Re: GiST insert algorithm rewrite

On Wed, Dec 1, 2010 at 4:00 AM, Heikki Linnakangas
<heikki(dot)linnakangas(at)enterprisedb(dot)com> wrote:
> On 01.12.2010 04:10, Robert Haas wrote:
>>
>> On Tue, Nov 30, 2010 at 10:26 AM, Heikki Linnakangas
>> <heikki(dot)linnakangas(at)enterprisedb(dot)com>  wrote:
>>>>
>>>> Does the current code cope with the corruption?
>>>
>>> It's not corruption, but "intended degradation". Yes, the current code
>>> copes
>>> with it, that's how GiST survives a crash. However, even with the current
>>> code, VACUUM will nag if it finds any invalid tuples with this message:
>>>
>>> ereport(NOTICE,
>>>        (errmsg("index \"%s\" needs VACUUM FULL or REINDEX to finish crash
>>> recovery",
>>>
>>> That's harmless, in the sense that all scans and inserts work fine, but
>>> scans might need to do more work than if the invalid tuple wasn't there.
>>>
>>> I don't think we need to go out of our way to support such degraded
>>> indexes
>>> in 9.1. If you see such notices in your logs, you should REINDEX anyway,
>>> before of after pg_upgrade. Let's just make sure that you get a
>>> reasonable
>>> error message in 9.1 if a scan or insert encounters such a tuple.
>>
>> I just don't want to take a risk of giving people unexpected wrong
>> answers.  It's not clear to me whether that's a risk here or not.
>
> You'll get an error if a scan encounters an invalid tuple.
>
> In the patch I posted, I just ripped out everything related to invalid
> tuples altogether. But we should add a check and ereport for that before
> commit.

All right, that seems like a reasonable backstop, if we're fairly sure
this won't be a common scenario.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company