Howard Wilkinson wrote:

Fedora-ds-1.1.1 on Fedora 7 + (the + is back ports from 8/9, all of the updates applied, and additional packages I have cross ported)

I have succeeded in getting a fault tolerant mesh configured that consists of 2 or more Multi-Master servers, a number of Hub (0+) and a number of consumers (0+).

I have done this by modifying mmr.pl to accept --host1_role and --host2_role parameters which can be set to supplier, hub, or consumer.

For all of the usual DCROOTs i.e. not o=NetscapeRoot I set the relationships up as implied i.e. supplier<->supplier for the Multi-Master Hosts, supplier<->hub, hub<->consumer. Where the site is too small for hub servers I have gone supplier<->consumer direct. Inter-site topology and hub grouping within sites is left as an exercise for the reader (me when it comes back to bite me...)

For the o=Netscape I have chosen to use supplier<->supplier relationships but to apply the same topology.

Sequence of events are:

* On first Master

     1. Install clean environment - erase rpm's delete residual
        files, install rpms, patch dirsrv-admin startup to work!
     2. Run setup-ds-admin.pl in silent mode, this adds schema
        files. The inf file has SlapdConfigMC=1, UseExistingMC=0
        and points ConfigDirectoryLdapURL to this host.
     3. Set up SSL certs using certutil commands and openssl
        supplied certificates from our CA.
     4. Restart dirsrv and dirsrv-admin
     5. Create 2nd and subsequent DCROOTS with default aci's and
        "standard" container entries
     6. Preload data into DCROOTS for users and other objects
        being migrated.

* On other servers - doing other masters first, followed by hubs
  and then consumers - carry out steps 1-5 above creating the
  o=NetscapeRoot DCROOT as well.
      o The inf file has SlapdConfigMC=1, UseExistingMC=1 and
        points ConfigDirectoryLdapUrl to the first Master
* Then run the mmr.pl script on each connection for each DCROOT
  starting with replicating the first master to all other masters,
  then to hubs, then other masters to hubs and finally hubs to
  consumers.
     1. For o=NetscapeRoot run mmr.pl as supplier<->supplier,
        otherwise honor the role played by each server.
     2. Replace entries in cn=UserDirectory, ou=Global
        Preferences, ou=<localdomain>, o=NetscapeRoot for
        nsDirectoryFailoverList with one for each server other
        than the first master which is mentioned in the
        nsDirectoryURL entry in the same object. *Is this the
        right sort of thing to do?*

Yes.

     1. On every host alter the cn=Pass Through
        Authentication,cn=plugins,cn=config object to have
        nssslapd-pluginarg0 to reference that host rather than the
        first master. *Is this correct on the consumers (or hubs)?*

Yes. Note that you can specify failover in pass through auth by using a special form of the ldap url. See *http://tinyurl.com/32kjqy*

     1. I am assuming that this is for authentication not for
        password modification purposes!

Right.

     1. Which brings up the question of where in the consumers and
        hubs do I put referrals to the Master(s)?

They are automatically set by the replication protocol. You should not have to do anything. If you attempt to modify a hub or consumer, your client should get LDAP Error 10 and a referral to a master.

     1. Edit adm.conf on each host to change the ldapurl to point
        to the local host.

Now assuming that this was the right thing to do I now need to set up referrals for writing to the system from the consumers and hubs back to the "site" masters. Where do I put this information?

I am also getting these errors logged on the first master!

Feb 28 22:00:35 bastion ns-slapd: auxpropfunc error invalid parameter supplied Feb 28 22:00:35 bastion ns-slapd: sql_select option missing Feb 28 22:00:35 bastion ns-slapd: auxpropfunc error no mechanism available

I think you can ignore these.

These are appearing about every 15 minutes. Anybody any idea where these are coming from?

I'm not sure, but the directory server does not support SASL auxprop with sql.

Finally the shutdown time for the dirsrv servers on the suppliers is extremely long - orders of minutes, what could be causing this?

Are they under load while shutting down? Can you post the shutdown sequence from the error log?

--

Howard Wilkinson

Phone:

+44(20)76907075

Coherent Technology Limited

Fax:

23 Northampton Square,

Mobile:

+44(7980)639379

United Kingdom, EC1V 0HL

Email:

howard@cohtech.com

---

-- Fedora-directory-users mailing list Fedora-directory-users@redhat.com https://www.redhat.com/mailman/listinfo/fedora-directory-users