[CodingAnarchy]

Closes #130. Part of #122.

Pre-release fixes for the core library: six bugs and the cleanups listed in the issue.

Fixes

1. Worker events without webhooks. Worker::with_event_manager, the event_manager field and the publish path in record_event were gated on webhooks. They now compile in every build, because the events module always does. Streaming-only builds and plain EventManager subscribers now get worker events. The worker event tests in worker_config_tests and payload_encryption_tests no longer need webhooks.
2. Config sections of gated features. webhooks, streaming, alerting and metrics on HammerworkConfig are now #[serde(default)]. When a feature is off, a private Option<serde::de::IgnoredAny> placeholder (skip_serializing) detects the section. from_file then returns HammerworkError::Config("[alerting] is set but hammerwork was built without the alerting feature"). from_env reads no sections, so it needs no check. Tests cover a file saved without the sections loading in any build, and the rejection message for each compiled-out feature.
3. TestQueue::set_throttle_config calls ThrottleConfig::validate. The shared throttle_configuration scenario now asserts the rejection on PG, MySQL and TestQueue.
4. config::SimpleEventFilter is deleted, along with its mentions in the CHANGELOG and the upgrade guide.
5. MySQL session-local time.
   • The key manager's MySQL SQL uses UTC_TIMESTAMP(6) instead of NOW(6) (12 sites).
   • Migration 021 now gives CURRENT_TIMESTAMP columns the expression default (UTC_TIMESTAMP(6)) (MySQL 8.0.13+, now documented). It drops ON UPDATE on hammerwork_queue_pause.updated_at; every library write to that table (pause_queue's insert and upsert) already set it explicitly, as 2.5.0 did.
   • The 021 guards also fire on a current_timestamp default or an on update extra, so a schema converted by the old 021 converges.
   • migration_datetime_tests now checks:
     • defaults are UTC from a +05:00 session;
     • updated_at is not auto-touched, and pause writes it;
     • a schema with the old defaults converges;
     • a re-run changes nothing.
6. WorkerConfig::validate rejects batch_size 0 or above MAX_DEQUEUE_BATCH_SIZE, and batch_concurrency 0. Worker::with_hammerwork_config now runs config.worker.validate() too.

Cleanups

• backend::dispatch_dequeue holds the 0/1/N dispatch and the MAX_DEQUEUE_BATCH_SIZE cap; PG, MySQL and TestQueue all use it. The caps inside claim_many are gone.
• On PG and MySQL, a single weighted claim goes through fill_weighted_batch with room 1. MySQL's dequeue_with_priority_weights_inner is removed, and a runnable_levels helper replaces the duplicated levels query. The behaviour is the same:
  • same candidates and seeds (weighted_seed(queue, attempt));
  • the same move to the next level when a level comes back empty;
  • the same error propagation, since with room 1 nothing is ever claimed before an error.
• Dead cfg(not(feature)) fallbacks are deleted, together with the always-true cfg(feature) attributes in those modules:
  • alerting.rs: the webhook and Slack stubs;
  • metrics.rs: the no-op collector and its test;
  • encryption/engine.rs: the "feature not enabled" branch;
  • webhooks.rs: the redundant attributes only.
• WebhookManager::new and new_default return Result, and try_new is removed.
• RateLimit::with_burst_limit returns Result and rejects 0. Worker::with_attempt_cap returns Result and rejects values below 1. Both follow the fallible style 3.0: remove legacy compatibility paths #90 used for RateLimit.

Not done

• TestQueue does not use weighted_seed. It seeds from its mock clock, which keeps its picks deterministic under a frozen clock. The backends' weighted_seed hashes wall-clock nanoseconds and the thread id, so switching would make TestQueue picks non-deterministic. That is not provably the same behaviour, so it was left as is.
• Other MySQL NOW(). It remains only in cargo-hammerwork/src/integration_tests/sql_query.rs:149,169 (MySQL test queries using DATE_SUB(NOW(), ...)). These were left for 3.0 pre-release: __private, requeue-stale scope, health lease rule, docs #131. hammerwork-web has no MySQL NOW().

Docs

• CHANGELOG [Unreleased]: the breaking changes and fixes, plus an updated 021 Database entry.
• Upgrade guide:
  • a new "Configuration sections and worker events" subsection under Feature flags;
  • notes under Batch mode, with_attempt_cap, MySQL 021 defaults and Zero rates (burst limit);
  • a new "WebhookManager::new returns a Result" subsection.
• ? added to doc examples in docs/*.md for the now-fallible builders.

Testing

• cargo fmt --all --check, cargo clippy --workspace --all-features --all-targets -D warnings, cargo check --examples --all-features and cargo doc --workspace --all-features --no-deps (with -D warnings) are all clean.
• cargo clippy -p hammerwork --lib --tests --no-default-features -D warnings is clean for every one of these feature sets: "", postgres, mysql, streaming, kafka, kinesis, google-pubsub, alerting, metrics, webhooks, full, test-queue, postgres,streaming, encryption and mysql,webhooks.
• PG suite: 2114 passed, 1 failed. MySQL suite: 233 passed, 1 failed. In both, the failure is queue_names_include_archived_queues, caused by leftover hunt_* / hunt3_q queues in the shared local databases. The database collation orders hunt_... before hunt3_q, while Rust byte order does the opposite. No code here touches that query.
• The shared local MySQL DB was re-migrated: the 021_mysql_datetime row was deleted and migrations rerun. All DATETIME defaults now read utc_timestamp(6), and no ON UPDATE remains.

🤖 Generated with Claude Code