v0.0.15

What's Changed

• Try harder to import redis-py by @0xDEC0DE in #13
• Signal job failures when recovering dead jobs by @bigjools in #16
• Support running coroutines in new asyncio worker
• Drop support for Python 3.5

Full Changelog: v0.0.14...v0.0.15

v0.0.14

• Add a max_concurrency parameter to tasks, allowing to limit the number of simultaneous jobs executed across the entire cluster of workers. Implemented by Julian Edwards.

v0.0.13

• Fix typo preventing the Django/Datadog integration from starting

v0.0.12

• Automatically detect dead workers (killed processes, power outage...) and re-enqueue the jobs that were running on them.
• Fix a bug that could result in jobs not being retried if they finished while the workers were shutting down.
• Fix the integration between Django, Datadog and Spinach. Users of these three libraries may have seen the Datadog integration not reporting tasks anymore.
• Remove the deprecated Raven integration, users are encouraged to use the Sentry SDK integration instead.
• Enable performance tracing of tasks with the Sentry SDK.

v0.0.11

• Provide a traceback when logging retry warnings
• Prevent jobs with incompatible task arguments from being scheduled

v0.0.10

• Add AbortException to stop a job from being retried
• Add new Sentry SDK integration
• Deprecate old Sentry Raven integration

v0.0.9

• Allow to safely retry enqueuing jobs
• Use a single Lua script to enqueue normal and future jobs: make scheduling a batch truly atomic
• Add Datadog APM integration

v0.0.8

• Switch back to sending signals from worker threads
• Allow to schedule a task using its function

v0.0.7

• Add Flask integration
• Add Django integration
• Terminate blocking workers on SIGTERM
• Start worker thread name at 1 instead of 0

v0.0.6

• Prevent worker thread from dying during callback in case of connection error to redis
• Officially support Python 3.7
• Reset periodic task score when periodicity is changed
• Add option to stop workers when their queue is empty
• Remove useless function wrapper in decorator
• Submit job execution results to broker from a single thread
• Only generate as many UUIDs as there are periodic tasks
• Log information about workers on startup