Re: problem with device validation

In article <[email protected]> you wrote:
> Peter T. Breuer a ?crit :
>> Why are you interested in [the validation flag]?

> in fact, the problem occurs in this situation :
> - primary has been stopped.
> - on secondary, enbd-client is running, but connection with server is lost
> - raid is started on secondary with only the local disk.
> - primary starts
> - enbd-server is started on primary by init script. connexion is established 
> with client.

Really? And the client has been sent SIGPWR?

> - when i try to hotadd nbd device to secondary raid, sometimes it fails to read 
> nbd superblock.

Well, if it fails to read it, the device is not there. That's a read of the
last 64K of the resource over the nbd server/client.

> I found that waiting nbd to be marked validated in nbdinfo 
> solves the problem.

Well, if it has validated, it has read the first sector of the
resource, and therefore the device is "there", and I suppose the last
64K of the resource can also be read!

> But maybe it's a wrong diagnosis and there is a better way 
> to be sure that enbd device is useable?

It sounds as though it's not connected yet. Are you sure SIGPWR is sent
to the client via the enbd-cstatd/sstatd pair? When the server starts
it is supposed to consult its list of extant connections from last time
and connect to the enbd-cstatd port on each recorded client and tell it
that is there. The enbd-cstatd on the client on receiving the news, will
consult the list of servers it is supposed to be connected to (is that
in enbd.conf? I don't recall) and if it is OK, will send the client
that is supposed to be connected to it a SIGPWR, thus forcing a
reconnect from scratch. Or it'll start a client if there is no client
(I think I recall).


Peter