In the last episode (Jul 13), Tim Bunce said:
> On Thu, Jul 13, 2000 at 02:51:40PM +0300, Michael Widenius wrote:
> > The problem is that BDB uses page locks to guard against multiple
> > transactions and this is what you are seeing.
> > On big databases, this is not normally noticeable, but it's notable
> > on small tables like in your example.
> And a standard workaround is to pad the row to be the size of a page
> by adding a dummy char field. (Assuming it's a fixed width table,
> else use multiple long int fields.)
But in Pavel's case, it doesn't matter how you pad, because the hanging
query was a "select *", which isn't a transaction. During a
transaction, there should only be a _write_ lock on the affected pages,
right? Reads from the table should use the rollback information to
present a view of the table as it was before the transaction was