v0.8.0

Upgrading needs a few manual NATS steps: see docs/upgrading.md.

Added

• Cosmo.configure with config.server_middleware: a middleware chain around every job execution
• config.client_middleware: a middleware chain around every job enqueue, to change the payload or stop the publish
• config.error_handlers: callables given every error Cosmo rescues, jobs and stream batches included
• config.logger = ... and config.log_level = ... in Cosmo.configure; a logger without trace no longer raises
• Built-in config defaults: Cosmo runs without config/cosmo.yml, and a user file is deep-merged over the defaults;
  cosmo --init writes them into the project
• config.replicas, config.dead (retention, or off), config.scheduled.enabled, and config.batches.expiry in
  Cosmo.configure for Cosmo's own service streams
• Cosmo::Job::Requeue: a middleware raises it to redeliver a job later instead of failing it
• Metrics tab in the web UI: per job class runs, failures, average execution and wait time, and a per-day chart;
  config.metrics.enabled and config.metrics.retention in Cosmo.configure
• The web UI's batches table shows each batch's progress and pending jobs
• The web UI's processes table groups workers by jobs, streams, or both

Changed

• Breaking: the scheduled and dead streams are Cosmo's, created by cosmo --setup: listing them under
  setup.jobs/consumers.jobs raises ConfigError, as does batch_expiry (now config.batches.expiry). Listing
  your own setup.jobs replaces the built-in default stream
• Breaking: the dead stream uses workqueue retention instead of limits. NATS can't change retention in place,
  so delete an existing one (nats stream rm dead, dropping its jobs) before cosmo --setup
• Breaking: Cosmo's internal streams and KV buckets are named _cosmo<domain>: _cosmostats is _cosmototals, and
  the cosmo_jobs_batches, cosmo_jobs_busy, cosmo_processes, and cosmo_jobs_limits buckets are _cosmobatches,
  _cosmobusy, _cosmoprocesses, and _cosmolimits; nothing is migrated automatically
• Breaking: batch counters moved from _cosmostats to their own _cosmobatches stream, which expires them after
  config.batches.expiry; batches still open during the upgrade lose their progress counters
• Breaking: Config.set and Config.to_h are removed: streams and settings go in cosmo.yml, code-level setup in
  Cosmo.configure, and command-line flags now win over both
• Breaking: Sentry is the Cosmo::Middleware::Sentry middleware, which only traces: require "cosmo/middleware/sentry",
  add it to the server middleware, and register Cosmo::Middleware::Sentry::ERROR_HANDLER in config.error_handlers to
  capture exceptions; cosmo/sentry/auto and Cosmo::Sentry::JobProcessorMiddleware are removed
• Breaking: jobs are enqueued through Cosmo::Job::Enqueuer.enqueue; Publisher.publish_job, Publisher.publish_batch,
  and Job::Data#to_args are removed, and Job::Data#subject returns the subject as a string
• The web UI's Failed total counts every failed execution, retries included, instead of only jobs that gave up
• A job class's limit: is enforced by Cosmo::Middleware::Limit, the first entry of the server middleware chain,
  instead of the job processor
• Stream processor errors are logged at error level instead of debug

Fixed

• Stream consumers sharing one NATS stream no longer share one empty-fetch backoff, which made an idle consumer hold
  back the busy ones; backoff is kept per consumer
• The jobs page passes page and limit on to the processes table