Re: [HACKERS] Restrict concurrent update/delete with UPDATE of partition key

On Sat, Nov 25, 2017 at 11:39 AM, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
> On Thu, Nov 23, 2017 at 5:18 PM, amul sul <sulamul(at)gmail(dot)com> wrote:
>> On Sat, Nov 11, 2017 at 1:05 AM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
>>> On Wed, Sep 27, 2017 at 7:07 AM, amul sul <sulamul(at)gmail(dot)com> wrote:
>>>>
[...]
> Few comments:
>
Thanks for looking at the patch, please find my comments inline:

> 1.
> @@ -1480,6 +1493,10 @@ ExecOnConflictUpdate(ModifyTableState *mtstate,
> ereport(ERROR,
> (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
> errmsg("could not serialize access due to concurrent update")));
> + if (!BlockNumberIsValid(BlockIdGetBlockNumber(&((hufd.ctid).ip_blkid))))
> + ereport(ERROR,
> + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
> + errmsg("tuple to be updated was already moved to an another
> partition due to concurrent update")));
>
> Why do you think we need this check in the OnConflictUpdate path? I
> think we don't it here because we are going to relinquish this version
> of the tuple and will start again and might fetch some other row
> version. Also, we don't support Insert .. On Conflict Update with
> partitioned tables, see[1], which is also an indication that at the
> very least we don't need it now.
>
Agreed, even though this case will never going to be anytime soon
shouldn't we have a check for invalid block id? IMHO, we should have
this check and error report or assert, thoughts?

> 2.
> @@ -2709,6 +2709,10 @@ EvalPlanQualFetch(EState *estate, Relation
> relation, int lockmode,
> ereport(ERROR,
> (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
> errmsg("could not serialize access due to concurrent update")));
> + if (!BlockNumberIsValid(BlockIdGetBlockNumber(&((hufd.ctid).ip_blkid))))
> + ereport(ERROR,
> + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
> + errmsg("tuple to be updated was already moved to an another
> partition due to concurrent update")));
>
> ..
> ..
> +++ b/src/backend/executor/nodeLockRows.c
> @@ -218,6 +218,11 @@ lnext:
> ereport(ERROR,
> (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
> errmsg("could not serialize access due to concurrent update")));
> + if (!BlockNumberIsValid(BlockIdGetBlockNumber(&((hufd.ctid).ip_blkid))))
> + ereport(ERROR,
> + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
> + errmsg("tuple to be locked was already moved to an another partition
> due to concurrent update")));
> +
>
> At some places after heap_lock_tuple the error message says "tuple to
> be updated .." and other places it says "tuple to be locked ..". Can
> we use the same message consistently? I think it would be better to
> use the second one.
>
Okay, will use "tuple to be locked"

> 3.
> }
> +
> /* tuple already deleted; nothing to do */
> return NULL;
>
> Spurious whitespace.
>
Sorry about that, will fix this.

> 4. There is no need to use *POC* in the name of the patch. I think
> this is no more a POC patch.
>
Understood.

Regards,
Amul