From: Mike Eisler (mike@eisler.com)
Date: 08/16/01-06:26:42 PM Z
Message-ID: <3B7C56B2.A05405F4@eisler.com> Date: Thu, 16 Aug 2001 16:26:42 -0700 From: Mike Eisler <mike@eisler.com> Subject: Re: Replication and migration requirements > Migration of data/metadata is one thing, but transient state > like locks/delegations seems terribly fragile and complex. > I recommend the punt. ... > I think we should try to keep this protocol independent of v4 > and make it a general protocol that's independent of the > data access protocol. I'd hate to see this protocol get larded > up with requirements unusual to any single file access protocol, > e.g. replication of filehandles or file open state. If these end up being requirements, I suspect that it will means a replication/migration protocol for only read-only filesets. The client side failover experiment is 2.6+ is read-only, and yet has been very successful. This is because we know from half a decade of operational experience that locking and preservation of metadata like filehandles, fileids on read-only filesets is not needed. It is easy for me to imagine that with writeable filesets, there will be many problems without migratable locking and metadata. -mre
This archive was generated by hypermail 2.1.2 : 03/04/05-01:49:01 AM Z CST