CHANGELOG.md

1.0.0.beta2 (2018-04-13)

• A schema upgrade to version 4 will be required for this release. See the migration doc for information if you're upgrading from a previous release.

  • Please note that this migration requires a rewrite of the jobs table, which makes it O(n) with the size of the table. If you have a very large backlog of jobs you may want to schedule downtime for this migration.

• Que's implementation has been changed from one in which worker threads hold their own PG connections and lock their own jobs to one in which a single thread (and PG connection) locks jobs through LISTEN/NOTIFY and batch polling, and passes jobs along to worker threads. This has many benefits, including:

  • Jobs queued for immediate processing can be actively distributed to workers with LISTEN/NOTIFY, which is more efficient than having workers repeatedly poll for new jobs.

  • When polling is necessary (to pick up jobs that are scheduled for the future or that need to be retried due to errors), jobs can be locked and fetched in batches, rather than one at a time.

  • Individual workers no longer need to monopolize their own (usually idle) connections while working jobs, so Ruby processes will require fewer Postgres connections.

  • PgBouncer or another external connection pool can be used for workers' connections (though not for the connection used to lock and listen for jobs).

• Other features introduced in this version include:

  • Much better support for all versions of ActiveJob.

    • In particular, you may (optionally) include Que::ActiveJob::JobExtensions into ApplicationJob to get support for all of Que's job helper methods.

  • Custom middleware that wrap running jobs and executing SQL statements are now supported.

  • Support for categorizing jobs with tags.

  • Support for configuring a maximum_retry_count on individual job classes.

  • Job configuration options are now inheritable, so job class hierarchies are more useful.

  • There are now built-in models for ActiveRecord and Sequel to allow inspecting the queue easily.

  • Jobs that have finished working may optionally be retained in the database indefinitely.

    • To keep a job record, replace the destroy calls in your jobs with finish. destroy will still delete records entirely, for jobs that you don't want to keep.

    • If you don't resolve a job yourself one way or another, Que will still destroy the job for you by default.

    • Finished jobs have a timestamp set in the finished_at column.

  • Jobs that have errored too many times will now be marked as expired, and won't be retried again.

    • You can configure a maximum_retry_count in your job classes, to set the threshold at which a job will be marked expired. The default is 15.

    • To manually mark a job as expired (and keep it in the database but not try to run it again) you can call expire helper in your job.

  • You can now set job priority thresholds for individual workers, to ensure that there will always be space available for high-priority jobs.

  • Que.job_states returns a list of locked jobs and the hostname/pid of the Ruby processes that have locked them.

  • Que.connection_proc= has been added, to allow for the easy integration of custom connection pools.

• In keeping with semantic versioning, the major version is being bumped since the new implementation requires some backwards-incompatible changes. These changes include:

  • Support for MRI Rubies before 2.2 has been dropped.

  • Support for Postgres versions before 9.5 has been dropped (JSONB and upsert support is required).

  • JRuby support has been dropped. It will be reintroduced whenever the jruby-pg gem is production-ready.

  • The que:work rake task has been removed. Use the que executable instead.

    • Therefore, configuring workers using QUE_* environment variables is no longer supported. Please pass the appropriate options to the que executable instead.

  • The mode setter has been removed.

    • To run jobs synchronously when they are enqueued (the old :sync behavior) you can set Que.run_synchronously = true.

    • To start up the worker pool (the old :async behavior) you should use the que executable to start up a worker process. There's no longer a supported API for running workers outside of the que executable.

  • The following methods are not meaningful under the new implementation and have been removed:

    • The Que.wake_interval getter and setter.

    • The Que.worker_count getter and setter.

    • Que.wake!

    • Que.wake_all!

  • Since Que needs a dedicated Postgres connection to manage job locks, running Que through a single PG connection is no longer supported.

    • It's not clear that anyone ever actually did this.

  • Que.worker_states has been removed, as the connection that locks a job is no longer the one that the job is using to run. Its functionality has been partially replaced with Que.job_states.

  • When using Rails, for simplicity, job attributes and keys in argument hashes are now converted to symbols when retrieved from the database, rather than being converted to instances of HashWithIndifferentAccess.

  • Arguments passed to jobs are now deep-frozen, to prevent unexpected behavior when the args are mutated and the job is reenqueued.

  • Since JSONB is now used to store arguments, the order of argument hashes is no longer maintained.

    • It wouldn't have been a good idea to rely on this anyway.

  • Calling Que.log() directly is no longer supported/recommended.

  • Features marked as deprecated in the final 0.x releases have been removed.

• Finally, if you've built up your own tooling and customizations around Que, you may need to be aware of some DB schema changes made in the migration to schema version #4.

  • The job_id column has been renamed id and is now the primary key. This makes it easier to manage the queue using an ActiveRecord model.

