Quantile trackers should not be used in production due to…

Error message

Quantile trackers should not be used in production due to the unbounded memory consumption.

What it means

Constructing a QuantileTracker emits a warning that it stores all observed values (landmark window mode), causing unbounded memory growth, and must not be used in production. It is intended for experimentation on bounded or short-lived streams.

Solutions

Switch to a bounded-memory tracker variant for production.
Restrict QuantileTracker to offline/bounded datasets or short-lived experiments.
Escalate the warning to an error outside experiments: `warnings.filterwarnings('error', message='Quantile trackers should not be used in production')`.

Example fix

// before
tracker = QuantileTracker(q=0.99)
// after
tracker = OrderStatisticsTracker(q=0.99)  # bounded-memory alternative

Defensive patterns

Strategy: validation

Validate before calling

import warnings
with warnings.catch_warnings(record=True) as w:
    warnings.simplefilter('always')
    tracker = QuantileTracker(q=0.5)
    if any('unbounded memory' in str(x.message) for x in w) and is_production:
        raise ValueError('QuantileTracker is unbounded; not allowed in production')

Prevention

Use bounded-memory tracker variants for long-running streams.
Restrict QuantileTracker to tests and offline analysis.
Filter that escalates this warning to an error outside experiments.

When it happens

Trigger: `QuantileTracker(q)` (or a specifiable-annotated instantiation in an anomaly detection config) with landmark window mode.

Common situations: Prototyping anomaly detection on unbounded production streams where memory grows indefinitely until OOM.

Related errors

AI-assisted analysis of apache/beam@12126d8942 (2026-09-13). Data as JSON: /api/errors/0f6cb5888695a213. Report an issue: GitHub.

Appendix: source

Thrown at sdks/python/apache_beam/ml/anomaly/univariate/quantile.py:192

        A list of calculated quantiles.
    """
    return self._master._get_helper(self._master._sorted_items, self._q)


@specifiable
class BufferedLandmarkQuantileTracker(BufferedQuantileTracker):
  """Landmark quantile tracker using a sorted list for quantile calculation.

  Warning:
    Landmark quantile trackers have unbounded memory consumption as they store
    all pushed values in a sorted list. Avoid using in production for
    long-running streams.

  Args:
    q: The quantile to calculate, a float between 0 and 1 (inclusive).
  """
  def __init__(self, q):
    warnings.warn(
        "Quantile trackers should not be used in production due to "
        "the unbounded memory consumption.")
    super().__init__(window_mode=WindowMode.LANDMARK, q=q)


@specifiable
class BufferedSlidingQuantileTracker(BufferedQuantileTracker):
  """Sliding window quantile tracker using a sorted list for quantile
  calculation.

  Warning:
    Maintains a sorted list of values within the sliding window to calculate
    the specified quantile. Memory consumption is bounded by the window size
    but can still be significant for large windows.

  Args:
    window_size: The size of the sliding window.
    q: The quantile to calculate, a float between 0 and 1 (inclusive).

View on GitHub (pinned to 12126d8942)