Question

I am doing a bulk insert of records into a database from a log file. Occasionally (~1 row out of every thousand) one of the rows violates the primary key and causes the transaction to fail. Currently, the user has to manually go through the file that caused the failure and remove the offending row before attempting to re-import. Given that there are hundreds of these files to import it is impractical.

My question: How can I skip the insertion of records that will violate the primary key constraint, without having to do a SELECT statement before each row to see if it already exists?

Note: I am aware of the very similar question #1054695, but it appears to be a SQL Server specific answer and I am using PostgreSQL (importing via Python/psycopg2).

Answer 1

You can also use SAVEPOINTs in a transaction.

Pythonish pseudocode is illustrate from the application side:

database.execute("BEGIN")
foreach data_row in input_data_dictionary:
    database.execute("SAVEPOINT bulk_savepoint")
    try:
        database.execute("INSERT", table, data_row)
    except:
        database.execute("ROLLBACK TO SAVEPOINT bulk_savepoint")
        log_error(data_row)
        error_count = error_count + 1
    else:
        database.execute("RELEASE SAVEPOINT bulk_savepoint")

if error_count > error_threshold:
    database.execute("ROLLBACK")
else:
    database.execute("COMMIT")

Edit: Here’s an actual example of this in action in psql based on a slight variation of the example in the documentation (SQL statements prefixed by “>”):

> CREATE TABLE table1 (test_field INTEGER NOT NULL PRIMARY KEY);
NOTICE:  CREATE TABLE / PRIMARY KEY will create implicit index "table1_pkey" for table "table1"
CREATE TABLE

> BEGIN;
BEGIN
> INSERT INTO table1 VALUES (1);
INSERT 0 1
> SAVEPOINT my_savepoint;
SAVEPOINT
> INSERT INTO table1 VALUES (1);
ERROR:  duplicate key value violates unique constraint "table1_pkey"
> ROLLBACK TO SAVEPOINT my_savepoint;
ROLLBACK
> INSERT INTO table1 VALUES (3);
INSERT 0 1
> COMMIT;
COMMIT
> SELECT * FROM table1;  
 test_field 
------------
          1
          3
(2 rows)

Note that the value 3 was inserted after the error, but still inside the same transaction!

The documentation for SAVEPOINT is at http://www.postgresql.org/docs/8.4/static/sql-savepoint.html.