Failfast and DataConnectors

[Rod Widdowson]

I'm moving this over from the users list.

The basic conversation is about data connectors, failover and fail fast.
Rather than rehash that conversation I'll state what I think we have right
now and what I think we need.

The default mode of operation has always been fail fast.  When we talk about
non fail fast we mean swallowing errors in initialization and allowing first
access to fail (presumably by rerunning the initialization).

Services can failfast or not.  Since Attribute resolution is presented as a
service that gives course granularity control only.
Data connectors can fail over.  This allows attribute resolution to proceed
even if a connector fails.  This is different, but not quite orthogonal to
failover.

The RDBMS and LDAP data connector have the capability to be configured via
native spring.  Hence if the underlying components support non-fail fast
operation a configuration can be made to do so.

The idea is to have a top level concept for all data connectors of fast
fail.  This would basically attempt to do the initialization in doInit, but
if that fails would continue to try to reinitialize on every call to
resolve, until things worked correctly.

We would need to be careful with the semantics of failover when combined
with non-fast fail.

Does this make sense?

R