All committers have signed the CLA.

Hello @vesalaakso-oura , thank you for this PR. We were wondering if you were aware of the "resample" expression available already in Grafana. It works a bit differently than this PR, but it may solve your needs. Take a look at a demo here and let me know what you think! https://play.grafana.org/d/krmmxss/resampling-demo?folderUid=ae9kn9tsqxe68c&orgId=1&from=now-6h&to=now&timezone=utc

Hello @vesalaakso-oura , thank you for this PR. We were wondering if you were aware of the "resample" expression available already in Grafana. It works a bit differently than this PR, but it may solve your needs. Take a look at a demo here and let me know what you think! https://play.grafana.org/d/krmmxss/resampling-demo?folderUid=ae9kn9tsqxe68c&orgId=1&from=now-6h&to=now&timezone=utc

Hello @gelicia, We believe resample and ASAP solve different problems. Resample changes timestamps to a consistent interval so series can line up for math but that can blur spikes or hide outliers. ASAP smooths noise while keeping trends and outliers visible (like Datadog's autosmoother). We believe ASAP and resample just complement each other.

Another possible option using SQL Expressions, keeps spike while smoothing. Might not be as robust as the autosmoother example, and takes a couple more params:

(Note, when there are spikes, both series have them, one just visually covers the other).

WITH w AS (
  SELECT
    time,
    a,
    __value__  AS val,

    /* rolling center & spread over the last 20 rows */
    AVG(__value__) OVER (
      PARTITION BY a
      ORDER BY time
      ROWS BETWEEN 20 PRECEDING AND CURRENT ROW
    ) AS mu,

    STDDEV_POP(__value__) OVER (
      PARTITION BY a
      ORDER BY time
      ROWS BETWEEN 20 PRECEDING AND CURRENT ROW
    ) AS sigma,

    /* neighbors for "extra-picky" local-extremum check */
    LAG(__value__, 1)  OVER (PARTITION BY a ORDER BY time) AS prev_val,
    LEAD(__value__, 1) OVER (PARTITION BY a ORDER BY time) AS next_val
  FROM A
)
SELECT
  time,
  a,
  val AS Value,
  CASE
    WHEN sigma IS NULL OR sigma = 0 THEN mu
    /* keep spikes/dips when z >= 2.5 and it's a local extremum (or at edges) */
    WHEN ABS(val - mu) >= 2 * sigma
         AND (
              prev_val IS NULL OR next_val IS NULL
              OR (val > prev_val AND val > next_val)
              OR (val < prev_val AND val < next_val)
         )
      THEN val
    ELSE mu
  END AS smoothed
FROM w
ORDER BY a, time;

Oh, and needs transform for viz panel:

Hi @kylebrandt , thanks a lot for taking the time to write up the SQL approach and sharing the example. I really appreciate the effort you put into it. 🙇

My main concern is scalability in larger orgs. If every team needs to ship or maintain their own SQL query, it quickly becomes a burden. It also makes it harder to convince teams migrating from datadog, where autosmoothing is just available out of the box. Having a built-in ASAP transformation would not only help internally but could also tip other organisations out there to consider moving away from datadog to the grafana stack.

ASAP itself is well researched. The original paper shows that ASAP can spot meaningful trends and outliers while suppressing noise, so having it as a built-in transformation would be an easier sell.

Hello @gelicia, is this a feature we can proceed with, or do you have any concerns or alternative ways to implement it?

@vesalaakso-oura thank you for your patience, adding a new transformation to the catalog is something we have to discuss internally so we are checking in with folks! I'll let you know what we decide.

This pull request has been automatically marked as stale because it has not had activity in the last 30 days. It will be closed in 2 weeks if no further activity occurs. Please feel free to give a status update or ping for review. Thank you for your contributions!