You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@spamassassin.apache.org by bu...@bugzilla.spamassassin.org on 2008/01/15 20:53:39 UTC

[Bug 5790] New: dnsbl test failing on solaris 8

http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790

           Summary: dnsbl test failing on solaris 8
           Product: Spamassassin
           Version: 3.2.4
          Platform: Sun
        OS/Version: Solaris
            Status: NEW
          Severity: normal
          Priority: P5
         Component: Regression Tests
        AssignedTo: dev@spamassassin.apache.org
        ReportedBy: schulz@adi.com


The dnsbl test fails almost all of the time on Solaris 8.  It will sometimes
pass and sometimes fail with only one or two subtests failing, usually just
after a reboot, but most of the time close to half the subtests fail.  When I
try it on Solaris 10 it passes every time.  I have tried this on 2 different
Solaris 8 machines and 2 Solaris 10 machines.  All machines are running the
same perl (5.8.5) and are pointing to the same DNS server.  Perl was built on
Solaris 8 and installed on all the machines.  The make test for Net::DNS
passes on all machines. Looking at the log I see that a huge number of dns
queries are made, perhaps more that can be handled.  None the less, the dnsbl
test passes every time on Solaris 10.
I will attach the test output and the log file.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-01-15 13:13 -------
> so the resolv.conf and nsswitch.conf files are *identical* on the 2 machines?

I just double checked and the files are identical on all 4 machines (the 2
Solaris 8 and the 2 Solaris 10).



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From Mark.Martinec@ijs.si  2008-01-28 15:35 -------
> The dns debug output went into the log file created by the test. Here it is.
> I have not yet looked at it more than briefly.

At a first glance it seems the missing replies are also missing from
Net::DNS debug output.

Could you please check if your kernel/ip stack has a small setting for
UDP buffers, say below 18 kB? Don't know how it works on Solaris,
on FreeBSD the command would be:  sysctl net.inet.udp.recvspace



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From Mark.Martinec@ijs.si  2008-01-18 17:21 -------
> I still wonder why there are 100 queries for the 23 subtests and why there
> are queries to other than the spamassassin servers.

There are probably several rules active, each covering some RBL zone.
Seems normal.

I checked a handful of queries which are reported as timed out in the log
(as also listed in #11), looking at the first tcpdump you posted, which
I understand corresponds to a failing case. Assuming the log and the dump
correspond to the same event. I could find all the replies corresponding
to the queries (matching port number and a dns query ID number) - and not
even all returning NXdomain, some returning normal replies. So it seems
the DNS packets were not lost, but somehow did not make it back to the
program. No idea why. Perhaps instrumenting Net::DNS with a couple of
low level prints could reveal more.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] [review] dnsbl test failing on solaris 8 (and other OS with small rx udp buff size)

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790


sidney@sidney.com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Status Whiteboard|ready for release, seeking  |ready for release
                   |check with Windows          |




------- Additional Comments From sidney@sidney.com  2008-02-11 05:40 -------
It works under Windows and Cygwin, go ahead and close this bug.




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-01-17 14:13 -------
Some of the extra queries can be explained.  Bluegill, seahorse & crayfish
are the servers listed in resolv.conf.  Dolphin is the machine running the
test and running tcpdump.  When I run tcpdump without doing the dnsbl test,
I see occasional bursts of queries doing address to name and name to addrress
queries for all 4 machines.  Also, mackerel is the mail machine and I forgot
to shut down the mail checker program.  It checks every few minutes.

I still wonder why there are 100 queries for the 23 subtests and why there
are queries to other than the spamassassin servers.  Looking at the logs,
it looks like there are responses for the queries listed as timed out.

