apache_beam.io.gcp.datastore.v1.adaptive_throttler module

class apache_beam.io.gcp.datastore.v1.adaptive_throttler.AdaptiveThrottler(window_ms, bucket_ms, overload_ratio)[source]

Bases: object

Implements adaptive throttling.

See https://landing.google.com/sre/book/chapters/handling-overload.html#client-side-throttling-a7sYUg for a full discussion of the use case and algorithm applied.

Initializes AdaptiveThrottler.

Parameters:
  • window_ms – int, length of history to consider, in ms, to set throttling.
  • bucket_ms – int, granularity of time buckets that we store data in, in ms.
  • overload_ratio – float, the target ratio between requests sent and successful requests. This is “K” in the formula in https://landing.google.com/sre/book/chapters/handling-overload.html.
MIN_REQUESTS = 1
throttle_request(now)[source]

Determines whether one RPC attempt should be throttled.

This should be called once each time the caller intends to send an RPC; if it returns true, drop or delay that request (calling this function again after the delay).

Parameters:now – int, time in ms since the epoch
Returns:bool, True if the caller should throttle or delay the request.
successful_request(now)[source]

Notifies the throttler of a successful request.

Must be called once for each request (for which throttle_request was previously called) that succeeded.

Parameters:now – int, time in ms since the epoch