[Linux-ha-dev] GFS task in HeartbeatTNGOpenIssue
gshi at ncsa.uiuc.edu
Fri Jul 23 14:25:26 MDT 2004
At 09:23 PM 7/23/2004 +0200, you wrote:
> Guochun Shi <gshi at ncsa.uiuc.edu> said:
>> >This consists of two parts:
>> there are three parts now :)
>> >a) Making us work with RH's membership, etc.
>> I have one concern using external membership, i.e. how do we handle
>> the "false" membership? One example is heartbeat crashed in one node,
>> that node is considered dead from heartbeat pointer of view however if
>> GFS membership module --called cluster management (CMAN) is still
>> running fine in that node, CMAN will deliver a "false" membership
>> which includes that node to heartbeat. How are we going to handle
>No, we all need to use the same membership; ie export the membership
>provided by whatever module to the others and not run multiple ones.
I am not talking about multiple memberships. I am concerned the membership
may not be correct, that if we use it --- like CRM in heartbeat --- it may crash the
program, e.g. CRM may want to move a resource to a node where heartbeat has
failed (but cman doesn't)
>> >and b) making RH's code work with any AIS membership layer,
>> Can we do anything in this task? Some pushing effort?
>Basically the same as above, we need to be able to export or import the
>AIS APIs too.
>> >and c) testing with GFS
>> You mean failover/failback with GFS using FileSystem script?
>> I think it should work as any other filesystem right now
>No, testing whether the integrated stack in fact works. GFS as a
>concurrent SAN-based cluster file system doesn't actually need to be
>controlled by a Filesystem resource.
More information about the Linux-HA-Dev