I did run tcpdump on the dns server and the test host at the same time.
They look to be the same just reversed.  I could attach them if you would
like to see them.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-01-16 12:04 -------
Created an attachment (id=4240)
 --> (http://issues.apache.org/SpamAssassin/attachment.cgi?id=4240&action=view)
dump when test passes

I continued running the test and got one of the rare cases where the test
passed with no errors.	Here is a tcpdump of that.  Note that I did not
invalidate the cache daemon cache this time.  After this test I immediately
ran the test again and it failed.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790


Mark.Martinec@ijs.si changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
Attachment #4255 is|0                           |1
           obsolete|                            |




------- Additional Comments From Mark.Martinec@ijs.si  2008-02-07 16:05 -------
Created an attachment (id=4256)
 --> (http://issues.apache.org/SpamAssassin/attachment.cgi?id=4256&action=view)
Slightly refined and hardened version of the 4255 patch, as in trunk

An updated patch to automatically increase SO_RCVBUF in DnsResolver.pm
if resolver's UDP socket buffer size is below 32k.

Rbl tests start failing below 17 kB buffer size, so 32 kB should
offer some margin. The default size is 57344 bytes on Solaris 10
(as reported by Tom in #19), is 42080 bytes on FreeBSD 6.2 and 7.0,
and is 124928 on some version of Linux that I tried, so 32 kB should
not be too much even for an elderly system.

Does the patch appear suitable for 3.2.5?




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From jeffc@surbl.org  2008-01-15 18:49 -------
Perhaps Solaris changed the way it handles resolv.conf and nsswitch.conf between
versions 8 and 10?  Sniffing DNS packets may show decisively where the DNS
traffic is going and could be informative.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8 (lack of system requirements document)

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790


Mark.Martinec@ijs.si changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|dnsbl test failing on       |dnsbl test failing on
                   |solaris 8                   |solaris 8 (lack of system
                   |                            |requirements document)
   Target Milestone|Undefined                   |3.2.5




------- Additional Comments From Mark.Martinec@ijs.si  2008-02-06 17:16 -------
> As to what to do with this bug, three things come to mind
> 1) Modify the test to do only the lookups really needed.  I expect that many
>    of the lookups are due the normal action of SpamAssassin when processing
>    the test messages which contain addresses that are not really part of the
>    test.

The number of DNS lookups triggered by this test message is pretty much
what can be expected from real life mail. It is not unusual that some
messages trigger even more lookups, although most messages do with
a bit less.

> 2) Leave the test as it is as it reveals an operating system limitation that
>    may well interfere with the proper operation of SpamAssassin.  Add some
>    information to the documention (wiki?) about how to fix this on releases
>    of Solaris older than Solaris 10 (I verified that Solaris 9 has the same
>    values as Solaris 8).

Right. This needs to be documented. Is there a system requirements
document somewhere? I checked the wiki and I find it pretty much
disorganized, stale and not helpful (I know, I know, the purpose
of a wiki is to go there and fix it...)

> 3) Just close the bug and let the bug be the documentation.

It is a (lack of) documentation bug anyway, should be resolved somehow.




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From Mark.Martinec@ijs.si  2008-02-07 07:01 -------
> How about adding to t/dnsbl.t in the BEGIN section something like
> onfail => sub {
>     warn "\n\nNote: A failure in dnsbl test may indicate ...

Reopened bug 3806 to be able to execute t/dnsbl.t.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From Mark.Martinec@ijs.si  2008-02-07 09:34 -------
Created an attachment (id=4255)
 --> (http://issues.apache.org/SpamAssassin/attachment.cgi?id=4255&action=view)
Automatically increase SO_RCVBUF in DnsResolver if below 32k

Tom, would you be so kind and see if this patch fixes the problem
even on systems with a small default udp receive buffer size.

The patch bumps up the buffer size on a resolver's socket to 32k
if smaller. On the debug output watch for something like:
  dbg: dns: UDP receive buffer size is 28000 bytes
  dbg: dns: UDP receive buffer size changed to 32768 bytes




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From Mark.Martinec@ijs.si  2008-01-16 16:55 -------
On a failing test a large proportion of NXDomain responses from a name server
never make it back to the querying host. Is there any misguided firewall
or router sitting between both hosts which might misinterpret a barrage of
NXDomain responses as some kind of a DoS attack and drop packets? Or an
equivalent DoS protection functionality on the host itself?

