[thelist] SOLVED? Local Director checking web server, not application server

[Tony Brandner]

We managed to put a little something together thanks to http probes.
A quick example of our commands:

# Configure a probe with a 10 second interval
probe virtual 192.168.192.36:80:0:tcp http 3
probe real 192.168.192.13:80:0:tcp http 3
probe real 192.168.192.14:80:0:tcp http 3

# Configure the HTTP probe for non-default settings:
probehttp virtual 192.168.192.36:80:0:tcp file /index.cfm request GET
probehttp real 192.168.192.13:80:0:tcp file /index.cfm request GET
probehttp real 192.168.192.14:80:0:tcp file /index.cfm request GET

# Configure the probe configuration to display the threshold for HTTP:
probeconfig http 3

We're still tweaking it, but it looks promising. Right now we're struggling
with setting it up so that when CFMX comes back online the Local Director
allows connections again.

Thanks,

Tony

-----Original Message-----

Thanks Scott.

We do indeed have access to Cisco docs, and I've forwarded this through to
our network admin.

I assume that since the LD is simply looking for packet acknowledgement, we
can probably point it to a JRUN port of some sort, something that will
respond in the same way as CFMX. Anyone on the list have any ideas?

We'll get started on those docs. Thanks for pointing us in the right
direction and giving us some hope. If I get a chance, I'll post any
solutions here.

Tony

-----Original Message-----

> The problem we have is that when CFMX crashes, the web server is
> still
> running. When the Local Director checks to see if the server is
> responding
> OK, it receives a positive from the server and continues to direct
> traffic
> to the server. Visitors to this server receive errors.
>

Depending on the version of LD software, there are some (I
understand) extended checking you can do. We have LDs running here,
and the basic availability checking isn't a poll from the LD; it
takes the real IP packet destined for the site, shoots it at the
server it wants to use, and if it gets that packet ACKed, then the
server is In Service. If the packed dies, it goes through the
out-of-service logic (which depends on how you have things set up
with wait times and such). So without going through the extended
settings, the LD is really only checking the server at the IP stack
level --far below the httpd service.

I know Cisco's docs are worse than MS to wade through, but do you
have access to them?

Scott