Bruce Momjian <[email protected]> wrote:
> I am not the author of the original ctid doc patch, but I believe the
> goal was to use ctid so we don't need to use needless index lookups for
> primary keys.
Yes, you are right, this was one of the goals. The other initial goals
were to stop people complaining about no order by and limit, and to
encourage batching. Avoiding deadlocks was very much discussed, but
not listed as one of the primary initial goals by the patch writer, if
memory serves. I was just thinking I prefer keeping ctid in lock free
examples with your new warning. ctid is safe without concurrency and
for batching to completion. There's an argument for moving the
deadlock stuff into the locking chapter because as soon as one does a
select and then an update in a concurrent world, one is guaranteed to
lose rows with ctid or any mutable key. If you think I'm overly
concerned, and the examples are fine with the locking and ctid and a
warning, I concede. The locking chapter does have good guidance on
ordering already.