Capturing tcpdump of the same event (failing test) on the DNS server's side
of the network, and comparing both tcpdumps could show if the NXDomain 
responses were sent from DNS but never arrived at a querying host, or
the DNS server never replied to these queries.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-01-31 11:37 -------
I found out how to increase the udp buffer size and that does fix the problem
with the dnsbl test.  The kind people on the Solaris news group pointed me to
the ndd command and the Solaris documentation web site.  There are 3 parameters
changed between Solaris 8 and Solaris 10. They are:

udp_xmit_hiwat  changed from 8192 to 57344
udp_recv_hiwat  changed from 8192 to 57344
udp_max_buf     changed from 262144 to 2097152

It is udp_recv_hiwat that is the limiting parameter and changing that fixed
the problem.  It can be changed with the command:

ndd -set /dev/udp udp_recv_hiwat 57344

To make this permanent, you have to create a custom startup script with
the ndd command in it.

Note that the manual says:
    Note that an application can use setsockopt(3SOCKET) SO_RCVBUF to
    change the size for an individual socket. In general, you do not need
    to change the default value.
But I expect that SpamAssassin can not do that itself.

As to what to do with this bug, three things come to mind
1) Modify the test to do only the lookups really needed.  I expect that many
   of the lookups are due the normal action of SpamAssassin when processing
   the test messages which contain addresses that are not really part of the
   test.
2) Leave the test as it is as it reveals an operating system limitation that
   may well interfere with the proper operation of SpamAssassin.  Add some
   information to the documention (wiki?) about how to fix this on releases
   of Solaris older than Solaris 10 (I verified that Solaris 9 has the same
   values as Solaris 8).
3) Just close the bug and let the bug be the documentation.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-02-08 06:41 -------
I just tried the new patch and it works.  On Solaris 8 the log message is:

dbg: dns: resolver socket rx buffer size changed from 8192 to 32768 bytes

and on Solaris 10:

dbg: dns: resolver socket rx buffer size is 57344 bytes



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-01-15 11:55 -------
Created an attachment (id=4238)
 --> (http://issues.apache.org/SpamAssassin/attachment.cgi?id=4238&action=view)
Test output and test log




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-01-17 11:07 -------
> On a failing test a large proportion of NXDomain responses from a name server
> never make it back to the querying host. Is there any misguided firewall
> or router sitting between both hosts which might misinterpret a barrage of
> NXDomain responses as some kind of a DoS attack and drop packets? Or an
> equivalent DoS protection functionality on the host itself?

Perhaps I am not familiar enough with tcpdump, but when I look at the tcpdump
output it looks to me that there are exactly the same number of responses as
there are queries.  So I don't quite understand how one can conclude that
there are large proportion of NXDomain responses from a name server that
never make it back to the querying host.  Note that tcpdump was running on
the machine running the dnsbl test (the querying host).  In any case, there
is nothing between the two machines.  All  the machines that I have been
testing on are connected to the same ethernet switch.  Also, there is no
special softwere running on the failing test machine.  Perhaps Solaris 8
has something like that built in.  In any case, looking at the logs it looks
to me as if the responses are reaching the test machine but are not being
received by SpamAssassin.  Perhaps Solaris 8 can not handle a large number
of responses in a short period of time.  I am wondering why there are so many
queries.  I had thought that the dnsbl test would only query the spamassassin
server.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-01-28 13:28 -------
Created an attachment (id=4247)
 --> (http://issues.apache.org/SpamAssassin/attachment.cgi?id=4247&action=view)
Test log with dns debug

The dns debug output went into the log file created by the test. Here it is.
I have not yet looked at it more than briefly.	I will look at it more
closely as soon as I can.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From jm@jmason.org  2008-01-15 12:29 -------
so the resolv.conf and nsswitch.conf files are *identical* on the 2 machines?



