Tasks

Overview

MultiFlexi introduces the Task concept to answer a question that Jobs alone cannot: “was the periodic obligation actually fulfilled within its window?”

A Task is the unit of obligation for one scheduling window. A Job is a single attempt at fulfilling it. When a RunTemplate has an hourly interval, 24 Tasks are produced per day. Each Task is fulfilled by the first successful Job. If a Job fails, retry Jobs are spawned within the same window until one succeeds or the budget is exhausted.

Headline metric: 23/24 tasks fulfilled on time, 1 late instead of raw job counts.

Data Model

RunTemplate  1 ── N  Task  1 ── N  Job

• RunTemplate — unchanged role; gains retry and SLA configuration fields.

• Task — one scheduling window. Holds 1..N Jobs. Fulfilled when one succeeds.

• Job — one execution attempt. Gains a task_id foreign key.

Task States

State	Meaning
open	Window open, no successful job yet.
running	A job is currently executing.
fulfilled	A job succeeded on or before the deadline.
fulfilled_late	A job succeeded after the deadline but within the window (requires allow_late = true).
failed	The retry budget was exhausted or the window expired without a successful job.
missed	The window expired with zero job attempts (scheduler was down).

State Machine

open
├─► running ──► fulfilled          (success ≤ deadline)
│           ──► fulfilled_late     (deadline < success ≤ window_end, allow_late=true)
│           ──► open               (job failed, retry budget remains)
│           ──► failed             (budget exhausted or window expired)
└─► missed                         (window_end reached, zero attempts)

Cadence vs Deadline

A periodic obligation has two independent time anchors:

• Cadence (interval) — how often a new Task is born. Defines Task identity.

• Deadline (SLA) — by when the result must be ready to be useful. Defaults to the window end, can be set earlier.

Example — daily bank-statement download with an 08:00 deadline:

Scenario	Window	Deadline	Retry budget
Minute interval	1 min	= window end	~0 (no real retry)
Hourly interval	1 h	= window end	60 min → 4–8 tries
Daily, morning	24 h	e.g. 08:00	08:00 − window start

RunTemplate Retry/SLA Configuration

Field	Default	Description
deadline_offset	null	Offset from window start (+3h) or absolute time-of-day (08:00). null = window end.
max_attempts	1	Maximum number of job attempts per task window.
retry_backoff	none	Strategy: none | fixed | linear | exponential.
retry_min_gap	0	Minimum seconds between retry attempts.
allow_late	false	Whether a post-deadline success counts as fulfilled_late.

If retry_min_gap × max_attempts > (deadline − window_start), the scheduler caps max_attempts and emits a warning.

REST API

Endpoint	Description
GET /tasks.json	List tasks. Query params: state, runtemplate_id, from, to, limit.
GET /task/{id}.json	Single task with embedded jobs array (attempt history).

The existing GET /job/{id}.json response now includes a task_id field.

CLI Commands

task:list

List tasks:

multiflexi-cli task:list [options]

Options:

Option	Description
--runtemplate=ID	Filter by RunTemplate ID
--state=STATE	Filter by state (open, running, fulfilled, …)
--from=DATETIME	Window start from
--to=DATETIME	Window end to
--limit=N	Limit results (default 50)
--format=text|json	Output format

task:get

Get a single task with its job history:

multiflexi-cli task:get --id=42
multiflexi-cli task:get --id=42 --format=json

See Also

• Job Lifecycle — Job state transitions

• Execution Architecture — How the scheduler daemon works