[KalimeroMK]

Stacked on #16 — review that one first. The base moves to master once it lands.

#16 made the checkpoint record which tasks had already run, by id. That left one record holding every run emitted at a due time, so its size grew with the tasks sharing that time, and persist() rewrote the whole list after every single run.

Measured over one due minute, with every task due at it:

tasks	largest stored value	written over the minute
10	341 B → 42 B	2.1 kB → 292 B
100	2.9 kB → 42 B	148 kB → 2.5 kB
1000	29 kB → 42 B	14.3 MB → 25 kB

Each task now keeps the time of its own last emitted run under a key of its own, and the schedule keeps the time it was last evaluated to the end. Writing is one value per run instead of a growing list, and a stored value is the same size whatever the schedule, so no backend limit on the size of a value is in reach.

The position of the schedule moves only once a tick has drained. A process that died partway leaves it where it was, so the runs it never started stay due — which is what lets a task be seeded strictly after its own record, and drops the pass that had to tell a re-discovered run from a fresh one.

A checkpoint written by either earlier format is read for its time alone, so an upgrade resumes after it without replaying.

Summary by Sourcery

Keep checkpoint state per task so schedule persistence remains bounded and interrupted runs resume correctly.

Enhancements:

• Store each task's last emitted run independently while retaining a separate schedule evaluation position, keeping checkpoint values bounded and reducing persistence to one value per run.
• Resume interrupted and upgraded schedules without replaying completed runs, while preserving runs left due after a partial tick.
• Use bounded hashed cache keys for per-task checkpoint records and validate the new checkpoint payloads.

Tests:

• Add coverage for per-task checkpoint storage, interrupted first ticks, malformed task records, cache key limits, and checkpoint write behavior.

[sourcery-ai]

Reviewer's Guide

Replaces the growing, shared checkpoint payload with independently persisted last-run timestamps per task plus a schedule-level drained position, allowing crash-safe resumption, bounded value sizes, one write per run, and backward-compatible upgrades.

Sequence diagram for per-task checkpoint persistence

sequenceDiagram
    participant Generator as TaskGenerator
    participant Checkpoint
    participant Cache
    participant Task

    Generator->>Checkpoint: lastRun(taskId)
    Checkpoint->>Cache: getMultiple(scheduleKey, taskKeys)
    Cache-->>Checkpoint: schedule position and task timestamps
    Checkpoint-->>Generator: task last-run time
    Generator->>Task: getNextRunDate(seedTime)
    Generator-->>Task: yield TaskContext
    Generator->>Checkpoint: save(taskId, runTime)
    Checkpoint->>Cache: set(taskKey(taskId), runTime)
    Generator->>Checkpoint: markTick(now)
    Checkpoint->>Cache: set(cacheKey(), schedule position)

Loading

Entity relationship diagram for checkpoint records

erDiagram
    SCHEDULE_CHECKPOINT {
        string schedule_key PK
        timestamp last_evaluated
    }
    TASK_CHECKPOINT {
        string task_key PK
        string task_id
        timestamp last_run
    }
    SCHEDULE_CHECKPOINT ||--o{ TASK_CHECKPOINT : owns

Loading

File-Level Changes

Change	Details	Files
Split checkpoint persistence into one schedule position and one last-run record per task.	Store task run times under hashed, PSR-16-safe keys. Persist each task run independently and persist the schedule’s drained-through position separately. Load all current task records in bulk while retaining compatibility with earlier checkpoint formats.	src/Generator/Checkpoint.php
Resume and generate runs from each task’s individual checkpoint, advancing the schedule position only after a tick drains.	Seed each task strictly after its own last run, or from the schedule position for tasks without a record. Remove shared-time emitted-run filtering and always mark a successfully drained tick. Pass the schedule’s task identifiers into checkpoint loading.	src/Generator/TaskGenerator.php
Validate and document the task collection as an ordered list.	Update the task return type annotation to list semantics.	src/Schedule.php
Expand coverage for per-task records, migration behavior, malformed task records, and updated write counts.	Verify task-specific checkpoint keys and constant-size schedule state. Verify invalid per-task payloads are rejected. Update expectations for the separate task and schedule writes while preserving legacy checkpoint upgrade coverage.	tests/Generator/TaskGeneratorTest.php
Document the new checkpoint storage and write-efficiency behavior.	Add the per-task checkpoint change and its scaling characteristics to the unreleased changelog.	CHANGELOG.md

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it. You can also reply to a
  review comment with @sourcery-ai issue to create an issue from it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time. You can also comment
  @sourcery-ai title on the pull request to (re-)generate the title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time exactly where you
  want it. You can also comment @sourcery-ai summary on the pull request to
  (re-)generate the summary at any time.
• Generate reviewer's guide: Comment @sourcery-ai guide on the pull
  request to (re-)generate the reviewer's guide at any time.
• Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
  pull request to resolve all Sourcery comments. Useful if you've already
  addressed all the comments and don't want to see them anymore.
• Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
  request to dismiss all existing Sourcery reviews. Especially useful if you
  want to start fresh with a new review - don't forget to comment
  @sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

[sourcery-ai]

Hey - I've reviewed your changes and they look great!

Sourcery assessment

Needs a human reviewer. If a per-task checkpoint is wrong or fails to advance as intended, the scheduler can run that task again after a restart; any side effects the task already performed would remain after reverting this change. The checkpoint records can be cleared or rebuilt, but effects from a duplicate run may not be reversible.

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

[samdark]

Two regressions need correction before merge:

1. In TaskGenerator.php:149, a new stateful scheduler can lose unstarted runs after its first due tick is interrupted. Query nextRunDate() at 00:00, start the due tick at 00:01, and interrupt it after the first task completes. No drained schedule position exists yet, so a fresh scheduler seeds the remaining task from 00:01 and skips its pending run. The base resumes that task. Persist the initial seed before processing can be interrupted and cover recovery without a prior idle tick.

2. In Checkpoint.php:244, appending the task hash breaks previously valid names on portable cache backends. daily_subscription_invoices used a 51-character key but now requests 68-character task keys, which a 64-character-limited cache rejects during scheduler construction. PSR-16 guarantees support only through 64 characters. Derive a bounded task key from the schedule/task identity and add a 64-character-limited cache regression.