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From Mark.Martinec@ijs.si  2008-02-07 10:21 -------
$ svn -m 'DnsResolver: bump up resolver receive buffer size to 32kB
          if smaller, bug 5790' ci
Sending lib/Mail/SpamAssassin/DnsResolver.pm
Committed revision 619543.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] [review] dnsbl test failing on solaris 8 (and other OS with small rx udp buff size)

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From Mark.Martinec@ijs.si  2008-02-10 17:04 -------
$ svn -m 'DnsResolver: bump up resolver receive buffer size to 32kB
          if smaller, bug 5790' ci
Sending lib/Mail/SpamAssassin/DnsResolver.pm
Committed revision 620364.

Committed to 3.2 to facilitate further testing. Keeping it open
for a little while longer, hoping for someone to try it on Windows.

I tested it on FreeBSD 6.2 and 7.0 through several settings of a
system-wide default buffer sizes.

If someone wants to try a simple command-line code simulation below
on new platforms, even without a SpamAssassin context, it would suffice:

perl -le 'use Socket; use IO::Socket::INET;
  $sock=IO::Socket::INET->new(PeerAddr=>"127.0.0.1", PeerPort=>53,
  Proto=>"udp") or die "Err: $!"; $s=$sock->sockopt(Socket::SO_RCVBUF);
  print $s; $sock->sockopt(Socket::SO_RCVBUF,32*1024) or die "Err2: $!";
  $s=$sock->sockopt(Socket::SO_RCVBUF); print $s'



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8 (lack of system requirements document)

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From jm@jmason.org  2008-02-07 01:31 -------
+1 to fixing the documentation.

I think this is sufficiently hard to diagnose that it deserves doc'ing in two
places: 1. in the INSTALL file and 2. in the online FAQ on the wiki.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From Mark.Martinec@ijs.si  2008-01-17 13:22 -------
> when I look at the tcpdump output it looks to me that there are
> exactly the same number of responses as there are queries.

You are right. I was comparing both files side by side in parallel
windows and this was my impression (the NXDomain in chunks on one
side but not on the other), but I was wrong.

The first file has 105 queries and as many replies,
the second dump has 89 queries and 89 replies.

A diff on the sorted list of queries from both files seems interesting,
although I can't see what it tells us:

$ pcregrep '> .*\.domain:' 0.lis | \
  perl -ne 'print "$1\n" if /\b([A-Z]+\?\s+\S+)/' | sort >00.lis
$ pcregrep '> .*\.domain:' 1.lis | \
  perl -ne 'print "$1\n" if /\b([A-Z]+\?\s+\S+)/' | sort >11.lis
$ diff -U1 00.lis 11.lis
-A? angler.adi.com.
-A? bluegill.adi.com.
-A? bluegill.adi.com.
-A? crayfish.adi.com.
-A? dolphin.adi.com.
-A? dolphin.adi.com.
+A? bluefish.adi.com.
-A? od-in-f164.google.com.
-A? od-in-f165.google.com.
+A? mackerel.adi.com.
-A? od-in-f167.google.com.
-A? pagead2.googlesyndication.com.
-A? seahorse.adi.com.
...
-NS? ebay.com.
+NS? comcast.net.
+NS? sourceforge.net.
...
-NS? w3.org.
-PTR? 101.2.168.192.in-addr.arpa.
-PTR? 126.2.168.192.in-addr.arpa.
-PTR? 131.2.168.192.in-addr.arpa.
-PTR? 134.2.168.192.in-addr.arpa.
-PTR? 164.161.233.64.in-addr.arpa.
-PTR? 165.161.233.64.in-addr.arpa.
+PTR? 144.2.168.192.in-addr.arpa.
-PTR? 167.161.233.64.in-addr.arpa.
-PTR? 95.2.168.192.in-addr.arpa.

