On 8/2/22 10:57, Rory Campbell-Lange wrote:
Special care must be taken with the "replica identity" of published
tables, as set out at
https://www.postgresql.org/docs/current/logical-replication-publication.html.
For example, you may need a unique identifying column for each source
table in addition to the normal row identifier to differentiate*this*
table's id 1 row from the*other* table's id 1 row, otherwise the
subscriber won't be able to identify the row to delete if this table's
id 1 row is deleted (for example).
Although this seems to work fine with native replication, the pglogical
extension has more knobs. For instance, the
pglogical.wait_for_subscription_sync_complete function is useful to ensure that
sync finishes when part of a migration.
We would literally just be merging bulk data rows that are considered
immutable, meaning they would never be updated or deleted. We would
replicate only inserts, not deletes, updates, etc.
Would the table identifier still be required in this case?
We have a half a dozen DB's that just collect call records, they are in
different locations. They get their local call data and store it into
their local table. We would want to aggregate all that data in a central
subscription database table for pure analytics/reporting purposes...
--
inoc.net!rblayzor
XMPP: rblayzor.AT.inoc.net
PGP: https://pgp.inoc.net/rblayzor/
