Well, I think I found the root cause. It's because of 7c15cef86d, which
changed the definition of sql_identifier so that it's a domain over name
instead of varchar. So we now have this:
SELECT typname, typlen FROM pg_type WHERE typname = 'sql_identifier':
-[ RECORD 1 ]--+---------------
typname | sql_identifier
typlen | -1
instead of this
-[ RECORD 1 ]--+---------------
typname | sql_identifier
typlen | 64
Unfortunately, that seems very much like a break of on-disk format, and
after pg_upgrade any table containing sql_identifier columns is pretty
much guaranteed to be badly mangled. For example, the first row from the
table used in the original report looks like this on PostgreSQL 11:
test=# select ctid, * from q_tbl_archiv limit 1;
-[ RECORD 1 ]----+--------------------------
ctid | (0,1)
table_name | _pg_foreign_data_wrappers
column_name | foreign_data_wrapper_name
ordinal_position | 5
col_qualifier | foreign_data_wrapper_name
id_column |
id_default |
while on PostgreSQL 12 after pg_upgrade it looks like this
test=# select ctid, table_name, column_name, ordinal_position from q_tbl_archiv limit 1;:
-[ RECORD 1 ]----+---------------------------------------------------------
ctid | (0,1)
table_name | 5_pg_foreign_data_wrappers5foreign_data_wrapper_name\x05
column_name | _data_wrapper_name
ordinal_position | 0
Not sure what to do about this :-(
regards
--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
