Authentication problem - solved, but maybe a bug in Mac OS X?

Question

Hi,
I've a rather small installation with only a handful of users configured on a Mac mini (Mac OS X Server, 10.6.8). All of them use the mail, calendar and addressbook server on the Mac, nothing more. They use it with Mac, iPhone and iPad. Everything worked fine for months but suddenly all of them were faced authentication problems: it was not possible to login on the imap server, the calendar server, the addressbook server. It was possible to login using the admin account on the server directly. Moreover, all users disappeared from the workgroup manager, however they still were available on the servers LDAP server and findable using ldapsearch.
First, I used to completely restart the server to solve the problem, but it reappeared after only few hours again.
Second, after understanding more about the authentication process, I found the "killall DirectoryService" was sufficient to solve the problem, but it still reappeared after few hours.
Then I found the, once the problem occured, there was nearly no more communication to the local LDAP server on port 389 on localhost. When everything was working fine, the was a lot of such communication, including queries for usernames, when a login attempt was made. I started a "tcpdump -n -i lo0 port 389" and waited for the problem again. After the problem occured, I found in the pcap files that there were a few final query attempts, actually attempts the open a port 389 TCP connection to the slapd running on localhost, which were answered with a TCP RST. Then, no more attempts were made until l restarted the DirectoryService. Using the logfile of the slapd I found that this happened exactly at the time the slapd was stopped and restarted. And - surprisingly for me - stopping and restarting the slapd happened exactly once an hour.
I then found that it happened exactly at the time the time machine backup process was started and indeed it was possible to trigger the event of restarting the slapd by manually starting a time machine backup.
(Indeed, I switched my backup strategy from SuperDuper to time machine the other day and maybe that was the time the problem occured for the first time. I know that time machine is not considered as the best backup strategy for a server but I wanted to try on my own.)
Google helped my to find a hint that time machine will actually stop and restart slapd - which is a generally a good idea, since otherwise a backup from some open database files would be made, which could work but may fail. So, I thing, someone of the developers thought about that problem too and has considered time machine for backups of a server.
However, a not running slapd can not answer queries from a DirectoryService and a stopping or starting process might indeed end up with TCP SYNs answered with TCP RST.
My solution was to disable time machine again and from that time the problem does not occur again.
I'm wondering why the DirectoryService process isn't starting to query the slapd again after a failed connection. Isn't this a bug? After this experience I consider time machine as not only the not preferred backup solution for a server but as completely incompatible with Mac OS X server - although, as I said, it seems that someone thought about backing up the LDAP database using time machine.
(On a Lion server this problem does not occur, the slapd will not be stopped and restarted when time machine is running. Moreover, I saw a com.apple.slapd.start notification in the slapd.log ... maybe this tells DirectoryService to try again.)
Cheers,
Wolfgang

Answer 1

Another problem I found with the MacOS X key bindings: the 6 key doesn't work!
In the config that ships with SQL Developer, I found this:
<Item class="oracle.javatools.util.Pair">
<first class="java.lang.String">DOCUMENT_6_CMD_ID</first>
<second class="oracle.ide.keyboard.KeyStrokes">
<data>
<Item class="javax.swing.KeyStroke">6</Item>
</data>
</second>
</Item>
which should be:
<Item class="oracle.javatools.util.Pair">
<first class="java.lang.String">DOCUMENT_6_CMD_ID</first>
<second class="oracle.ide.keyboard.KeyStrokes">
<data>
<Item class="javax.swing.KeyStroke">meta 6</Item>
</data>
</second>
</Item>