I think that the best solution is to make the partition logic
persistent such that the arbitrator works even in the system restart
PS. While waiting for such a feature one could easily add a config
parameter that restarts even in this partitioned case.
2004-09-15 kl. 20.24 skrev Crouch, Luke H.:
> the db nodes only come up when there is sufficient number of nodes to
> come up. since your cluster was only a 2 node cluster, the first db
> node was waiting for the 2nd one to come up because the 2nd node
> contained the active fragment of half the database. in a situation
> where the 2 nodes are both up and one goes down, the other knows to
> make itself active on both fragments of the db.
> I think it's a good question about bringing up just one node after a
> 2-node crash. and the first thing I thought of was the same - changing
> the config.ini to a one node config, restarting with -i and then
> restoring a backup. but that gives you a lot of down-time and then
> more downtime when you want to go back up to a 2-node setup.
> I'm interested to hear what the MySQL guys would say about this...
>> -----Original Message-----
>> From: Yannis Tsakiris [mailto:gtsakiris@stripped]
>> Sent: Wednesday, September 15, 2004 9:10 AM
>> To: cluster@stripped
>> Subject: DB node is too slow
>> Hello people,
>> In a ndb cluster with two DB nodes (2 replicas, 1 node
>> group), I have noticed that if I try to start only one of the
>> two DB nodes, its start up process is progressing very slowly.
>> It stays a lot of time at 'phase #1' and then it stucks in
>> 'phase #2'. I waited for about ten minutes but it never
>> passed phase 2.
>> If I start the two DB nodes together they are coming up
>> pretty fast. The same happens when one of the two DB nodes is
>> already started, and I start the other.
>> What causes the DB node to progress so slow when it runs
>> alone? What if both my DB nodes crash and (for any reason) I
>> am unable to start both of them (so they start in a
>> reasonable amount of time)? Do I have to change my
>> configuration to one-db-node in such case?
>> Looking forward for your suggestions.
>> Yannis Tsakiris
> MySQL Cluster Mailing List
> For list archives: http://lists.mysql.com/cluster
> To unsubscribe:
Mikael Ronström, Senior Software Architect
MySQL AB, www.mysql.com