apache/beam · error · ValueError

window fn (%s) does not have a determanistic coder (%s)

Error message

window fn (%s) does not have a determanistic coder (%s)

What it means

Raised in the Windowing constructor when the provided window function's window coder reports is_deterministic() == False. Beam serializes window objects (e.g. as group-by keys during shuffle), and a non-deterministic coder would produce inconsistent bytes for equal windows, breaking correctness. Note the message contains the long-standing typo 'determanistic'.

Source

Thrown at sdks/python/apache_beam/transforms/core.py:3894

      environment_id: Environment where the current window_fn should be
        applied in.
    """
    global AccumulationMode, DefaultTrigger  # pylint: disable=global-variable-not-assigned
    # pylint: disable=wrong-import-order, wrong-import-position
    from apache_beam.transforms.trigger import AccumulationMode
    from apache_beam.transforms.trigger import DefaultTrigger

    # pylint: enable=wrong-import-order, wrong-import-position
    if triggerfn is None:
      triggerfn = DefaultTrigger()
    if accumulation_mode is None:
      if triggerfn == DefaultTrigger():
        accumulation_mode = AccumulationMode.DISCARDING
      else:
        raise ValueError(
            'accumulation_mode must be provided for non-trivial triggers')
    if not windowfn.get_window_coder().is_deterministic():
      raise ValueError(
          'window fn (%s) does not have a determanistic coder (%s)' %
          (windowfn, windowfn.get_window_coder()))
    self.windowfn = windowfn
    self.triggerfn = triggerfn
    self.accumulation_mode = accumulation_mode
    self.allowed_lateness = Duration.of(allowed_lateness)
    self.environment_id = environment_id
    self.timestamp_combiner = (
        timestamp_combiner or TimestampCombiner.OUTPUT_AT_EOW)
    self._is_default = (
        self.windowfn == GlobalWindows() and
        self.triggerfn == DefaultTrigger() and
        self.accumulation_mode == AccumulationMode.DISCARDING and
        self.timestamp_combiner == TimestampCombiner.OUTPUT_AT_EOW and
        self.allowed_lateness == 0)

  def __repr__(self):
    return "Windowing(%s, %s, %s, %s, %s)" % (

View on GitHub (pinned to 12126d8942)

Solutions

1. Implement a deterministic custom Coder for your window type (stable byte encoding for equal windows) and return it from get_window_coder().
2. Replace float fields with int/Decimal or fixed-point representations in the window object.
3. Sort any collection fields before encoding.
4. Prefer built-in window functions (GlobalWindows, FixedWindows, SlidingWindows, Sessions) which ship deterministic coders.

Example fix

// before
class MyWindow:  # coder inferred from float fields -> non-deterministic
  ...
pc | beam.WindowInto(MyWindowFn())
// after
class MyWindowFn(UserDefinedWindowFn):
  def get_window_coder(self):
    return DeterministicMyWindowCoder()  # stable encoding
pc | beam.WindowInto(MyWindowFn())

coder = windowfn.get_window_coder()
assert coder.is_deterministic(), f'{windowfn} coder {coder} is not deterministic'

try:
    pc | beam.WindowInto(custom_windowfn)
except ValueError as e:
    log.error('Window coder issue: %s', e)

When it happens

Trigger: Passing a custom WindowFn whose get_window_coder() returns a non-deterministic coder (e.g. a coder over unsorted iterables or float fields) into beam.WindowInto or the Windowing transform.

Common situations: Custom IntervalWindow/GlobalWindow subclasses coded with an inferred coder based on floats or dicts; third-party window functions; coders affected by Python hash randomization for sets.

Understand the failure class

Background: "Must be a positive integer", "Invalid value", "Unsupported": the invalid-argument-value error family, when a library rejects the value you pass — this error's family across 35 libraries.

Related errors

• Window coders must be deterministic.
• window_coder should not be None
• Unknown PaneInfo encoding 0x" + encoding.toString(16)
• Expected that the coder is deterministic
• the keyCoder of a GroupByEncryptedKey must be deterministic

AI-assisted analysis of apache/beam@12126d8942 (2026-09-13). Data as JSON: /api/errors/d6620dbc7169edd8. Report an issue: GitHub.