From: | Gregory Stark <stark(at)enterprisedb(dot)com> |
---|---|
To: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
Cc: | pgsql-hackers(at)postgresql(dot)org |
Subject: | Re: Tricky bugs in concurrent index build |
Date: | 2006-08-25 09:05:52 |
Message-ID: | 87odu9w5fz.fsf@enterprisedb.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> writes:
> At the moment it may be moot, because I've realized that validate_index
> doesn't work anyway. It is scanning the index and then assuming that
> any tuple inserted into the index subsequent to that scan will still be
> INSERT_IN_PROGRESS when the heapscan reaches it.
EGADS
Boy I feel stupid now. In fairness I think what happened is that the original
plan was, like your new plan, based on snapshots. And I only switched to using
HeapSatisfiesVacuum after several iterations. I guess there were some
assumptions in the original thinking that I never revisited.
Because of the way the AM API works changing how the initial heap scan works
is a bit of a pain. It would require either having some global state or
passing the concurrent flag through the AM methods or alternatively having a
whole new AM method.
I'll have to read (and reread) your description again in the morning
--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com
From | Date | Subject | |
---|---|---|---|
Next Message | Teodor Sigaev | 2006-08-25 09:13:13 | Re: tsvector/tsearch equality and/or portability issue |
Previous Message | ITAGAKI Takahiro | 2006-08-25 08:51:10 | Re: autovacuum cannot start when connection is full |