I want to apply pagination on a table with huge data. All I want to know a better option than using OFFSET in SQL Server.
Here is my simple query:
SELECT *
FROM TableName
ORDER BY Id DESC
OFFSET 30000000 ROWS
FETCH NEXT 20 ROWS ONLY
You can use Keyset Pagination for this. It's far more efficient than using Rowset Pagination (paging by row number).
In Rowset Pagination, all previous rows must be read, before being able to read the next page. Whereas in Keyset Pagination, the server can jump immediately to the correct place in the index, so no extra rows are read that do not need to be.
For this to perform well, you need to have a unique index on that key, which includes any other columns you need to query.
In this type of pagination, you cannot jump to a specific page number. You jump to a specific key and read from there. So you need to save the unique ID of page you are on and skip to the next. Alternatively, you could calculate or estimate a starting point for each page up-front.
One big benefit, apart from the obvious efficiency gain, is avoiding the "missing row" problem when paginating, caused by rows being removed from previously read pages. This does not happen when paginating by key, because the key does not change.
Here is an example:
Let us assume you have a table called TableName with an index on Id, and you want to start at the latest Id value and work backwards.
You begin with:
SELECT TOP (#numRows)
*
FROM TableName
ORDER BY Id DESC;
Note the use of ORDER BY to ensure the order is correct
In some RDBMSs you need LIMIT instead of TOP
The client will hold the last received Id value (the lowest in this case). On the next request, you jump to that key and carry on:
SELECT TOP (#numRows)
*
FROM TableName
WHERE Id < #lastId
ORDER BY Id DESC;
Note the use of < not <=
In case you were wondering, in a typical B-Tree+ index, the row with the indicated ID is not read, it's the row after it that's read.
The key chosen must be unique, so if you are paging by a non-unique column then you must add a second column to both ORDER BY and WHERE. You would need an index on OtherColumn, Id for example, to support this type of query. Don't forget INCLUDE columns on the index.
SQL Server does not support row/tuple comparators, so you cannot do (OtherColumn, Id) < (#lastOther, #lastId) (this is however supported in PostgreSQL, MySQL, MariaDB and SQLite).
Instead you need the following:
SELECT TOP (#numRows)
*
FROM TableName
WHERE (
(OtherColumn = #lastOther AND Id < #lastId)
OR OtherColumn < #lastOther
)
ORDER BY
OtherColumn DESC,
Id DESC;
This is more efficient than it looks, as SQL Server can convert this into a proper < over both values.
The presence of NULLs complicates things further. You may want to query those rows separately.
On very big merchant website we use a technic compound of ids stored in a pseudo temporary table and join with this table to the rows of the product table.
Let me talk with a clear example.
We have a table design this way :
CREATE TABLE S_TEMP.T_PAGINATION_PGN
(PGN_ID BIGINT IDENTITY(-9 223 372 036 854 775 808, 1) PRIMARY KEY,
PGN_SESSION_GUID UNIQUEIDENTIFIER NOT NULL,
PGN_SESSION_DATE DATETIME2(0) NOT NULL,
PGN_PRODUCT_ID INT NOT NULL,
PGN_SESSION_ORDER INT NOT NULL);
CREATE INDEX X_PGN_SESSION_GUID_ORDER
ON S_TEMP.T_PAGINATION_PGN (PGN_SESSION_GUID, PGN_SESSION_ORDER)
INCLUDE (PGN_SESSION_ORDER);
CREATE INDEX X_PGN_SESSION_DATE
ON S_TEMP.T_PAGINATION_PGN (PGN_SESSION_DATE);
We have a very big product table call T_PRODUIT_PRD and a customer filtered it with many predicates. We INSERT rows from the filtered SELECT into this table this way :
DECLARE #SESSION_ID UNIQUEIDENTIFIER = NEWID();
INSERT INTO S_TEMP.T_PAGINATION_PGN
SELECT #SESSION_ID , SYSUTCDATETIME(), PRD_ID,
ROW_NUMBER() OVER(ORDER BY --> custom order by
FROM dbo.T_PRODUIT_PRD
WHERE ... --> custom filter
Then everytime we need a desired page, compound of #N products we add a join to this table as :
...
