workhorse 2.0.0.rc1 → 2.0.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 84d0f28bfeb67292f3de373115ada4ef7c2cff27b29e847af29326733ba89841
-  data.tar.gz: f69f1d82c2c9c15f47f87d8f8390c31e655596636383b9cadefc8c240d8f43b6
+  metadata.gz: 0ca6c8e15a44efebbcbdc4af242c10397aeba1a20baa3b3c45af4025a657bafe
+  data.tar.gz: f342e2b53fb79ea99982101e5fc6912baacfe2fa19b6ad97cc3dd9a5922214f1
 SHA512:
-  metadata.gz: 00c490f9e4aa2b57066b8b974e38286509cb156cb27f79dc7d7a355c9797338eaa8dcf33a5dc9c07e2f3bbe37878d0be5e82b7f098da9e45194ad0fe5b064d6f
-  data.tar.gz: e0116907c47ae31d9fa5a5626e331a503286667966aa974255e8d9f816c363fe3a9e55406e5f79c9a3da7e82abe4b0648b505de3e279992a58e1864d847edaef
+  metadata.gz: e3c898ab631105ce71355032551dd70c94c4a357b17314258756c982c7f06e9e12cc89732c351030a20b5709ff1cfb99a78b123377572637e4ac03d0a0c165bb
+  data.tar.gz: 5b90517edcd626f792a1861cac9a513b83ff7846a92bcc00c289c03596a0b3255141f59ce3a51e9b1d724d2f425c648c4fb6dfac76a43a17ef757c49fc73509b

data/CHANGELOG.md

@@ -1,65 +1,13 @@
 # Workhorse Changelog
 
-## 2.0.0.rc1 - 2026-09-30
-
-Sitrox reference: #154443.
-
-### Added
-
-* Oracle 12c or later is supported again, and is now covered by CI against
-  `activerecord-oracle_enhanced-adapter` rather than only by hand. The global
-  lock uses `DBMS_LOCK`, and the generated migrations leave out the index
-  prefix lengths Oracle rejects while naming every index within the 30
-  characters it allows before 12.2. See
-  [Database support](README.md#database-support).
-
-  2.0.0.rc0 dropped Oracle; anyone who took that release up and needs it can
-  move straight to this one. Grant the schema execute permission on
-  `DBMS_LOCK`, as described under Database support.
-
-### Fixed
-
-* On Oracle, jobs are picked up in priority order. 1.x limited the rows it
-  selected by filtering on `ROWNUM`, which Oracle numbers before sorting, so
-  both the job taken from each queue and the jobs a poll picked up were an
-  arbitrary subset regardless of `priority`. Limiting now uses
-  `FETCH FIRST … ROWS ONLY`, which applies after the sort and is why 12c is
-  now the minimum.
-
-* A poll that finds the global lock taken now waits for it with a polling
-  interval below half a second. The lock timeout follows the interval, and
-  MySQL and Oracle take it as whole seconds, so they rounded it down to not
-  waiting at all - only MariaDB honours a fraction. Each such poll gave up
-  at once and counted towards `Workhorse.max_global_lock_fails`, so several
-  workers contending for the lock could trigger its alarm about a crashed
-  worker. The timeout is now rounded up to whole seconds.
-
-* A soft restart (`USR1`) no longer wedges a worker whose `tmp/pids` does not
-  exist yet, as on a fresh checkout or deployment. The worker touched its
-  shutdown file there after it had stopped accepting jobs, so failing on it
-  left a worker that took no jobs and never exited. The directory is now
-  created when missing, and a shutdown file that still cannot be written no
-  longer holds up the shutdown. The heartbeat and the memory-limit shutdown,
-  which write to the same directory, create it as well.
-
-* `Workhorse.clean_stuck_jobs` works on Oracle. It is built on splitting
-  `locked_by` into host, PID and random component, which was written with
-  MySQL's `substring_index` and so raised on Oracle for as long as the option
-  has existed — visible only to an installation that turned it on, as it is
-  off by default.
-
-## 2.0.0.rc0 - 2026-09-29
+## 2.0.0 - 2026-10-14
 
 Sitrox reference: #154443.
 
