Re: Commit Timestamp and LSN Inversion issue

On 11/6/24 06:23, Amit Kapila wrote:

> I think we avoid calling hook/callback functions after holding a lock
> (spinlock or LWLock) as the user may do an expensive operation or
> acquire some other locks in those functions which could lead to
> deadlocks or impact the concurrency. So, it would be better to
> directly call an inline function to perform the required operation.

This is a valid concern. The reason why I kept it as a hook is because 
ReserveXLogInsertLocation() has no knowledge that this is allocating WAL 
space for a commit record. Only the caller does. We certainly need to be 
extremely careful what any such hook function is doing. Acquiring 
additional locks is definitely not acceptable. But I am not sure we 
should burden this function with specialized knowledge about what it is 
reserving WAL space for.


Regards, Jan