titan/atlas-iac
3
0
Fork 0
Code Issues Pull Requests 9 Actions Packages Projects Releases Wiki Activity

hermes: recover confirmation for the published Soteria task

Browse Source
This commit is contained in:
jenkins 2026-09-13 20:41:00 -05:00
parent 73618ca8f8
commit e07cbdfce8
2 changed files with 164 additions and 31 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

106
services/hermes/scripts/release_soteria_publication_retry.py
View File

@ -9,6 +9,7 @@ import sqlite3
from pathlib import Path
from typing import Any
from gitea_api_policy import _draft_title
import scm_broker_client
import supervisor_state
@ -27,6 +28,8 @@ SOURCE_RAW_SHA = "13de550c15ecfdf6684022b07a823da1c51586f959e9d9d2b3aea67433a908
FAILED_RESULT_SHA = "0395eaed346eab364990de9b6d6c9a791c3d015a6b026f2473a9c80e8a1d2548"
FAILED_EVENT_ID = 209
RETRY_AFTER = 1789346240
CONFIRM_RUN = "11"
CONFIRM_EVENT_ID = 213
def _value(task: Any, name: str) -> Any:
@ -34,30 +37,30 @@ def _value(task: Any, name: str) -> Any:
return task.get(name) if isinstance(task, dict) else getattr(task, name, None)
def _pool_guard(database: Path) -> None:
"""Require exactly run 10's preserved, unpublished mediator result."""
def _pool_guard(database: Path, run_id: str) -> None:
"""Require one exact terminal mediator result and no competing assignment."""
connection = sqlite3.connect(f"file:{database}?mode=ro", uri=True)
try:
row = connection.execute(
"SELECT payload_json,result_json,result_digest,state,worker_ordinal,attempt "
"FROM assignments WHERE board=? AND task_id=? AND run_id=?",
(BOARD, CHILD, FAILED_RUN),
(BOARD, CHILD, run_id),
).fetchone()
active = connection.execute(
"SELECT 1 FROM assignments WHERE board=? AND task_id=? AND run_id<>? "
"AND state IN ('assigned','running','result')",
(BOARD, CHILD, FAILED_RUN),
(BOARD, CHILD, run_id),
).fetchone()
finally:
connection.close()
if row is None or active is not None or tuple(row[3:]) != ("finalized", ORDINAL, 1):
raise ValueError("run 10 is not the exact terminal publication attempt")
raise ValueError("publication attempt is not the exact terminal run")
if not isinstance(row[0], str) or not isinstance(row[1], str):
raise ValueError("run 10 evidence is malformed")
raise ValueError("publication attempt evidence is malformed")
try:
payload, result = json.loads(row[0]), json.loads(row[1])
except json.JSONDecodeError as error:
raise ValueError("run 10 evidence is malformed") from error
raise ValueError("publication attempt evidence is malformed") from error
resume = payload.get("scm_resume") if isinstance(payload, dict) else None
structured = result.get("structured") if isinstance(result, dict) else None
source = resume.get("source") if isinstance(resume, dict) else None
@ -74,11 +77,11 @@ def _pool_guard(database: Path) -> None:
or result.get("capacity_failure") is not False
or result.get("scm_submission") is not None
):
raise ValueError("run 10 is not the retained rejected publication")
raise ValueError("publication attempt is not the retained rejected publication")
def _native_guard() -> None:
"""Require event 209 to remain the private child’s current blocked outcome."""
def _native_guard(run_id: str, event_id: int, status: str, event_kind: str) -> None:
"""Require one exact native terminal event and private root/child chain."""
from hermes_cli import kanban_db
with kanban_db.scoped_current_board(BOARD):
@ -98,19 +101,19 @@ def _native_guard() -> None:
connection.close()
if (
task is None
or _value(task, "status") != "blocked"
or _value(task, "status") != status
or _value(task, "current_run_id") is not None
or _value(task, "block_kind") != "capability"
or ROOT not in {str(parent) for parent in parents}
or tuple(event or ()) != (FAILED_EVENT_ID, int(FAILED_RUN), "blocked")
or tuple(latest or ()) != (int(FAILED_RUN), "blocked", "blocked")