JOIN S_TEMP.T_PAGINATION_PGN
ON PGN_SESSION_GUID = #SESSION_ID
AND 1 + (PGN_SESSION_ORDER / #N) = #DESIRED_PAGE_NUMBER
AND PGN_PRODUCT_ID = dbo.T_PRODUIT_PRD.PRD_ID
All the indexes will do the job !
Of course, regularly we have to purge this table and this is why we have a scheduled job which deletes the rows whose sessions were generated more than 4 hours ago :
DELETE FROM S_TEMP.T_PAGINATION_PGN
WHERE PGN_SESSION_DATE < DATEADD(hour, -4, SYSUTCDATETIME());
In the same spirit as SQLPro solution, I propose:
WITH CTE AS
(SELECT 30000000 AS N
UNION ALL SELECT N-1 FROM CTE
WHERE N > 30000000 +1 - 20)
SELECT T.* FROM CTE JOIN TableName T ON CTE.N=T.ID
ORDER BY CTE.N DESC
Tried with 2 billion lines and it's instant !
Easy to make it a stored procedure...
Of course, valid if ids follow each other.
I am using Oracle (Enterprise Edition 10g) and I have a query like this:
SELECT * FROM (
SELECT * FROM MyTable
ORDER BY MyColumn
) WHERE rownum <= 10;
MyColumn is indexed, however, Oracle is for some reason doing a full table scan before it cuts the first 10 rows. So for a table with 4 million records the above takes around 15 seconds.
Now consider this equivalent query:
SELECT MyTable.*
FROM
(SELECT rid
FROM
(SELECT rowid as rid
FROM MyTable
ORDER BY MyColumn
)
WHERE rownum <= 10
)
INNER JOIN MyTable
ON MyTable.rowid = rid
ORDER BY MyColumn;
Here Oracle scans the index and finds the top 10 rowids, and then uses nested loops to find the 10 records by rowid. This takes less than a second for a 4 million table.
My first question is why is the optimizer taking such an apparently bad decision for the first query above?
An my second and most important question is: is it possible to make the first query perform better. I have a specific need to use the first query as unmodified as possible. I am looking for something simpler than my second query above. Thank you!
Please note that for particular reasons I am unable to use the /*+ FIRST_ROWS(n) */ hint, or the ROW_NUMBER() OVER (ORDER BY column) construct.
If this is acceptable in your case, adding a WHERE ... IS NOT NULL clause will help the optimizer to use the index instead of doing a full table scan when using an ORDER BY clause:
SELECT * FROM (
SELECT * FROM MyTable
WHERE MyColumn IS NOT NULL
-- ^^^^^^^^^^^^^^^^^^^^
ORDER BY MyColumn
) WHERE rownum <= 10;
The rational is Oracle does not store NULL values in the index. As your query was originally written, the optimizer took the decision of doing a full table scan, as if there was less than 10 non-NULL values, it should retrieve some "NULL rows" to "fill in" the remaining rows. Apparently it is not smart enough to check first if the index contains enough rows...
With the added WHERE MyColumn IS NOT NULL, you inform the optimizer that you don't want in any circumstances any row having NULL in MyColumn. So it can blindly use the index without worrying about hypothetical rows having NULL in MyColumn.
For the same reason, declaring the ORDER BY column as NOT NULL should prevent the optimizer to do a full table scan. So, if you can change the schema, a cleaner option would be:
ALTER TABLE MyTable MODIFY (MyColumn NOT NULL);
See http://sqlfiddle.com/#!4/e3616/1 for various comparisons (click on view execution plan)
I want a random selection of rows in PostgreSQL, I tried this:
select * from table where random() < 0.01;
But some other recommend this:
select * from table order by random() limit 1000;
I have a very large table with 500 Million rows, I want it to be fast.
Which approach is better? What are the differences? What is the best way to select random rows?
Fast ways
Given your specifications (plus additional info in the comments),
You have a numeric ID column (integer numbers) with only few (or moderately few) gaps.
Obviously no or few write operations.
Your ID column has to be indexed! A primary key serves nicely.
The query below does not need a sequential scan of the big table, only an index scan.
