[DRBD-user] DRBD suitable for Fast Failover ?

Digimer lists at alteeve.ca
Fri May 29 17:21:18 CEST 2015

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On 29/05/15 05:38 AM, DRBD User wrote:
> 2556

?

> So a Master/Slave Configuration would make no difference (without GFS2 ...) ?
> I assume fencing is required too ...
> 
> Or may be using DRBD in standalone mode ?

Fencing is needed, regardless of the FS or applications on top of DRBD.
This is how you properly avoid split brains. The actual fence process
depends on communicating with external devices, asking them to do
something, waiting for that to be done and confirming success. Even the
fastest fence devices take time.

In my experience, you will want two fence methods, for proper
resiliency, and this means that a failure in the primary fence method
(ie IPMI) can take ~5 seconds to detect, then you start the fence
process over on the backup fence method (ie PDUs).

The fastest fence method I've used is APC/Schneider brand AP7900
switched PDUs. They tend to fence and confirm in a couple of seconds.
However, relying on them alone makes the failure of the PDU cause
fencing to fail entirely if it is needed later, leaving your cluster
blocked (exactly why I recommend dual fence methods).

-- 
Digimer
Papers and Projects: https://alteeve.ca/w/
What if the cure for cancer is trapped in the mind of a person without
access to education?



More information about the drbd-user mailing list