"""Dedicated Docker worker: database-backed queue, serialized claim, retry + restart recovery."""
import json
import os
import signal
import time
from datetime import datetime, timedelta, timezone
from . import db, enterprise, builder

running=True

def stop(*_):
    global running
    running=False

def claim():
    with db.connect() as c:
        c.execute('BEGIN IMMEDIATE')
        job=c.execute("""SELECT id FROM build_jobs
            WHERE state='queued' AND (retry_at IS NULL OR retry_at<=?)
            ORDER BY created_at,id LIMIT 1""",(db.now(),)).fetchone()
        if not job:return None
        c.execute("UPDATE build_jobs SET state='running',attempts=attempts+1,started_at=?,retry_at=NULL WHERE id=?",(db.now(),job['id']))
        return dict(c.execute('SELECT * FROM build_jobs WHERE id=?',(job['id'],)).fetchone())

def run_once():
    job=claim()
    if job is None:return None
    try:
        if job['kind']=='build': result=builder.build(job['site_id'],False)
        else:raise ValueError('Unsupported job')
        db.execute("UPDATE build_jobs SET state='done',result_json=?,finished_at=? WHERE id=?",
            (json.dumps(result,ensure_ascii=False),db.now(),job['id']))
    except Exception as exc:
        maximum=max(1,int(os.getenv('SITEFORGE_JOB_MAX_ATTEMPTS','3')))
        if job['attempts']<maximum:
            delay=min(300,5*(2**max(job['attempts']-1,0)))
            retry_at=(datetime.now(timezone.utc)+timedelta(seconds=delay)).isoformat(timespec='seconds')
            db.execute("""UPDATE build_jobs SET state='queued',error=?,started_at=NULL,
                       finished_at=NULL,retry_at=? WHERE id=?""",
                       (str(exc)[:1200],retry_at,job['id']))
        else:
            db.execute("UPDATE build_jobs SET state='failed',error=?,finished_at=?,retry_at=NULL WHERE id=?",
                       (str(exc)[:1200],db.now(),job['id']))
    return job['id']

def main():
    db.init();enterprise.setup()
    signal.signal(signal.SIGTERM,stop);signal.signal(signal.SIGINT,stop)
    # A crash before the build record is committed is safe to retry: build IDs
    # are generated inside builder.build and output directories are immutable.
    maximum=max(1,int(os.getenv('SITEFORGE_JOB_MAX_ATTEMPTS','3')))
    db.execute("""UPDATE build_jobs SET state='queued',error='Worker interrupted; retry scheduled',
               started_at=NULL,finished_at=NULL,retry_at=?
               WHERE state='running' AND attempts<?""",(db.now(),maximum))
    db.execute("""UPDATE build_jobs SET state='failed',error='Worker interrupted; retry limit reached',
               finished_at=?,retry_at=NULL WHERE state='running'""",(db.now(),))
    while running:
        if not run_once():time.sleep(float(os.getenv('SITEFORGE_JOB_POLL_SECONDS','1.5')))

if __name__=='__main__':main()