scripts/workqueue.py

#!/usr/bin/env python3
"""Leased local-first OpenAlex work queue and reviewed-record publisher."""
from __future__ import annotations
import argparse, csv, datetime as dt, hashlib, json, os, random, re, shutil, tempfile, time, uuid
from pathlib import Path
from urllib.parse import urlsplit

ROOT = Path(__file__).resolve().parents[1]
STATE = 'planning/workqueue_state.json'
CSV = 'planning/openalex_workqueue.csv'
SUMMARY = 'planning/workqueue_summary.json'
FIELDS = ['record_id','work_id','collection','group','title','authors','year','work_type','pages','pdf_url','source_url','identifier','status','checked_at','sha256','verification_note','source_notes']
ACCEPT_TYPES = {'textbook','handbook','scholarly_book','edited_collection','book_chapter','doctoral_thesis','research_thesis','thesis_framing_text','habilitation_thesis','research_report'}


def utcnow(): return dt.datetime.now(dt.timezone.utc)
def stamp(t): return t.astimezone(dt.timezone.utc).isoformat(timespec='seconds').replace('+00:00','Z')
def parse_time(s):
    if not s: return None
    return dt.datetime.fromisoformat(s.replace('Z','+00:00')).astimezone(dt.timezone.utc)
def read_jsonl(p): return [json.loads(x) for x in Path(p).read_text().splitlines() if x.strip()]
def atomic(path, data):
    path=Path(path); path.parent.mkdir(parents=True,exist_ok=True)
    fd,tmp=tempfile.mkstemp(prefix='.'+path.name+'.',dir=path.parent)
    try:
        with os.fdopen(fd,'w',encoding='utf-8',newline='') as f: f.write(data); f.flush(); os.fsync(f.fileno())
        os.replace(tmp,path)
    finally:
        if os.path.exists(tmp): os.unlink(tmp)
def json_dump(x): return json.dumps(x,ensure_ascii=False,indent=2)+'\n'
def load_state(root):
    p=root/STATE
    if not p.exists(): return {'schema_version':1,'tasks':{}}
    s=json.loads(p.read_text())
    if s.get('schema_version')!=1 or not isinstance(s.get('tasks'),dict): raise ValueError('Invalid workqueue state')
    return s

def jobs(root): return {str(j['subfield_id']):j for j in read_jsonl(root/'planning/subfield_jobs.jsonl')}
def mapping(root):
    p=root/'planning/collection_subfield_map.json'
    if not p.exists(): return {}
    x=json.loads(p.read_text())
    return x.get('collections',{})
def refresh(root, state, now=None):
    atomic(root/STATE,json_dump(state))
    from workqueue_counts import render
    render(root,now=now)

def eligible_tasks(root, state, now):
    collmap=mapping(root); mixed_medical={str(sid) for name,ent in collmap.items() if name=='medical_sciences' for sid in ent.get('subfield_ids',[])}; rows=[]
    queue_counts={}
    cp=root/CSV
    if cp.exists():
        with cp.open(newline='') as f: queue_counts={str(r['subfield_id']):r for r in csv.DictReader(f)}
    for sid,j in jobs(root).items():
        # The mixed medical collection is intentionally outside this subfield-level queue.
        if str(sid) in mixed_medical or j.get('suggested_collection')=='medical_sciences': continue
        t=state['tasks'].get(str(sid),{})
        expiry=parse_time(t.get('lease_expires_at',''))
        if expiry and expiry>now: continue
        cool=parse_time(t.get('cooldown_until',''))
        if cool and cool>now: continue
        row=queue_counts.get(str(sid),{})
        if row.get('availability','available')!='available': continue
        count=int(row.get('valid_pdf_count',j.get('valid_pdf_count',0)) or 0)
        last=t.get('last_search_at','') or row.get('last_search_at','')
        rows.append((count,parse_time(last) or dt.datetime.min.replace(tzinfo=dt.timezone.utc),int(sid),sid,j))
    return sorted(rows)

