Thread: Parallel Insert and Delete operation

Yelai, Ramkumar IN BLR STS worte:
> Sent: Wednesday, October 31, 2012 12:40 PM
> To: pgsql-general@postgresql.org
> Subject: [GENERAL] Parallel Insert and Delete operation
>
> Hi All,
>
> Please clarify me the following example.
>
> I have 2 tables
>
> Table1  - ( it has one primary key and few  columns )
> Table2  - ( it has one primary key and few columns.  It has one
foreign key, which refers table1
> primary key ).
>
> I have 2 operations, which are written in pl/pgsql procedure.
>
> Operation1() - Inserts the records to table2 at every hour basis.
> Operation2() - Delete the records from Table 1 and Table2 based on the
primary key.
>
> What if both operations are running at the time for the same primary
key.
>
> what I have to take care to run these two operations perfectly without
creating inconsistency in
> database.

With the foreign key in place there can be no entry in table2
that does not have a corresponding entry in table1.

Concurrency is solved with locks, so one of the concurrent
operations might have to wait until the other one is done.

That is handled by the database system automatically.

Yours,
Laurenz Albe

Ramkumar Yelai wrote:
[is worried that a database might become inconsistent if conflicting
INSERTs and DELETEs occur]
> @Albe  - I  got you first point. The second point is little skeptical
because postgres could have been
> avoided this lock by using MVCC. Please correct me if I am wrong?

Which lock could have been avoided?

PostgreSQL locks rows when the data change.
That has little to do with MVCC.

If you INSERT into a table that has a foreign key, the
referenced row in the referenced table gets a SHARE lock
that conflicts with the EXCLUSIVE lock required for
a DELETE.
So they cannot execute concurrently.

Yours,
Laurenz Albe

Ramkumar Yelai wrote:
[is worried that a database might become inconsistent if conflicting INSERTs and DELETEs occur]
> @Albe  - I  got you first point. The second point is little skeptical
because postgres could have been
> avoided this lock by using MVCC. Please correct me if I am wrong?

Which lock could have been avoided?

PostgreSQL locks rows when the data change.
That has little to do with MVCC.

If you INSERT into a table that has a foreign key, the referenced row in the referenced table gets a SHARE lock that
conflictswith the EXCLUSIVE lock required for a DELETE. 
So they cannot execute concurrently.

Yours,
Laurenz Albe

Thanks very much Albe.

I am not aware of that, delete will lock the table.