An intrinsic part of working with time-series data is that the relevance of data can diminish over time. As new data accumulates, old data becomes less valuable and is rarely, if ever, updated. So, you often want to delete old raw data to save disk space.

This is particularly helpful when combined with continuous aggregates. The raw data is downsampled into a continuous aggregate and then a data retention policy can drop older raw data that's no longer needed.

In this image, dropping data on the underlying hypertable doesn't affect the continuous aggregate. Your continuous aggregate is unaffected as long as you do not refresh the continuous aggregate for time periods where you dropped data:

Continuous aggregate with refresh and retention policies

There are two ways to drop historic data from a hypertable:

  • Automatic data retention policy
  • Manually dropping chunks

Automated data retention policies drop data according to a schedule and defined rules. These policies are "set it and forget it" in nature, meaning less hassle for maintenance and upkeep.

For example, many stock trading apps don't need raw trade data once continuous aggregates have been created for various time buckets that show the high, low, open, and close values. To save disk space consumed by raw data that is rarely queried, you may want to continually remove stock trade data in the underlying hypertable stocks_real_time after the trade timestamp is older than three weeks ago.

  1. Use the add_retention_policy() function to add an automatic retention policy to the stocks_real_time table:

    SELECT add_retention_policy('stocks_real_time', INTERVAL '3 weeks');

    When you run this command, all data older than 3 weeks is dropped from stocks_real_time, and a recurring retention policy is created. No data is dropped from your continuous aggregate, stocks_real_time_daily.

  2. To see information about your retention policies and verify job statistics, query the Timescale informational views:


    The results look like this:

    job_id|application_name |schedule_interval|max_runtime|max_retries|retry_period|proc_schema |proc_name |owner |scheduled|config |next_start |hypertable_schema |hypertable_name |
    1|Telemetry Reporter [1] | 24:00:00| 00:01:40| -1| 01:00:00|_timescaledb_internal|policy_telemetry |postgres |true | |2022-05-04 21:52:45.304 -0400| | |
    1000|Refresh Continuous Aggregate Policy [1000]| 01:00:00| 00:00:00| -1| 01:00:00|_timescaledb_internal|policy_refresh_continuous_aggregate|tsdbadmin|true |{"end_offset": "00:01:00", "start_offset": "02:00:00", "mat_hypertable_id": 3}|2022-05-04 16:21:36.704 -0400|_timescaledb_internal|_materialized_hypertable_3|
  3. Or you can try this query:

    SELECT * FROM timescaledb_information.job_stats;

    The results look like this:

    hypertable_schema |hypertable_name |job_id|last_run_started_at |last_successful_finish |last_run_status|job_status|last_run_duration|next_start |total_runs|total_successes|total_failures|
    _timescaledb_internal|_materialized_hypertable_3| 1000|2022-05-04 15:21:36.443 -0400|2022-05-04 15:21:36.704 -0400|Success |Scheduled | 00:00:00.260945|2022-05-04 16:21:36.704 -0400| 1978| 1978| 0|
    | | 1|2022-05-03 21:52:45.068 -0400|2022-05-03 21:52:45.304 -0400|Success |Scheduled | 00:00:00.235434|2022-05-04 21:52:45.304 -0400| 109| 108| 1|

To manually remove data on a once-off basis, use the Timescale function drop_chunks().

This function takes similar arguments to the data retention policy. However, in addition to letting you drop data older than a particular interval, it also lets you drop data that is newer than a particular interval. This means you can drop data from an interval that is bounded on both ends.

To drop all data older than three weeks, run:

SELECT drop_chunks('stocks_real_time', INTERVAL '3 weeks');

To drop all data older than two weeks but newer than three weeks old:

SELECT drop_chunks(
older_than => INTERVAL '2 weeks',
newer_than => INTERVAL '3 weeks'

For more details and best practices on data retention and automated data retention policies, see the Data Retention docs.


Found an issue on this page?

Report an issue!