Re: Minimal logical decoding on standbys

Hi,

On 4/4/23 1:43 PM, Amit Kapila wrote:
> On Tue, Apr 4, 2023 at 3:14 PM Drouvot, Bertrand
> <bertranddrouvot.pg@gmail.com> wrote:
>>
> 
> 
> +static inline bool
> +LogicalReplicationSlotXidsConflict(ReplicationSlot *s, TransactionId xid)
> +{
> + TransactionId slot_xmin;
> + TransactionId slot_catalog_xmin;
> +
> + slot_xmin = s->data.xmin;
> + slot_catalog_xmin = s->data.catalog_xmin;
> +
> + return (((TransactionIdIsValid(slot_xmin) &&
> TransactionIdPrecedesOrEquals(slot_xmin, xid)) ||
> 
> For logical slots, slot->data.xmin will always be an
> InvalidTransactionId. It will only be set/updated for physical slots.
> So, it is not clear to me why in this and other related functions, you
> are referring to and or invalidating it.
> 

I think you're right that invalidating/checking only on the catalog xmin is
enough for logical slot (I'm not sure how I ended up taking the xmin into account but
that seems useless indeed).

I'll submit a new version to deal with the catalog xmin only, thanks!

Regards,

-- 
Bertrand Drouvot
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com