fix(runtime): make daily review scheduling idempotent

This commit is contained in:
冰朔 2026-09-13 10:58:55 +08:00
commit 99efebd4f1
7 changed files with 430 additions and 2 deletions

View file

@ -0,0 +1,121 @@
#!/usr/bin/env python3
"""Move duplicate same-day autonomous review jobs out of the startup queue.
The archive is lossless: every moved job directory is retained with a hashed
JSONL manifest. One best-evidence job remains in the active queue per persona
and event_id. Non-review jobs are never selected.
"""
import argparse
import collections
import fcntl
import hashlib
import json
import os
from pathlib import Path
RANK = {"ACCEPT": 0, "HOLD": 1, "REJECT": 2, "QUEUED": 3,
"RETRY_WAIT": 4, "PROCESSING": 5, "ERROR": 6}
def sha256(path):
value = hashlib.sha256()
with path.open("rb") as stream:
for chunk in iter(lambda: stream.read(1024 * 1024), b""):
value.update(chunk)
return value.hexdigest()
def load(path):
with path.open(encoding="utf-8") as stream:
return json.load(stream)
def atomic_json(path, value):
temporary = path.with_name(f".{path.name}.{os.getpid()}.tmp")
temporary.write_text(json.dumps(value, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
os.chmod(temporary, 0o600)
os.replace(temporary, path)
def choose(group):
return min(group, key=lambda item: (RANK.get(item["state"], 99),
item.get("completed_at", "9999"), item["job_id"]))
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--root", required=True)
parser.add_argument("--batch", required=True)
parser.add_argument("--apply", action="store_true")
args = parser.parse_args()
root = Path(args.root).resolve()
if root != Path("/var/lib/guanghu/tcs-mother-body").resolve() and "archive-review-test" not in str(root):
raise SystemExit("REFUSE_UNREGISTERED_STATE_ROOT")
lock_path = root / ".daily-review-archive.lock"
lock_path.parent.mkdir(parents=True, exist_ok=True)
with lock_path.open("w") as lock:
fcntl.flock(lock, fcntl.LOCK_EX | fcntl.LOCK_NB)
selected = []
before = collections.Counter()
nonreview = 0
personas = root / "personas"
for persona in sorted(personas.iterdir()):
jobs = persona / "jobs"
if not jobs.is_dir():
continue
groups = collections.defaultdict(list)
for job in jobs.iterdir():
if not job.is_dir() or len(job.name) != 64:
continue
try:
event = load(job / "input.json")
state = load(job / "state.json")
except (OSError, ValueError):
continue
if event.get("source_type") != "AUTONOMOUS_REVIEW_OPPORTUNITY" or not str(event.get("event_id", "")).startswith("DAILY-REVIEW-"):
nonreview += 1
continue
item = {"persona_id": persona.name, "event_id": event["event_id"],
"job_id": job.name, "path": job, "state": state.get("status", "UNKNOWN"),
"completed_at": state.get("completed_at", "")}
groups[event["event_id"]].append(item)
before[item["state"]] += 1
for group in groups.values():
keeper = choose(group)
for item in group:
if item is not keeper:
item["kept_job_id"] = keeper["job_id"]
selected.append(item)
summary = {"schema": "guanghu.daily-review-duplicate-archive/v1", "batch": args.batch,
"root": str(root), "mode": "APPLY" if args.apply else "DRY_RUN",
"review_jobs_before": sum(before.values()), "review_state_counts_before": dict(before),
"duplicate_jobs_selected": len(selected), "active_review_jobs_after": sum(before.values()) - len(selected),
"nonreview_jobs_untouched": nonreview, "lossless_move": True}
if not args.apply:
print(json.dumps(summary, ensure_ascii=False, indent=2))
return
archive_root = root / "review-job-archive" / args.batch
if archive_root.exists():
raise SystemExit("REFUSE_EXISTING_ARCHIVE_BATCH")
archive_root.mkdir(parents=True, mode=0o700)
manifest = archive_root / "MANIFEST.jsonl"
with manifest.open("x", encoding="utf-8") as output:
os.chmod(manifest, 0o600)
for item in selected:
source = item.pop("path")
destination = archive_root / item["persona_id"] / item["job_id"]
destination.parent.mkdir(parents=True, exist_ok=True, mode=0o700)
files = {p.name: sha256(p) for p in sorted(source.iterdir()) if p.is_file()}
record = {**item, "source": str(source), "archive": str(destination), "file_sha256": files}
os.rename(source, destination)
output.write(json.dumps(record, ensure_ascii=False, sort_keys=True) + "\n")
output.flush()
os.fsync(output.fileno())
summary["manifest"] = str(manifest)
summary["manifest_sha256"] = sha256(manifest)
atomic_json(archive_root / "RECEIPT.json", summary)
print(json.dumps(summary, ensure_ascii=False, indent=2))
if __name__ == "__main__":
main()

