> Cherokee reports that information source poll is exhausted when all of them
> are off-line.  Most probably they are off-line because it couldn't connect
> to them (back-end server overload maybe?).

Sorry to be persistent with proxy issues.

Our backend servers are fine - far from overloaded - dual Opterons
with 16GB of RAM and gigabit ethernet. They are mostly idle - some
spikes on some page generations but nothing to be concerned about. For
better perspective, pages we generate range from 10 milliseconds to
around 400milliseconds -  on average around 100ms. They are fast, many
are seemingly instant.  I am sure other than Java garbage process that
the backends servers are always fast and responding just fine. Garbage
collection can/will cause pauses that last seconds depending on what
is actually in RAM and how far up RAM has gone consumption wise.

Having said this, Cherokee is choking on something and it isn't
garbage collection. We have something odd also going on where the
backends are NOT being detected again either. This is when we finally
get a total failure from Cherokee and it happily sends out 503 errors.

I believe the problem only occurs when Cherokee is ran for some time.
My guess is 12 hours, but if load goes up less.  It could be some
cumulative issue that builds over time - seems reasonable.

Rather stumped, as I haven't seen anything odd to pinpoint the problem.
_______________________________________________
Cherokee mailing list
[email protected]
http://lists.octality.com/listinfo/cherokee

Reply via email to