Explicit timestamp conversion very slow

From: "Hamby, Mark CONTRACTOR HAMBYM" <MARK(dot)HAMBY(at)LEAVENWORTH(dot)ARMY(dot)MIL>
To: pgsql-bugs(at)postgresql(dot)org
Subject: Explicit timestamp conversion very slow
Date: 2002-09-10 14:18:24
Message-ID: C3F752D3A4C7D41192190008C791CC3D03626D72@kshgnc06.leavenworth.army.mil
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-bugs

NOTICE: May not be considered a bug. More of an observation.

When specifying 'timestamp' to explicitly convert a string constant to a

timestamp, the query is more than 2000 time slower than the same query

without 'timestamp'. Output and more info below. Last two queries are

without and with the 'timestamp' specification.

No one may care, but I thought I would report it.

Thanks,

Mark Hamby

/* PG VERSION */

select version();

version

------------------------------------------------------------------

PostgreSQL 7.2.1 on sparc-sun-solaris2.7, compiled by GCC 2.95.2

(1 row)

/* THE DATA TABLE */

\d truth_units

Table "truth_units"

Column | Type | Modifiers

----------------+-----------------------------+-----------

time | timestamp without time zone |

unit_id | integer |

unit | character varying(13) |

color | character(1) |

subtype | integer |

symbol | integer |

echelon | integer |

hhq | integer |

in_combat | integer |

utm | character varying(11) |

latitude | real |

longitude | real |

direction | integer |

shooting | integer |

strength | integer |

speed | integer |

posture | integer |

radar_deployed | integer |

radar_on | integer |

Indexes: truth_units_unique

/* THE INDEX */

\d truth_units_unique

Index "truth_units_unique"

Column | Type

---------+-----------------------------

time | timestamp without time zone

unit_id | integer

unit | character varying(13)

btree

/* NUMBER OF DATA ROWS*/

select count(*) from truth_units;

count

--------

717334

(1 row)

/* NORMAL QUERY USES INDEX AND IS QUICK. */

explain analyze

select count(*)

from truth_units

where time = '2002-08-15 10:00'

;

psql:timestamp.sql:22: NOTICE: QUERY PLAN:

Aggregate (cost=13460.55..13460.55 rows=1 width=0) (actual
time=379.59..379.59 rows=1 loops=1)

-> Index Scan using truth_units_unique on truth_units
(cost=0.00..13451.58 rows=3587 width=0) (actual time=0.53..182.56 rows=3982
loops=1)

Total runtime: 380.00 msec

EXPLAIN

/* QUERY WITH EXPLICIT TIMESTAMP CONVERSION */

/* DOES NOT USE INDEX AND IS VERY SLOW. */

/* DOES ANYONE CARE? */

explain analyze

select count(*)

from truth_units

where time = timestamp '2002-08-15 10:00'

;

psql:timestamp.sql:32: NOTICE: QUERY PLAN:

Aggregate (cost=23484.98..23484.98 rows=1 width=0) (actual
time=775731.52..775731.53 rows=1 loops=1)

-> Seq Scan on truth_units (cost=0.00..23476.01 rows=3587 width=0)
(actual time=303737.26..775252.14 rows=3982 loops=1)

Total runtime: 775731.86 msec

EXPLAIN

Responses

Browse pgsql-bugs by date

  From Date Subject
Next Message Tom Lane 2002-09-10 16:15:42 Re: make unsuccessful on mac os x 10.2
Previous Message Mats Lofkvist 2002-09-10 07:36:29 Re: _bt_check_unique checks every row in table when doing update??