First, get estimates for the main query:
SELECT count(*) AS ct -- optional
, min(id) AS min_id
, max(id) AS max_id
, max(id) - min(id) AS id_span
FROM big;
The only possibly expensive part is the count(*) (for huge tables). Given above specifications, you don't need it. An estimate to replace the full count will do just fine, available at almost no cost:
SELECT (reltuples / relpages * (pg_relation_size(oid) / 8192))::bigint AS ct
FROM pg_class
WHERE oid = 'big'::regclass; -- your table name
Detailed explanation:
Fast way to discover the row count of a table in PostgreSQL
As long as ct isn't much smaller than id_span, the query will outperform other approaches.
WITH params AS (
SELECT 1 AS min_id -- minimum id <= current min id
, 5100000 AS id_span -- rounded up. (max_id - min_id + buffer)
)
SELECT *
FROM (
SELECT p.min_id + trunc(random() * p.id_span)::integer AS id
FROM params p
, generate_series(1, 1100) g -- 1000 + buffer
GROUP BY 1 -- trim duplicates
) r
JOIN big USING (id)
LIMIT 1000; -- trim surplus
Generate random numbers in the id space. You have "few gaps", so add 10 % (enough to easily cover the blanks) to the number of rows to retrieve.
Each id can be picked multiple times by chance (though very unlikely with a big id space), so group the generated numbers (or use DISTINCT).
Join the ids to the big table. This should be very fast with the index in place.
Finally trim surplus ids that have not been eaten by dupes and gaps. Every row has a completely equal chance to be picked.
Short version
You can simplify this query. The CTE in the query above is just for educational purposes:
SELECT *
FROM (
SELECT DISTINCT 1 + trunc(random() * 5100000)::integer AS id
FROM generate_series(1, 1100) g
) r
JOIN big USING (id)
LIMIT 1000;
Refine with rCTE
Especially if you are not so sure about gaps and estimates.
WITH RECURSIVE random_pick AS (
SELECT *
FROM (
SELECT 1 + trunc(random() * 5100000)::int AS id
FROM generate_series(1, 1030) -- 1000 + few percent - adapt to your needs
LIMIT 1030 -- hint for query planner
) r
JOIN big b USING (id) -- eliminate miss
UNION -- eliminate dupe
SELECT b.*
FROM (
SELECT 1 + trunc(random() * 5100000)::int AS id
FROM random_pick r -- plus 3 percent - adapt to your needs
LIMIT 999 -- less than 1000, hint for query planner
) r
JOIN big b USING (id) -- eliminate miss
)
TABLE random_pick
LIMIT 1000; -- actual limit
We can work with a smaller surplus in the base query. If there are too many gaps so we don't find enough rows in the first iteration, the rCTE continues to iterate with the recursive term. We still need relatively few gaps in the ID space or the recursion may run dry before the limit is reached - or we have to start with a large enough buffer which defies the purpose of optimizing performance.
Duplicates are eliminated by the UNION in the rCTE.
The outer LIMIT makes the CTE stop as soon as we have enough rows.
This query is carefully drafted to use the available index, generate actually random rows and not stop until we fulfill the limit (unless the recursion runs dry). There are a number of pitfalls here if you are going to rewrite it.
