PyPI History

• Moving lock factory used in publisher client to the Batch implementation (#4628).
• Use a UUID (rather than a sentinel object) on Future (#4634).
• Apply scopes to explicitly provided credentials if needed (#4594). Fixes #4479. This feature comes as part of google-api-core==0.1.3.

• Upgrading to google-api-core==0.1.3 which depends on the latest grpcio==1.8.2 (#4642). This fixes #4600. For details, see related gRPC bug and fix.

PyPI: https://pypi.org/project/google-cloud-pubsub/0.30.1/

• Dropping redundant Policy._paused data member (#4568).
• Removing redundant "active" check in policy (#4603).
• Adding a Consumer.active property (#4604).
• Making it impossible to call Policy.open() on an already opened policy (#4606).
• Bug fix (#4575): Fix bug with async publish for batches. There were two related bugs. The first: if a batch exceeds the max_messages from the batch settings, then the commit() will fail. The second: when a "monitor" worker calls commit() after max_latency seconds, a failure can occur if a new message is added to the batch during the commit. To fix, the following changes were implemented:
  • Adding a "STARTING" status for Batch.commit() (#4614). This fixes the issue when the batch exceeds max_messages.
  • Adding extra check in Batch.will_accept for the number of messages (#4612).
  • Moving will_accept() check out of PublisherClient.batch() factory (#4613).
  • Checking Batch.will_accept in thread-safe way (#4616).
• Breaking API change: As part of #4613, changing PublisherClient.batch() to no longer accept a message (since the will_accept check needs to happen in a more concurrency friendly way). In addition, changing the create argument so that it means "create even if batch already exists" rather than "create if missing".

• Add more explicit documentation for Message.attributes (#4601).
• Make Message.__repr__ a bit prettier / more useful (#4602).

PyPI: https://pypi.org/project/google-cloud-pubsub/0.30.0/

• Bug fix: Restore previous behavior of the subscription lease maintenance worker. This was accidentally "stopped" in 0.29.3 due to a change in implementation that went from an active boolean to an "inactive" / stopped boolean, so True became False and vice-versa (#4564).

PyPI: https://pypi.org/project/google-cloud-pubsub/0.29.4/

• In subscription consumer thread: Making sure the request generator attached to an inactive bidirectional streaming pull is stopped before spawning a new request generator. This way we have a (fairly strong) guarantee that requests in the queue don't get sent into an inactive stream (#4503, #4554).
• Adding pause / resume to subscription consumer thread and using these methods during flow control. The previous implementation tried to close the subscription (which involved 3 worker threads and 10 executors in a thread pool) and then re-open a new subscription. But, this was not entirely possible to shut down correctly from within one of the worker threads. Instead, we only pause the worker (of the 3) that is pulling new responses from the bidirectional streaming pull (#4558).
• Bug fix (#4516): Using max where min was used by mistake to ensure the number of bytes tracked for subscription flow control remained non-negative (#4514).
• Raising TypeError if SubscriberClient.subscribe receives a non-callable callback (#4497).
• Shutting down thread pool executor when closing a subscriber policy (#4522).
• Renaming Policy.on_callback_request to Policy.dispatch_callback and making the behavior much less dynamic (#4511).
• Make sure subscription consumer thread doesn't try to join itself when exiting in error (#4540).

• Upgrading google-api-core dependency to latest revision (0.1.2) since we rely on the latest version of the concurrent.futures backport to provide the thread_name_prefix argument for thread pool executor (#4521, #4559).

PyPI: https://pypi.org/project/google-cloud-pubsub/0.29.3/

• Bug fix (#4463): Making a subscription consumer actually stop running after encountering an exception (#4472, #4498). This bug is the only reason for the 0.29.2 release.
• Thread Changes
  • Added names to all threads created directly by Pub / Sub (#4474, #4476, #4480). Also removing spaces and colons from thread names (#4476).
• Logging changes
  • Adding debug logs when lease management exits (#4484)
  • Adding debug logs when QueueCallbackThread exits (#4494). Instances handle the processing of messages in a subscription (e.g. to ack).
  • Using a named logger in publisher.batch.thread (#4473)
  • Adding newlines before logging protobuf payloads (#4471)

PyPI: https://pypi.org/project/google-cloud-pubsub/0.29.2/

• Bug fix (#4234): Adding retries for connection UNAVAILABLE. This bug made the Pub / Sub client mostly unusable for subscribers to topics that don't have a steady stream of messages. After ~2 minutes of inactivity, the gRPC connection would timeout and raise UNAVAILABLE locally, i.e. not due to a response from the backend. (#4444)
• Updating autogenerated packages (#4438)

• Fixing broken examples in quick start (#4398)
• Fixing broken example in README (#4402, h/t to @mehmetboraezer)
• Updating old/dead link to usage doc in README (#4406, h/t to @mehmetboraezer)

• Dropping dependency on google-cloud-core in exchange for google-api-core (#4438)

PyPI: https://pypi.org/project/google-cloud-pubsub/0.29.1/

• Honor max_messages always (#4262)
• Add futures for subscriptions (#4265)
• Set gRPC message options and keepalive (#4269)

• Added link to "Python Development Environment Setup Guide" in project README (#4187, h/t to @michaelawyu)

• Upgrading to google-cloud-core >= 0.28.0 and adding dependency on google-api-core (#4221, #4280)
• Deferring to google-api-core for grpcio and googleapis-common-protosdependencies (#4096, #4098)

PyPI: https://pypi.org/project/google-cloud-pubsub/0.29.0/