You should always use separate connection for separate threads as the drivers are not thread safe in this manner. A connection pool could help you, as it allows connection reuse in a safe manner.
You could also do a query-dispatch pattern - if I understand your problem correctly -, where one thread is responsible for the querying and starts of N threads which might update the database - all of which have separate connection objects.
You might also consider doing a batch update via the PreparedStatement, so threads don't stumble upon each other in terms of row and table locks, using the following structure:
- 1 query thread
- NCPU processing threads
- 1 batch update thread
Like a mini db fork-join.
Edit
Examples on how to do batch update with Pstmt:
PreparedStatement pstmt = connection.prepareStatement(
"UPDATE table SET field=? WHERE id=?");
for (int i = 0; i < 100; i++) {
pstmt.setInt(1, i * i);
pstmt.setInt(2, i);
pstmt.addBatch();
}
pstmt.executeBatch();
pstmt.close();
Or you could query a Queue in the loop where the update requests arrive from the processing threads:
class WhatToUpdate {
public int id;
public int value;
}
Queue<WhatToUpdate> queue = new LinkedBlockingQueue<WhatToUpdate>();
PreparedStatement pstmt = connection.prepareStatement(
"UPDATE table SET field=? WHERE id=?");
int poisons = THE_NUMBER_OF_PROCESSING_THREADS;
while (true) {
WhatToUpdate record == queue.take();
if (record == null) { // poison pill
if (--poisons <= 0) {
break;
}
}
pstmt.setInt(1, record.value);
pstmt.setInt(2, record.id);
pstmt.addBatch();
}
pstmt.executeBatch();
pstmt.close();