WSREP initial position

[Acampada Leipzig]

Hi, we are trying to build a 4-node Galera environment in 4 machines.

2 nodes are joining correctly, but the 2 others not. When looking at
the log we see that the nodes that connect successfully join with this
initial WSREP position:
120518 11:10:03 [Note] WSREP: Initial position:
15266fd2-9a2b-11e1-0800-be986ec198cb:55287

while the nodes that do not connect join like this:
120518 11:14:29 [Note] WSREP: Read nil XID from storage engines,
skipping position init

and when assigning an initial position in a working node in a
successful node:
120518 11:10:03 [Note] WSREP: Assign initial position for
certification: 55287, protocol version: -1
120518 11:10:03 [Note] WSREP: Setting initial position to
15266fd2-9a2b-11e1-0800-be986ec198cb:55287

while in a node that does not connect we receive:
120518 11:14:29 [Note] WSREP: Assign initial position for
certification: -1, protocol version: -1
120518 11:14:29 [Note] WSREP: Setting initial position to
00000000-0000-0000-0000-000000000000:-1

All the wsrep.cnf files have the same configuration while 192.168.X.X
is the IP of the master node:
wsrep_cluster_address="gcomm://192.168.X.X:4567"

All have the same version of Galera and the library is loaded
correctly. Any ideas what are we missing?

[Alex Yurchenko]

Hi,

From the logs that you submitted, it follows that the nodes that join
successfully had joined successfully in the past, and the nodes that
don't - had not. Which is not that revealing ;)

The nodes that show

> 120518 11:14:29 [Note] WSREP: Setting initial position to
> 00000000-0000-0000-0000-000000000000:-1

will have to request state snapshot transfer (SST) later in the logs.

The fact that they don't join most likely means that SST fails. For the
reasons you should check both joiner and donor longs.

Regards,
Alex

--
Alexey Yurchenko,
Codership Oy, www.codership.com
Skype: alexey.yurchenko, Phone: +358-400-516-011