 ### Breaking changes
 
-* **Support for Oracle is dropped.** Workhorse supports MySQL and MariaDB
-  only. The Oracle branches of the global lock, the row limiting and the
-  generated migrations are gone, along with
-  `Workhorse::Poller::ORACLE_LOCK_MODE` and `ORACLE_LOCK_HANDLE`. It was never
-  covered by CI, so it only ever had manual verification. An Oracle
-  installation has no upgrade path and should stay on 1.x. See
+* Oracle 12c is the minimum, as job selection now limits its rows with
+  `FETCH FIRST … ROWS ONLY` (see the priority fix below). See
   [Database support](README.md#database-support).
 
 * A forked daemon worker no longer runs the `at_exit` handlers registered by
@@ -108,7 +56,9 @@ Sitrox reference: #154443.
 * The composite indexes `[state, perform_at]`, `[state, priority, created_at]`
   and `[state, expires_at]` in the generated `jobs` migration, replacing the
   single-column index on `state`. `rails generate workhorse:install` now emits
-  two migrations rather than one.
+  two migrations rather than one. On Oracle, the generated migrations leave
+  out the index prefix lengths Oracle rejects and name every index within the
+  30 characters it allows before 12.2.
 
 * `fugit` as a runtime dependency, for parsing cron expressions.
 
@@ -127,8 +77,40 @@ Sitrox reference: #154443.
 
 * `Workhorse::DbJob#reset!` accepts `expired` as the terminal state it is.
 
+* Oracle is covered by CI against `activerecord-oracle_enhanced-adapter`
+  rather than only by hand.
+
 ### Fixed
 
+* On Oracle, jobs are picked up in priority order. 1.x limited the rows it
+  selected by filtering on `ROWNUM`, which Oracle numbers before sorting, so
+  both the job taken from each queue and the jobs a poll picked up were an
+  arbitrary subset regardless of `priority`. Limiting now uses
+  `FETCH FIRST … ROWS ONLY`, which applies after the sort and is why 12c is
+  now the minimum.
+
+* A poll that finds the global lock taken now waits for it with a polling
+  interval below half a second. The lock timeout follows the interval, and
+  MySQL and Oracle take it as whole seconds, so they rounded it down to not
+  waiting at all - only MariaDB honours a fraction. Each such poll gave up
+  at once and counted towards `Workhorse.max_global_lock_fails`, so several
+  workers contending for the lock could trigger its alarm about a crashed
+  worker. The timeout is now rounded up to whole seconds.
+
+* A soft restart (`USR1`) no longer wedges a worker whose `tmp/pids` does not
+  exist yet, as on a fresh checkout or deployment. The worker touched its
+  shutdown file there after it had stopped accepting jobs, so failing on it
+  left a worker that took no jobs and never exited. The directory is now
+  created when missing, and a shutdown file that still cannot be written no
+  longer holds up the shutdown. The heartbeat and the memory-limit shutdown,
+  which write to the same directory, create it as well.
+
+* `Workhorse.clean_stuck_jobs` works on Oracle. It is built on splitting
+  `locked_by` into host, PID and random component, which was written with
+  MySQL's `substring_index` and so raised on Oracle for as long as the option
+  has existed — visible only to an installation that turned it on, as it is
+  off by default.
+
 * A deadlock between shutting a worker down and the poller posting a job.
   `Worker#shutdown` held the worker's mutex while waiting for the poller
   thread, which could be waiting for that same mutex in `Worker#perform`. The

data/VERSION

@@ -1 +1 @@
-2.0.0.rc1
+2.0.0

data/workhorse.gemspec

@@ -1,14 +1,14 @@
 # -*- encoding: utf-8 -*-
-# stub: workhorse 2.0.0.rc1 ruby lib
+# stub: workhorse 2.0.0 ruby lib
 
