[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]

Re: descriptor lost by authorities?

On Wed, Apr 01, 2009 at 06:59:22AM -0500, Scott Bennett wrote:
>      In the past several hours, traffic through my tor node has almost stopped.
> The last time an exit was taken through my node was over three and a half hours
> ago.  I see that the descriptor for my node, MYCROFTsOtherChild, is no longer
> in the consensus document nor was it ever distributed back to my node in
> directory updates.

Hm. moria1 isn't voting for it either, currently.

The last descriptor I have for it is from
published 2009-03-31 11:24:33
(GMT) which matches your first log lines.

>      The relevant info-level log entries from the last two descriptor
> updates posted are below.  Timestamps are for CDT.  The first update did work,
> but the next one at roughly 18 hours later appears to have been ignored or
> discarded by the authorities.  I would like to know why.
> Mar 31 06:24:37.522 [info] directory_post_to_dirservers(): Uploading an extrainfo (length 1978)
> Apr 01 00:25:03.710 [info] directory_post_to_dirservers(): Uploading an extrainfo (length 1936)

I'd like to know why also. :) There's no hint in moria1's logs about
rejecting it for whatever reason. Can you post more of your info-level
logs around the "Apr 01 00:25:03.710" timeframe? (If it's large, no need
to spam all 1000+ people on this list.)

The best theory I've heard so far for this phenomenon is that a brief
network hiccup around the time of publishing will make you fail to upload,
and you won't try for another 18 hours (meaning 12 hours of being missing
from the consensus).

We have some partial plans for having relays recognize when their upload
failed: see directories_have_accepted_server_descriptor() and look for
"rejected = 1" in connection_dir_client_reached_eof(), both in
but for various reasons we haven't finished that feature yet.

It would be good to confirm if this is what's going wrong, or if it's
something else. I'm going to put yet more debugging log messages in,
both on the relay side and the authority side, to help track this down.