62575 - DNS: timeout error takes too long.

Status: VERIFIED INVALID

(Core :: Networking, defect, P3)

Description

[Cesar Eduardo Barros]

From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; Linux 2.4.0-test10 i686; en-US; m18) Gecko/20001208
BuildID:    2000120808

When I try to go to a host I didn´t go since starting up my box, mozilla is
prone to timeout its DNS requests (showing an annoying ¨host not found¨).
Retrying always works.

Reproducible: Sometimes
Steps to Reproduce:
1. Find a machine with a local DNS server with no forwarders and a slow link
with 200+ms latency and lots of packet loss
2. Restart the DNS server
3. Try to load a page in a hostname with lots of dots and a slow link to its
nameserver

Actual Results:  A timeout (saying "the host does not exist", bad mozilla) on
the first try, followed by immediate success on the next try (since it's already
on the cache on the local DNS server).

Expected Results:  Waited a bit longer (maybe twice as much).

This problem happens because the local DNS server has to ask the server for
every component in the host name (a sample sequence would be root, edu,
udel.edu, eecis.udel.edu). Since the link has lots of packet loss (20% is the
average), sometimes the packet or its answer is lost, which means the local
server has to wait for a timeout. This sometimes happen more than once. This is
enough to delay the recursive answer for so long that mozilla gives up. Next
time, not only the final answer is already in the cache, but all the
intermediate steps are too. I get this behavior about five times a day (when I
have a bad luck). Doubling the timeout would fix it.

Comment 1

[Matthias Versen [:Matti]]

I agree with the reporter.
Mozilla timeouts much faster than NS4.75.
I saw this sometimes while my network is very busy (download).

Comment 2

[Keyser Sose]

Yeah I see this a lot too. Marking NEW so someone up above will look at it.
Maybe an option to set it higher or something.

Comment 3

[neeti]

dns bug to darin

Comment 4

[Darin Fisher]

This is not a mozilla bug... we just call gethostbyname for XP_UNIX DNS lookups.
You should therefore configure your /etc/resolv.conf file with the desired
timeout.  On linux, take a look at [man resolv.conf] for details.

Marking INVALID.

Comment 5

[Matthias Versen [:Matti]]

darin: My OS isn´t Linux.
I use Win2k and I see this sometimes with a busy network.

Comment 6

[Darin Fisher]

My point is that I don't see how this could be anything but the responsibility
of the host environment, be it linux or windows.  All we do is call
gethostbyname, which is implemented by the OS.  If this is incorrectly timing
out then you should look to the OS for ways to increase the associated timeout
value.

Comment 7

[Brendan Eich [:brendan]]

20% packet loss is horrendous, and tuning your DNS client and server code to
help cope is the least thing to do.

Someone says this happens more with Mozilla than with 4.x on the same platform,
yet both presumably use the platform's gethostbyname or equivalent.  If someone
can show reproducible steps to make Mozilla timeout where 4.x does not, please
reopen.  Otherwise I agree with Darin: the problem lies outside of Mozilla code.

/be

Comment 8

[benc]

VERIFIED/invalid.

The DNS error for a non-responsive server should be a timeout, not hostname not
found. That is discussedin bug 164715.

The reason we might take longer in some cases is that:

If you send a fqdn, it is not literalized, so it might be treated as a pqdn,
causing multiple requests (bug 124565).

If you send a hostname, domain guessing runs and walks through all your search
domains (bug 40082).

Since this describes a distinct problem related two several unrelated problems,
I'm just going to let it go as INVALID.