FreeBSD 6.1 Tor issues (Once More, with Feeling)

[Chuck Swiger]

Peter Thoenen wrote:
> --- Fabian Keil <freebsd-listen at fabiankeil.de> wrote:
[ ... ]
>> Is anyone on this list running a Tor node on FreeBSD 6.1-RELEASE
>> or later with similar or higher load?
> 
> I am hitting the same issue still Fabian.  I had that PR closed as
> "works for me" with insignificant testing.  I am still crashing (as
> before) but maybe only once every week or two instead of every couple
> hours with 6.1 RELEASE.  The PR really should be reopened.  Couple
> other folk have emailed me with similiar issues offline (and also spoke
> with it about me on IRC).

Well, having several people show similar problems will help track the issue 
down, if only by letting us examine common aspects (ie, this happens on SMP 
systems, it happens when people are using PF, or IPFW, it only happens to 
people using vr0, or rl0, or some other specific NIC, etc).

> I am still 99% sure this is NOT A TOR ISSUE!!!  I have spoken with many
> tor users on other platforms and the actual developers and this is not
> seen by any of them.  I can also recreate this crash NOT running tor
> but just generating a heavy load with freenet and i2p.

It's probably not a TOR issue, no.  I gather that you've already run the 
manufacturer's hardware diagnostics and something like prime95 or memtest86 
overnight or longer than 24 hours (ideally)...

> My gut feeling is still a network code regression between 5.x -> 6.x with the stack
> rewrite. I am at a loss how to troubleshoot this anymore (as noted in
> the PR and my earlier email).  I truly hope somebody (e.g. a developer)
> can shed some light on this issue or troubleshoot it.

It would also be interesting to know whether you can revert to running FreeBSD 
5.5 on the same hardware under the same workload and have it stay up for longer.

Put your dmesg(s), kernel config files, /etc/make.conf, and best efforts at 
logging the issue (serial console, running vmstat whatever or sysctl -a kern 
via cron periodicly to a file), on a webpage someplace, and try to cross-link 
with other people showing the same problem.  Post that URL to a PR and/or the 
mailing lists...

-- 
-Chuck