def check_lease(state,sid,owner,lease_id,generation,now):
    t=state['tasks'].get(str(sid),{})
    if t.get('owner')!=owner or t.get('lease_id')!=lease_id or t.get('generation')!=generation or not parse_time(t.get('lease_expires_at','')) or parse_time(t['lease_expires_at'])<=now:
        raise ValueError('Lease is stale, expired, or owned by another worker')
    return t

def task_state(state,sid):
    t=state['tasks'].setdefault(str(sid),{})
    defaults={'generation':0,'owner':'','lease_id':'','lease_expires_at':'','last_search_at':'','completed_rounds':0,'last_new_entries':0,'last_new_pdfs':0,'cooldown_until':''}
    for k,v in defaults.items(): t.setdefault(k,v)
    return t

def do_claim(root, owner, now=None):
    now=now or utcnow(); s=load_state(root); choices=eligible_tasks(root,s,now)
    if not choices: return {'claimed':False,'reason':'no eligible task'}
    *_,sid,j=choices[0]; t=task_state(s,sid)
    gen=int(t.get('generation',0))+1; lid=str(uuid.uuid4())
    t.update(generation=gen,owner=owner,lease_id=lid,lease_expires_at=stamp(now+dt.timedelta(minutes=30)))
    refresh(root,s,now)
    return {'claimed':True,'subfield_id':sid,'generation':gen,'owner':owner,'lease_id':lid,'lease_expires_at':t['lease_expires_at'],'job':j}

def do_lease(root,action,sid,owner,lid,generation,now=None):
    now=now or utcnow(); s=load_state(root); t=check_lease(s,sid,owner,lid,int(generation),now)
    if action=='renew': t['lease_expires_at']=stamp(now+dt.timedelta(minutes=30))
    else: t.update(owner='',lease_id='',lease_expires_at='')
    refresh(root,s,now); return {'action':action,'subfield_id':str(sid),'lease_expires_at':t.get('lease_expires_at','')}

# Backend remains injectable to permit deterministic fake-CAS testing.
class HFBackend:
    def __init__(self, repo, scratch):
        from huggingface_hub import HfApi
        self.api=HfApi(); self.repo=repo; self.scratch=Path(scratch)
    def head(self): return self.api.repo_info(self.repo,repo_type='dataset').sha
    def server_now(self):
        import requests
        r=requests.get('https://huggingface.co/api/datasets/'+self.repo+'?workqueue_nonce='+str(uuid.uuid4()),timeout=20,headers={'Cache-Control':'no-cache','Pragma':'no-cache'}); r.raise_for_status()
        value=r.headers.get('Date')
        if not value: raise RuntimeError('Hugging Face HTTP Date unavailable')
        return dt.datetime.strptime(value,'%a, %d %b %Y %H:%M:%S %Z').replace(tzinfo=dt.timezone.utc)
    def snapshot(self,revision,dest,allow_patterns=None):
        from huggingface_hub import snapshot_download
        snapshot_download(self.repo,repo_type='dataset',revision=revision,local_dir=str(dest),cache_dir=str(self.scratch/'hf-cache'),token=None,allow_patterns=allow_patterns)
        shutil.rmtree(Path(dest)/'.cache',ignore_errors=True)
    def lease_snapshot(self,revision,dest):
        # Lease heartbeats need only counts/state inputs, not every collection export.
        self.snapshot(revision,dest,allow_patterns=[
            'collections.json','data/verified_links.jsonl',
            'planning/subfield_jobs.jsonl','taxonomy/openalex_subfields.csv',
            STATE,CSV,SUMMARY,'planning/collection_subfield_map.json',
            'planning/*search_progress.jsonl'])
    def commit(self, base, files, message):
        from huggingface_hub import CommitOperationAdd
        ops=[CommitOperationAdd(path_in_repo=k,path_or_fileobj=str(v)) for k,v in sorted(files.items())]
        return self.api.create_commit(repo_id=self.repo,repo_type='dataset',operations=ops,commit_message=message,parent_commit=base)

