ACE: How to differ server crash from HTTP status 5xx

[shumshi]

For business logic, I must regards as server crash as a normal case, that can not make the serverfarm's state be PROBE_FAILED.

[shumshi]

My application is a controller module, it should be designed as fail-safe, that means if the controller crash, it will do NOT effect the system.

if controller module return 500, the system will be suspended, otherwise(other HTTP status code, even the controller crash) the system will be continue in service.

Unfortunately, I did not find any good method to differ server crash from HTTP_STATUS_500, because both cases will make the serverfarm be PROBE_FAILED state.

[shumshi]

To implements my application requiement,  I think out some methods, all of them avoid the term as subject, but meet other problems:

if ACE does not support the feature as the subject, then we can convert the issue to another requirment to ACE using a logic.

some_serverfarm

  probe1

  probe2

Now ACE only support (probe1 & probe2)  or (proble1 | probe2)

Does ACE support (probe1 & ( ! probe2 )) ??

If that, probe2 is my controller probe, probe2 except status 500 599, if the controller server crash,  the NOT logic can protect the some_serverfram state as OK.

[Gilles Dufour]

we do not have an out-of-the-box solution for what you want.  We do not support probe1 and !probe2.

But you can implement a TCL script to do what you need.

I have to say however, this is a very weird requirement to allow traffic to go to a dead rserver.

Gilles.

[shumshi]

gdufour,

Thank your response, I wrote a script probe according HTTP_PROBE_SCRIPT_EXAMPLE.

Unfortunately, the performance is very bad. I config 30 http_script_probes, then them cost whole ACE CPU.

is it normal?

if ACE TCL performance is really not good, I also think out some other solution to avoid the issue, but meet other issue, I need your help

is there event-driver model to trigger the probe?

that also will help reduce those probes cost in most of time, only when rserver crash, them cost high CPU in short time, it is accepted

Or how TCL script access serverfarm's state directly(do NOT use invoke show command, that is also bad performance)??

if that, we can only config one script probe, then other SF can AND logic the CUSTOM_SF.

I appreciate any help

Thanks again