  • Finished jobs are now kept in the DB, unless you explicitly call destroy. If you want to query the DB for only jobs that haven't finished yet, add a WHERE finished_at IS NULL condition to your query, or use the not_finished scope on one of the provided ORM models.

  • There is now an expired_at timestamp column, which is set when a job reaches its maximum number of retries and will not be attempted again.

  • Due to popular demand, the default queue name is now "default" rather than an empty string. The migration will move pending jobs under the "" queue to the "default" queue.

  • The last_error column has been split in two, to last_error_message and last_error_backtrace. These two columns are now limited to 500 and 10,000 characters, respectively. The migration will split old error data correctly, and truncate it if necessary.

  • Names for queues and job classes are now limited to 500 characters, which is still far longer than either of these values should reasonably be.

  • There is now a data JSONB column which is used to support various ways of organizing jobs (setting tags on them, etc).

For a detailed list of the changes between each beta release of 1.0.0, see the beta Changelog.

0.14.3 (2018-03-02)

• Recorded errors now always include the error class, so that empty error messages can still be helpful. ( joehorsnell)

• Recorded error messages are now truncated to the first 500 characters.

0.14.2 (2018-01-05)

• Deprecate the Que.disable_prepared_statements= accessors.

• Add Que.use_prepared_statements= configuration accessors.

• Update the generated Rails migration to declare a version. (NARKOZ)

0.14.1 (2017-12-14)

• Fix a bug with typecasting boolean values on Rails 5+.

0.14.0 (2017-08-11)

• Fix incompatibility with Rails 5.1.

• Drop support for waking an in-process worker when an ActiveRecord transaction commits.

0.13.1 (2017-07-05)

• Fix issue that caused error stacktraces to not be persisted in most cases.

0.13.0 (2017-06-08)

• Fix recurring JSON issues by dropping MultiJson support. Previously MultiJson was detected and used automatically, and now it's just ignored and stdlib JSON used instead, so this shouldn't require any code changes.

0.12.3 (2017-06-01)

• Fix incompatibility with MultiJson introduced by the previous release.

0.12.2 (2017-06-01)

• Fix security vulnerability in parsing JSON from the DB (by specifying create_additions: false). This shouldn't be a concern unless you were passing untrusted user input in your job arguments. (hmac)

0.12.1 (2017-01-22)

• Fix incompatibility with Rails 5.0. (#166) (nbibler, thedarkone)

0.12.0 (2016-09-09)

• The error_handler configuration option has been renamed to error_notifier, which is more descriptive of what it's actually supposed to do. You can still use error_handler for configuration, but you'll get a warning.

• Introduced a new framework for handling errors on a per-job basis. See the docs for more information. (#106, #147)

0.11.6 (2016-07-01)

• Fix for operating in nested transactions in Rails 5.0. (#160) (greysteil)

0.11.5 (2016-05-13)

• Fix error when running que -v. (#154) (hardbap)

0.11.4 (2016-03-03)

• Fix incompatibility with ActiveRecord 5.0.0.beta3. (#143, #144) (joevandyk)

0.11.3 (2016-02-26)

• Fixed bug with displaying the current version of the que executable. (#122) (hardbap)

• Output to STDOUT when running via the executable or rake task is no longer buffered. This prevented logging in some cases. (#129) (hmarr)

• Officially added support for Ruby 2.2 and 2.3.

• String literals are now frozen on Ruby 2.3.

0.11.2 (2015-09-09)

• Fix Job class constantizing when ActiveSupport isn't loaded. (#121) (godfat)

0.11.1 (2015-09-04)

• The rake que:work rake task that was specific to Rails has been deprecated and will be removed in Que 1.0. A deprecation warning will display when it is run.

0.11.0 (2015-09-04)

• A command-line program has been added that can be used to work jobs in a more flexible manner than the previous rake task. Run que -h for more information.

• The worker pool will no longer start automatically in the same process when running the rails server - this behavior was too prone to breakage. If you'd like to recreate the old behavior, you can manually set Que.mode = :async in your app whenever conditions are appropriate (classes have loaded, a database connection has been established, and the process will not be forking).

• Add a Que.disable_prepared_transactions= configuration option, to make it easier to use tools like pgbouncer. (#110)

• Add a Que.json_converter= option, to configure how arguments are transformed before being passed to the job. By default this is set to the Que::INDIFFERENTIATOR proc, which provides simple indifferent access (via strings or symbols) to args hashes. If you're using Rails, the default is to convert the args to HashWithIndifferentAccess instead. You can also pass it the Que::SYMBOLIZER proc, which will destructively convert all keys in the args hash to symbols (this will probably be the default in Que 1.0). If you want to define a custom converter, you will usually want to pass this option a proc, and you'll probably want it to be recursive. See the implementations of Que::INDIFFERENTIATOR and Que::SYMBOLIZER for examples. (#113)

