Slave connecting to maser always stay connected?

[Mathew Rowley]

I have a setup with 2 masters (nway multimaster) and 4 slaves each having
syncrepl set up to connect to the 2 masters. The question I have, with the
following configuration (on the slaves) should they always be connected to
the masters? Netstat is showing that the connections are intermittent:

type=refreshAndPersist retry="60 +"


Netstat on the masters shows that the connections from the slaves will vary
­ sometimes 0, sometimes 4 ­ seems to be somewhat random. We are trying to
determine if there is something wrong in our network. Thanks for the help.


MAT

[Quanah Gibson-Mount]

--On Tuesday, March 30, 2010 12:23 PM -0600 Mathew Rowley

Post by Mathew Rowley
I have a setup with 2 masters (nway multimaster) and 4 slaves each having
syncrepl set up to connect to the 2 masters. The question I have, with
the following configuration (on the slaves) should they always be
connected to the masters? Netstat is showing that the connections are
type=refreshAndPersist retry="60 +"
Netstat on the masters shows that the connections from the slaves will
vary – sometimes 0, sometimes 4 – seems to be somewhat random. We
are trying to determine if there is something wrong in our network.
Thanks for the help.

Yes, that's what refreshAndPersist implies. The connections from the
replica should stay connected to the provider, unless the network is flaky,
a firewall is timing them out, etc. Or you stop/restart the slapd on
either side. ;)

--Quanah

--

Quanah Gibson-Mount
Principal Software Engineer
Zimbra, Inc
--------------------
Zimbra :: the leader in open source messaging and collaboration

[Mathew Rowley]

Can you elaborate on ³a firewall is flakey²? The hosts are physically
diverse, each side has 1 master, and 2 slaves, all 3 on the same subnet.
Even those connections are on and off. Is there any logging that can
explain why this would be happening?

MAT

Post by Quanah Gibson-Mount
--On Tuesday, March 30, 2010 12:23 PM -0600 Mathew Rowley

Post by Mathew Rowley
I have a setup with 2 masters (nway multimaster) and 4 slaves each having
syncrepl set up to connect to the 2 masters. The question I have, with
the following configuration (on the slaves) should they always be
connected to the masters? Netstat is showing that the connections are
type=refreshAndPersist retry="60 +"
Netstat on the masters shows that the connections from the slaves will
vary ­ sometimes 0, sometimes 4 ­ seems to be somewhat random. We
are trying to determine if there is something wrong in our network.
Thanks for the help.

Yes, that's what refreshAndPersist implies. The connections from the
replica should stay connected to the provider, unless the network is flaky,
a firewall is timing them out, etc. Or you stop/restart the slapd on
either side. ;)
--Quanah
--
Quanah Gibson-Mount
Principal Software Engineer
Zimbra, Inc
--------------------
Zimbra :: the leader in open source messaging and collaboration

[David Hawes]

Post by Quanah Gibson-Mount
--On Tuesday, March 30, 2010 12:23 PM -0600 Mathew Rowley

Post by Mathew Rowley
I have a setup with 2 masters (nway multimaster) and 4 slaves each having
syncrepl set up to connect to the 2 masters. The question I have, with
the following configuration (on the slaves) should they always be
connected to the masters? Netstat is showing that the connections are
type=refreshAndPersist retry="60 +"
Netstat on the masters shows that the connections from the slaves will
vary – sometimes 0, sometimes 4 – seems to be somewhat random. We
are trying to determine if there is something wrong in our network.
Thanks for the help.

Yes, that's what refreshAndPersist implies. The connections from the
replica should stay connected to the provider, unless the network is
flaky, a firewall is timing them out, etc. Or you stop/restart the
slapd on either side. ;)

... or (possibly) your accesslog purges:

http://www.openldap.org/lists/openldap-software/201003/msg00066.html

Are you using delta-syncrepl?

[Quanah Gibson-Mount]

Post by David Hawes

Post by Quanah Gibson-Mount
Yes, that's what refreshAndPersist implies. The connections from the
replica should stay connected to the provider, unless the network is
flaky, a firewall is timing them out, etc. Or you stop/restart the
slapd on either side. ;)

http://www.openldap.org/lists/openldap-software/201003/msg00066.html
Are you using delta-syncrepl?

I suggest you file an ITS on this issue (I can confirm I've seen it too).

--Quanah

--

Quanah Gibson-Mount
Principal Software Engineer
Zimbra, Inc
--------------------
Zimbra :: the leader in open source messaging and collaboration

[Quanah Gibson-Mount]

--On Wednesday, March 31, 2010 9:36 AM -0700 Quanah Gibson-Mount

Post by Quanah Gibson-Mount

Post by David Hawes
Are you using delta-syncrepl?

I suggest you file an ITS on this issue (I can confirm I've seen it too).

I tracked down the issue and filed an ITS on it:
<http://www.openldap.org/its/index.cgi/?findid=6519>

--Quanah


--

Quanah Gibson-Mount
Principal Software Engineer
Zimbra, Inc
--------------------
Zimbra :: the leader in open source messaging and collaboration

[Quanah Gibson-Mount]

--On Tuesday, March 30, 2010 7:15 PM -0600 Mathew Rowley

Can you elaborate on "a firewall is flakey"? The hosts are
physically diverse, each side has 1 master, and 2 slaves, all 3 on the
same subnet. Even those connections are on and off. Is there any
logging that can explain why this would be happening?

A firewall that times out sessions, for example...

I would use packet level debugging (You have to start slapd -d packet or
slapd -d packets, I forget which (or maybe even both work!)) to see what's
going on.

--Quanah

--

Quanah Gibson-Mount
Principal Software Engineer
Zimbra, Inc
--------------------
Zimbra :: the leader in open source messaging and collaboration