Hide Forgot
When mounting a volume, the servername is specified to retrieve the volume configuration. If that server is down, the client cannot proceed. If the client could handle round-robin dns responses, it could sequentially try the responses until it succeeds, allowing the client to start even with a down server. For instance: host www.google.com returns www.google.com is an alias for www.l.google.com. www.l.google.com has address 74.125.53.106 www.l.google.com has address 74.125.53.147 www.l.google.com has address 74.125.53.99 www.l.google.com has address 74.125.53.103 www.l.google.com has address 74.125.53.104 www.l.google.com has address 74.125.53.105 If this host was used to mount the volume: mount -t glusterfs www.google.com:searchdata the client would first query 74.125.53.106. If it did not succeed in retrieving the vol file for searchdata, it would then try 74.125.53.147, etc.
Check the current behavior, see what all needs to change for bringing in RR DNS for initial getspec.
CHANGE: http://review.gluster.com/239 (Enables usage of 'volfile-max-fetch-attempts' option of glusterfsd.) merged in master by Vijay Bellur (vijay)
With the enabling of this option, gluster attempts to establish the connection the given number of times. If a server has multiple ips, each connection uses a different ip from the list provided by dns resolution. So this in a way solves the problem.
CHANGE: http://review.gluster.com/455 (This is a followup patch to e378778686fca24934e76c41e6496b12162a5420) merged in master by Vijay Bellur (vijay)