[prev in list] [next in list] [prev in thread] [next in thread] 

List:       postgresql-general
Subject:    Re: [GENERAL] Curious case of huge simple btree indexes bloat.
From:       Tom Lane <tgl () sss ! pgh ! pa ! us>
Date:       2015-05-31 17:18:21
Message-ID: 2088.1433092701 () sss ! pgh ! pa ! us
[Download RAW message or body]

Maxim Boguk <maxim.boguk@gmail.com> writes:
> On the one of databases under my support I found very curious case of the
> almost endless index bloat (index size stabilises around 100x of the
> original size).

> The table have 5 indexes and they all have the same bloating behaviour
> (growth to almost 100x and stabilisation around that amount). An original
> index size 4-8Mb (after manual reindex), over time of the 5 days they all
> monotonically growth to 300-900MB. In the same time table size staying
> pretty constant at 30-50Mb (and amount of rows in the same don't vary
> widely and stays between 200k and 500k).

At least for the index you gave stats for, it seems like it's stabilizing
at one index entry per page.  This is a known possible pathological
behavior if the application's usage involves heavy decimation of original
entries; say, you insert sequential timestamps and then later remove all
but every one-thousandth one, leaving at most one live entry on every
index page.  Btree can recover the totally-empty leaf pages but it has no
provision for merging non-empty leaf pages, so those all stay as they are
indefinitely.

It would be pretty unusual for all the indexes on a table to be used like
that, though.

			regards, tom lane


-- 
Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general
[prev in list] [next in list] [prev in thread] [next in thread]