Clustering the SP

[Cantor, Scott]

On 1/22/14, 4:30 AM, "Martin Haase" <Martin.Haase at DAASI.de> wrote:
>
>thanks for this one. No, we had not reported the bug yet to Ubuntu, but
>looks like we're forced to now. The application (moodle) ist clustered
>via some mySQL DBs and an haProxy on each machine. There's another
>CGI-based application clustered via shared NFS. Both of them are
>protected by the SP. As far as I can see, Apache worker is not an
>option, it's PHP again, where worker is (at least used to be, we'd have
>to check) not thread safe.

That means it's likely to be your problem, not a bug. The stack trace
suggested a mutex exhaustion issue, so short of maybe increasing kernel
resource counts, you have to avoid the thread explosion. If you use PHP,
then you need to get it out of process to avoid their bugs.

>What about the idea I mentioned, a failover via the TCPListener, do you
>see this is an option?

No. You will not find that it scales, the remoting was not meant to work
across machines. It slows applications with a lot of requests per page to
a crawl, a factor of 10 or more. Unless you don't use the SP session for
anything, which is the best choice actually. Use it to establish the app
session and then get it out of the way.

-- Scott