Re: Restart after poweroutage

Tom Lane wrote:
> Adrian Klaver <aklaver(at)comcast(dot)net> writes:
>> On Sunday 24 September 2006 09:17 am, Tom Lane wrote:
>>> Jon Lapham <lapham(at)jandr(dot)org> writes:
>>>> FATAL: pre-existing shared memory block (key 5432001, ID 65536) is
>>>> still in use
>>> This is extremely odd, because a shared memory block could not possibly
>>> have survived a reboot.
>
>> In the its a small world department I experienced the same problem shortly
>> after reading this message.
>
> I spent quite some time today trying to duplicate this failure (by
> pulling the plug on an up-to-date Fedora Core 5 machine). No luck.
> I suppose there is some contributing factor on your machines that
> we haven't identified yet ...

For what it is worth, I created a FC5 VMware installation and loaded my
database data into it. I simulated a bunch of power outages by telling
VMware to power off the vm. Is this a good simulation of a power
outage, or is there something inherently flawed about using a VM to test
this?

To stress test, I turned off the power while actively processing db
operations, such as: loading data into a database, create a new
database, delete contents of a large table within a transaction.

Anyway, in every case I could not reproduce the issue, postgresql came
back flawlessly. I am willing to run any power outage simulation tests
people may have.

--
-**-*-*---*-*---*-*---*-----*-*-----*---*-*---*-----*-----*-*-----*---
Jon Lapham <lapham(at)jandr(dot)org> Rio de Janeiro, Brasil
Personal: http://www.jandr.org/
***-*--*----*-------*------------*--------------------*---------------