Re: [GENERAL] Replication of databases (esp. postgres)

Dustin,


thank you for your answer.

On Mon, 15 Feb 1999, dustin sallings wrote:

> // {Replication of r/w Databases in a two pass scheme w/ renumbering}
>     Just one note.  Renumbering isn't all that important.  I did one
> that had a non-numeric unique ID for identity type columns.

I thought of a field named db_id. This field combined with
a unique (at least to the local database) counter should give a
unique index to each relation.


The rule to follow is: Every record in a relation with the
same db_id and the same rec_id belongs together.


But think of this situation:

DB 1 gets a relation.

table1        table2
record1       record1
              record2

Now this relation gets replicated to DB 2.

After this is done, DB 2 adds a record3 to
table2 this relation.

Question: Which db_id should DB 2 use?

His own? In this way record3 wouldn't belong
to this relation.

The id of DB 1? In this way the data may
not be unique to DB 1 as DB 1 might enter
the same values for record3 for a completely
other relation and a collision is created.

So i dropped the idea of having a db_id field
to identify the record of a relation w/o
renumbering the relations by a master.


have a nice day! t++