for a specific response to determine if the
> web
> server is actually up. (since the server may be up but Apache could have
> died).

Me too. It is in the works. There will be a pluggable module interface
where you can write your own Service Level Checks in C and they will be
loaded and run by wackamole. Of course, several useful samples (like
HTTP GETs) will accompany the feature.

> Also, when I un-plugged the Ethernet cable from the server, neither of
> the 2
> servers I was testing with seemed to notice.

You don't have two NICs do you? I think some massaging of Spread
configuration and/or wackamole code needs to take place in this case.
If Spread talks over on network and you pull the plug on the other,
wackamole will not notice. You can get around this with Service Level
Checks (when they are implemented).

I will have to think about the two multiple NIC problem. Basically,
right now, you need Spread to talk over the same ethernet interface that
the VIPs sit on.

So, there needs to be a clear difference between a machines REAL ip
address and the VIPs. They should not overlap at all. If the machine
is running and Spread is running and wackamole is not, Spread should be
talking over the ethernet interface you intend to put the VIPs on. This
way, if the machine interface it up, Spread can talk and wackamole
works. Wackamole should never manage the machines REAL IP.

Assuming that it is installed and operating correctly, give it a second
or two to failover. You can tune this down in Spread, but I don't
advise it.

--
Theo Schlossnagle
1024D/82844984/95FD 30F1 489E 4613 F22E 491A 7E88 364C 8284 4984
2047R/33131B65/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7
wackamole www tests? [ In reply to ]
Hi

> Me too. It is in the works.
Fantastic.

> You don't have two NICs do you?
No

However, you could implement a heartbeat sent via Spread as one of the
plugable modules to solve this issue and the dual NIC issue. However just
the HTTP GET will I'm sure be sufficient for most uses.

Once you have the plugable modules then I think the Wackamole and Backhand
combination will probably make for the best HA system around.

Regards
Anthony
wackamole www tests? [ In reply to ]
On Tuesday, October 2, 2001, at 03:09 PM, Anthony Walker wrote:
> However, you could implement a heartbeat sent via Spread as one of the
> plugable modules to solve this issue and the dual NIC issue.

Spread has its own "heartbeat." It has the idea of daemon membership,
so if any of the Spread daemons fall out of communication with each
other, then the wackamole will release/acquire IPs accordingly. If you
only have one NIC you should not be having any problems at all.

--
Theo Schlossnagle
1024D/82844984/95FD 30F1 489E 4613 F22E 491A 7E88 364C 8284 4984
2047R/33131B65/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7
wackamole www tests? [ In reply to ]
Hi,

It is important to point the base idea behind Wackamole. It bases its decision
on membership notifications of the Spread Toolkit (http://www.spread.org).

It is not useful to send heartbeats via Spread to check livelihood of Spread
programs. If Spread reports membership then this is it. Sending heartbeats through
Spread to check livelihood of Spread user programs that are not "there" as far as
Spread is concerned, will *always* fail.

The question is: do you trust Spread membership notifications enough to acquire
and release IP addresses? If the spread configuration is set correctly and the network
is not unusual, I would think so. Otherwise, (e.g. when the machines are very very busy
and Spread is not scheduled) I would recommend some parameter changes to the
out-of-the-box Spread. Doing that may require somewhat deeper knowledge of Spread itself.

You could improve Wackamole using some application-level information (such as HTTP GET)
with some coordination between the Wackamole instances using Spread Agreed order service.
I would not break the Wackamole/Spread current model of agreement because that completely
defies the consistency of the result and will definitely lead to cases where multiple
guys acquire the same IP address or some address not covered at all.

Cheers,

:) Yair.

Anthony Walker wrote:
>
> Hi
>
> > Me too. It is in the works.
> Fantastic.
>
> > You don't have two NICs do you?
> No
>
> However, you could implement a heartbeat sent via Spread as one of the
> plugable modules to solve this issue and the dual NIC issue. However just
> the HTTP GET will I'm sure be sufficient for most uses.
>
> Once you have the plugable modules then I think the Wackamole and Backhand
> combination will probably make for the best HA system around.
>
> Regards
> Anthony
>
> _______________________________________________
> wackamole-users mailing list
> wackamole-users@lists.backhand.org
> http://lists.backhand.org/mailman/listinfo/wackamole-users