def sensitive(s): return bool(re.search(r'hf_[A-Za-z0-9]{20,}|/e/(?:home|fscratch)/|-----BEGIN .*PRIVATE KEY-----|[?&](?:token|access_token|api_key|x-amz-signature)=',str(s),re.I))
def has_secret(s): return bool(re.search(r'hf_[A-Za-z0-9]{20,}|-----BEGIN .*PRIVATE KEY-----|[?&](?:token|access_token|api_key|x-amz-signature)=',str(s),re.I))
def verify_record(record, proof, scratch):
    if set(record)!=set(FIELDS): raise ValueError('Record must contain exactly the 17 SCHEMA fields')
    if any(not isinstance(record[k],str) for k in FIELDS if k not in {'year','pages'}): raise ValueError('Public text fields must be strings')
    if record['status']!='downloaded_and_screened' or record['work_type'] not in ACCEPT_TYPES: raise ValueError('Only screened books, chapters, theses, and reports can be accepted')
    if not record['title'] or not record['pdf_url'] or not record['source_url'] or not record['verification_note'].strip(): raise ValueError('Missing title, public URLs, or editorial verification note')
    if record['year'] is not None and (type(record['year']) is not int or not 1000<=record['year']<=9999): raise ValueError('Invalid publication year')
    if type(record['pages']) is not int or record['pages']<1: raise ValueError('Invalid PDF page count')
    for k in ('pdf_url','source_url'):
        u=urlsplit(record[k]);
        if u.scheme not in {'http','https'} or not u.netloc or u.username or u.password or sensitive(record[k]): raise ValueError('Unsafe public URL')
    sha=record['sha256']
    if not re.fullmatch('[0-9a-f]{64}',sha): raise ValueError('Invalid PDF SHA256')
    if record.get('work_id')!='sha256:'+sha or not record.get('record_id'): raise ValueError('New record identity must be sha256:<complete PDF hash>')
    if any(sensitive(record[k]) for k in FIELDS if isinstance(record.get(k),str)): raise ValueError('Public record contains a private path, token, or secret pattern')
    p=proof
    if not isinstance(p,dict) or p.get('decision')!='accept' or p.get('complete') is not True or not (p.get('academic_context') is True or isinstance(p.get('academic_context'),str) and p['academic_context'].strip()): raise ValueError('Private proof must affirm academic context, completeness, and a positive decision')
    if p.get('public_access') is not True or p.get('source_verified') is not True or not isinstance(p.get('source_check_notes'),str) or not p['source_check_notes'].strip(): raise ValueError('Proof must record public access, source identity, and source-check notes')
    if p.get('pdf_url')!=record['pdf_url'] or p.get('source_url')!=record['source_url']: raise ValueError('Proof must bind the screened PDF and source URLs to the public record')
    if type(p.get('pages')) is not int or p['pages']!=record['pages']: raise ValueError('Proof page count mismatch')
    samples=p.get('reviewed_pages')
    if not isinstance(samples,list) or not samples or any(not isinstance(x,dict) for x in samples): raise ValueError('Reviewed first/middle/last page probes required')
    if any(type(x.get('page')) is not int or not 1<=x['page']<=record['pages'] for x in samples): raise ValueError('Reviewed page numbers must be valid one-based PDF pages')
    nums={x['page'] for x in samples}
    required={1,record['pages']//2+1,record['pages']}
    if not required.issubset(nums): raise ValueError('Proof must review the one-based first, middle, and last pages')
    if any(not isinstance(x.get('notes'),str) or not x['notes'].strip() for x in samples): raise ValueError('Each reviewed page needs actual screening notes')
    for k,v in p.items():
        if k not in {'local_pdf','sample_file','evidence_file'} and isinstance(v,str) and has_secret(v): raise ValueError('Private evidence contains a credential or secret pattern')
    actual=Path(p.get('local_pdf',''))
    sr=Path(scratch).resolve(); actual=actual.resolve(strict=True)
    if sr not in actual.parents or actual.suffix.lower()!='.pdf': raise ValueError('Proof PDF must be a local PDF under --scratch-dir')
    if actual.stat().st_size<8: raise ValueError('Invalid PDF bytes or unsafe path')
    with actual.open('rb') as f: header=f.read(1024)
    if b'%PDF-' not in header: raise ValueError('Invalid PDF bytes or unsafe path')
    h=hashlib.sha256()
    with actual.open('rb') as f:
        for b in iter(lambda:f.read(1024*1024),b''): h.update(b)
    if h.hexdigest()!=sha or p.get('sha256')!=sha: raise ValueError('PDF hash mismatch')
    from pypdf import PdfReader
    reader=PdfReader(str(actual),strict=False)
    if reader.is_encrypted and reader.decrypt('')==0: raise ValueError('Encrypted PDF cannot be screened')
    if len(reader.pages)!=record['pages']: raise ValueError('Parsed PDF page count mismatch')

