Space partitions: Using space partitions is highly recommended for to achieve efficient scale-out performance. For regular hypertables that exist only on a single node, additional partitioning can be used for specialized use cases and not recommended for most users.

Space partitions use hashing: Every distinct item is hashed to one of N buckets. Remember that we are already using (flexible) time intervals to manage chunk sizes; the main purpose of space partitioning is to enable parallelization across multiple data nodes (in the case of distributed hypertables) or across multiple disks within the same time interval (in the case of single-node deployments).

In a distributed hypertable, space partitioning enables inserts to be parallelized across data nodes, even while the inserted rows share timestamps from the same time interval, and thus increases the ingest rate. Query performance also benefits by being able to parallelize queries across nodes, particularly when full or partial aggregations can be “pushed down” to data nodes (e.g., as in the query avg(temperature) FROM conditions GROUP BY hour, location when using location as a space partition). Please see our for more information.

Parallelizing disk I/O on a single node

Parallel I/O can benefit in two scenarios: (a) two or more concurrent queries should be able to read from different disks in parallel, or (b) a single query should be able to use query parallelization to read from multiple disks in parallel.

1. Use a RAID setup across multiple physical disks, and expose a single logical disk to the hypertable (i.e., via a single tablespace).

2. For each physical disk, add a separate tablespace to the database. TimescaleDB allows you to actually add multiple tablespaces to a single hypertable (although under the covers, a hypertable’s chunks are spread across the tablespaces associated with that hypertable).

We recommend a RAID setup when possible, as it supports both forms of parallelization described above (i.e., separate queries to separate disks, single query to multiple disks in parallel). The multiple tablespace approach only supports the former. With a RAID setup, no spatial partitioning is required.

That said, when using space partitions, we recommend using 1 space partition per disk.

TimescaleDB does not benefit from a very large number of space partitions (such as the number of unique items you expect in partition field). A very large number of such 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.

Required Arguments

Optional Arguments

Returns

The should be specified as follows:

• If the column is some other integer type, this length should be an integer that reflects the column’s underlying semantics (e.g., the chunk_time_interval should be given in milliseconds if this column is the number of milliseconds since the UNIX epoch).

WARNING:Supporting more than one additional dimension is currently experimental. For any production environments, users are recommended to use at most one “space” dimension.

Sample Usage

First convert table conditions to hypertable with just time partitioning on column time, then add an additional partition key on location with four partitions:

Now in a multi-node example for distributed hypertables with a cluster of one access node and two data nodes, configure the access node for access to the two data nodes. Then, convert table conditions to a distributed hypertable with just time partitioning on column time, and finally add a space partitioning dimension on with two partitions (as the number of the attached data nodes).