1.3.1-cloud

Version 1.3.1 with more permissive permissions to work correctly when provided by a managed environment. Use the regular version of the release, unless you have specific reason to use this one.

1.3.1

Internal memory usage improvements.

1.3.0-cloud

Version 1.3.0 with more permissive permissions to work correctly when provided by a managed environment. Use the regular version of the release, unless you have specific reason to use this one.

1.3.0

This release stabilizes many in-progress features, improves stats_agg, and introduces a new experimental feature, timevector function pipelines, to ease working with time-series data.

Stabilizations

• counter_agg helper functions for Prometheus-style resetting monotonic counters.
• hyperloglog efficient approximate COUNT DISTINCT.
• stats_agg two-step aggregate for common statistics.

Experimental Features

• timevector function pipelines. A compact and readable way to perform a sequence of analytic operations such as the following one for volatility

  timevector(ts, val) -> sort() -> delta() -> abs() -> sum()

• -> accessor for Toolkit types enables syntax like stats_agg(data) -> average()
• to_epoch() wrapper for extract ('EPOCH' FROM timestamp) that makes it work more like an inverse of to_timestamp(DOUBLE PRECISION) #251

Migration Warning

• This release changes the textual I/O format for Toolkit types. We are uncertain if we will need to do so again in the future. Due to this we currently only support dump/restore within a single version of the extension.

New Contributors

• @jonatas made their first contribution in #237
• @burmecia made their first contribution in #251

Full Changelog: 1.2.0...1.3.0

1.2.0-cloud

Version 1.2.0 with more permissive permissions to work correctly when provided by a managed environment. Use the regular version of the release, unless you have specific reason to use this one.

1.2.0

This release adds some bugfixes and optimizations and includes new and refined experimental features.

Experimental Features

• Refinements to hyperloglog including a function to report relative error and fixing the functionality of rollup.
• Introduction of a TopN approximation API. Presently this will only work for integer data, but expect to see further refinements that greatly expand this behavior.
• New map_series and map_data pipeline elements for the time series API that allow uses to provide custom transforms of their time series data. Additionally introduced a |>> pipeline operator for an even more streamlined interface into the new mapping functionality.

Bugfixes/Optimizations

• Made a pass through all toolkit functions to correctly label behavior as immutable and parallel safe. This should improve the optimizations Postgres can apply to toolkit plans, particularly when run in a Timescale multinode cluster.
• Improved handling of internal data structures to reduce extraneous copies of data.

1.1.0 Cloud Edition

Version 1.1.0 with more permissive permissions to work correctly when provided by a managed environment. Use the regular version of the release, unless you have specific reason to use this one.

1.1.0

This release contains new experimental features and bugfixes.

Experimental Features

• hyperloglog has been updated to use Hyperloglog++ under the hood. This does not change the user-facing API but should improve the accuracy of hyperloglog() estimates. This is the last major change expected for hyperloglog() and is now a candidate for stabilization pending user feedback.
• We've started experimenting with the pipeline API. While it's still very much a work in progress, it's at a point where the high-level concepts should be understandable. For example, a pipeline that outputs the daily change of a set of data, interpolating away any gaps in daily data, could look like

  SELECT timeseries(time, val)
      |> sort()
      |> resample_to_rate('trailing_average', '24 hours', true)
      |> fill_holes('interpolate')
      |> delta()
  FROM ...

  It's still early days for this API and it is not yet polished, but we would love feedback about its direction.

Bugfixes

• Fixed a small memory leak in aggregation functions. This could have leaked ≈8 bytes per aggregate call.

---

As always we love feedback! All in-progress features are tracked in Feature Progress, and please stop by our Discussions or issue tracker if you have any questions, comments, or feedback!

1.0.0 Cloud Edition

Version 1.0.0 with more permissive permissions to work correctly when provided by a managed environment. Use the regular version of the release, unless you have specific reason to use this one.

1.0.0

This release renames the extension to TimescaleDB Toolkit, starts stabilizing functionality, and includes internal improvements, and new experimental features.

Stabilizes:

• Percentile-approximation algorithms including percentile_agg(), uddsketch() and tdigest() along with their associated functions. These are especially useful for computing percentiles in continuous aggregates.
• Time-weighted average along with it's associated functions. This eases taking the average over an irregularly spaced dataset that only includes changepoints

New Experimental Features:

• stats_agg() eases the analysis of more sophisticated bucketed statistics, such as rolling averages. (Docs are forthcoming, until then fell free to peruse the design discussion doc.
• timeseries which will serve as a building block for many pipelines, and unifies the output of lttb and ASAP.

Internal Improvements

• uddsketch's on-disk layout has be reworked to store buckets compressed. This can result in an orders-of-magnitude reduction in it's storage requirements.
• uddsketch's textual format has been reworked to be more readable.
• functions that take in a uddsketch or t-digest have been reworked to be 0-copy when applicable, improving the performance of such functions by 10-100x.