Insert data

You can insert data into a distributed hypertable with an INSERT statement. The syntax looks the same as for a regular hypertable or PostgreSQL table. For example:

INSERT INTO conditions(time, location, temperature, humidity)
  VALUES (NOW(), 'office', 70.0, 50.0);

Optimize data insertion

Distributed hypertables have higher network load than regular hypertables, because they must push inserts from the access node to the data nodes. You can optimize your insertion patterns to reduce load.

Insert data in batches

Reduce load by batching your INSERT statements over many rows of data, instead of performing each insertion as a separate transaction.

The access node first splits the batched data into smaller batches by determining which data node each row should belong to. It then writes each batch to the correct data node.

Optimize insert batch size

When inserting to a distributed hypertable, the access node tries to convert INSERT statements into more efficient COPY operations between the access and data nodes. But this doesn't work if:

  • The INSERT statement has a RETURNING clause and
  • The hypertable has triggers that could alter the returned data

In this case, the planner uses a multi-row prepared statement to insert into each data node. It splits the original insert statement across these sub-statements. You can view the plan by running an EXPLAIN on your INSERT statement.

In the prepared statement, the access node can buffer a number of rows before flushing them to the data node. By default, the number is 1000. You can optimize this by changing the timescaledb.max_insert_batch_size setting, for example to reduce the number of separate batches that must be sent.

The maximum batch size has a ceiling. This is equal to the maximum number of parameters allowed in a prepared statement (currently 32,767) divided by the number of columns in each row. For example, if you have a distributed hypertable with 10 columns, the highest you can set the batch size is 3276.

For more information on changing timescaledb.max_insert_batch_size, see the section on TimescaleDB configuration.

Use a copy statement instead

COPY can perform better than INSERT on a distributed hypertable. But it doesn't support some features, such as conflict handling using the ON CONFLICT clause.

To copy from a file to your hypertable, run:

COPY <HYPERTABLE> FROM '<FILE_PATH>';

When doing a COPY, the access node switches each data node to copy mode. It then streams each row to the correct data node.

Found an issue on this page?

Report an issue!

Keywords

Related Content