[Fedora-directory-users] Failover between masters

[Coe, Colin C. (Unix Engineer)]

Hi all

We are currently using Sun's Directory server and have had some problems
with clients failing over to the other master if one fails.  The clients
are a minxute of RHEL 3 WS and Solaris 8 (SPARC), and the Sun Directory
servers are both Solars 9 (SPARC) running Directory One 5.1.

/etc/ldap.conf
host 1.1.1.1 2.2.2.2
port 636
ldap_version 3
base o=unix,dc=company,dc=com
scope sub
timelimit 5
bind_timelimit 3
ssl on
pam_filter objectclass=posixAccount
pam_login_attribute uid
pam_member_attribute memberUid
pam_password crypt
idle_timelimit 3600

/etc/openldap/ldap.conf
BASE o=unix,dc=company,dc=com
HOST ldap1.company.com ldap2.company.com
PORT 636
SASL_SECPROPS "noanonymous,noplain"
SIZELIMIT 0
TIMELIMIT 0
DEREF never
TLS_CACERT      /etc/ssl/ldap/cacert.pem
TLS_REQCERT     demand

We're using the bog standard nscd daemons provided by the OS vendors.
We also use IDSync to synchronise user passwords from AD to LDAP but not
from LDAP to AD.

What we're finding is if ldap1 dies for some reason, the clients don't
failover to ldap2.

We don't know if the problem is client side or server side.  Would
Fedora Directory Server, set up in a similar manner, also not failover
properly?  While we're prepared to look at Fed DS, there is a feeling
that it too will behave in the same manner, given they are both forks of
the same project.

Comments?

Thanks

CC

NOTICE: This email and any attachments are confidential. 
They may contain legally privileged information or 
copyright material. You must not read, copy, use or 
disclose them without authorisation. If you are not an 
intended recipient, please contact us at once by return 
email and then delete both messages and all attachments.

[Rich Megginson]

Coe, Colin C. (Unix Engineer) wrote:
> Hi all
>
> We are currently using Sun's Directory server and have had some 
> problems with clients failing over to the other master if one fails.  
> The clients are a minxute of RHEL 3 WS and Solaris 8 (SPARC), and the 
> Sun Directory servers are both Solars 9 (SPARC) running Directory One 5.1.
>
> /etc/ldap.conf
> host 1.1.1.1 2.2.2.2
> port 636
> ldap_version 3
> base o=unix,dc=company,dc=com
> scope sub
> timelimit 5
> bind_timelimit 3
> ssl on
> pam_filter objectclass=posixAccount
> pam_login_attribute uid
> pam_member_attribute memberUid
> pam_password crypt
> idle_timelimit 3600
>
> /etc/openldap/ldap.conf
> BASE o=unix,dc=company,dc=com
> HOST ldap1.company.com ldap2.company.com
> PORT 636
> SASL_SECPROPS "noanonymous,noplain"
> SIZELIMIT 0
> TIMELIMIT 0
> DEREF never
> TLS_CACERT      /etc/ssl/ldap/cacert.pem
> TLS_REQCERT     demand
>
> We're using the bog standard nscd daemons provided by the OS vendors.  
> We also use IDSync to synchronise user passwords from AD to LDAP but 
> not from LDAP to AD.
>
> What we're finding is if ldap1 dies for some reason, the clients don't 
> failover to ldap2.
>
> We don't know if the problem is client side or server side.  Would 
> Fedora Directory Server, set up in a similar manner, also not failover 
> properly?

