Daniel Thevessen created BEAM-13599:
---------------------------------------
Summary: Overflow in Python Datastore RampupThrottlingFn
Key: BEAM-13599
URL: https://issues.apache.org/jira/browse/BEAM-13599
Project: Beam
Issue Type: Bug
Components: io-py-gcp
Affects Versions: 2.35.0, 2.34.0, 2.33.0, 2.32.0
Reporter: Daniel Thevessen
```
File
"/usr/local/lib/python3.8/site-packages/apache_beam/io/gcp/datastore/v1new/rampup_throttling_fn.py",
line 74, in _calc_max_ops_budget
max_ops_budget = int(self._BASE_BUDGET / self._num_workers * (1.5**growth))
RuntimeError: OverflowError: (34, 'Numerical result out of range') `[while
running 'Write to Datastore/Enforce throttling during ramp-up-ptransform-483']
```
An intermediate value is a float dependent on start time, meaning it can run
into overflows in long-running pipelines (usually on the ~6th day).
`max_ops_budget` should either clip to float(inf) or INT_MAX, or short-circuit
the throttling decision
[here]([https://github.com/apache/beam/blob/ea65a054f2fcb6349478d19609a773f66bbfa20e/sdks/python/apache_beam/io/gcp/datastore/v1new/rampup_throttling_fn.py#L87)]
since it will long be irrelevant by then.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
