QueueKeeperCheck.Config (queue_keeper_check v0.2.1)

Thresholds and safety limits. All check durations are in seconds. Pass overrides as keyword lists:

QueueKeeperCheck.run(MyApp.Repo, backlog: [max_depth: 5_000], hanging: [default_budget: 120])

Check defaults:

check	option	default	meaning
`:orphaned`	`after`	3600	executing longer than this, far beyond any expected runtime (matches Oban Lifeline's default `rescue_after` of 60 min). Measures execution age only: the job may be orphaned by a node crash, or may still be running
`:hanging`	`default_budget`	300	executing longer than the worker's budget
`:hanging`	`budgets`	`%{}`	per-worker budgets, e.g. `%{"MyApp.VideoWorker" => 1800}`
`:backlog`	`max_depth`	1000	due jobs (available, or scheduled/retryable whose time has passed) per queue
`:backlog`	`max_latency`	300	age of the oldest due job per queue
`:stalled`	`window`	60	due jobs older than `backlog.max_latency` and no job attempted in this queue within the window
`:stalled`	`lookback`	3600	completed jobs count as recent activity only if they were scheduled at most this many seconds before the queue's oldest due job (bounds the query; see `QueueKeeperCheck.Checks.stalled/4`)
`:retry_storm`	`window`	300	look-back window
`:retry_storm`	`min_retries`	20	retrying/retried jobs per worker (see below)
`:retry_storm`	`include_completed`	`false`	also count completed jobs on attempt >= 2 attempted in the window. Costs a full scan of `oban_jobs` on large tables (no index on `attempted_at`), so it is off by default
`:discarded`	`window` / `max`	3600 / 5	jobs discarded in the window, per worker
`:cancelled`	`window` / `max`	3600 / 10	jobs cancelled in the window, per worker
`:cron_missing`	`expected`	`%{}` (off)	worker => max seconds between insertions, e.g. `%{"MyApp.Nightly" => 90_000}`

:retry_storm counts, per worker, within the window:

retryable jobs whose last (failed) attempt started in the window (attempted_at; the query also bounds scheduled_at, which Oban sets after the attempt, so it can use the composite index),
executing jobs on attempt >= 2 attempted in the window (attempted_at),
discarded / cancelled jobs on attempt >= 2 discarded / cancelled in the window (discarded_at / cancelled_at),
with include_completed: true, completed jobs on attempt >= 2 attempted in the window (attempted_at).

A job that failed before the window and is still waiting out a long backoff is not counted. (Up to v0.2.0 every retryable job counted, whatever the window.)

A finding is :critical when the value is at least twice its threshold; :orphaned, :stalled and :cron_missing findings are always :critical.

Safety options (top level, not per check):

option	default	meaning
`:statement_timeout`	5000	ms. Per statement server-side limit (Postgres `statement_timeout`, MySQL `max_execution_time`); also the basis of the client-side timeout on every engine
`:lock_timeout`	1000	ms. Postgres `lock_timeout`: give up rather than queue behind a lock (e.g. a migration)
`:check_timeout`	30000	ms. Client-side budget for one whole check (Postgres transaction / MySQL checkout holding one connection); the connection is dropped if exceeded
`:read_only`	`true`	Postgres: run each check in a transaction marked `transaction_read_only`, so the server rejects any write
`:prefix`	`"public"` (Postgres), none elsewhere	schema of `oban_jobs` (a database name on MySQL). Must be a plain identifier, `[A-Za-z_][A-Za-z0-9_]{0,62}`, or `false` for no prefix (use the connection's `search_path`)

Invalid or unknown options raise ArgumentError before any query runs.

Summary

Functions

build(opts)

Merge user options over the defaults, validating every value. Raises ArgumentError.

defaults()

guard(opts)

Safety options (:statement_timeout, :lock_timeout, :read_only) merged over defaults.

guard_defaults()

valid_prefix?(prefix)

Validate an Oban prefix. Ecto already quotes it as an identifier (and rejects quote characters); this is a stricter allow-list so a prefix can only ever be a plain name.