Re: 5.1.6-alpha NDB: Could not get apply status share

[jim shnider]

I may be able to conclude this conversation with myself. :-)

I was able to get a multi-data-node configuration working. This is exactly the same setup as discussed in the last two posts. To make it go you just:

Bring everything down:
(on Xeon)
[root]# support-files/mysql.server stop
[root]# killall ndbd
[root]# killall ndb_mgmd
(on Athlon)
[root]# killall ndbd

Flush the ndb_fs on both nodes:
(on Xeon)
[root]# rm -rf /var/lib/mysql-cluster/ndb_2_fs
(on Athlon)
[root]# rm -rf /var/lib/mysql-cluster/ndb_3_fs

Bring everything up (in this order):
(on Xeon)
[root]# bin/ndb_mgmd
[root]# bin/ndbd --initial
(on Athlon)
[root]# bin/ndbd --initial
(on Xeon)
[root]# support-files/mysql.server start

When everything looked pristine from the start and came up synchronously, I didn't have any trouble. I was able to (in this order):
- create a table then add and verify data
- drop node3 then add and verify data
- bring node3 back then add and verify data
- drop node2 then add and verify data
- bring node 2 back then verify data

The only caveat to this looks a little like the problem from Gabe Harriman's post. Node 2 did not finish its startup until node 3 came up. Then both registered as 'started' simultaneously. From my past trials, I would believe that starting the api node before that point would have produced catastrophy.

It would be a valuable lesson to others setting up mysql-clusters on 5.0.19 that, while the system is designed to handle lost nodes at run-time, at startup it appears that everything must be present and functional.

Good luck!