def safe_rel(path):
    p=Path(path)
    if p.is_absolute() or '..' in p.parts or not p.parts or sensitive(str(path)): raise ValueError('Unsafe repository path')
    return p.as_posix()

def cas_conflict(exc):
    status=getattr(exc,'status_code',None) or getattr(getattr(exc,'response',None),'status_code',None)
    return status in (409,412)

def collection_def(root,j):
    coll=j.get('suggested_collection')
    if not coll: raise ValueError('Job has no suggested collection for this subfield')
    defs=json.loads((root/'collections.json').read_text()); existing=next((x for x in defs if x['collection']==coll),None)
    if existing: return existing
    group=j.get('suggested_group'); path=safe_rel(j.get('suggested_public_path',''))
    # The job supplies the prepared public definition; the mapped mixed medical folder is excluded above.
    if not all((coll,group,path)): raise ValueError('Mapped job lacks suggested group or path')
    return {'collection':coll,'group':group,'path':safe_rel(path),'display_name':j.get('subfield_name'),'verified_entries':0,'legacy_candidates':0,'openalex_subfield_id':str(j['subfield_id']),'openalex_field_id':str(j['field_id'])}

def publish(root, backend, sid, owner, lid, generation, records, scratch, now_fn=None, attempts=5):
    scratch=Path(scratch).resolve()
    for r,p in records: verify_record(r,p,scratch)
    for attempt in range(attempts):
        base=backend.head(); now=backend.server_now()
        with tempfile.TemporaryDirectory(prefix='workqueue-snapshot-',dir=scratch) as td:
            snap=Path(td)/'repo'; backend.snapshot(base,snap)
            before={p.relative_to(snap).as_posix():hashlib.sha256(p.read_bytes()).hexdigest() for p in snap.rglob('*') if p.is_file()}
            s=load_state(snap); t=check_lease(s,sid,owner,lid,generation,now)
            lease_guard=dict(t)
            # Recheck against latest snapshot canonical rows and make all output changes in this isolated tree.
            from catalog import read_rows, render as catalog_render, validate
            verified=read_rows(snap/'data/verified_links.jsonl'); legacy=read_rows(snap/'data/legacy_candidates.jsonl')
            from workqueue_counts import build as build_counts
            prior_rows,_=build_counts(snap,now=now)
            prior_sid_count=next((int(x['valid_pdf_count']) for x in prior_rows if str(x['subfield_id'])==str(sid)),0)
            j=jobs(snap)[str(sid)]; cdef=collection_def(snap,j)
            defs=json.loads((snap/'collections.json').read_text())
            if not any(c['collection']==cdef['collection'] for c in defs): defs.append(cdef)
            keys={(x['collection'],x['sha256']) for x in verified}
            added=0
            for rec,_ in records:
                key=(rec['collection'],rec['sha256'])
                if rec['collection']!=cdef['collection']: raise ValueError('Record collection does not match claimed subfield')
                if key in keys: continue
                verified.append(rec); keys.add(key); added+=1
            atomic(snap/'data/verified_links.jsonl',''.join(json.dumps(x,ensure_ascii=False)+'\n' for x in verified))
            atomic(snap/'collections.json',json_dump(defs))
            # Round closes even with zero accepts. Coverage counters reflect actual accepted deltas.
            t['last_search_at']=stamp(now); t['completed_rounds']=int(t.get('completed_rounds',0))+1; t['last_new_entries']=added
            t['cooldown_until']=stamp(now+dt.timedelta(hours=24)); t.update(owner='',lease_id='',lease_expires_at='')
            atomic(snap/STATE,json_dump(s))
            after_rows,_=build_counts(snap,now=now)
            after_sid_count=next((int(x['valid_pdf_count']) for x in after_rows if str(x['subfield_id'])==str(sid)),prior_sid_count)
            t['last_new_pdfs']=max(0,after_sid_count-prior_sid_count)
            atomic(snap/STATE,json_dump(s))
            catalog_render(snap)
            from workqueue_counts import render as counts_render
            counts_render(snap,now=now)
            if hasattr(__import__('catalog'),'render_readme_counts'):
                __import__('catalog').render_readme_counts(snap)
            if not validate(snap): raise ValueError('Catalogue validation failed')
            changed={}
            allowed={'collections.json','README.md','data/verified_links.jsonl','data/verified_links.csv','data/release_summary.json',STATE,CSV,SUMMARY,cdef['group']+'/README.md'}