Wrap into function
For repeated use with the same table with varying parameters:
CREATE OR REPLACE FUNCTION f_random_sample(_limit int = 1000, _gaps real = 1.03)
RETURNS SETOF big
LANGUAGE plpgsql VOLATILE ROWS 1000 AS
$func$
DECLARE
_surplus int := _limit * _gaps;
_estimate int := ( -- get current estimate from system
SELECT (reltuples / relpages * (pg_relation_size(oid) / 8192))::bigint
FROM pg_class
WHERE oid = 'big'::regclass);
BEGIN
RETURN QUERY
WITH RECURSIVE random_pick AS (
SELECT *
FROM (
SELECT 1 + trunc(random() * _estimate)::int
FROM generate_series(1, _surplus) g
LIMIT _surplus -- hint for query planner
) r (id)
JOIN big USING (id) -- eliminate misses
UNION -- eliminate dupes
SELECT *
FROM (
SELECT 1 + trunc(random() * _estimate)::int
FROM random_pick -- just to make it recursive
LIMIT _limit -- hint for query planner
) r (id)
JOIN big USING (id) -- eliminate misses
)
TABLE random_pick
LIMIT _limit;
END
$func$;
Call:
SELECT * FROM f_random_sample();
SELECT * FROM f_random_sample(500, 1.05);
Generic function
We can make this generic to work for any table with a unique integer column (typically the PK): Pass the table as polymorphic type and (optionally) the name of the PK column and use EXECUTE:
CREATE OR REPLACE FUNCTION f_random_sample(_tbl_type anyelement
, _id text = 'id'
, _limit int = 1000
, _gaps real = 1.03)
RETURNS SETOF anyelement
LANGUAGE plpgsql VOLATILE ROWS 1000 AS
$func$
DECLARE
-- safe syntax with schema & quotes where needed
_tbl text := pg_typeof(_tbl_type)::text;
_estimate int := (SELECT (reltuples / relpages
* (pg_relation_size(oid) / 8192))::bigint
FROM pg_class -- get current estimate from system
WHERE oid = _tbl::regclass);
BEGIN
RETURN QUERY EXECUTE format(
$$
WITH RECURSIVE random_pick AS (
SELECT *
FROM (
SELECT 1 + trunc(random() * $1)::int
FROM generate_series(1, $2) g
LIMIT $2 -- hint for query planner
) r(%2$I)
JOIN %1$s USING (%2$I) -- eliminate misses
UNION -- eliminate dupes
SELECT *
FROM (
SELECT 1 + trunc(random() * $1)::int
FROM random_pick -- just to make it recursive
LIMIT $3 -- hint for query planner
) r(%2$I)
JOIN %1$s USING (%2$I) -- eliminate misses
)
TABLE random_pick
LIMIT $3;
$$
, _tbl, _id
)
USING _estimate -- $1
, (_limit * _gaps)::int -- $2 ("surplus")
, _limit -- $3
;
END
$func$;
Call with defaults (important!):
SELECT * FROM f_random_sample(null::big); --!
Or more specifically:
SELECT * FROM f_random_sample(null::"my_TABLE", 'oDD ID', 666, 1.15);
About the same performance as the static version.
Related:
Refactor a PL/pgSQL function to return the output of various SELECT queries - chapter "Various complete table types"
Return SETOF rows from PostgreSQL function
Format specifier for integer variables in format() for EXECUTE?
INSERT with dynamic table name in trigger function
This is safe against SQL injection. See:
Table name as a PostgreSQL function parameter
SQL injection in Postgres functions vs prepared queries
Possible alternative
I your requirements allow identical sets for repeated calls (and we are talking about repeated calls) consider a MATERIALIZED VIEW. Execute above query once and write the result to a table. Users get a quasi random selection at lightening speed. Refresh your random pick at intervals or events of your choosing.
Postgres 9.5 introduces TABLESAMPLE SYSTEM (n)
Where n is a percentage. The manual:
The BERNOULLI and SYSTEM sampling methods each accept a single
argument which is the fraction of the table to sample, expressed as a
percentage between 0 and 100. This argument can be any real-valued expression.
Bold emphasis mine. It's very fast, but the result is not exactly random. The manual again:
The SYSTEM method is significantly faster than the BERNOULLI method
when small sampling percentages are specified, but it may return a
less-random sample of the table as a result of clustering effects.
The number of rows returned can vary wildly. For our example, to get roughly 1000 rows:
SELECT * FROM big TABLESAMPLE SYSTEM ((1000 * 100) / 5100000.0);
Related:
Fast way to discover the row count of a table in PostgreSQL
Or install the additional module tsm_system_rows to get the number of requested rows exactly (if there are enough) and allow for the more convenient syntax:
SELECT * FROM big TABLESAMPLE SYSTEM_ROWS(1000);
See Evan's answer for details.
But that's still not exactly random.
You can examine and compare the execution plan of both by using
EXPLAIN select * from table where random() < 0.01;
EXPLAIN select * from table order by random() limit 1000;
A quick test on a large table1 shows, that the ORDER BY first sorts the complete table and then picks the first 1000 items. Sorting a large table not only reads that table but also involves reading and writing temporary files. The where random() < 0.1 only scans the complete table once.
For large tables this might not what you want as even one complete table scan might take to long.
A third proposal would be
select * from table where random() < 0.01 limit 1000;
This one stops the table scan as soon as 1000 rows have been found and therefore returns sooner. Of course this bogs down the randomness a bit, but perhaps this is good enough in your case.
