Having now flirted with recreating BLOBs (and even DBs) with matching OIDs,
I find myself thinking it may be a waste of effort for the moment. A
modified version of the system used by Pavel Janik in pg_dumplo may be
substantially more reliable than my previous proposal:
To Dump
-------
Dump all LOs by looking in pg_class for relkind='l'.
Don't bother cross-referencing with actual table entries, since we are
trying to do a backup rather than a consistency check.
The dump will consist of the LO and it's original OID.
To Load
-------
Create a temporary table, lo_xref, with appropriate indexes
Reload the LOs, storing old & new oid in lo_xref.
Now, disable triggers and sequentially search through all tables that have
one or more oid columns: for each oid column, see if the column value is in
lo_xref, if it is, update it with the new value.
For large databases, this system will rely heavily on lo_xref, so my main
worries are:
1. How are temp tables stored? (eg. if in memory this is a problem -
Dec/Rdb stores temp tables in memory).
2. Are there any limitation on indexes of temp tables (I seem to be able to
create them at least - Dec/Rdb won't even let you do that).
----------------------------------------------------------------
Philip Warner | __---_____
Albatross Consulting Pty. Ltd. |----/ - \
(A.C.N. 008 659 498) | /(@) ______---_
Tel: (+61) 0500 83 82 81 | _________ \
Fax: (+61) 0500 83 82 82 | ___________ |
Http://www.rhyme.com.au | / \| | --________--
PGP key available upon request, | /
and from pgp5.ai.mit.edu:11371 |/
