ITEM: DO6453L

Q and A on HAGEO

Response:

Env: 4.1.4 SP  HAGEO 2.1

Customer rep: Primitivo

Problem: Customer has 4 HAGEO questions: (Cascading configuration)

Action taken: 

> In what circumstance will a site isolation occurr.

- When all Geo networks are disabled.
- Geo seconary network or DBFS shows remote nodes funtioning.

> When disks on one site have failed on does the GEOmirror device hang?

A remote disk failure will be returned to the application as a write 
failure with sync and mwc (I'd have to check on async).  If this is not 
what you want the disk error should be promoted to a node failure in 
order to break the mirror and start staleness markings.  On a node 
failure the I/O will hang until the gmddown is performed during the 
node_down event for sync and mwc.

> When dbfs calls, it can give three answers, I'm OK, I'm not OK, and 
> I'm not sure. When does the "I'm not OK" and "I'm not sure" occur?

The "I'm not OK" occurs when dbfs connects to the remote machine and 
determines that the cluster is down or when the remote utility cannot 
determine the state of the cluster (e.g. the clinfo daemon is down).  
The "I'm not sure" occurs when dbfs gets a timeout waiting for a 
response.  dbfs retries a few times after an "I'm not sure" response 
before assuming the cluster is down.

> If Geo primary and secondary networks fail, is there a way to ping 
> machines on the public network, to prevent the non-dominant machine 
> from being taken down.

If the Geo network is down, one of the nodes (either the local node or 
the remote peer) has to be brought down to prevent data divergence. 
It doesn't matter whether we can ping the remote node or not.  

> What would happen if there was a type of double data divergence. i.e. 
> if concurrent access was being used and somehow data was written to 
> both disks. He could clean one of the stale maps but that would not 
> sync up the other side. 

I explained that if this unlikely occurence ever happened which it 
probably wouldn't, he could take a master copy and do a 
"gmddirty -l gmd_name" and then sync that up with the other side so that 
it would be totolly clean.




Support Line: Q and A on HAGEO ITEM: DO6453L

Dated: September 1997 Category: N/A

This HTML file was generated 99/06/24~13:30:15

Comments or suggestions?
Contact us