RedisTimeSeries 1.8 release notes

Added a time-weighted average aggregator, gap filling, ability to control bucket timestamps, ability to control alignment for compaction rules, new reducer types, and ability to include the latest (possibly partial) raw bucket samples when retrieving compactions

Requirements

RedisTimeSeries v1.8.13 requires:

  • Minimum Redis compatibility version (database): 6.0.16
  • Minimum Redis Enterprise Software version (cluster): 6.2.8

v1.8.13 (March 2024)

This is a maintenance release for RedisTimeSeries 1.8.

Update urgency: LOW: No need to upgrade unless there are new features you want to use.

Details

  • Bug fixes:

    • LibMR#51 Crash on SSL initialization failure (MOD-5647)
  • Improvements:

    • #1593 More detailed LibMR error messages

v1.8.12 (December 2023)

This is a maintenance release for RedisTimeSeries 1.8.

Update urgency: SECURITY: There are security fixes in the release.

Details:

  • Security and privacy:

    • #1506 Don’t expose internal commands (MOD-5643)
  • Bug fixes:

    • #1494 Potential crash when using an invalid argument value
  • Improvements:

    • #1516 Added support for CBL-Mariner 2

v1.8.11 (July 2023)

This is a maintenance release for RedisTimeSeries 1.8.

Update urgency: MODERATE: Program an upgrade of the server, but it's not urgent.

Details:

  • Bug fixes:

    • #1486 When using LATEST, results may contain samples earlier than fromTimestamp (TS.RANGE, TS.REVRANGE, TS.MRANGE, and TS.MREVRANGE)
    • #1471 Potential crash on TS.MRANGE when aggregating millions of time series
    • #1469 Potential memory leak in TS.MRANGE after eviction
  • Performance enhancements:

    • #1476 Significant performance improvement when using multiple label filters (TS.MGET, TS.MRANGE, TS.MREVRANGE, and TS.QUERYINDEX) (MOD-5338)

v1.8.10 (April 2023)

This is a maintenance release for RedisTimeSeries 1.8.

Update urgency: MODERATE: Program an upgrade of the server, but it's not urgent.

Details:

  • Bug fixes:

    • #1455 TS.ADD - optional arguments are not replicated (MOD-5110)

v1.8.9 (March 2023)

This is a maintenance release for RedisTimeSeries 1.8.

Update urgency: MODERATE: Program an upgrade of the server, but it's not urgent.

Details:

  • Bug fixes:

    • #1421 Potential crash after deleting from a time series with an AVG compaction (MOD-4972)
    • #1422 Incorrectly return an error when deleting from a time series with a compaction and with no expiry

v1.8.8 (March 2023)

This is a maintenance release for RedisTimeSeries 1.8

Update urgency: MODERATE: Program an upgrade of the server, but it's not urgent.

Details:

  • Bug fixes:

    • #1290 Potential crash when using FILTER_BY_TS
    • #1397 Memory leak when trying to create an already existing key (MOD-4724, RED-93418)

v1.8.5 (January 2023)

This is a maintenance release for RedisTimeSeries 1.8.

Update urgency: HIGH: There is a critical bug that may affect a subset of users. Upgrade!

Details:

  • Bug fixes:

    • #1388 Potential crash when upgrading from v1.6 to 1.8 if there are compactions with min or max aggregation (MOD-4695)

v1.8.4 (December 2022)

This is a maintenance release for RedisTimeSeries 1.8.

Update urgency: HIGH: There is a critical bug that may affect a subset of users. Upgrade!

Details:

  • Bug fixes:

    • #1360 Potential crash when upgrading from v1.6 to 1.8 if there are compactions with min or max aggregation (MOD-4559)
    • #1370 Potential crash when using TS.REVRANGE or TS.MREVRANGE with aggregation
    • #1347 When adding samples with TS.ADD or TS.MADD using * as timestamp, the timestamp could differ between master and replica shards
  • Improvements:

v1.8 GA (v1.8.3) (November 2022)

This is the General Availability release of RedisTimeSeries 1.8.

Highlights

RedisTimeSeries 1.8 introduces seven highly requested features, performance improvements, and bug fixes.

What's new in 1.8

  • Optionally retrieving aggregation results for the latest (still open) bucket for compactions

    Till version 1.8, when a time series is a compaction, TS.GET, TS.MGET, TS.RANGE, TS.REVRANGE, TS.MRANGE, TS.MREVRANGE did not report the compacted value of the latest bucket. The reason is that the data in the latest bucket of a compaction is still partial. A bucket is ‘closed’ and compacted only upon the arrival of data that ‘opens’ a ‘new latest’ bucket.

    There are use cases, however, where the compaction of the latest bucket should be retrieved as well. For example, a user may want to receive the count of events since the start of the decade, and the retention period for raw data is only one month. Till version 1.8, the user would have to run two queries - one on a compaction and one on the latest raw data, and then sum the results. Since version 1.8, by specifying LATEST, it is possible to retrieve the latest (possibly partial) bucket as well.

    To report aggregations for the latest bucket, use the new optional LATEST flag to TS.GET, TS.MGET, TS.RANGE, TS.REVRANGE, TS.MRANGE, and TS.REVRANGE.

