Solaris

I never wrote what reborg wrote was incorrect. I wrote one blog statement quoted there was misleading and you definitely demonstrated it to be misleading.

I strongly disagree about your "fallacy" point. Many Solaris administrators undersize their swap misunderstanding the crash dump requirements.
Please don't insist misleading people. Most Solaris installations, I bet more than 99.9%, use the swap partition as their crash dump storage area.
I'm glad you finally get it. I try hard to explain but failed to convinced you ...

Rereading the thread...thanks to Perderabo for clarifiying the dump/savecore distinction. I ran the two together which isn't really correct, I was implying the overall process for going from panic to having a saved core, and mostly to the coredump part of the process.

Ignroance of alternatives is not the same a necessity.

In the context of this thread I think that it is clear that the comment is accurate.

While the vast majority of systems will use swap as a dump area I *know* that it is not 99.9% of systems, also it still does not make it necessary which was the original point.

One other thing I also delberately avoided mentioning earlier so as not to muddy the water even further. Neo was sort of correct, despite misunderstanding the behaviour. The dump device can be in a filesystem if a file is used.

Code:

-bash-3.00# mkdir -p /dumps/crashes
-bash-3.00# mkfile 2g /dumps/crashes/dumpfile
-bash-3.00# dumpadm -d /dumps/crashes/dumpfile
      Dump content: kernel pages
       Dump device: /dumps/crashes/dumpfile (dedicated)
Savecore directory: /var/crash/statler
  Savecore enabled: yes

In this situation the file is a valid input for the -f savecore option.

Hmm, I stand corrected

I first checked on my system but the command failed:

Code:

dumpadm : dumps not supported on /dump/crashes/dumpfile

I checked further and found I was hitting a bug still unfixed as I was using ZFS to store the dumpfile. So indeed, it is possible to have a crash dump written to a file located on UFS, and safely written given the fact the filesystem regular functions are bypassed.

It is unfortunate the manual page doesn't clearly tell about this possibility.

Next time I'll read the kernel source code before posting

How can you *know* it ?

Do you have a private backdoor to access all Solaris installations and make statistics ?

I have to be careful here so please forgive the vagueness becasue I'm running into NDA territory. The company I work for consumes more than 5% of Suns total system output internally, additionally we are the a top OEM of Sun servers globally and have been in both of these situations for quite a number of years. None of the systems we deploy internally or externally use swap as a dump device. This install base accounts for much more than .2% of all Solaris installations according to Sun figures.

Sorry to put you straight with such an obscure dump device I think this may actually be a deliberate omission from the manpage. I haven't tried it on ZFS for a long time. The original Solaris Express bug for this was written in early 2004. Was it even fixed to allow a zvol to be used?

Also for reasons of being pedantic for others reading this thread, I will add a note. It appears you are using either OpenSolaris or a Nevada built. Technically neither of these is Solaris despite using SunOS 5.11 at their core. Features and functions from these are not necessarily ever going to be seen in Solaris proper, though usually they will; equally not all of the Solaris features will always be in these product.