and here is a list of timed-out queries from your log:

timing: 31.761 X dns:A:14.35.17.212.iadb.isipp.com.
timing: 31.768 X dns:A:226.149.120.193.zen.spamhaus.org.
timing: 31.771 X dns:A:18.13.119.61.zen.spamhaus.org.
timing: 31.774 X dns:A:134.88.73.210.zen.spamhaus.org.
timing: 31.777 X dns:A:98.3.137.144.zen.spamhaus.org.
timing: 31.792 X dns:TXT:14.35.17.212.sb.dnsbltest.spamassassin.org.
timing: 31.796 X dns:TXT:226.149.120.193.sb.dnsbltest.spamassassin.org.
timing: 31.799 X dns:TXT:18.13.119.61.sb.dnsbltest.spamassassin.org.
timing: 31.802 X dns:TXT:134.88.73.210.sb.dnsbltest.spamassassin.org.
timing: 31.806 X dns:TXT:98.3.137.144.sb.dnsbltest.spamassassin.org.
timing: 31.811 X dns:TXT:14.35.17.212.dnsbltest.spamassassin.org.
timing: 31.814 X dns:TXT:226.149.120.193.dnsbltest.spamassassin.org.
timing: 31.817 X dns:TXT:18.13.119.61.dnsbltest.spamassassin.org.
timing: 31.820 X dns:TXT:134.88.73.210.dnsbltest.spamassassin.org.
timing: 31.824 X dns:TXT:98.3.137.144.dnsbltest.spamassassin.org.
timing: 31.829 X dns:A:226.149.120.193.dnsbltest.spamassassin.org.
timing: 31.832 X dns:A:18.13.119.61.dnsbltest.spamassassin.org.




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-02-07 14:26 -------
The test now passes without my changing the buffer size. The log output
has several of

[4371] dbg: dns: UDP receive buffer size is 8192 bytes
[4371] dbg: dns: UDP receive buffer size changed to 32768 bytes

in it, one for each server that it verifies before the test is actually run.

I suppose that now there will be no message with 'blah blah' in it.  Oh well.
Of course it would have had to be 'blah blah blah' to be correct!



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From felicity@apache.org  2008-01-15 12:04 -------
*** Bug 5791 has been marked as a duplicate of this bug. ***



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From Mark.Martinec@ijs.si  2008-01-27 16:10 -------
> > Perhaps instrumenting Net::DNS with a couple of low level prints could
> > reveal more.
> 
> Probably what is needed but beyond what I can do without help.

Looks like the Net::DNS is already instrumented with some debugging
printouts, which unfortunately go to stdout (not to stderr), but
may still be useful when spamassassin is called from a command line.
To get there enabled, add debug=>1 as an argument to a call to
Net::DNS::Resolver->new in Mail/SpamAssassin/DnsResolver.pm
in sub load_resolver, e.g.:

$self->{res} = Net::DNS::Resolver->new(force_v4=>$force_ipv4, debug=>1);




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-01-30 14:17 -------
> Could you please check if your kernel/ip stack has a small setting for
> UDP buffers, say below 18 kB? Don't know how it works on Solaris,
> on FreeBSD the command would be:  sysctl net.inet.udp.recvspace

I asked on the Solaris news group and was pointed to the netperf utility,
which I had to go and get.  According to netperf, Solaris 8 has a buffer
size of 8192 bytes and Solaris 10 has a buffer size of 57344.  That would
explain it, especially when all the answers come back in a burst.  It may
also explain why the test sometimes passes.  If the dns server has to go
get the answers then the responses could come in slowly enough for the
dnsbl test to keep the buffer from filling up.

I am trying to find out how to increase the buffer size so that I can find
out if we really know what is going on.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From jm@jmason.org  2008-02-08 02:01 -------
(In reply to comment #28)
> Does the patch appear suitable for 3.2.5?