Edit: Besides of this considerations, you might check out the already asked questions for this. Using the query [postgresql] random returns quite a few hits.
quick random row selection in Postgres
How to retrieve randomized data rows from a postgreSQL table?
postgres: get random entries from table - too slow
And a linked article of depez outlining several more approaches:
http://www.depesz.com/index.php/2007/09/16/my-thoughts-on-getting-random-row/
1 "large" as in "the complete table will not fit into the memory".
postgresql order by random(), select rows in random order:
These are all slow because they do a tablescan to guarantee that every row gets an exactly equal chance of being chosen:
select your_columns from your_table ORDER BY random()
select * from
(select distinct your_columns from your_table) table_alias
ORDER BY random()
select your_columns from your_table ORDER BY random() limit 1
If you know how many rows are in the table N:
offset by floored random is constant time. However I am NOT convinced that OFFSET is producing a true random sample. It's simulating it by getting 'the next bunch' and tablescanning that, so you can step through, which isn't quite the same as above.
SELECT myid FROM mytable OFFSET floor(random() * N) LIMIT 1;
Roll your own constant Time Select Random N rows with periodic table scan to be absolutely sure of a random:
If your table is huge then the above table-scans are a show stopper taking up to 5 minutes to finish.
To go faster you can schedule a behind the scenes nightly table-scan reindexing which will guarantee a perfectly random selection in an O(1) constant-time speed, except during the nightly reindexing table-scan, where it must wait for maintenance to finish before you may receive another random row.
--Create a demo table with lots of random nonuniform data, big_data
--is your huge table you want to get random rows from in constant time.
drop table if exists big_data;
CREATE TABLE big_data (id serial unique, some_data text );
CREATE INDEX ON big_data (id);
--Fill it with a million rows which simulates your beautiful data:
INSERT INTO big_data (some_data) SELECT md5(random()::text) AS some_data
FROM generate_series(1,10000000);
--This delete statement puts holes in your index
--making it NONuniformly distributed
DELETE FROM big_data WHERE id IN (2, 4, 6, 7, 8);
--Do the nightly maintenance task on a schedule at 1AM.
drop table if exists big_data_mapper;
CREATE TABLE big_data_mapper (id serial, big_data_id int);
CREATE INDEX ON big_data_mapper (id);
CREATE INDEX ON big_data_mapper (big_data_id);
INSERT INTO big_data_mapper(big_data_id) SELECT id FROM big_data ORDER BY id;
--We have to use a function because the big_data_mapper might be out-of-date
--in between nightly tasks, so to solve the problem of a missing row,
--you try again until you succeed. In the event the big_data_mapper
--is broken, it tries 25 times then gives up and returns -1.
CREATE or replace FUNCTION get_random_big_data_id()
RETURNS int language plpgsql AS $$
declare
response int;
BEGIN
--Loop is required because big_data_mapper could be old
--Keep rolling the dice until you find one that hits.
for counter in 1..25 loop
SELECT big_data_id
FROM big_data_mapper OFFSET floor(random() * (
select max(id) biggest_value from big_data_mapper
)
) LIMIT 1 into response;
if response is not null then
return response;
end if;
end loop;
return -1;
END;
$$;
--get a random big_data id in constant time:
select get_random_big_data_id();
--Get 1 random row from big_data table in constant time:
select * from big_data where id in (
select get_random_big_data_id() from big_data limit 1
);
┌─────────┬──────────────────────────────────┐
│ id │ some_data │
├─────────┼──────────────────────────────────┤
│ 8732674 │ f8d75be30eff0a973923c413eaf57ac0 │
└─────────┴──────────────────────────────────┘
--Get 4 random rows from big_data in constant time:
select * from big_data where id in (
select get_random_big_data_id() from big_data limit 3
);
┌─────────┬──────────────────────────────────┐
│ id │ some_data │
├─────────┼──────────────────────────────────┤
│ 2722848 │ fab6a7d76d9637af89b155f2e614fc96 │
│ 8732674 │ f8d75be30eff0a973923c413eaf57ac0 │
│ 9475611 │ 36ac3eeb6b3e171cacd475e7f9dade56 │
└─────────┴──────────────────────────────────┘
--Test what happens when big_data_mapper stops receiving
--nightly reindexing.
delete from big_data_mapper where 1=1;
select get_random_big_data_id(); --It tries 25 times, and returns -1
--which means wait N minutes and try again.
