[CUWiN-Dev] dangling pointers finally secured
Bill Comisky
bcomisky at pobox.com
Wed Dec 28 15:32:37 CST 2005
On Wed, 28 Dec 2005, Bill Comisky wrote:
> On Tue, 27 Dec 2005, Bill Comisky wrote:
>
>> On Mon, 19 Dec 2005, David Young wrote:
>>
>> > In the last few days I have found & fixed bugs involving dangling
>> > pointers
>> > that I believe were responsible for 99% of hslsd crashes.
>>
>> Just checked the status of how r3754 + 19-Dec snapshot is doing on our
>> Lawndale network. Neither of our tower gateways has been down since the
>> upgrade (5+ days). One gateway has 0 /var/core/hslsd.* files (woot!) and
>> the other had four core files (a vast reduction from previous revs). I
>> used the -S -S options to tar up the source/object files, and copied the
>> core files to here if you want to look:
>>
>> http://flatiron.cntwireless.org/cuwin/r3754/
>
> Turns out the the other gateway had generated a few core files too, but they
> had been deleted by a cron job I was using to free space on /mfs. It deleted
> the core files when it didn't really need to because it was looking at the
> capacity based on the output of df, which can be misleading because the size
> of /mfs is now dynamic since switching to tmpfs.
>
> Bill
BTW, we were seeing the /mfs usage vary quite a bit because of the verbose
logging to /var/log/daemon with the etx_any, etxpack_any, and hellow set
in hslsd_flags in in /etc/rc.d/route_conf. With a log rotation interval
of 10 minutes the daemon file was getting up to almost 10MB before
rotating. The load average was up around 2 as well, split between hslsd
and syslog (and gzip for compressing /var/log/daemon). I've turned these
off for our latest build; the logging should probably get reduced for the
distributed 0.6.0 release right?
bill
--
Bill Comisky
bcomisky at pobox.com
More information about the CU-Wireless-Dev
mailing list