            folder=cdef['path']
            allowed.update({folder+'/'+name for name in ('README.md','links.csv','links.jsonl','legacy_candidates.csv','legacy_candidates.jsonl','legacy_candidates.md')})
            for f in snap.rglob('*'):
                if not f.is_file() or '.git' in f.parts: continue
                rel=f.relative_to(snap).as_posix()
                if rel not in allowed: continue
                digest=hashlib.sha256(f.read_bytes()).hexdigest()
                if before.get(rel)!=digest: changed[rel]=f
            # Restrict staged data to exact allowed outputs and commit from this HEAD.
            # Lease ownership and expiry are checked again against live server time immediately before CAS.
            if parse_time(lease_guard.get('lease_expires_at',''))<=backend.server_now(): raise ValueError('Lease expired before publish commit')
            try:
                commit_result=backend.commit(base,changed,'Publish reviewed OpenAlex workqueue records')
                return {'published':True,'new_entries':added,'round_closed':True,'base_commit':base,'commit':getattr(commit_result,'oid',None) or getattr(commit_result,'commit_id',None)}
            except Exception as exc:
                if not cas_conflict(exc): raise
                if attempt+1==attempts: raise
        # fresh-head retry on CAS races only
        time.sleep(min(2.0,0.15*(2**attempt))+random.random()*0.2)
    raise RuntimeError('CAS retries exhausted')

def main(argv=None):
    ap=argparse.ArgumentParser(); ap.add_argument('action',choices=['status','claim','renew','release','publish']); ap.add_argument('--root',type=Path,default=ROOT); ap.add_argument('--repo',default='laion/academic-textbook-links'); ap.add_argument('--owner','--worker-id',dest='owner'); ap.add_argument('--subfield-id'); ap.add_argument('--lease-id'); ap.add_argument('--generation',type=int); ap.add_argument('--proofs',type=Path); ap.add_argument('--scratch-dir',type=Path,required=True); ap.add_argument('--apply',action='store_true'); args=ap.parse_args(argv)
    root=args.root.resolve()
    scratch=args.scratch_dir.resolve()
    if not scratch.is_dir() or scratch==root or root in scratch.parents: ap.error('--scratch-dir must be an existing private directory outside the checkout')
    if args.action=='status': print(json.dumps(json.loads((root/SUMMARY).read_text()) if (root/SUMMARY).exists() else {'state':load_state(root)},ensure_ascii=False,indent=2)); return
    owner=args.owner or str(uuid.uuid4())
    try:
        if str(uuid.UUID(owner))!=owner.lower(): raise ValueError()
    except (ValueError,AttributeError): ap.error('--owner must be an opaque UUID (or omit it to generate one)')
    if args.action=='claim':
        if not args.apply:
            print(json.dumps({'dry_run':True,'candidate':eligible_tasks(root,load_state(root),utcnow())[0][3] if eligible_tasks(root,load_state(root),utcnow()) else None})); return
        result=remote_lease(args.repo,root,'claim',{'owner':owner},scratch=scratch)
        if result['claimed']: result['owner']=owner
        print(json.dumps(result,ensure_ascii=False)); return
    if not all((args.subfield_id,args.owner,args.lease_id,args.generation is not None)): ap.error('lease actions require --subfield-id --owner --lease-id --generation')
    if args.action in {'renew','release'}:
        if not args.apply: print(json.dumps({'dry_run':True,'action':args.action,'subfield_id':args.subfield_id})); return
        print(json.dumps(remote_lease(args.repo,root,args.action,vars(args),scratch=scratch))); return
    if not args.scratch_dir or not args.proofs: ap.error('publish requires --scratch-dir and --proofs')
    proof_path=args.proofs.resolve()
    if scratch not in proof_path.parents: ap.error('--proofs must be stored under --scratch-dir')
    items=json.loads(proof_path.read_text()); records=[(x['record'],x['proof']) for x in items]
    for record,proof in records: verify_record(record,proof,scratch)
    if not args.apply:
        print(json.dumps({'dry_run':True,'subfield_id':args.subfield_id,'reviewed_records':len(records),'unique_pdf_hashes':len({r['sha256'] for r,_ in records})})); return
    backend=HFBackend(args.repo,scratch); print(json.dumps(publish(root,backend,args.subfield_id,args.owner,args.lease_id,args.generation,records,scratch),ensure_ascii=False))

