honker 0.4.0

Durable queues, streams, pub/sub, and cron scheduler on SQLite. One file, zero servers.

honker (Python)

Python binding for Honker: durable queues, streams, pub/sub, and time-trigger scheduling on SQLite.

Full docs live in the main repo and docs site:

• Main repo
• Docs

Install

pip install honker

The Python wheel includes Honker's SQLite loadable extension when the package is built through the release/proof workflow. To load Honker into your own sqlite3 or ORM connection:

import honker
import sqlite3

conn = sqlite3.connect("app.db")
honker.load_extension(conn)
conn.execute("SELECT honker_bootstrap()")

If a client needs the raw pieces instead, use extension_info():

path, entrypoint = honker.extension_info()

For local source builds, build and copy the extension before building the wheel:

git clone https://github.com/russellromney/honker
cd honker
cargo build --release -p honker-extension
scripts/copy-python-extension.sh

Quick start

import honker

db = honker.open("app.db")
q = db.queue("emails")

q.enqueue({"to": "[e-mail]"})

async for job in q.claim("worker-1"):
    send_email(job.payload)
    job.ack()

Delayed jobs use run_at:

import time

q.enqueue({"to": "[e-mail]"}, run_at=int(time.time()) + 10)

Recurring schedules are registered on a Scheduler. Build the schedule with crontab(expr) or every_s(n), then run the scheduler loop (it enqueues onto the target queue on each boundary; a normal worker claims the jobs):

from honker import Scheduler, crontab, every_s

sched = Scheduler(db)
sched.add(name="fast", queue="emails", schedule=every_s(1), payload={"kind": "tick"})
sched.add(name="nightly", queue="emails", schedule=crontab("0 3 * * *"))

await sched.run()  # acquires the leader lock and fires due tasks

add() is keyword-only and schedule must be a CronSchedule (from crontab or every_s), not a raw string. Supported schedule expressions:

• 5-field cron: crontab("0 3 * * *")
• 6-field cron: crontab("*/2 * * * * *")
• interval: every_s(1) (use every_s, not crontab("@every …"))

Notes

• claim() wakes on database updates and on due deadlines like run_at.
• schedule is the canonical recurring-schedule name.
• cron still works as a compatibility alias in older call sites.
• Construct db.queue(name) handles outside an open db.transaction(). First open for a name runs schema init in its own write transaction; doing that inside an outer transaction() raises RuntimeError (would otherwise deadlock the single writer). Create the handle first, then pass tx= to enqueue.

For streams, notify/listen, tasks, SQL extension usage, and full scheduler docs, see the main repo and docs site.