Support for multiple masters for automatic failover

[breaklee]

i want to use one single entry for multiple master.

is it support about this?

[phal0r]

Hi @breaklee,

right now, you can configure only one master server. According to the seaweed fs wiki (https://github.com/chrislusf/seaweedfs/wiki/Failover-Master-Server) it should be quite easy, to change that behaviour, since all traffic rerouting is done between the master servers. The part we need to implement to the client, is to check the cluster status on startup and add all peer addresses to an array. When the currently used master node goes done, we can just switch to the next peer.

I will change the title of this issue to the implementation story. If you want you can implement it and send an PR, otherwise you will be informed here, when we implement this feature.

[breaklee]

thanks for information.

and i have other questions.

A: > weed server -master.port=9333 -dir=./1 -volume.port=8080
B: > weed server -master.port=9334 -dir=./2 -volume.port=8081 -master.peers=localhost:9333
C: > weed server -master.port=9335 -dir=./3 -volume.port=8082 -master.peers=localhost:9334

there are 3 masters and 3 volume servers.

and now A: is a master.

some clients save and get files using A.
and now, suppose A has gone down. and B: elected new master.

and now, client can save and get using B right?
this is ok. and no problems.

but i want to get some files saved on A.

is it possible to replication A, B, C volume server?

@phal0r

if it's possible, everything will be fine for me! :)

[phal0r]

These are 2 different things. Cluster failover is when one master goes down and a new one is elected. Replication is when your files are redundantly saved on 2 or more volume servers.

You can start master and volume servers separately via "weed volume" and "weed master". In your scenario when the whole server A goes down, your files on volume server A won't be accesible, unless you enabled replication to the volume servers on B and/or C. Considering volume and master servers would be started separately and master A goes down, this client would not function anymore, since it does not know yet about the other master servers. This is an enhancement that needs to be implemented.