It wouldn't make any difference.  I'm pretty sure failover is a properly 
of the client.  Are you sure you have the multiple hosts configured 
correctly in your ldap.conf files?
> While we're prepared to look at Fed DS, there is a feeling that it too 
> will behave in the same manner, given they are both forks of the same 
> project.
>
> Comments?
>
> Thanks
>
> CC
>
> NOTICE: This email and any attachments are confidential. 
> They may contain legally privileged information or 
> copyright material. You must not read, copy, use or 
> disclose them without authorisation. If you are not an 
> intended recipient, please contact us at once by return 
> email and then delete both messages and all attachments.
>
> --
> Fedora-directory-users mailing list
> [e-mail]
> https://www.redhat.com/mailman/listinfo/fedora-directory-users

[Coe, Colin C. (Unix Engineer)]

See inline comments
> Coe, Colin C. (Unix Engineer) wrote:
> > Hi all
> >
> > We are currently using Sun's Directory server and have had some
> > problems with clients failing over to the other master if
>
> one fails.
> > The clients are a minxute of RHEL 3 WS and Solaris 8
>
> (SPARC), and the
> > Sun Directory servers are both Solars 9 (SPARC) running
>
> Directory One 5.1.
> > /etc/ldap.conf
> > host 1.1.1.1 2.2.2.2
> > port 636
> > ldap_version 3
> > base o=unix,dc=company,dc=com
> > scope sub
> > timelimit 5
> > bind_timelimit 3
> > ssl on
> > pam_filter objectclass=posixAccount
> > pam_login_attribute uid
> > pam_member_attribute memberUid
> > pam_password crypt
> > idle_timelimit 3600
> >
> > /etc/openldap/ldap.conf
> > BASE o=unix,dc=company,dc=com
> > HOST ldap1.company.com ldap2.company.com
> > PORT 636
> > SASL_SECPROPS "noanonymous,noplain"
> > SIZELIMIT 0
> > TIMELIMIT 0
> > DEREF never
> > TLS_CACERT      /etc/ssl/ldap/cacert.pem
> > TLS_REQCERT     demand
> >
> > We're using the bog standard nscd daemons provided by the
>
> OS vendors.
> > We also use IDSync to synchronise user passwords from AD to
>
> LDAP but
> > not from LDAP to AD.
> >
> > What we're finding is if ldap1 dies for some reason, the
>
> clients don't
> > failover to ldap2.
> >
> > We don't know if the problem is client side or server side.  Would
> > Fedora Directory Server, set up in a similar manner, also
>
> not failover
> > properly?
>
> It wouldn't make any difference.  I'm pretty sure failover is 
> a properly 
> of the client.  Are you sure you have the multiple hosts configured 
> correctly in your ldap.conf files?

No, I'm not 100% sure that the clients are set right.  My sanitised
/etc/ldap and /etc/openldap/ldap.conf are shown above.  Can you suggest
any improvements to them?
> > While we're prepared to look at Fed DS, there is a feeling
>
> that it too
> > will behave in the same manner, given they are both forks
>
> of the same
> > project.
> >
> > Comments?
> >
> > Thanks
> >
> > CC

NOTICE: This email and any attachments are confidential. 
They may contain legally privileged information or 
copyright material. You must not read, copy, use or 
disclose them without authorisation. If you are not an 
intended recipient, please contact us at once by return 
email and then delete both messages and all attachments.

[Rich Megginson]