def remote_lease(repo,root,action,args,attempts=5,backend=None,scratch=None):
    if scratch is None: raise ValueError('Scratch directory is required for every network operation')
    scratch=Path(scratch).resolve(); backend=backend or HFBackend(repo,scratch)
    for attempt in range(attempts):
        base=backend.head(); now=backend.server_now()
        with tempfile.TemporaryDirectory(prefix='workqueue-lease-',dir=scratch) as td:
            snap=Path(td)/'repo'
            if hasattr(backend,'lease_snapshot'): backend.lease_snapshot(base,snap)
            else: backend.snapshot(base,snap)
            before={p.relative_to(snap).as_posix():hashlib.sha256(p.read_bytes()).hexdigest() for p in snap.rglob('*') if p.is_file()}
            state=load_state(snap)
            if action=='claim':
                choices=eligible_tasks(snap,state,now)
                if not choices:return {'claimed':False,'reason':'no eligible task'}
                *_,sid,job=choices[0]; task=task_state(state,sid)
                generation=int(task.get('generation',0))+1; lease_id=str(uuid.uuid4())
                task.update(generation=generation,owner=args['owner'],lease_id=lease_id,lease_expires_at=stamp(now+dt.timedelta(minutes=30)))
                result={'claimed':True,'subfield_id':sid,'generation':generation,'owner':args['owner'],'lease_id':lease_id,'lease_expires_at':task['lease_expires_at'],'job':job}
            else:
                task=check_lease(state,args['subfield_id'],args['owner'],args['lease_id'],args['generation'],now)
                lease_guard=dict(task)
                if action=='renew':task['lease_expires_at']=stamp(now+dt.timedelta(minutes=30))
                else:task.update(owner='',lease_id='',lease_expires_at='')
                result={'action':action,'subfield_id':str(args['subfield_id']),'lease_expires_at':task.get('lease_expires_at','')}
            atomic(snap/STATE,json_dump(state))
            from workqueue_counts import render
            render(snap,now=now)
            changed={p.relative_to(snap).as_posix():p for p in [snap/STATE,snap/CSV,snap/SUMMARY] if p.exists() and before.get(p.relative_to(snap).as_posix())!=hashlib.sha256(p.read_bytes()).hexdigest()}
            try:
                if action!='claim' and parse_time(lease_guard.get('lease_expires_at',''))<=backend.server_now(): raise ValueError('Lease expired before state commit')
                backend.commit(base,changed,'Update OpenAlex workqueue lease')
                return result
            except Exception as exc:
                if not cas_conflict(exc): raise
                if attempt+1==attempts:raise
        time.sleep(min(2,.15*2**attempt)+random.random()*.2)
    raise RuntimeError('CAS retries exhausted')
if __name__=='__main__': main()