• When using Que with ActiveRecord, workers now call ActiveRecord::Base.clear_active_connections! between jobs. This cleans up connections that ActiveRecord leaks when it is used to access mutliple databases. (#116)

• If it exists, use String#constantize to constantize job classes, since ActiveSupport's constantize method behaves better with Rails' autoloading. (#115, #120) (joevandyk)

0.10.0 (2015-03-18)

• When working jobs via the rake task, Rails applications are now eager-loaded if present, to avoid problems with multithreading and autoloading. (#96) (hmarr)

• The que:work rake task now uses whatever logger Que is configured to use normally, rather than forcing the use of STDOUT. (#95)

• Add Que.transaction() helper method, to aid in transaction management in migrations or when the user's ORM doesn't provide one. (#81)

0.9.2 (2015-02-05)

• Fix a bug wherein the at_exit hook in the railtie wasn't waiting for jobs to finish before exiting.

• Fix a bug wherein the que:work rake task wasn't waiting for jobs to finish before exiting. (#85) (tycooon)

0.9.1 (2015-01-11)

• Use now() rather than 'now' when inserting jobs, to avoid using an old value as the default run_at in prepared statements. (#74) (bgentry)

0.9.0 (2014-12-16)

• The error_handler callable is now passed two objects, the error and the job that raised it. If your current error_handler is a proc, as recommended in the docs, you shouldn't need to make any code changes, unless you want to use the job in your error handling. If your error_handler is a lambda, or another callable with a strict arity requirement, you'll want to change it before upgrading. (#69) (statianzo)

0.8.2 (2014-10-12)

• Fix errors raised during rollbacks in the ActiveRecord adapter, which remained silent until Rails 4.2. (#64, #65) (Strech)

0.8.1 (2014-07-28)

• Fix regression introduced in the que:work rake task by the mode / worker_count disentangling in 0.8.0. (#50)

0.8.0 (2014-07-12)

• A callable can now be set as the logger, like Que.logger = proc { MyLogger.new }. Que uses this in its Railtie for cleaner initialization, but it is also available for public use.

• Que.mode= and Que.worker_count= now function independently. That is, setting the worker_count to a nonzero number no longer sets mode = :async (triggering the pool to start working jobs), and setting it to zero no longer sets mode = :off. Similarly, setting the mode to :async no longer sets the worker_count to 4 from 0, and setting the mode to :off no longer sets the worker_count to 0. This behavior was changed because it was interfering with configuration during initialization of Rails applications, and because it was unexpected. (#47)

• Fixed a similar bug wherein setting a wake_interval during application startup would break worker awakening after the process was forked.

0.7.3 (2014-05-19)

• When mode = :sync, don't touch the database at all when running jobs inline. Needed for ActiveJob compatibility (#46).

0.7.2 (2014-05-18)

• Fix issue wherein intermittent worker wakeups would not work after forking (#44).

0.7.1 (2014-04-29)

• Fix errors with prepared statements when ActiveRecord reconnects to the database. (dvrensk)

• Don't use prepared statements when inside a transaction. This negates the risk of a prepared statement error harming the entire transaction. The query that benefits the most from preparation is the job-lock CTE, which is never run in a transaction, so the performance impact should be negligible.

0.7.0 (2014-04-09)

• JobClass.queue(*args) has been deprecated and will be removed in version 1.0.0. Please use JobClass.enqueue(*args) instead.

• The @default_priority and @default_run_at variables have been deprecated and will be removed in version 1.0.0. Please use @priority and @run_at instead, respectively.

• Log lines now include the process pid - its omission in the previous release was an oversight.

• The Pond gem is now supported as a connection. It is very similar to the ConnectionPool gem, but creates connections lazily and is dynamically resizable.

0.6.0 (2014-02-04)

• A schema upgrade to version 3 is required for this release. See the migration doc for information if you're upgrading from a previous release.

• You can now run a job's logic directly (without enqueueing it) like MyJob.run(arg1, arg2, other_arg: arg3). This is useful when a job class encapsulates logic that you want to invoke without involving the entire queue.

• You can now check the current version of Que's database schema with Que.db_version.

• The method for enqueuing a job has been renamed from MyJob.queue to MyJob.enqueue, since we were beginning to use the word 'queue' in a LOT of places. MyJob.queue still works, but it may be removed at some point.

• The variables for setting the defaults for a given job class have been changed from @default_priority to @priority and @default_run_at to @run_at. The old variables still work, but like Job.queue, they may be removed at some point.

• Log lines now include the machine's hostname, since a pid alone may not uniquely identify a process.

• Multiple queues are now supported. See the docs for details. (chanks, joevandyk)

• Rubinius 2.2 is now supported. (brixen)

• Job classes may now define their own logic for determining the retry interval when a job raises an error. See error handling for more information.