Coe, Colin C. (Unix Engineer) wrote:
> See inline comments
> > Coe, Colin C. (Unix Engineer) wrote:
> > > Hi all
> > >
> > > We are currently using Sun's Directory server and have had some
> > > problems with clients failing over to the other master if
> >
> > one fails.
> > > The clients are a minxute of RHEL 3 WS and Solaris 8
> >
> > (SPARC), and the
> > > Sun Directory servers are both Solars 9 (SPARC) running
> >
> > Directory One 5.1.
> > > /etc/ldap.conf
> > > host 1.1.1.1 2.2.2.2
> > > port 636
> > > ldap_version 3
> > > base o=unix,dc=company,dc=com
> > > scope sub
> > > timelimit 5
> > > bind_timelimit 3
> > > ssl on
> > > pam_filter objectclass=posixAccount
> > > pam_login_attribute uid
> > > pam_member_attribute memberUid
> > > pam_password crypt
> > > idle_timelimit 3600
> > >
> > > /etc/openldap/ldap.conf
> > > BASE o=unix,dc=company,dc=com
> > > HOST ldap1.company.com ldap2.company.com
> > > PORT 636
> > > SASL_SECPROPS "noanonymous,noplain"
> > > SIZELIMIT 0
> > > TIMELIMIT 0
> > > DEREF never
> > > TLS_CACERT      /etc/ssl/ldap/cacert.pem
> > > TLS_REQCERT     demand
> > >
> > > We're using the bog standard nscd daemons provided by the
> >
> > OS vendors.
> > > We also use IDSync to synchronise user passwords from AD to
> >
> > LDAP but
> > > not from LDAP to AD.
> > >
> > > What we're finding is if ldap1 dies for some reason, the
> >
> > clients don't
> > > failover to ldap2.
> > >
> > > We don't know if the problem is client side or server side.  Would
> > > Fedora Directory Server, set up in a similar manner, also
> >
> > not failover
> > > properly?
> >
> > It wouldn't make any difference.  I'm pretty sure failover is 
> > a properly 
> > of the client.  Are you sure you have the multiple hosts configured 
> > correctly in your ldap.conf files?
>
> No, I'm not 100% sure that the clients are set right.  My sanitised
> /etc/ldap and /etc/openldap/ldap.conf are shown above.  Can you suggest
> any improvements to them?

I don't know.  I'm not familiar with failover configuration.
> > > While we're prepared to look at Fed DS, there is a feeling
> >
> > that it too
> > > will behave in the same manner, given they are both forks
> >
> > of the same
> > > project.
> > >
> > > Comments?
> > >
> > > Thanks
> > >
> > > CC
>
> NOTICE: This email and any attachments are confidential. 
> They may contain legally privileged information or 
> copyright material. You must not read, copy, use or 
> disclose them without authorisation. If you are not an 
> intended recipient, please contact us at once by return 
> email and then delete both messages and all attachments.
>
> --
> Fedora-directory-users mailing list
> [e-mail]
> https://www.redhat.com/mailman/listinfo/fedora-directory-users

[Josh Kelley]

On 3/28/07, Coe, Colin C. (Unix Engineer)  wrote:
> No, I'm not 100% sure that the clients are set right.  My sanitised
> /etc/ldap and /etc/openldap/ldap.conf are shown above.  Can you suggest
> any improvements to them?

We're using RHEL 3 and CentOS 4 with ldap.conf files pretty much like
you described, and failover works.

The only difference I see is that in /etc/openldap/ldap.conf, instead of
HOST ldap1.company.com ldap2.company.com
we use
URI ldaps://ldap1.company.com ldaps://ldap2.company.com
But that shouldn't make any difference.

On Fedora 6, instead of setting up /etc/ldap.conf as
Host 1.1.1.1 2.2.2.2
we instead have to use
uri ldaps://1.1.1.1/ ldaps://2.2.2.2/

I'm assuming that the new version of nss_ldap parses the config file
differently but haven't bothered tracking down details.  (nss_ldap is
version 207 on RHEL 3, 226 on CentOS 4, and 253 on Fedora 6.)

Josh Kelley

[George Holbert]

> What we're finding is if ldap1 dies for some reason, the clients don't 
> failover to ldap2.
>
> We don't know if the problem is client side or server side.

When ldap1 dies, do you see any activity in ldap2's access log?  If not, 
you know the clients aren't making the switch to ldap2.

On one of your Linux LDAP clients, try doing this while ldap1 is down:
# service nscd stop
# strace getent passwd
Among the tons of output should be some indication of what LDAP servers 
are being tried.