  • Optionally retrieving aggregation results for empty buckets

    The commands TS.RANGE, TS.REVRANGE, TS.MRANGE, and TS.MREVRANGE have an optional [AGGREGATION aggregator bucketDuration] parameter. When this parameter is specified, raw reports are aggregated per bucket.

    Till version 1.8, results were not reported for empty buckets. With EMPTY, it is now possible to report aggregations for empty buckets as well.

    The two primary reasons for wanting to retrieve values for empty buckets:

    • It is easier to align results from similar queries over multiple time series

    • It is easier to use the retrieved results with some external tools (e.g., charting tools)

    For the sum and count aggregators, the value 0 is reported for empty buckets.

    For the min, max, range, avg, first, std.p, and std.s aggregators, the value NaN (not a number) is reported.

    For the last aggregator and the new twa aggregator, the EMPTY flag is used for gap filling (see below).

    To report aggregations for empty buckets, use the new optional EMPTY flag in TS.RANGE, TS.REVRANGE, TS.MRANGE, and TS.MREVRANGE.

    Regardless of the values of fromTimestamp and toTimestamp, no data is reported for empty buckets that end before the earliest sample or begin after the latest sample in the time series.

  • A new aggregator: time-weighted average

    When a time series holds discrete samples of a continuous signal (e.g., temperature), using avg to estimate the average value over a given timeframe would produce a good estimate only when the signal is sampled at constant intervals. If, however, samples are available at non-constant intervals (e.g., when some samples are missing), the twa aggregator produces a more accurate estimate by averaging samples over time instead of simply averaging the samples.

    A graph showing the difference between average and time-weighted average.

    This is an extreme example: the signal in the diagram above has 4 samples in its ‘high’ value and 13 samples in its ‘low’ value. However, the period in each of those states is about the same. It is easy to see that the simple average (avg) of all the 17 samples does not represent the signal’s average over time.

    Time-weighted average (twa) adds weight to each sample. The weight is proportional to the time interval that the sample represents. In the diagram, the time-weighted average over the whole period assigns appropriate weight to each sample, so the result represents the signal’s average value over the whole period much more accurately.

  • Gap filling: optionally interpolate or repeat last value for empty buckets

    Gap filling is performed when using EMPTY together with either last or twa aggregator.

    Using EMPTY with the twa aggregator allows us to estimate the average of a continuous signal even for buckets where no samples were collected (gap-filling).

    A graph that illustrates gap-filling.

    Consider we want to use TS.RANGE to calculate the average value of each bucket (p1, p2, p3 in the diagram above). Using avg, the value reported for bucket p2 would be NaN, as this bucket contains no samples. If we use EMPTY with twa, on the other hand, the average value for bucket p2 would be calculated based on the linear interpolation of the value left of p2 and the value right of p2.

    When sampling a continuous signal, we can use this ‘gap-filling’ capability to calculate the average value of the signal over equal-width buckets without concern about bucket alignment or missing samples.

    Using EMPTY with the last aggregator allows filling empty buckets by repeating the value of the previous sample. This is useful, for example, when values in the time series represent stock prices and the price has not been changed during a bucket’s timeframe.

  • Ability to control how bucket timestamps are reported

    Till version 1.8, TS.RANGE, TS.REVRANGE, TS.MRANGE, and TS.MREVRANGE returned the start time of each reported bucket as its timestamp.

    Changing this behavior and reporting each bucket’s start time, end time, or mid-time is now possible. This is required in many use cases. For example, when drawing bars in trading applications, annotating each bar with the end timestamp of the bucket it represents is very common.

    The way bucket timestamps are reported can be specified with the new optional BUCKETTIMESTAMP parameter of TS.RANGE, TS.REVRANGE, TS.MRANGE, and TS.MREVRANGE.

    - or low: the bucket's start time (default)

    + or high: the bucket's end time

    ~ or mid: the bucket's mid-time (rounded down if not an integer)

  • Ability to control alignment for compaction rules

    Till version 1.8, compaction rules could not be aligned. One could specify a compaction rule with 24-hour buckets, and as a result, each bucket would aggregate events from midnight till the next midnight. The first bucket always started at the epoch and all other buckets were aligned accordingly.

    But what if we want to aggregate daily events from 06:00 to 06:00 the next day? We can now specify alignment for compaction rules.

    Alignment can be specified with the new optional alignTimestamp parameter of TS.CREATERULE and the COMPACTION_POLICY configuration parameter. Specifying alignTimestamp ensures that there is a bucket that starts exactly at alignTimestamp and all other buckets are aligned accordingly. alignTimestamp is expressed in milliseconds. The default value is 0 (aligned with the epoch).

  • New reducers

    Till version 1.8, only the sum, min, and max could be used as reducer types.

    It is now possible, for example, to calculate the maximal temperature per timeframe for each sensor and then report the average temperature (avg reducer) over groups of sensors (grouped by a given label's value).

    This can be specified with the new reducer types (TS.MRANGE and TS.MREVRANGE: avg, range, count, std.p, std.s, var.p, and var.s.

Note:
New RDB version (v7). RDB files created with v1.8.3 are not backward compatible.
RATE THIS PAGE
Back to top ↑