0.5.0 (2014-01-14)

• When running a worker pool inside your web process on ActiveRecord, Que will now wake a worker once a transaction containing a queued job is committed. (joevandyk, chanks)

• The que:work rake task now has a default wake_interval of 0.1 seconds, since it relies exclusively on polling to pick up jobs. You can set a QUE_WAKE_INTERVAL environment variable to change this. The environment variable to set a size for the worker pool in the rake task has also been changed from WORKER_COUNT to QUE_WORKER_COUNT.

• Officially support Ruby 1.9.3. Note that due to the Thread#kill problems (see "Remove Que.stop!" below) there's a danger of data corruption when running under 1.9, though.

• The default priority for jobs is now 100 (it was 1 before). Like always (and like delayed_job), a lower priority means it's more important. You can migrate the schema version to 2 to set the new default value on the que_jobs table, though it's only necessary if you're doing your own INSERTs - if you use MyJob.queue, it's already taken care of.

• Added a migration system to make it easier to change the schema when updating Que. You can now write, for example, Que.migrate!(version: 2) in your migrations. Migrations are run transactionally.

• The logging format has changed to be more easily machine-readable. You can also now customize the logging format by assigning a callable to Que.log_formatter=. See the new doc on logging) for details. The default logger level is INFO - for less critical information (such as when no jobs were found to be available or when a job-lock race condition has been detected and avoided) you can set the QUE_LOG_LEVEL environment variable to DEBUG.

• MultiJson is now a soft dependency. Que will use it if it is available, but it is not required.

• Remove Que.stop!.

  Using Thread#raise to kill workers is a bad idea - the results are unpredictable and nearly impossible to spec reliably. Its purpose was to prevent premature commits in ActiveRecord/Sequel when a thread is killed during shutdown, but it's possible to detect that situation on Ruby 2.0+, so this is really better handled by the ORMs directly. See the pull requests for Sequel and ActiveRecord.

  Now, when a process exits, if the worker pool is running (whether in a rake task or in a web process) the exit will be stalled until all workers have finished their current jobs. If you have long-running jobs, this may take a long time. If you need the process to exit immediately, you can SIGKILL without any threat of commiting prematurely.

0.4.0 (2014-01-05)

• Que.wake_all! was added, as a simple way to wake up all workers in the pool.

• Que.sleep_period was renamed to the more descriptive Que.wake_interval.

• When queueing a job, Que will wait until the current transaction commits and then wake a background worker, if possible. This allows newly queued jobs to be started immediately instead of waiting for a worker to wake up and poll, which may be up to Que.wake_interval seconds.

  This feature currently only works with Sequel, since there doesn't seem to be a clean way to do it on ActiveRecord (if anyone can figure one out, please let me know). Note that if you're using ActiveRecord, you can always manually trigger a single worker to wake up and check for work by manually calling Que.wake! after your transaction completes.

• Add Que.job_stats, which queries the database and returns statistics on the different job classes - for each class, how many are queued, how many are currently being worked, what is the highest error_count, and so on.

• Add Que.worker_states, which queries the database and returns all currently-locked jobs and info on their workers' connections - what and when was the last query they ran, are they waiting on locks, and so on.

• Have Que only clear advisory locks that it has taken when locking jobs, and not touch any that may have been taken by other code using the same connection.

• Add Que.worker_count, to retrieve the current number of workers in the pool of the current process.

• Much more internal cleanup.

0.3.0 (2013-12-21)

• Add Que.stop!, which immediately kills all jobs being worked in the process.

  This can leave database connections and such in an unpredictable state, and so should only be used when the process is exiting.

• Use Que.stop! to safely handle processes that exit while Que is running.

  Previously, a job that was in the middle of a transaction when the process was killed with SIGINT or SIGTERM would have had its work committed prematurely.

• Clean up internals and hammer out several race conditions.

0.2.0 (2013-11-30)

• Officially support JRuby 1.7.5+. Earlier versions may work.

  JRuby support requires the use of the jruby-pg gem, though that gem seems to currently be incompatible with ActiveRecord, so the ActiveRecord adapter specs don't pass (or even run). It works fine with Sequel and the other adapters, though.

• Officially support Rubinius 2.1.1+. Earlier versions may work.

• Use multi_json so we always use the fastest JSON parser available. (BukhariH)

• :sync mode now ignores scheduled jobs (jobs queued with a specific run_at).

0.1.0 (2013-11-18)

• Initial public release, after a test-driven rewrite.

  Officially support Ruby 2.0.0 and Postgres 9.2+.

  Also support ActiveRecord and bare PG::Connections, in or out of a ConnectionPool.

  Added a Railtie for easier setup with Rails, as well as a migration generator.

0.0.1 (2013-11-07)

• Copy-pasted from an app of mine. Very Sequel-specific. Nobody look at it, let's pretend it never happened.