I think so.

what OSes have you tried it on? I'm slightly worried about portability (although
the eval {} scope should help there).



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From sidney@sidney.com  2008-02-08 12:35 -------
Seems to work fine on Mac OS 10.5.1 (Intel).
+1 for including in 3.2.5

I would like to see what happens under Windows and under Cygwin. I'm not set up
to try that out right now.




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From jm@jmason.org  2008-01-16 01:38 -------
yes, a packet trace would be very informative.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-01-16 11:39 -------
Created an attachment (id=4239)
 --> (http://issues.apache.org/SpamAssassin/attachment.cgi?id=4239&action=view)
Tcpdump output

Well, it was time I had tcpdump available, so I went and got it.  I first
invalidated the cache daemon hosts cache with 'nscd -i hosts' and then ran
tcpdump with 'tcpdump -s 200 port domain > /tmp/pass1'.  There was no other
activity on the machine that would have caused name lookups.  If you need me
to run it with different switches or run it also on Solaris 10, let me know.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From jm@jmason.org  2008-02-07 03:58 -------
(In reply to comment #22)
> How about adding to t/dnsbl.t in the BEGIN section something like
> 
> onfail => sub {
>     warn "\n\nNote: A failure in dnsbl test may indicate that the OS UDP buffer
> size blah blah point to correct documentation\n\n";
>   }

yeah, that too ;)



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] [review] dnsbl test failing on solaris 8 (and other OS with small rx udp buff size)

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790


Mark.Martinec@ijs.si changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED




------- Additional Comments From Mark.Martinec@ijs.si  2008-02-11 06:12 -------
> It works under Windows and Cygwin, go ahead and close this bug.

Thanks. I also tried it with IPv6 sockets. Closing. 





------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] dnsbl test failing on solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From schulz@adi.com  2008-01-22 13:00 -------
> Assuming the log and the dump
> correspond to the same event. I could find all the replies corresponding
> to the queries (matching port number and a dns query ID number) - and not
> even all returning NXdomain, some returning normal replies. So it seems
> the DNS packets were not lost, but somehow did not make it back to the
> program.

The log and dump were from different runs but with the same subtests failing.
Just to be sure I just made a matched set and now that I see how to read the
logs I checked for lost packets.  For every response that was claimed to be
not answered I found both the query and the response.  The whole set of
queries and responses happened within a 2 second window.  It does seem that
the packets are not delivered to the test program.  That is why I was
wondering about the number of queries.  It may be that something in Solaris 8
does not like such a large number of packets is that short of a time.

> Perhaps instrumenting Net::DNS with a couple of low level prints could
> reveal more.

Probably what is needed but beyond what I can do without help.




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790


jm@jmason.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|dnsbl test failing on       |update INSTALL docs to note
                   |solaris 8 (lack of system   |DNSBL usage bug in Solaris 8
                   |requirements document)      |






------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] [review] dnsbl test failing on solaris 8 (and other OS with small rx udp buff size)

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790


Mark.Martinec@ijs.si changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|update INSTALL docs to note |[review] dnsbl test failing
                   |DNSBL usage bug in Solaris 8|on solaris 8 (and other OS
                   |                            |with small rx udp buff size)
  Status Whiteboard|                            |ready for release, seeking
                   |                            |check with Windows






------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From jm@jmason.org  2008-02-08 02:02 -------
+1, if you're looking for votes to apply to 3.2.5



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug 5790] update INSTALL docs to note DNSBL usage bug in Solaris 8

Posted by bu...@bugzilla.spamassassin.org.
http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5790





------- Additional Comments From sidney@sidney.com  2008-02-07 03:47 -------
How about adding to t/dnsbl.t in the BEGIN section something like

onfail => sub {
    warn "\n\nNote: A failure in dnsbl test may indicate that the OS UDP buffer
size blah blah point to correct documentation\n\n";
  }





------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.