Skip to content

Latest commit

 

History

History
181 lines (128 loc) · 7.57 KB

_dimensions_info.md

File metadata and controls

181 lines (128 loc) · 7.57 KB

import CreateHypertablePolicyNote from "versionContent/_partials/_create-hypertable-columnstore-policy-note.mdx";

Dimension info

To create a _timescaledb_internal.dimension_info instance, you call add_dimension to an existing hypertable.

Samples

Hypertables must always have a primary range dimension, followed by an arbitrary number of additional dimensions that can be either range or hash, Typically this is just one hash. For example:

SELECT add_dimension('conditions', by_range('time'));
SELECT add_dimension('conditions', by_hash('location', 2));

For incompatible data types such as jsonb, you can specify a function to the partition_func argument of the dimension build to extract a compatible data type. Look in the example section below.

Custom partitioning

By default, $TIMESCALE_DB calls $PG's internal hash function for the given type. You use a custom partitioning function for value types that do not have a native $PG hash function.

You can specify a custom partitioning function for both range and hash partitioning. A partitioning function should take a anyelement argument as the only parameter and return a positive integer hash value. This hash value is not a partition identifier, but rather the inserted value's position in the dimension's key space, which is then divided across the partitions.

by_range()

Create a by-range dimension builder. You can partition by_range on it's own.

Samples

  • Partition on time using CREATE TABLE

    The simplest usage is to partition on a time column:

    CREATE TABLE conditions (
   time        TIMESTAMPTZ       NOT NULL,
   location    TEXT              NOT NULL,
   device      TEXT              NOT NULL,
   temperature DOUBLE PRECISION  NULL,
   humidity    DOUBLE PRECISION  NULL
) WITH (
   tsdb.hypertable
);

    This is the default partition, you do not need to add it explicitly.

  • Extract time from a non-time column using create_hypertable

    If you have a table with a non-time column containing the time, such as a JSON column, add a partition function to extract the time:

    CREATE TABLE my_table (
   metric_id serial not null,
   data jsonb,
);

CREATE FUNCTION get_time(jsonb) RETURNS timestamptz AS $$
  SELECT ($1->>'time')::timestamptz
$$ LANGUAGE sql IMMUTABLE;

SELECT create_hypertable('my_table', by_range('data', '1 day', 'get_time'));
Arguments
Name Type Default Required Description
column_name NAME - ✔ Name of column to partition on.
partition_func REGPROC - ✖ The function to use for calculating the partition of a value.
partition_interval ANYELEMENT - ✖ Interval to partition column on.

If the column to be partitioned is a:

  • TIMESTAMP, TIMESTAMPTZ, or DATE: specify partition_interval either as an INTERVAL type or an integer value in microseconds.

  • Another integer type: specify partition_interval as an integer that reflects the column's underlying semantics. For example, if this column is in UNIX time, specify partition_interval in milliseconds.

The partition type and default value depending on column type is:

Column Type Partition Type Default value
TIMESTAMP WITHOUT TIMEZONE INTERVAL/INTEGER 1 week
TIMESTAMP WITH TIMEZONE INTERVAL/INTEGER 1 week
DATE INTERVAL/INTEGER 1 week
SMALLINT SMALLINT 10000
INT INT 100000
BIGINT BIGINT 1000000

by_hash()

The main purpose of hash partitioning is to enable parallelization across multiple disks within the same time interval. Every distinct item in hash partitioning is hashed to one of N buckets. By default, $TIMESCALE_DB uses flexible range intervals to manage chunk sizes.

Parallelizing disk I/O

You use Parallel I/O in the following scenarios:

  • Two or more concurrent queries should be able to read from different disks in parallel.
  • A single query should be able to use query parallelization to read from multiple disks in parallel.

For the following options:

  • RAID: use a RAID setup across multiple physical disks, and expose a single logical disk to the hypertable. That is, using a single tablespace.

    Best practice is to use RAID when possible, as you do not need to manually manage tablespaces in the database.

  • Multiple tablespaces: for each physical disk, add a separate tablespace to the database. $TIMESCALE_DB allows you to add multiple tablespaces to a single hypertable. However, although under the hood, a hypertable's chunks are spread across the tablespaces associated with that hypertable.

    When using multiple tablespaces, a best practice is to also add a second hash-partitioned dimension to your hypertable and to have at least one hash partition per disk. While a single time dimension would also work, it would mean that the first chunk is written to one tablespace, the second to another, and so on, and thus would parallelize only if a query's time range exceeds a single chunk.

When adding a hash partitioned dimension, set the number of partitions to a multiple of number of disks. For example, the number of partitions P=N*Pd where N is the number of disks and Pd is the number of partitions per disk. This enables you to add more disks later and move partitions to the new disk from other disks.

$TIMESCALE_DB does not benefit from a very large number of hash partitions, such as the number of unique items you expect in partition field. A very large number of hash partitions leads both to poorer per-partition load balancing (the mapping of items to partitions using hashing), as well as much increased planning latency for some types of queries.

Samples
CREATE TABLE conditions (
   "time"      TIMESTAMPTZ       NOT NULL,
   location    TEXT              NOT NULL,
   device      TEXT              NOT NULL,
   temperature DOUBLE PRECISION  NULL,
   humidity    DOUBLE PRECISION  NULL
) WITH (
   tsdb.hypertable
   tsdb.chunk_interval='1 day'
);

SELECT add_dimension('conditions', by_hash('location', 2));
Arguments
Name Type Default Required Description
column_name NAME - ✔ Name of column to partition on.
partition_func REGPROC - ✖ The function to use to calcule the partition of a value.
number_partitions ANYELEMENT - ✔ Number of hash partitions to use for partitioning_column. Must be greater than 0.

Returns

by_range and by-hash return an opaque _timescaledb_internal.dimension_info instance, holding the dimension information used by this function.