v3.2.0

• Use timeout when waiting for queues to empty at shutdown

v3.1.0: - Avoid crashing flush() when CreateLogStream API call is throttled …

• Avoid crashing flush() when CreateLogStream API call is throttled (#192)

  • Replace deprecated function datatime.utcnow() (#196)

v3.0.1

• Truncate messages based on bytes, not unicode characters (#181)

  • Test suite improvements (#180, #182)

v3.0.0

• Use repr to represent all JSON fields of unknown types by
  default. Previously, when passing a mapping (dictionary) as a log
  message, watchtower would replace datetime objects with their
  ".isoformat()" string representation, and would replace all other
  non-JSON-serializable objects with null. The new behavior is to
  use the output of repr() to represent these non-JSON-serializable
  objects. This change may cause your logger to log more data than you
  intended, which is why it triggers a major version bump. If you use
  watchtower to log sensitive information or objects with large repr
  strings, you are advised to examine your log messages to see if any
  newly visible data should be sanitized. If you need to customize
  this behavior, you can pass a custom JSON default serializer using
  the json_serialize_default keyword argument.

v2.1.1

• Use correct default log stream name

v2.1.0

• Deconflict log streams from process pools in default log stream name

  • Documentation and CI improvements

v2.0.1

• Update empty message filtering to catch formatted string case (#162)

v2.0.0

• Rename log_group to log_group_name for consistency with the Boto3 API

  • Rename stream_name to log_stream_name for consistency with the Boto3 API

  • Introduce the ability to pass a Boto3 logs client and remove the ability to pass Boto3 sessions

  • Document the ability to pass a Boto3 configuration profile name for declarative configs

  • Remove the Django customization, which was deprecated and unneeded.
    Django can use watchtower directly via its logging configuration as documented in the readme

  • Introduce configurable log formatters. Special thanks to @terencehonles for starting this work (#117, #138)

  • Use logging.Handler.handleError to correctly handle errors while processing log records (#149)

  • Move stream name determining logic to separate method (#148)

  • Reset internal state on fork to prevent deadlocks in worker threads (#139)

  • Drop Python 3.5 support

  • Expand documentation

  • Update test and release infrastructure

v1.0.6

• Catch OperationAbortedException in _idempotent_create. Fixes #136

v1.0.5

• Don't crash if nextSequenceToken is missing. Fixes #134