Why not report ERROR when "concurrent insert in progress within table"?

From: Meng Qingzhong <mengqingzhong(dot)gm(at)gmail(dot)com>
To: pgsql-hackers(at)postgresql(dot)org
Subject: Why not report ERROR when "concurrent insert in progress within table"?
Date: 2021-01-03 13:49:51
Message-ID: CAL+MxYn4hhVJKsmYgvZEMd8c4eamsg=BFVP1gGRLvsYPg==Cvw@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Hi, all,

When using 'create index concurrently', we use a MVCC snaptshot, we should
not see tuples that are HEAPTUPLE_INSERT_IN_PROGRESS. I think we should
report an ERROR in this case. But, we report WARNING now.

In the early days, it indeed reports ERROR, but the behaviour is changed by
commit 1ddc2703a936d03953657f43345460b9242bbed1.

I am not very clear about it. Could anyone explain it?

The following are the related codes:

Earlier PG versions: the code is in function IndexBuildHeapScan
case HEAPTUPLE_DELETE_IN_PROGRESS:
/*
* As with INSERT_IN_PROGRESS case, this is unexpected
* unless it's our own deletion or a system catalog.
*/
Assert(!(heapTuple->t_data->t_infomask & HEAP_XMAX_IS_MULTI));
xwait = HeapTupleHeaderGetXmax(heapTuple->t_data);
if (!TransactionIdIsCurrentTransactionId(xwait))
{
if (!is_system_catalog)
elog(WARNING, "concurrent delete in progress within table
\"%s\"",
RelationGetRelationName(heapRelation));

Newest PG version: the code in function heapam_index_build_range_scan

Thanks all!

Regards,
Meng Qingzhong

Browse pgsql-hackers by date

  From Date Subject
Next Message Tomas Vondra 2021-01-03 14:35:39 Re: Handing off SLRU fsyncs to the checkpointer
Previous Message Amit Kapila 2021-01-03 13:34:40 Re: [Patch] Optimize dropping of relation buffers using dlist