Ruby
Postgres partitioning as easy as pie

pgslice

Postgres partitioning as easy as pie. Works great for both new and existing tables, with zero downtime and minimal app changes.

🍊 Battle-tested at Instacart

Install

pgslice is a command line tool. To install, run:

gem install pgslice

This will give you the pgslice command.

Steps

1. Ensure the table you want to partition has been created. We’ll refer to this as <table>.

2. Specify your database credentials

   export PGSLICE_URL=postgres://localhost/myapp_development

3. Create an intermediate table

   pgslice prep <table> <column> <period>

   Period can be day or month.

   This creates a table named <table>_intermediate with the appropriate trigger for partitioning.

4. Add partitions

   pgslice add_partitions <table> --intermediate --past 3 --future 3

   This creates child tables that inherit from the intermediate table.

   Use the --past and --future options to control the number of partitions.

5. Optional, for tables with data - Fill the partitions in batches with data from the original table

   pgslice fill <table>

   Use the --batch-size and --sleep options to control the speed.

   To sync data across different databases, check out pgsync.

6. Swap the intermediate table with the original table

   pgslice swap <table>

   The original table is renamed <table>_retired and the intermediate table is renamed <table>.

7. Fill the rest (rows inserted between the first fill and the swap)

   pgslice fill <table> --swapped

8. Archive and drop the original table

Adding Partitions

To add partitions, use:

pgslice add_partitions <table> --future 3

Add this as a cron job to create a new partition each day or month.

# day
0 0 * * * pgslice add_partitions <table> --future 3 --url ...

# month
0 0 1 * * pgslice add_partitions <table> --future 3 --url ...

Add a monitor to ensure partitions are being created.

SELECT 1 FROM
    pg_catalog.pg_class c
INNER JOIN
    pg_catalog.pg_namespace n ON n.oid = c.relnamespace
WHERE
    c.relkind = 'r' AND
    n.nspname = 'public' AND
    c.relname = '<table>_' || to_char(NOW() + INTERVAL '3 days', 'YYYYMMDD')
    -- for months, use to_char(NOW() + INTERVAL '3 months', 'YYYYMM')

Additional Commands

To undo prep (which will delete partitions), use:

pgslice unprep <table>

To undo swap, use:

pgslice unswap <table>

Sample Output

pgslice prints the SQL commands that were executed on the server. To print without executing, use the --dry-run option.

$ pgslice prep visits created_at month
BEGIN;

CREATE TABLE visits_intermediate (LIKE visits INCLUDING ALL);

CREATE FUNCTION visits_insert_trigger()
    RETURNS trigger AS $$
    BEGIN
        EXECUTE 'INSERT INTO visits_' || to_char(NEW.created_at, 'YYYYMM') || ' VALUES ($1.*)' USING NEW;
        RETURN NULL;
    END;
    $$ LANGUAGE plpgsql;

CREATE TRIGGER visits_insert_trigger
    BEFORE INSERT ON visits_intermediate
    FOR EACH ROW EXECUTE PROCEDURE visits_insert_trigger();

COMMIT;

$ pgslice add_partitions visits --intermediate --past 1 --future 1
BEGIN;

CREATE TABLE visits_201608
    (CHECK (created_at >= '2016-08-01'::date AND created_at < '2016-09-01'::date))
    INHERITS (visits_intermediate);

ALTER TABLE visits_201608 ADD PRIMARY KEY (id);

CREATE INDEX ON visits_201608 USING btree (user_id);

CREATE TABLE visits_201609
    (CHECK (created_at >= '2016-09-01'::date AND created_at < '2016-10-01'::date))
    INHERITS (visits_intermediate);

ALTER TABLE visits_201609 ADD PRIMARY KEY (id);

CREATE INDEX ON visits_201609 USING btree (user_id);

CREATE TABLE visits_201610
    (CHECK (created_at >= '2016-10-01'::date AND created_at < '2016-11-01'::date))
    INHERITS (visits_intermediate);

ALTER TABLE visits_201610 ADD PRIMARY KEY (id);

CREATE INDEX ON visits_201610 USING btree (user_id);

COMMIT;

$ pgslice fill visits
/* 1 of 3 */
INSERT INTO visits_intermediate (id, user_id, ip, created_at)
    SELECT id, user_id, ip, created_at FROM visits
    WHERE id > 0 AND id <= 10000 AND created_at >= '2016-08-01'::date AND created_at < '2016-11-01'::date

/* 2 of 3 */
INSERT INTO visits_intermediate (id, user_id, ip, created_at)
    SELECT id, user_id, ip, created_at FROM visits
    WHERE id > 10000 AND id <= 20000 AND created_at >= '2016-08-01'::date AND created_at < '2016-11-01'::date

/* 3 of 3 */
INSERT INTO visits_intermediate (id, user_id, ip, created_at)
    SELECT id, user_id, ip, created_at FROM visits
    WHERE id > 20000 AND id <= 30000 AND created_at >= '2016-08-01'::date AND created_at < '2016-11-01'::date

$ pgslice swap visits
BEGIN;

ALTER TABLE visits RENAME TO visits_retired;

ALTER TABLE visits_intermediate RENAME TO visits;

ALTER SEQUENCE visits_id_seq OWNED BY visits.id;

COMMIT;

$ pgslice add_partitions visits --future 2
BEGIN;

CREATE TABLE visits_201611
    (CHECK (created_at >= '2016-11-01'::date AND created_at < '2016-12-01'::date))
    INHERITS (visits);

ALTER TABLE visits_201611 ADD PRIMARY KEY (id);

CREATE INDEX ON visits_201611 USING btree (user_id);

COMMIT;

App Changes

This set up allows you to read and write with the original table name with no knowledge it’s partitioned. However, there are a few things to be aware of.

Reads

When possible, queries should include the column you partition on to limit the number of partitions the database needs to check. For instance, if you partition on created_at, try to include it in queries:

SELECT * FROM
    visits
WHERE
    user_id = 123 AND
    -- for performance
    created_at >= '2016-09-01' AND created_at < '2016-09-02'

For this to be effective, ensure constraint_exclusion is set to partition (default value) or on.

SHOW constraint_exclusion;

Writes

If you use INSERT statements with a RETURNING clause (as frameworks like Rails do), you’ll no longer receive the id of the newly inserted record back. If you need this, you can either:

1. Insert directly into the partition
2. Get the value after the insert with SELECT CURRVAL('sequence_name')

One Off Tasks

You can also use pgslice to reduce the size of a table without partitioning by creating a new table, filling it with a subset of records, and swapping it in.

pgslice prep <table> --no-partition
pgslice fill <table> --start 1000 # starting primary key
pgslice swap <table>

Upgrading

Run:

gem install pgslice

To use master, run:

gem install specific_install
gem specific_install ankane/pgslice

Reference

• PostgreSQL Manual
• PostgreSQL Wiki

TODO

• Command to sync index changes with partitions
• Disable indexing for faster fill
• ETA for fill

Related Projects

Also check out:

• PgHero - A performance dashboard for Postgres
• pgsync - Sync Postgres data to your local machine

Contributing

Everyone is encouraged to help improve this project. Here are a few ways you can help:

• Report bugs
• Fix bugs and submit pull requests
• Write, clarify, or fix documentation
• Suggest or add new features