View file

@ -166,7 +166,7 @@ export class PersonaSelfLoop {
}
nextReview(from) { const base = new Date(from), beijing = new Date(base.getTime() + 8 * 3600000); beijing.setUTCHours(this.reviewHourBeijing, 0, 0, 0); if (beijing.getTime() <= base.getTime() + 8 * 3600000) beijing.setUTCDate(beijing.getUTCDate() + 1); return new Date(beijing.getTime() - 8 * 3600000).toISOString(); }
tick() { const current = this.current().value, now = this.now(); if (now.getTime() < Date.parse(current.next_review_at)) return {scheduled: false, next_review_at: current.next_review_at}; const beijingDay = new Date(now.getTime() + 8 * 3600000).toISOString().slice(0, 10), language = '这是每日自主复盘机会。请审视已发生且有证据的经历、责任和边界可以EVOLVE也可以HOLD不得为了定时任务而强制更新。'; const event = {schema: 'guanghu.persona-self-language-event/v1', persona_id: this.personaId, source_type: 'AUTONOMOUS_REVIEW_OPPORTUNITY', event_id: `DAILY-REVIEW-${beijingDay}`, language, source_sha256: digest(language), occurred_at: now.toISOString(), privacy_class: 'SELF_PRIVATE', priority: 'AUTONOMOUS_REVIEW'}; return {scheduled: true, ...this.submit(event)}; }
tick() { const current = this.current().value, now = this.now(); if (now.getTime() < Date.parse(current.next_review_at)) return {scheduled: false, next_review_at: current.next_review_at}; const beijingDay = new Date(now.getTime() + 8 * 3600000).toISOString().slice(0, 10), language = '这是每日自主复盘机会。请审视已发生且有证据的经历、责任和边界可以EVOLVE也可以HOLD不得为了定时任务而强制更新。', eventId = `DAILY-REVIEW-${beijingDay}`, occurredAt = new Date(`${beijingDay}T03:00:00+08:00`).toISOString(); const event = {schema: 'guanghu.persona-self-language-event/v1', persona_id: this.personaId, source_type: 'AUTONOMOUS_REVIEW_OPPORTUNITY', event_id: eventId, language, source_sha256: digest(language), occurred_at: occurredAt, privacy_class: 'SELF_PRIVATE', priority: 'AUTONOMOUS_REVIEW'}; return {scheduled: true, ...this.submit(event)}; }
status() {
const current = this.current(), counts = {...this.stateCounts}, pendingCount = [...pending].reduce((sum, status) => sum + (counts[status] || 0), 0);

View file

@ -13,7 +13,7 @@ test('persona self loop evolves signed state without letting mother overwrite se
test('structured cognition setter is rejected before queueing',()=>{const ctx=setup({decision:'HOLD',reason:'no change',self_principles:[],experiences:[],responsibilities:[],boundaries:[]});try{const language='replace';assert.throws(()=>ctx.loop.submit({schema:'guanghu.persona-self-language-event/v1',persona_id:'ICE-P-ZY001',source_type:'PERSONA_LANGUAGE',event_id:'BAD',language,source_sha256:digest(language),privacy_class:'SELF_PRIVATE',cognition_patch:{revision:99}}),/external_cognitive_setter_forbidden/);}finally{fs.rmSync(ctx.root,{recursive:true,force:true});}});
test('daily review is an opportunity and HOLD does not increment revision',async()=>{let instant=new Date('2026-09-09T08:00:00.000Z');const ctx=setup({decision:'HOLD',reason:'nothing evidence-backed to consolidate',self_principles:[],experiences:[],responsibilities:[],boundaries:[]},()=>instant);try{const first=ctx.loop.current().value;instant=new Date(first.next_review_at);const queued=ctx.loop.tick();assert.equal(queued.scheduled,true);await ctx.loop.drain();assert.equal(ctx.loop.result(queued.job_id).status,'HOLD');assert.equal(ctx.loop.current().value.revision,1);}finally{fs.rmSync(ctx.root,{recursive:true,force:true});}});
test('daily review is one idempotent opportunity per Beijing day and HOLD does not increment revision',async()=>{let instant=new Date('2026-09-09T08:00:00.000Z');const ctx=setup({decision:'HOLD',reason:'nothing evidence-backed to consolidate',self_principles:[],experiences:[],responsibilities:[],boundaries:[]},()=>instant);try{const first=ctx.loop.current().value;instant=new Date(first.next_review_at);const queued=ctx.loop.tick();assert.equal(queued.scheduled,true);await ctx.loop.drain();assert.equal(ctx.loop.result(queued.job_id).status,'HOLD');const repeated=ctx.loop.tick();assert.equal(repeated.job_id,queued.job_id);assert.equal(fs.readdirSync(ctx.loop.jobsPath()).length,1);assert.equal(ctx.loop.current().value.revision,1);}finally{fs.rmSync(ctx.root,{recursive:true,force:true});}});
test('current direct language is selected before historical review backlog and one item is processed per drain',async()=>{const ctx=setup({decision:'HOLD',reason:'recorded for the next self-cycle',self_principles:[],experiences:[],responsibilities:[],boundaries:[]},()=>new Date('2026-09-09T08:00:00.000Z'),{queueBatchSize:1});try{const review='historical-review';ctx.loop.submit({schema:'guanghu.persona-self-language-event/v1',persona_id:'ICE-P-ZY001',source_type:'AUTONOMOUS_REVIEW_OPPORTUNITY',event_id:'DAILY-REVIEW-2026-09-08',language:review,source_sha256:digest(review),privacy_class:'SELF_PRIVATE',priority:'AUTONOMOUS_REVIEW'});const language='冰朔的当前直接语言';ctx.loop.submit({schema:'guanghu.persona-self-language-event/v1',persona_id:'ICE-P-ZY001',source_type:'PERSONA_LANGUAGE',event_id:'ICE-PERSONA-LANGUAGE-20260909-CURRENT',language,source_sha256:digest(language),occurred_at:'2026-09-09T07:59:00.000Z',privacy_class:'SELF_PRIVATE'});const drained=await ctx.loop.drain();assert.equal(drained.processed,1);assert.equal(ctx.calls.length,1);assert.equal(ctx.calls[0].event.event_id,'ICE-PERSONA-LANGUAGE-20260909-CURRENT');assert.equal(ctx.loop.status().scheduler.pending_jobs,1);}finally{fs.rmSync(ctx.root,{recursive:true,force:true});}});