The migration jobs: AshOban's triggers on MishkaCmsMigration.MigrationRun generate the two workers, MishkaCmsMigration.Workers.DryRunWorker (:run_dry_run) and MishkaCmsMigration.Workers.ImportWorker (:run_import), each three attempts of an hour, run as the admin who started the run (MishkaCmsMigration.Workers.Helpers.Actor). What a job does is MishkaCmsMigration.Workers.Helpers.Pipeline. One run of a site is dry running or importing at a time, held by the run's :one_active_per_site identity. Oban stops an attempt that runs longer than its hour and tries the job again; an import's next attempt skips what the last one wrote and recorded, so a large site goes on where it stopped.

This module attaches the handler that fails the run of a job Oban discards (attach/0), the one ending AshOban's on_error cannot reach.

Telemetry

Each attempt of a job, and the New import form's connection check, tell :telemetry what they do, so a host can attach its own handlers for metrics or logs. Every event's name starts with :mishka_cms_migration. A reason is in words (MishkaCmsMigration.Report.reason/1), and no event carries a connection or its secret. Durations are in :native time units.

• [:mishka_cms_migration, :run, :start] — an attempt of a dry run or an import begins (:telemetry.span/3). Measurements: system_time, monotonic_time. Metadata: run_id, site_id, source (the source's id), source_host and mode (:dry_run or :import).
• [:mishka_cms_migration, :run, :stop] — the attempt ended. Measurements: duration, monotonic_time and, for a finished one, its report's created, updated, skipped and failed counts. Metadata: the start's, and result: :finished; :cancelled when the run no longer waited for its job; or :failed, with its reason, which AshOban tries again and, after the last attempt, fails the run on.
• [:mishka_cms_migration, :run, :exception] — the attempt raised past the walk's own rescue. Measurements: duration, monotonic_time. Metadata: the start's, and kind, reason and stacktrace.
• [:mishka_cms_migration, :kind, :listed] — the source answered the last page of a kind. Measurements: count, the rows it listed, and pages. Metadata: the run's, and kind.
• [:mishka_cms_migration, :item, :settled] — an item was written and recorded, recorded as skipped or failed, or, in a dry run, reported. Measurements: duration. Metadata: the run's, and the item's kind, source_id, outcome (:created, :updated, :skipped or :failed) and reason.
• [:mishka_cms_migration, :connect, :refused] — a source refused to connect, at the form's check or in a job (MishkaCmsMigration.Source.connect/4). Measurements: system_time. Metadata: source, source_host, site_id and reason, and from a job the run's run_id and mode.

A handler runs in the process that emits the event, the job's or the form's, so it should be quick, and one that raises is detached by :telemetry:

:telemetry.attach(
  "imports-to-metrics",
  [:mishka_cms_migration, :run, :stop],
  &MyApp.Metrics.handle_migration/4,
  nil
)