root server, bounced mail & propogation speeds

[Joseph S D Yao]

On Wed, Jul 12, 2000 at 05:24:08PM +0000, Thomas Duterme wrote:
> My registrar really messed up and changed my nameservers to 2 non-existent
> names and IP addresses. I discovered this once mail stopped flowing into my
> site.  I immediately took action and got NS changed back, but as we all
> know, it takes 24 + hours for all root servers to roll over.  I have a
> couple questions for the list which I haven't been able to figure out:
> 
> 1) Why do the root servers all change over at different times?  I saw that
> it took almost 24 hours for all the servers to change over.   What rules
> are there for regulating this process.

None of which I'm aware.  The information is loaded into the A server,
and the others download it on a daily basis, I believe.  I suspect that
they would each try to schedule it during a time of least use, since
each GTLD is pretty large.  But each root and GTLD name server is run
by a different organization, and probably has its own maintenance
schedule, run in loose co-ordination with the others.

I have read that VeriSign/NSI has declared that it is going to take
over management of all root servers from the current "haphazard"
management by volunteer organizations - universities, societies, and
other companies.  Despite the fact that they are running their servers
well enough, this concept scares me.

> 2) Considering that the false nameservers my registar advertised propogated
> through the internet, what determines the cache on other name servers  (ie.
> there is no soa record or ttl for the domain since those false IP addresses
> don't even have a machine attached) 

The records in the GTLD server will have TTLs attached.  I believe that
no RR exists without a TTL.

> 3) What about mail?  Since my domain will have been acting funky for the
> entire time it takes Name service to propogate through, what happens to all
> the mail sent.  (please don't just say bounce)  I'm aware that some will
> bounce but don't most MTA's run on a retransmission algorithm..(I know
> qmail does)  Is this widely different per MTA?  How long is the avg. time
> an MTA will try before it gives up requeing the mail and giving the fatal
> bounce.

The better MTAs [sendmail, postfix] have this variable.  However, some
will bounce.  If the falsely posted mail server did not exist, then the
mail may be queued.  Some mail servers may continue to try to send it
to the false mail server, however.  Others will re-query for a good
mail server.  If the falsely posted mail server existed, then mail will
have been bounced.

I think you need to assume that almost all e-mail sent during that
period bounced, and just be happy if any made it through.

OBTW, if you are interested in a pointer to problems in qmail's
re-transmission algorithm, I can send you a pointer privately.

-- 
Joe Yao				jsdy at cospo.osis.gov - Joseph S. D. Yao
COSPO/OSIS Computer Support					EMT-B
-----------------------------------------------------------------------
This message is not an official statement of COSPO policies.