Martijn van Oosterhout <kleptog@svana.org> writes:
> On Fri, Nov 11, 2005 at 05:45:28PM +0800, Lincoln Yeoh wrote:
>> Assuming the transactions don't explicitly do any locks ...
> Every transaction takes locks, on every table it accesses. Shared lock,
> but locks anyway. UPDATEs take stronger locks, so any UPDATE may cause
> other queries to wait until you COMMIT or ABORT.
Also, the mere existence of an old open transaction restricts VACUUM's
ability to reclaim dead rows.
regards, tom lane