 Gem::Specification.new do |s|
   s.name = "workhorse".freeze
-  s.version = "2.0.0.rc1".freeze
+  s.version = "2.0.0".freeze
 
   s.required_rubygems_version = Gem::Requirement.new(">= 0".freeze) if s.respond_to? :required_rubygems_version=
   s.require_paths = ["lib".freeze]
   s.authors = ["Sitrox".freeze]
-  s.date = "2026-09-30"
+  s.date = "2026-10-14"
   s.files = [".github/workflows/ruby.yml".freeze, ".gitignore".freeze, ".releaser_config".freeze, ".rubocop.yml".freeze, "CHANGELOG.md".freeze, "FAQ.md".freeze, "Gemfile".freeze, "LICENSE".freeze, "README.md".freeze, "RUBY_VERSION".freeze, "Rakefile".freeze, "VERSION".freeze, "bin/rubocop".freeze, "lib/active_job/queue_adapters/workhorse_adapter.rb".freeze, "lib/generators/workhorse/install_generator.rb".freeze, "lib/generators/workhorse/templates/bin/workhorse.rb".freeze, "lib/generators/workhorse/templates/config/initializers/workhorse.rb".freeze, "lib/generators/workhorse/templates/create_table_jobs.rb".freeze, "lib/generators/workhorse/templates/create_table_workhorse_schedules.rb".freeze, "lib/workhorse.rb".freeze, "lib/workhorse/active_job_extension.rb".freeze, "lib/workhorse/daemon.rb".freeze, "lib/workhorse/daemon/shell_handler.rb".freeze, "lib/workhorse/db_job.rb".freeze, "lib/workhorse/enqueuer.rb".freeze, "lib/workhorse/jobs/cleanup_succeeded_jobs.rb".freeze, "lib/workhorse/jobs/detect_late_schedules_job.rb".freeze, "lib/workhorse/jobs/detect_stale_jobs_job.rb".freeze, "lib/workhorse/jobs/run_active_job.rb".freeze, "lib/workhorse/jobs/run_rails_op.rb".freeze, "lib/workhorse/notifiers/base.rb".freeze, "lib/workhorse/notifiers/file_system.rb".freeze, "lib/workhorse/notifiers/none.rb".freeze, "lib/workhorse/notifiers/redis.rb".freeze, "lib/workhorse/performer.rb".freeze, "lib/workhorse/poller.rb".freeze, "lib/workhorse/pool.rb".freeze, "lib/workhorse/schedule.rb".freeze, "lib/workhorse/schedules.rb".freeze, "lib/workhorse/scoped_env.rb".freeze, "lib/workhorse/worker.rb".freeze, "test/active_job/queue_adapters/workhorse_adapter_test.rb".freeze, "test/lib/db_schema.rb".freeze, "test/lib/jobs.rb".freeze, "test/lib/test_helper.rb".freeze, "test/workhorse/daemon_test.rb".freeze, "test/workhorse/db_job_test.rb".freeze, "test/workhorse/enqueuer_test.rb".freeze, "test/workhorse/notifier_test.rb".freeze, "test/workhorse/performer_test.rb".freeze, "test/workhorse/poller_test.rb".freeze, "test/workhorse/pool_test.rb".freeze, "test/workhorse/schedule_test.rb".freeze, "test/workhorse/worker_test.rb".freeze, "test/workhorse/yjit_test.rb".freeze, "workhorse.gemspec".freeze]
   s.homepage = "https://github.com/sitrox/workhorse".freeze
   s.licenses = ["MIT".freeze]

metadata

@@ -1,13 +1,13 @@
 --- !ruby/object:Gem::Specification
 name: workhorse
 version: !ruby/object:Gem::Version
-  version: 2.0.0.rc1
+  version: 2.0.0
 platform: ruby
 authors:
 - Sitrox
 bindir: bin
 cert_chain: []
-date: 2026-09-30 00:00:00.000000000 Z
+date: 2026-10-14 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: activesupport