SQL/MX 3.2.1 Management Manual (H06.26+, J06.15+)

Table 22 Consequences of Network Node Loss, Reconfiguration 1
Consequences of Losing This Node on the Other Nodes
CAT_1 and
CAT_2
CAT_1 and
CAT_2CAT_1 and
CAT_2 User
Data Present?Node
Applications
Present?
Metadata
Present?
The same consequences occur as were identified in the node \A
row in Table 21 (page 276).
NoNoYes\A
The same consequences occur as were identified in the node \B
row in Table 21 (page 276). In addition, none of the applications
that contain queries to CAT_1 or CAT_2 can be executed.
YesYesYes\B
The same consequences occur as were identified in the node \C
row in Table 21 (page 276).
NoNoYes\C
Clearly, the configuration described in Table 22 (page 277) has less local autonomy than the
configuration described in Table 21 (page 276). Because node \B represents a single point of
failure, this configuration is not advisable in a geographically distributed environment. However,
in a locally distributed environment where all three nodes are in the same computer room and a
reliable high-speed network interconnect is used, the risks of this configuration are reduced.
Depending on the interrelationships between the involved catalogs, you might want to implement
the configuration described in Table 23 (page 277). In this configuration, CAT_1 is created on node
\A and CAT_2 on node \B, so their respective object metadata remain on these two nodes.
Table 23 Consequences of Network Node Loss, Reconfiguration 2
Consequences of Losing This Node on the Other Nodes
CAT_1 and
CAT_2
CAT_1 and
CAT_2
CAT_1 and
CAT_2 User
Node
Applications
Present?
Metadata
Present?
Data
Present?
From nodes \B and \C, you cannot perform DDL or utility
operations on any of the CAT_2 catalog’s objects, which
Yes for those
parts that
Yes for
CAT_1
Yes\A
have partitions on node \A. You cannot perform DDL or utilitymainly access
CAT_1 operations on CAT_1. You cannot perform dynamically
compiled or automatically recompiled DML operations on
objects in CAT_1. You cannot perform statically-compiled
DML operations that require data stored on node \A or late
name resolution of ANSI names in CAT_1.
From nodes \A and \C, you cannot perform DDL or utility
operations on any of the CAT_1 catalog’s objects, which
Yes for those
parts that
Yes for
CAT_2
Yes\B
have partitions on node \B. You cannot perform DDL or utilitymainly access
CAT_2 operations on CAT_2. You cannot perform dynamically
compiled or automatically recompiled DML operations on
any objects in CAT_2. You cannot perform statically compiled
DML operations that require data stored on node \B or late
name resolution of ANSI names in CAT_2.
From nodes \A and \B, you cannot perform DDL or utility
operations on objects partitioned across node \C. You cannot
NoNoYes\C
perform statically compiled DML operations that require data
stored on node \C.
If CAT_1 and CAT_2 have few or no interrelations and if the database objects in the two catalogs
are rarely or never accessed by the same application programs or queries, a configuration like
that shown in Table 23 (page 277) might make good sense. In this configuration, an outage of
node \A does not affect metadata access by applications and queries that access object data in
CAT_2. Likewise, an outage of node \B does not prevent metadata access by applications and
queries that access object data in CAT_1.
Managing a Network-Distributed SQL/MX Database 277