Hi all, i'm new to glusterfs and got some questions about replication. After reading documentation/examples, I choosed using a 2-server replication setup. (Config files copied from here: http://gluster.org/docs/index.php/Automatic_File_Replication_(Mirror)_across _Two_Storage_Servers) I'm starting with some tests now, and maybe I missunderstand the replication feature. If I put some files through the client to the servers, they are repliacted nice. The files are available on server1 and server2 too. But if I simulate a crash (shuting down server1 glusterfsd, and remove files from server1 export directory - then bring server1 glusterfsd up again), the files from server2 export directory would be not resynced to server1 export directory. This could be fatal, because any client could read from server1 (and doesn't found any files). Any suggestions? Regards, sven
Hi Sven,> But if I simulate a crash (shuting down server1 glusterfsd, > and remove files from server1 export directory - then bring > server1 glusterfsd up again), the files from server2 export > directory would be not resynced to server1 export directory.You should not read the export directly. Self-healing only occurs upon a read attempt via the client mount point. For more info, see: http://www.gluster.org/docs/index.php/Understanding_AFR_Translator#Frequently_Asked_Questions Regards, Jeff.
----- "Sven Kummer" <mail at proxworx.org> wrote:> I removed the files in the export directory on server1 only for > simulating a hdd crash (files lost) and want the server1-daemon to resync it's > export directory with the other server.Sven, What is happening in your case is: http://www.gluster.org/docs/index.php/Understanding_AFR_Translator#Self-heal_of_a_file_that_does_not_exist_on_the_first_subvolume If you happen to lose all files on server1, you can force self-heal to happen by doing: On server2 backend root at server2 # find /path/to/export/dir > file-list.txt On mountpoint: root at client1 # cat file-list.txt | xargs stat Vikas -- Engineer - http://gluster.com/