[OpenAFS] failover 1.6.0pre2 dafs
Sun, 27 Feb 2011 20:45:24 +0100
2011-02-27 20:15 keltezéssel, Derrick Brashear írta:
> 2011/2/27 Gémes Géza <email@example.com>:
>> I plan to run my two (1.6.0) openafs servers (both are vl pt vol and dafs
>> servers) in a failover configuration (the failover supervised by a
>> redhat-cluster installation). The data is on a SAN which is attached to both
>> I plan to define a service for both nodes with its own ip address and
>> partitions. On normal operation node1 is going to have ip address
>> 192.168.0.1 and all the partitions from vicepa to vicep... and node2 is
>> going to have ip address 192.168.0.2. If any of the nodes would fail the
>> service would get transfered to the other node, thus receiving its
>> partitions and ip address. With Netrestrict files will restrict the ip
>> addresses the pt vl and fileserver components can bind to those defined at
>> the clustering (192.168.0.1 and 192.168.0.2). The startup of any of the
>> services (on any of the nodes) would mean a complete restart of all of the
>> services controlled by bos. I also plan to put fsstate.dat of the service
>> running normally on node1 on shared storage.
> that only works for a hot failover. if the server dies, you have no
> data to read back.
> also, you need to make sure the uuid is the same, and that only one
> server can ever answer
> (only one ip address is allowed to have a reachable afs server)
>> Please tell me if I've overlooked anything. Won't get the the vl and pt
>> server confused from running on both addresses declared in CellServDB in
>> case of failover?
> that's a bad way to handle the database servers at all. CellServDB has
> no mechanism to list 2 addresses for one server, and if you list both,
> they are not the same server.
> if you have exactly one server, listed twice, the one with the higher
> IP will simply never be able to be written to.
> is allocating more servers an impossibility?
Maybe I wasn't clear about that part the one server would get to be
listed twice only if it would get failed over, in which case one server
would run with both addresses.
If that is not doable (without major problems) then I could separate the
pt and vl servers from the fileserver machine. If the failover for the
fileserver and the volumeserver seems fine.
Another question: if I would allocate 2 or 3 separate pt and vl server
machines what would happen if one or two of them would go off-line, from
the point of view of afs clients?