Re: Killing dead index tuples before they get vacuumed - Mailing list pgsql-hackers

From Tom Lane
Subject Re: Killing dead index tuples before they get vacuumed
Date
Msg-id 331.1022104581@sss.pgh.pa.us
Whole thread Raw
In response to Re: Killing dead index tuples before they get vacuumed  (Manfred Koizar <mkoi-pg@aon.at>)
List pgsql-hackers
Manfred Koizar <mkoi-pg@aon.at> writes:
> So it looks save to use this bit for marking dead tuples.  Wouldn't it
> be even possible to simply reset LP_USED instead of setting
> LP_DELETED?

Mmmm ... I don't think so.  That would cause the tuple to actually
disappear from the perspective of the index AM internals, which seems
like a bad idea.  (For example, if another backend had an indexscan
stopped on that same tuple, it would fail to re-find its place when it
tried to continue the indexscan.)

> Yet another idea: set ItemIdData.lp_len = 0 for killed index tuples.

See above.  This is *not* a substitute for vacuuming.
        regards, tom lane


pgsql-hackers by date:

Previous
From: "Dann Corbit"
Date:
Subject: Re: A more precise polygon_overlap()
Next
From: Tom Lane
Date:
Subject: Re: ksqo?