Shadow set recovery and repair are handled by
volume processing, which replaces mount verification for shadow sets.
Membership failure decisions are made by the host system. Device errors
that result in inaccessibility of physical member units first utilize
the class driver's connection walking algorithm. If that fails,
a local decision is made on the shadow set membership. The rules are:
If some, but not all,
members of the set are accessible, then the local node sequentially
adjusts the membership and notifies the other hosts.
If no members are accessible,
no modifications to the set membership are made.
There are two types of volume processing: active
and passive. Active volume processing handles error processing on
a local node. Triggered by a failed I/O operation, active volume processing
also controls mount verification functions, member removal, and failover.
Passive volume processing is triggered by lock messages or by a cluster
event. Passive volume processing revalidates shadow set membership,
ensures that the shadow set reflects changes made outside the shadow
set, and handles the following functions:
Member additions from
other nodes
Member removals from other
nodes
A new node mounting the
shadow set
A node dismounting the
shadow set
A system crash on a node
that has the shadow set mounted
For more information, see the Volume Shadowing for OpenVMS manual.