Coe, Colin C. (Unix Engineer) wrote:
> Hi all
>
> We are currently using Sun's Directory server and have had some 
> problems with clients failing over to the other master if one fails.  
> The clients are a minxute of RHEL 3 WS and Solaris 8 (SPARC), and the 
> Sun Directory servers are both Solars 9 (SPARC) running Directory One 5.1.
>
> /etc/ldap.conf
> host 1.1.1.1 2.2.2.2
> port 636
> ldap_version 3
> base o=unix,dc=company,dc=com
> scope sub
> timelimit 5
> bind_timelimit 3
> ssl on
> pam_filter objectclass=posixAccount
> pam_login_attribute uid
> pam_member_attribute memberUid
> pam_password crypt
> idle_timelimit 3600
>
> /etc/openldap/ldap.conf
> BASE o=unix,dc=company,dc=com
> HOST ldap1.company.com ldap2.company.com
> PORT 636
> SASL_SECPROPS "noanonymous,noplain"
> SIZELIMIT 0
> TIMELIMIT 0
> DEREF never
> TLS_CACERT      /etc/ssl/ldap/cacert.pem
> TLS_REQCERT     demand
>
> We're using the bog standard nscd daemons provided by the OS vendors.  
> We also use IDSync to synchronise user passwords from AD to LDAP but 
> not from LDAP to AD.
>
> What we're finding is if ldap1 dies for some reason, the clients don't 
> failover to ldap2.
>
> We don't know if the problem is client side or server side.  Would 
> Fedora Directory Server, set up in a similar manner, also not failover 
> properly?  While we're prepared to look at Fed DS, there is a feeling 
> that it too will behave in the same manner, given they are both forks 
> of the same project.
>
> Comments?
>
> Thanks
>
> CC

[Steve Rigler]

I've never seen it work adequately with RHEL 3 & 4  or Solaris 8 clients
(solaris 9 seems to work fine).  We use Piranha (which also distributes
the load nicely) to get around it.

-Steve

On Thu, 2007-03-29 at 08:20 +0800, Coe, Colin C. (Unix Engineer) wrote:
> Hi all
>
> We are currently using Sun's Directory server and have had some
> problems with clients failing over to the other master if one fails.
> The clients are a minxute of RHEL 3 WS and Solaris 8 (SPARC), and the
> Sun Directory servers are both Solars 9 (SPARC) running Directory One
> 5.1.
>
> /etc/ldap.conf 
> host 1.1.1.1 2.2.2.2 
> port 636 
> ldap_version 3 
> base o=unix,dc=company,dc=com 
> scope sub 
> timelimit 5 
> bind_timelimit 3 
> ssl on 
> pam_filter objectclass=posixAccount 
> pam_login_attribute uid 
> pam_member_attribute memberUid 
> pam_password crypt 
> idle_timelimit 3600
>
> /etc/openldap/ldap.conf 
> BASE o=unix,dc=company,dc=com 
> HOST ldap1.company.com ldap2.company.com 
> PORT 636 
> SASL_SECPROPS "noanonymous,noplain" 
> SIZELIMIT 0 
> TIMELIMIT 0 
> DEREF never 
> TLS_CACERT      /etc/ssl/ldap/cacert.pem 
> TLS_REQCERT     demand
>
> We're using the bog standard nscd daemons provided by the OS vendors.
> We also use IDSync to synchronise user passwords from AD to LDAP but
> not from LDAP to AD.
>
> What we're finding is if ldap1 dies for some reason, the clients don't
> failover to ldap2.
>
> We don't know if the problem is client side or server side.  Would
> Fedora Directory Server, set up in a similar manner, also not failover
> properly?  While we're prepared to look at Fed DS, there is a feeling
> that it too will behave in the same manner, given they are both forks
> of the same project.
>
> Comments?
>
> Thanks
>
> CC
>
> NOTICE: This email and any attachments are confidential. 
> They may contain legally privileged information or 
> copyright material. You must not read, copy, use or 
> disclose them without authorisation. If you are not an 
> intended recipient, please contact us at once by return 
> email and then delete both messages and all attachments.
> --
> Fedora-directory-users mailing list
> [e-mail]
> https://www.redhat.com/mailman/listinfo/fedora-directory-users