0

I have a Primary/Secondary 2 nodes DRBD setting as shown

global { usage-count no; }
common { syncer { rate 4M; } }
resource r0 {
        protocol B;
        startup {
                wfc-timeout  15;
                degr-wfc-timeout 60;
        }
        net {
                cram-hmac-alg sha1;
                shared-secret "whateverblahblahblah";
                max-buffers 4096;
                max-epoch-size 4096;
        }
        handlers {
                split-brain "/usr/lib/drbd/notify-split-brain.sh admin@mydomain.com";
        }
        on node01.chofert.com {
                device /dev/drbd0;
                disk /dev/sdb1;
                address <public_IP_1>:7788;
                meta-disk internal;
        }
        on node02.chofert.com {
                device /dev/drbd0;
                disk /dev/sdb1;
                address <public_IP_2>:7788;
                meta-disk internal;
        }
}

I moved to Protocol B as nodes are in different networks and B it is enough for my needs. And rate limited to 4Mb as total bandwith between servers is 12Mb (low, I know, but writing is really really few bytes a day).

Trouble is, probably due to a nightly process which eats up CPU, DRBD crashes around 2 hours. After that, despite of quite low intense writing, at least 2 or 3 full partition re-syncs take place a day.

My question is: how DRBD works upon replication? I mean why do I see full partition re-syncs instead of just the few bytes written?

May just 1 sector of difference affect DRBD proper behavior?

PARTITION AT NODE1:

Disk /dev/sdb1: 8588 MB, 8588886528 bytes
255 heads, 63 sectors/track, 1044 cylinders, total 16775169 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk identifier: 0x00000000

PARTITION AT NODE2

Disk /dev/sdb1: 8588 MB, 8588886016 bytes
64 heads, 32 sectors/track, 8191 cylinders, total 16775168 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk identifier: 0x00000000
slm
  • 7,615
  • 16
  • 56
  • 76
shrek
  • 1
  • 2

2 Answers2

0

:(

No one knows? Ok it is not easy to test on the production infrastructure that I have but, as every little detail counts, I guess I have to resize the disks somehow to discard the trouble of the 1-sector different

shrek
  • 1
  • 2
0

Ooook I will answer myself:

After no few testing and work, I can confirm :: Partitions/Disks involved in a DRBD replication brick, must be absolutely identical in their geometry. That's all ! So If you use Virtual environment, the most reliable thing to do is just clone right the same file used for the virtual volume.

maybe this is explicitely told somewhere at DRBD wiki; me myself I did not find it.

shrek
  • 1
  • 2