Hello, Claudio,

> Nickolay Samofatov wrote:
>> 3. DPM/VIO code has a lot of places where deadlocks are possible.
>> Many places in DPM rely on fixed 1 second timeouts to detect and
>> process deadlocks. The problem is that deadlocks are now getting
>> reported before 1 second timeout and page locking retry loop is
>> looping too quickly. If I add manual 1-second timeout in those loops
>> everything will work as it worked before.

> But I assume this is only a temporary fix while v1.5 is released, while you
> rework more code, right?

I'm testing more simple and elegant solution now. It seems it makes
retry-wait loop very responsive under contention (lock ownership
switch delay is several ns on Linux, ~40 ms on windows as compared
to several seconds or total lock-up on Windows XP earlier).

>> 4. There is a bug that if locking problem happens during
>> transaction-level savepoint backout during rollback engine reports
>> BUGCHECK(290) instead of declaring transaction as DEAD and continue
>> working.

> Mark it as dead? Is it safe or is it the only alternative?

It is safe alternative to doing BUGCHECK in this case.

> If tpb_no_auto_undo is used, you'll never encounter such case, I guess, since
> there's no list of actions to undo, hence the actions inside the txn can't
> be undone to mark it as committed.

True.

> BTW, what happens if tpb_no_auto_undo is specified at txn start time and
> explicit requests for savepoints are made: some strange condition or the
> explicit savepoints are granted and work as expected?

Explicit savepoints are granted and work as expected.

> Claudio Valderrama C.

Nickolay Samofatov