NetBase SQL Shadowing Handbook

Troubleshooting NetBase SQL
____________________________________________________________
4-12 QUEST Software, Inc. ver A0195
SQL ERRORS
The export and the import processes may incur errors when reading records from the log file or
when updating the shadow copy of the DBE. These errors are printed to the console using
SQLEXPLAIN. A number of errors that may occur on the shadow machine are not recoverable.
In general, these are errors that indicate that the DBEs are out of sync. If the process receives
one of these errors, it prints a message and terminates.
ERROR RECOVERY
If a machine that is exporting or importing updates gets a datacomm failure, the SQL processes
will stay up and will attempt to reestablish communications with the remote node.
If a machine crashes, much the same is true. The SQL processes on the other machines will stay
up attempting to reestablish a connection. The machine that crashed will be restarted, NetBase
will be restarted, and NetBase will automatically start up the SQL processes. These processes
will connect to the remote nodes and SQL shadowing will continue.
In either case updates cannot be lost because the way that WSL works insures that shadowing
starts with the last complete update applied on the shadow machine. SQL is secure enough that
the system crash is unlikely to hurt the DBE.
If a DBE does become corrupt, or the disc fails, or the DBEs go out of sync, then the DBEs must
be resync'ed. The procedures for doing this depend on the type of failure and on the particular
system design. In general, what is required is to unload all the tables that make up the out-of-
sync partition from the master copy of the DBE. At the same time the checkpoint record must be
retrieved from the master copy's WSL log file. Then those same tables are loaded with the data
from the master copy before the logfile is updated with the checkpoint record from the master.
Remember that the partition number is a logical construct. Nothing in SQL or WSL associates a
particular table with a particular partition number. Speaking in those terms assumes that no two
master machines update the same table.