Mailing List Archive

All,

On 11/29/22 09:20, Christopher Schultz wrote:
> All,
>
> I've recently come back to bug 64338
> (https://bz.apache.org/bugzilla/show_bug.cgi?id=64338) which is an
> enhancement I filed back in 2020 with a patch. I got some good feedback
> and I have an updated patch (not yet posted) but I'm finding that while
> there are no errors or crashes, the behavior I'm seeing is just not what
> I'm expecting.
>
> Could someone with mod_proxy dev experience have a quick look at the
> existing patch on the bug and my most-recent comment?
>
> It looks to me like the balancer worker doesn't have the expected
> status. Maybe I'm looking in the wrong place?

*facepalm*

I got confused between my localhost and remote hosting environments. I
had reconfigured my localhost and was testing the results in the remote.

:D

Nothing to see here.

Except, I now have a question about how mod_proxy works versus mod_jk
(which is where I'm coming from, trying to get mod_proxy to have
feature-parity with mod_jk before I switch).

When we set a mod_jk worker to "disabled", it means that the worker will
veto any new requests that don't include a session identifier. Any
session identifier present (valid or not) will be routed through the
appropriate worker for that session identifier. (With Java-based app
servers, this is a JSESSIONID=sessionid.workername cookie or
;jsessionid=sessionid.workername URL parameter.)

I'm finding that when I put my (one and only one) mod_proxy worker into
"drain" or "disabled" mode, I get a 503 response instead of my (perhaps
incorrectly) expected usual pass-through to the back-end server.

Here's my httpd configuration:

<Proxy "balancer://myapp">
# worker.template.connection_pool_timeout=60
# worker.template.socket_timeout=300
BalancerMember "ajp://localhost:8215" timeout=300
ping=5 ttl=60
</Proxy>

ProxyPassMatch /myapp/(.+) balancer://myapp/myapp/$1
ProxyPassReverse /myapp/(.+) balancer://myapp/myapp/$1


Hmm. It looks like maybe I need to set the following:

- worker name
- session stickiness

(I have these in my localhost environment and not in my remote one. I'll
ensure they have the same config.)

Thanks!
-chris

On 11/29/22 6:04 PM, Christopher Schultz wrote:
> All,
>
> On 11/29/22 09:20, Christopher Schultz wrote:
>> All,
>>
>> I've recently come back to bug 64338 (https://bz.apache.org/bugzilla/show_bug.cgi?id=64338) which is an enhancement I filed back
>> in 2020 with a patch. I got some good feedback and I have an updated patch (not yet posted) but I'm finding that while there are
>> no errors or crashes, the behavior I'm seeing is just not what I'm expecting.
>>
>> Could someone with mod_proxy dev experience have a quick look at the existing patch on the bug and my most-recent comment?
>>
>> It looks to me like the balancer worker doesn't have the expected status. Maybe I'm looking in the wrong place?
>
> *facepalm*
>
> I got confused between my localhost and remote hosting environments. I had reconfigured my localhost and was testing the results
> in the remote.
>
> :D
>
> Nothing to see here.
>
> Except, I now have a question about how mod_proxy works versus mod_jk (which is where I'm coming from, trying to get mod_proxy to
> have feature-parity with mod_jk before I switch).
>
> When we set a mod_jk worker to "disabled", it means that the worker will veto any new requests that don't include a session
> identifier. Any session identifier present (valid or not) will be routed through the appropriate worker for that session
> identifier. (With Java-based app servers, this is a JSESSIONID=sessionid.workername cookie or ;jsessionid=sessionid.workername URL
> parameter.)
>
> I'm finding that when I put my (one and only one) mod_proxy worker into "drain" or "disabled" mode, I get a 503 response instead
> of my (perhaps incorrectly) expected usual pass-through to the back-end server.

When exactly do you get a 503?

Request with session id and worker info, worker in drain mode?
Request with session id and worker info, worker in disabled mode?
Request without session id and worker info, worker in drain mode?
Request without session id and worker info, worker in disabled mode?

>
> Here's my httpd configuration:
>
>         <Proxy "balancer://myapp">
>                 # worker.template.connection_pool_timeout=60
>                 # worker.template.socket_timeout=300
>                 BalancerMember "ajp://localhost:8215" timeout=300 ping=5 ttl=60
>         </Proxy>
>
>         ProxyPassMatch /myapp/(.+) balancer://myapp/myapp/$1
>         ProxyPassReverse /myapp/(.+) balancer://myapp/myapp/$1
>
>
> Hmm. It looks like maybe I need to set the following:
>
> - worker name
> - session stickiness

Yes. You need to set the route parameter of the BalancerMember to the worker name and the sticksession parameter on the
ProxyPassMatch or in the Proxy block via ProxySet.

Regards

Rüdiger