[Gluster-users] Problem with quorum on a replicated volume with 2 bricks and 3 nodes in trusted pool

Ravishankar N ravishankar at redhat.com
Fri Oct 9 14:17:33 UTC 2015 


On 10/09/2015 05:20 PM, Adrian Gruntkowski wrote:
> Hello everyone,
>
> I'm trying to setup a quorum on my cluster and hit an issue where 
> taking down one node blocks writing
> on the affected volume. The thing is, I have 3 servers where 2 volumes 
> are setup in a cross-over manner,
> like this:
>
> [Server1: vol1]<--->[Server2: vol1 vol2]<--->[Server3: vol2].
>
> The trusted pool contains 3 servers so AFAIK taking down, for example, 
> "Server3" shouldn't take down "vol2",
> but it does with "quorum not met" message in the logs
>
> [2015-10-09 11:12:55.386736] C 
> [rpc-clnt-ping.c:161:rpc_clnt_ping_timer_expired] 
> 0-system_mail1-client-1: server 172.16.11.112:49152 
> <http://172.16.11.112:49152> has not responded in the last 42 seconds, 
> disconnecting.
> [2015-10-09 11:12:55.387213] E [rpc-clnt.c:362:saved_frames_unwind] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libglusterfs.so.0(_gf_log_callingfn+0x19a)[0x7f950b98340a] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(saved_frames_unwind+0x1cf)[0x7f950b74e4df] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(saved_frames_destroy+0xe)[0x7f950b74e5fe] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(rpc_clnt_connection_cleanup+0x9c)[0x7f950b74fdcc] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(rpc_clnt_notify+0x48)[0x7f950b750578] 
> ))))) 0-system_mail1-client-1: forced unwinding frame type(GlusterFS 
> 3.3) op(LOOKUP(27)) called at 2015-10-09 11:12:00.087425 (xid=0x517)
> [2015-10-09 11:12:55.387238] W [MSGID: 114031] 
> [client-rpc-fops.c:2971:client3_3_lookup_cbk] 0-system_mail1-client-1: 
> remote operation failed. Path: / 
> (00000000-0000-0000-0000-000000000001) [Transport endpoint is not 
> connected]
> [2015-10-09 11:12:55.387429] E [rpc-clnt.c:362:saved_frames_unwind] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libglusterfs.so.0(_gf_log_callingfn+0x19a)[0x7f950b98340a] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(saved_frames_unwind+0x1cf)[0x7f950b74e4df] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(saved_frames_destroy+0xe)[0x7f950b74e5fe] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(rpc_clnt_connection_cleanup+0x9c)[0x7f950b74fdcc] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(rpc_clnt_notify+0x48)[0x7f950b750578] 
> ))))) 0-system_mail1-client-1: forced unwinding frame type(GlusterFS 
> 3.3) op(LOOKUP(27)) called at 2015-10-09 11:12:07.374032 (xid=0x518)
> [2015-10-09 11:12:55.387591] E [rpc-clnt.c:362:saved_frames_unwind] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libglusterfs.so.0(_gf_log_callingfn+0x19a)[0x7f950b98340a] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(saved_frames_unwind+0x1cf)[0x7f950b74e4df] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(saved_frames_destroy+0xe)[0x7f950b74e5fe] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(rpc_clnt_connection_cleanup+0x9c)[0x7f950b74fdcc] 
> (--> 
> /usr/lib/x86_64-linux-gnu/libgfrpc.so.0(rpc_clnt_notify+0x48)[0x7f950b750578] 
> ))))) 0-system_mail1-client-1: forced unwinding frame type(GF-DUMP) 
> op(NULL(2)) called at 2015-10-09 11:12:13.381487 (xid=0x519)
> [2015-10-09 11:12:55.387614] W [rpc-clnt-ping.c:204:rpc_clnt_ping_cbk] 
> 0-system_mail1-client-1: socket disconnected
> [2015-10-09 11:12:55.387624] I [MSGID: 114018] 
> [client.c:2042:client_rpc_notify] 0-system_mail1-client-1: 
> disconnected from system_mail1-client-1. Client process will keep 
> trying to connect to glusterd until brick's port is available
> [2015-10-09 11:12:55.387635] W [MSGID: 108001] 
> [afr-common.c:4043:afr_notify] 0-system_mail1-replicate-0: 
> Client-quorum is not met
> [2015-10-09 11:12:55.387959] I [socket.c:3362:socket_submit_request] 
> 0-system_mail1-client-1: not connected (priv->connected = 0)
> [2015-10-09 11:12:55.387972] W [rpc-clnt.c:1571:rpc_clnt_submit] 
> 0-system_mail1-client-1: failed to submit rpc-request (XID: 0x51a 
> Program: GlusterFS 3.3, ProgVers: 330, Proc: 27) to rpc-transport 
> (system_mail1-client-1)
> [2015-10-09 11:12:55.387982] W [MSGID: 114031] 
> [client-rpc-fops.c:2971:client3_3_lookup_cbk] 0-system_mail1-client-1: 
> remote operation failed. Path: /images 
> (a63d0ff2-cb42-4cee-9df7-477459539788) [Transport endpoint is not 
> connected]
> [2015-10-09 11:12:55.388653] W [MSGID: 114031] 
