shared mem and panics when out of PV Entries

[Andrew Kinney]

On 26 Mar 2003, at 14:05, Terry Lambert wrote:

> It's to ensure an allocation of a PV Entry *DOES NOT FAIL*, even
> when you *are* completely starved.  That's the whole point: to
> move the KVA space pressure off onto some *other* system that
> *also* does not preallocate its resources, and expect that that
> failure will turn into a benign "Apache: fork failed" log message,
> instead of being a panic.
> 
> We want to get rid of the panic.
> 
> Once we know what works around something, we can fix it later,
> but the important part is to know what the something *is*, first.
> 
> Make sense?

Ok.  Now it makes perfect sense.  I guess I didn't understand the 
point of preallocating via MINPV until now.

My thanks to you and Mr. Sysoev for all the assistance  you've 
provided.

In particular, thanks for the in depth lesson in memory 
management.  It's not like there's much in the way of 
documentation out there, so all the information and explanations 
that you have provided have given me much more insight into the 
BSD memory mangement system than I had before.

At this point, I'm not so certain that I was right with regards to the 
root cause of the panic as you have provided a very thorough 
argument to the contrary.  I do think it is sound advice to take the 
approach you have suggested even if for no other reason than to 
determine definitively what is the root cause of the panic.

So, I think I'll do that little preallocation trick you showed me, turn 
on crash dumps (4GB crash dumps are a royal pain, but useful in 
this case), and unleash Apache.  If you're right, then I'll get some 
other panic or non-panic error message at some point.

I'll let you know how it goes.  It may be a few days until I get back 
to the list with results since I have to schedule this kind of work in 
advance due to the downtime involved.

As far as MySQL goes, I'll look into compiling it with FreeBSD 
native threads and replacing the prepackaged binary we're currently 
using.  That should allow me to bump up KVA_PAGES without 
incident if I understand correctly.  That appears to be much easier 
than recoding the Linux threads module.  

Sincerely,
Andrew Kinney
President and
Chief Technology Officer
Advantagecom Networks, Inc.
http://www.advantagecom.net