Adapted from: https://www.gab.lc/articles/bigdata_postgresql_order_by_random
Alternatively if all the above is too much work.
A simpler good 'nuff solution for constant time select random row is to make a new column on your big table called big_data.mapper_int make it not null with a unique index. Every night reset the column with a unique integer between 1 and max(n). To get a random row you "choose a random integer between 0 and max(id)" and return the row where mapper_int is that. If there's no row by that id, because the row has changed since re-index, choose another random row. If a row is added to big_data.mapper_int then populate it with max(id) + 1
Alternatively TableSample to the rescue:
If you have postgresql version > 9.5 then tablesample can do a constant time random sample without a heavy tablescan.
https://wiki.postgresql.org/wiki/TABLESAMPLE_Implementation
--Select 1 percent of rows from yourtable,
--display the first 100 rows, order by column a_column
select * from yourtable TABLESAMPLE SYSTEM (1)
order by a_column
limit 100;
TableSample is doing some stuff behind the scenes that takes some time and I don't like it, but is faster than order by random(). Good, fast, cheap, choose any two on this job.
Starting with PostgreSQL 9.5, there's a new syntax dedicated to getting random elements from a table :
SELECT * FROM mytable TABLESAMPLE SYSTEM (5);
This example will give you 5% of elements from mytable.
See more explanation on the documentation: http://www.postgresql.org/docs/current/static/sql-select.html
The one with the ORDER BY is going to be the slower one.
select * from table where random() < 0.01; goes record by record, and decides to randomly filter it or not. This is going to be O(N) because it only needs to check each record once.
select * from table order by random() limit 1000; is going to sort the entire table, then pick the first 1000. Aside from any voodoo magic behind the scenes, the order by is O(N * log N).
The downside to the random() < 0.01 one is that you'll get a variable number of output records.
Note, there is a better way to shuffling a set of data than sorting by random: The Fisher-Yates Shuffle, which runs in O(N). Implementing the shuffle in SQL sounds like quite the challenge, though.
select * from table order by random() limit 1000;
If you know how many rows you want, check out tsm_system_rows.
tsm_system_rows
module provides the table sampling method SYSTEM_ROWS, which can be used in the TABLESAMPLE clause of a SELECT command.
This table sampling method accepts a single integer argument that is the maximum number of rows to read. The resulting sample will always contain exactly that many rows, unless the table does not contain enough rows, in which case the whole table is selected. Like the built-in SYSTEM sampling method, SYSTEM_ROWS performs block-level sampling, so that the sample is not completely random but may be subject to clustering effects, especially if only a small number of rows are requested.
First install the extension
CREATE EXTENSION tsm_system_rows;
Then your query,
SELECT *
FROM table
TABLESAMPLE SYSTEM_ROWS(1000);
Here is a decision that works for me. I guess it's very simple to understand and execute.
SELECT
field_1,
field_2,
field_2,
random() as ordering
FROM
big_table
WHERE
some_conditions
ORDER BY
ordering
LIMIT 1000;
If you want just one row, you can use a calculated offset derived from count.
select * from table_name limit 1
offset floor(random() * (select count(*) from table_name));
One lesson from my experience:
offset floor(random() * N) limit 1 is not faster than order by random() limit 1.
I thought the offset approach would be faster because it should save the time of sorting in Postgres. Turns out it wasn't.
I think the best and simplest way in postgreSQL is:
SELECT * FROM tableName ORDER BY random() LIMIT 1
A variation of the materialized view "Possible alternative" outlined by Erwin Brandstetter is possible.
Say, for example, that you don't want duplicates in the randomized values that are returned. An example use case is to generate short codes which can only be used once.
The primary table containing your (non-randomized) set of values must have some expression that determines which rows are "used" and which aren't — here I'll keep it simple by just creating a boolean column with the name used.
Assume this is the input table (additional columns may be added as they do not affect the solution):
id_values id | used
----+--------
1 | FALSE
2 | FALSE
3 | FALSE
4 | FALSE
5 | FALSE
...
