Another more dynamic way (what I did sometimes back at Sun) is to use JINI auto-discovery/registry capabilities to keep a central controller with up-to-date information about registered services configurations.
Worked great. And I agree: pushing configs is messy and error-prone. Cos On Thu, Mar 11, 2010 at 05:45AM, Allen Wittenauer wrote: > On 3/11/10 4:53 AM, "Segel, Mike" <mse...@navteq.com> wrote: > > Then push them out respective of their machine types? > > > ... and this is the fundamental problem. > > I'm still a big fan of using LDAP to help us solve this problem. Rather > than push configs, read them from an LDAP store. Roughly: > > cn=mygrid > .. define common configs > > cn=datanode1,cn=mygrid > ... datanode properties > hosts: hostname1,hostname2, @netgroup1, ... Lists of hosts/netgroups this > set of properties applies to > > > cn=datanode2, > ... datanode properties > hosts:hostname4, hostname5, @netgroup2 ... 2nd set of hosts > > Etc. > > Pushing this is part of the problem. If Hadoop was smart enough to read its > configs from a central store such as LDAP, it would make things much much > much easier. > > See https://issues.apache.org/jira/browse/HADOOP-5670 for more discussion. >
