machinery

This releases contains some bug fixes and a new feature:

• merged support for opentracing (for example, with jaeger)
• refactored reflection logic, added some improvements and fixed couple small bugs

This releases contains some bug fixes and a new feature:

• fixed precision bug with large integers (caused by JSON encoding & decoding since Javascript objects only support floating point numbers, therefor integers get converted to floats and then back to integers when being transferred in JSON format)
• a custom retry logic can now be implemented via returning a custom retry error (tasks.NewErrRetryTaskLater)

This release contains few bigger changes as well as several small improvements and bug fixes.

• added support for slice task arguments
• fixed several issues with AWS SQS broker
• fixed Redis URL parsing

Config loading would result in AMQP config being present even for non AMQP brokers because the default config contains it. This has been fixed so non AMQP configs are loaded correctly with AMQP config object being nil.

Better documentation, tests and README for AWS SQS.

Machinery now supports AWS SQS broker.

AdjustRoutingKey has been fixed to check broker type and only rewrite routing key to binding key if the broker is AMQP and exchange is direct. Ignore this logic for Redis / other non AMQP brokers.

Tasks should be acked AFTER processing not before. This was fixed before but reverted by mistake.

See here: https://www.rabbitmq.com/confirms.html

Delivered tasks will be re-queued in case worker connection is lost. But otherwise ack should be sent after task is processed, not in advance.

Delayed tasks with other than default queue as routing key did not work with Redis broker as once delayed task was fetched from the delayed_tasks queue, it was processed by the current worker process instead of being published to a proper queue (which might be different than current worker instance).

Refactored and simplified config creation methods. Do not store config in internal variable anymore (was causing issues with multi config setups).

Added a non blocking LaunchAsync method to worker struct as alternative to blocking Launch.

First Ctrl+C starts graceful shutdown (i.e. wait for running tasks to finish before shutting down). Second Ctrl+C will just abruptly end the program without waiting for tasks.

When there are running tasks and you call worker.Quit() method, the process will wait for all running tasks to be processed first before shutting down.

Updated dep cli tool and all vendored dependencies.

Fixed gometalinter issue on Travis CI so builds now pass successfully. Usage of native go tool vet is enforced now via gometalinter.json config file.

When a task has OnSuccess callback, the returned arguments are appended to the callback task instead of prepended.

Small MongoDB Backend Improvement to allow storing task results as bson.Json.

Error channel in SendGroup could potentially have up to number of tasks * 2 size. Channel size increased to accomodate for this eventuality.

Fixed a bug in SendGroup pooling logic which meant all tasks would be sent synchronously irrespective of the concurrency parameter.

When using AMQP as backend, x-expires parameter is set when declaring a queue in order to set/update task state. This ensures task state queues will eventually get cleanup up.