Populate the ID_VALUES table as needed. Then, as described by Erwin, create a materialized view that randomizes the ID_VALUES table once:
CREATE MATERIALIZED VIEW id_values_randomized AS
SELECT id
FROM id_values
ORDER BY random();
Note that the materialized view does not contain the used column, because this will quickly become out-of-date. Nor does the view need to contain other columns that may be in the id_values table.
In order to obtain (and "consume") random values, use an UPDATE-RETURNING on id_values, selecting id_values from id_values_randomized with a join, and applying the desired criteria to obtain only relevant possibilities. For example:
UPDATE id_values
SET used = TRUE
WHERE id_values.id IN
(SELECT i.id
FROM id_values_randomized r INNER JOIN id_values i ON i.id = r.id
WHERE (NOT i.used)
LIMIT 1)
RETURNING id;
Change LIMIT as necessary -- if you need multiple random values at a time, change LIMIT to n where n is the number of values needed.
With the proper indexes on id_values, I believe the UPDATE-RETURNING should execute very quickly with little load. It returns randomized values with one database round-trip. The criteria for "eligible" rows can be as complex as required. New rows can be added to the id_values table at any time, and they will become accessible to the application as soon as the materialized view is refreshed (which can likely be run at an off-peak time). Creation and refresh of the materialized view will be slow, but it only needs to be executed when new id's added to the id_values table need to be made available.
Add a column called r with type serial. Index r.
Assume we have 200,000 rows, we are going to generate a random number n, where 0 < n <= 200, 000.
Select rows with r > n, sort them ASC and select the smallest one.
Code:
select * from YOUR_TABLE
where r > (
select (
select reltuples::bigint AS estimate
from pg_class
where oid = 'public.YOUR_TABLE'::regclass) * random()
)
order by r asc limit(1);
The code is self-explanatory. The subquery in the middle is used to quickly estimate the table row counts from https://stackoverflow.com/a/7945274/1271094 .
In application level you need to execute the statement again if n > the number of rows or need to select multiple rows.
I know I'm a little late to the party, but I just found this awesome tool called pg_sample:
pg_sample - extract a small, sample dataset from a larger PostgreSQL database while maintaining referential integrity.
I tried this with a 350M rows database and it was really fast, don't know about the randomness.
./pg_sample --limit="small_table = *" --limit="large_table = 100000" -U postgres source_db | psql -U postgres target_db
I apologize in advance for my long-winded question and if the formatting isn't up to par (newbie), here goes.
I have a table MY_TABLE with the following schema -
MY_ID | TYPE | REC_COUNT
1 | A | 1
1 | B | 3
2 | A | 0
2 | B | 0
....
The first column corresponds to an ID, the second is some type and 3rd some count. NOTE that the MY_ID column is not the primary key, there could be many records having the same MY_ID.
I want to write a stored procedure which will take an array of IDs and return the subset of them that match the following criteria -
the ID should match the MY_ID field of at least 1 record in the table and at least 1 matching record should not have TYPE = A OR REC_COUNT = 0.
This is the procedure I came up with -
PROCEDURE get_id_subset(
iIds IN ID_ARRAY,
oMatchingIds OUT NOCOPY ID_ARRAY
)
IS
BEGIN
SELECT t.column_value
BULK COLLECT INTO oMatchingIds
FROM TABLE(CAST(iIds AS ID_ARRAY)) t
WHERE EXISTS (
SELECT /*+ NL_SJ */ 1
FROM MY_TABLE m
WHERE (m.my_id = t.column_value)
AND (m.type != 'A' OR m.rec_count != 0)
);
END get_id_subset;
But I really care about performance and some IDs could match 1000s of records in the table. There is an index on the MY_ID and TYPE column but no index on the REC_COUNT column. So I was thinking if there are more than 1000 rows that have a matching MY_ID field then I'll just return the ID without applying the TYPE and REC_COUNT predicates. Here's this version -
PROCEDURE get_id_subset(
iIds IN ID_ARRAY,
oMatchingIds OUT NOCOPY ID_ARRAY
)
IS
BEGIN
SELECT t.column_value
BULK COLLECT INTO oMatchingIds
FROM TABLE(CAST(iIds AS ID_ARRAY)) t, MY_TABLE m
WHERE (m.my_id = t.column_value)
AND ( ((SELECT COUNT(m.my_id) FROM m WHERE 1) >= 1000)
OR EXISTS (m.type != 'F' OR m.rec_count != 0)
);
END get_id_subset;
But this doesn't compile, I get the following error on the inner select -
PL/SQL: ORA-00936: missing expression
Is there another way of writing this? The inner select needs to work on the joined table.