> [client-rpc-fops.c:2971:client3_3_lookup_cbk] 0-system_mail1-client-1: 
> remote operation failed. Path: (null) 
> (00000000-0000-0000-0000-000000000000) [Transport endpoint is not 
> connected]
> [2015-10-09 11:13:03.245909] I [MSGID: 108031] 
> [afr-common.c:1745:afr_local_discovery_cbk] 
> 0-system_mail1-replicate-0: selecting local read_child 
> system_mail1-client-0
> [2015-10-09 11:13:04.734547] W [fuse-bridge.c:1937:fuse_create_cbk] 
> 0-glusterfs-fuse: 1253: /x => -1 (Read-only file system)
> [2015-10-09 11:13:10.419069] E [socket.c:2332:socket_connect_finish] 
> 0-system_mail1-client-1: connection to 172.16.11.112:24007 
> <http://172.16.11.112:24007> failed (Connection timed out)
> [2015-10-09 11:12:55.387447] W [MSGID: 114031] 
> [client-rpc-fops.c:2971:client3_3_lookup_cbk] 0-system_mail1-client-1: 
> remote operation failed. Path: / 
> (00000000-0000-0000-0000-000000000001) [Transport endpoint is not 
> connected]
>
> (Another weird thing is that glusterfs version reported in logs is 
> 3.7.3, when the Debian packages installed
> on my system are for 3.7.3 - don't know if it's meant to be that way).
>
> Below is the output of "gluster volume info" on one of the servers 
> (there are 4 volumes in my actual setup):
>
> Volume Name: data_mail1
> Type: Replicate
> Volume ID: c2833dbe-aaa5-49d0-91d3-5abb44efb48c
> Status: Started
> Number of Bricks: 1 x 2 = 2
> Transport-type: tcp
> Bricks:
> Brick1: cluster-rep:/GFS/data/mail1
> Brick2: mail-rep:/GFS/data/mail1
> Options Reconfigured:
> cluster.quorum-count: 2
> auth.allow: 127.0.0.1,172.16.11.*,172.16.12.*
> performance.readdir-ahead: on
> cluster.server-quorum-type: server
> cluster.quorum-type: fixed
> cluster.server-quorum-ratio: 51%
> Volume Name: data_www1
> Type: Replicate
> Volume ID: 385a7052-3ab5-42c2-93bc-6e10c4e7c0f1
> Status: Started
> Number of Bricks: 1 x 2 = 2
> Transport-type: tcp
> Bricks:
> Brick1: cluster-rep:/GFS/data/www1
> Brick2: web-rep:/GFS/data/www1
> Options Reconfigured:
> cluster.quorum-count: 2
> auth.allow: 127.0.0.1,172.16.11.*,172.16.12.*
> performance.readdir-ahead: on
> cluster.server-quorum-type: server
> cluster.quorum-type: fixed
> cluster.server-quorum-ratio: 51%
> Volume Name: system_mail1
> Type: Replicate
> Volume ID: 82dc0617-d855-4bf0-b5e5-c4147ca15779
> Status: Started
> Number of Bricks: 1 x 2 = 2
> Transport-type: tcp
> Bricks:
> Brick1: cluster-rep:/GFS/system/mail1
> Brick2: mail-rep:/GFS/system/mail1
> Options Reconfigured:
> cluster.quorum-count: 2
> auth.allow: 127.0.0.1,172.16.11.*,172.16.12.*
> performance.readdir-ahead: on
> cluster.server-quorum-type: server
> cluster.quorum-type: fixed
> cluster.server-quorum-ratio: 51%
> Volume Name: system_www1
> Type: Replicate
> Volume ID: 83868eb5-7b32-4e80-882c-e83361b267b9
> Status: Started
> Number of Bricks: 1 x 2 = 2
> Transport-type: tcp
> Bricks:
> Brick1: cluster-rep:/GFS/system/www1
> Brick2: web-rep:/GFS/system/www1
> Options Reconfigured:
> cluster.quorum-count: 2
> auth.allow: 127.0.0.1,172.16.11.*,172.16.12.*
> performance.readdir-ahead: on
> cluster.server-quorum-type: server
> cluster.quorum-type: fixed
> cluster.server-quorum-ratio: 51%
>
> I have also tried switching between "cluster.qourum-type" = 
> fixed|auto, with the same result.
>

Client-quorum (cluster.quorum-type and cluster.quorum-count) options 
apply to AFR (the replicate translator).
You have set it to 'fixed' with a count of '2'. This means that both 
bricks of the volume(s) need to be online to meet quorum. If only one 
brick is up, the volume becomes read-only.  For replica 2 volumes, it 
doesn't make much sense to enable client-quorum without losing high 
availability (i.e volume still being writable when only one brick is up).


> What am I missing? Is there a way to add a "fake brick" to meet the 
> quorum requirements without
> holding 3rd replica of the data?
You can try using arbiter-volumes [1] that are a good compromise between 
replica-2 and normal replica-3 volumes.
Also  server-quorum (cluster.server-quorum-type and 
cluster.server-quorum-ratio) doesn't help much in avoiding data 
split-brains, so IMO, you might as well disable it.
-Ravi

[1] 
https://github.com/gluster/glusterfs-specs/blob/master/done/Features/afr-arbiter-volumes.md

>
> --
> Regards,
> Adrian
>
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20151009/7debb3fd/attachment.html>

More information about the Gluster-users mailing list