Bug 1117509

Description of problem:

When you do a gluster peer detach <peer-name>, you would expect to delete all 
records for a peer? Unfortunately on 3.4.2 I can see that it does not happen. Even though glusterd would report temporarily that the peer has been detached, restarting glusterd would cause the peer to be added back.

I could see that this had to do with the peer record not getting deleted from /var/lib/glusterd/peers/ through the peer detach command. To illustrate,

root@server1]$ gluster peer detach server2
peer detach: success

root@server1]$ cat /var/lib/glusterd/peers/5226a3d7-9a35-44b6-b9ca-d7
uuid=bd8c9f46-0fb7-4958-b0ca-9f5fab18e5ec
state=3
hostname1=192.168.24.81
 
root@server1]$ gluster p status
peer status: No peers present

root@server1]$ /etc/init.d/glusterd restart
[ OK ]

root@server1]$ gluster p status
uuid=bd8c9f46-0fb7-4958-b0ca-9f5fab18e5ec
state=3
hostname1=192.168.24.81

Now, if I change detach and also rm -f the peer record from /var/lib/glusterd, then after restart the record of the peer would disappear (correctly). That is to say, the following:

a) gluster peer detach server2
b) rm -f /var/lib/glusterd/peers/5226a3d7-9a35-44b6-b9ca-d7

IMHO peer detach needs to correct itself so that peer does not get added back automatically.

Version-Release number of selected component (if applicable):


How reproducible:
100%

Steps to Reproduce:
1. Create a twin replicated cluster comprising server1 and server2.
2. Detach server2 on server1 using gluster peer detach server2. [Note: For my testing, I had also powered down server2].
3. restart /etc/init.d/glusterd on server1.
4. Execute gluster peer status on server1 and check for server2 in peer list.

Actual results:

Before restart of glusterd. gluster correctly reports that no peer is present. But after restart of glusterd, it re-inits with the record of the deleted peer which causes the latter to show up in the peer list.

Expected results:

Deleted peer should not feature in gluster peer status output.

Additional info:
None.