And to clarify, I'm OK with the result set being different for this query. My assumption is that since there is an index on the my_id column, doing count(*) would be much cheaper than actually applying the rec_count predicate to 10000s of rows since there is no index on that column. Am I wrong?
I don't see your second query as being much if any improvement over the first. At best, the first subquery has to hit 1000 matching records in order to determine if the count is less than 1000, so I don't think it will save lots of work. Also it changes the actual result, and it's not clear from your description if you're saying that's OK as long as it's more efficient. (And if it is OK, then the business logic is very unclear -- why do the other conditions matter at all, if they don't matter when there's lots of records?)
You ask, "will the group by be applied before or after the predicate". I'm not clear what part of the query you're talking about, but logically speaking the order is always
Where predicates
Group By
Having predicates
The optimizer can change the order in which things are actually evaluated, but the result must always be logically equivalent to the above order of evaluation (barring optimizer bugs).
1000s of records is really not that much. Have you actually encountered a case where performance of the first query is unacceptable?
For either query, it may be better to rewrite the correlated EXISTS subquery as a non-correlated IN subquery. You need to test this.
You need to show actual execution plans to get more useful feedback.
Edit
For the kind of short-circuiting you're talking about, I think you need to rewrite your subquery (from the initial version of the query) like this (sorry, my first attempt at this wouldn't work because I tried to access a column from the top-level table in a sub-sub-query):
WHERE EXISTS (
SELECT /*+ NL_SJ */ 1
FROM MY_TABLE m
WHERE (m.my_id = t.column_value)
AND rownum <= 1000
HAVING MAX( CASE WHEN m.type != 'A' OR m.rec_count != 0 THEN 1 ELSE NULL END ) I S NOT NULL
OR MAX(rownum) >= 1000
)
That should force it to hit no more than 1,000 records per id, then return a row if either at least one row matches the conditions on type and rec_count, or the 1,000-record limit was reached. If you view the execution plan, you should expect to see a COUNT STOPKEY operation, which shows that Oracle is going to stop running a query block after a certain number of rows are returned.
I'm just getting into optimizing queries by logging slow queries and EXPLAINing them. I guess the thing is... I'm not sure exactly what kind of things I should be looking for.... I have the query
SELECT DISTINCT
screenshot.id,
screenshot.view_count
FROM screenshot_udb_affect_assoc
INNER JOIN screenshot ON id = screenshot_id
WHERE unit_id = 56
ORDER BY RAND()
LIMIT 0, 6;
Looking at these two elements.... where should I focus on optimization?
id select_type table type possible_keys key key_len ref rows Extra
1 SIMPLE screenshot ALL PRIMARY NULL NULL NULL 504 Using temporary; Using filesort
1 SIMPLE screenshot_udb_affect_assoc ref screenshot_id screenshot_id 8 source_core.screenshot.id,const 3 Using index; Distinct
To begin with please refrain using ORDER BY RAND(). This in particular degrades performance when the table size is large.
For example, even with limit 1 , it generates number of random numbers equal to the row count, and would pick the smallest one. This might be inefficient if table size is large or bound to grow. Detailed discussion on this can be found at: http://www.titov.net/2005/09/21/do-not-use-order-by-rand-or-how-to-get-random-rows-from-table/
Lastly, also ensure that your join columns are indexed.
Try:
SELECT s.id,
s.view_count
FROM SCREENSHOT s
WHERE EXISTS(SELECT NULL
FROM SCREENSHOT_UDB_AFFECT_ASSOC x
WHERE x.screenshot_id = s.id)
ORDER BY RAND()
LIMIT 6
Under 100K records, it's fine to use ORDER BY RAND() -- over that, and you want to start looking at alternatives that scale better. For more info, see this article.
I agree with kuriouscoder, refrain from using ORDER BY RAND(), and make sure each of the following fields are indexed in a single index:
screenshot_udb_affect_assoc.id
screenshot.id
screenshot.unit_id
do this using code like:
create index Index1 on screenshot(id):