or tuple(event or ()) != (event_id, int(run_id), event_kind)
or tuple(latest or ()) != (int(run_id), "blocked", "blocked")
):
raise ValueError("native task changed after run 10")
def _receipt_guard() -> None:
def _receipt_guard(run_id: str) -> dict[str, Any]:
"""Verify the sealed source-8 receipt and its normalized title provenance."""
receipt = supervisor_state.publication_retry(BOARD, CHILD, FAILED_RUN)
receipt = supervisor_state.publication_retry(BOARD, CHILD, run_id)
if not isinstance(receipt, dict):
raise ValueError("publication receipt is unavailable")
source = receipt.get("source")
@ -134,7 +137,7 @@ def _receipt_guard() -> None:
"SELECT raw_result_sha256,reconstruction FROM publication_retry_provenance "
"WHERE board=? AND child_task_id=?", (BOARD, CHILD)
).fetchone()
pending = (SOURCE_RUN, ORDINAL, FAILED_RUN, "", 1, RETRY_AFTER, "9")
pending = (SOURCE_RUN, ORDINAL, run_id, "", 1, RETRY_AFTER, "9")
released = (SOURCE_RUN, ORDINAL, "", "", 1, RETRY_AFTER, "9")
if (
tuple(retry or ()) not in {pending, released}
@ -144,6 +147,7 @@ def _receipt_guard() -> None:
or not provenance[1].startswith("normalized-title-from-sealed-receipt-sha256:")
):
raise ValueError("publication retry state changed after run 10")
return receipt
def _remote_guard() -> None:
@ -167,15 +171,41 @@ def _remote_guard() -> None:
raise ValueError("pull request changed after run 10")
def _published_remote_guard(receipt: dict[str, Any]) -> None:
"""Require the already-published PR head and prose to equal sealed evidence."""
try:
pull = json.loads(scm_broker_client.read("/api/v1/repos/titan/soteria/pulls/3"))
except (json.JSONDecodeError, OSError) as error:
raise ValueError("published pull request metadata is unavailable") from error
head = pull.get("head") if isinstance(pull, dict) else None
base = pull.get("base") if isinstance(pull, dict) else None
title, body = receipt.get("title"), receipt.get("body")
if (
not isinstance(title, str)
or not isinstance(body, str)
or not isinstance(pull, dict)
or pull.get("state") != "open"
or pull.get("merged") is not False
or not isinstance(head, dict)
or head.get("ref") != BRANCH
or head.get("sha") != PRESERVED_HEAD
or not isinstance(base, dict)
or base.get("ref") != "main"
or pull.get("title") != _draft_title(title)
or pull.get("body") != body
):
raise ValueError("published pull request changed after run 11")
def release(pool_database: Path) -> bool:
"""CAS-release run 10’s issue fence without changing its retry budget/history.
A retry after a crash before native unblock accepts the already-released
state. Any other state is evidence that another actor changed the card.
"""
_pool_guard(pool_database)
_native_guard()
_receipt_guard()
_pool_guard(pool_database, FAILED_RUN)
_native_guard(FAILED_RUN, FAILED_EVENT_ID, "blocked", "blocked")
_receipt_guard(FAILED_RUN)
_remote_guard()
with supervisor_state._connect(BOARD) as connection:
changed = connection.execute(
@ -196,15 +226,49 @@ def release(pool_database: Path) -> bool:
raise ValueError("publication retry state changed before CAS release")
def confirm_published_release(pool_database: Path) -> bool:
"""Release run 11 only after its PR publish is independently visible.
This intentionally does not reopen native work. The caller must separately
use the native triage specification API after a fixed mediator is deployed.
"""
_pool_guard(pool_database, CONFIRM_RUN)
_native_guard(CONFIRM_RUN, CONFIRM_EVENT_ID, "triage", "block_loop_detected")
receipt = _receipt_guard(CONFIRM_RUN)
_published_remote_guard(receipt)
with supervisor_state._connect(BOARD) as connection:
changed = connection.execute(
"UPDATE publication_retries SET issued_run_id='' WHERE board=? AND child_task_id=? "
"AND source_run_id=? AND source_ordinal=? AND issued_run_id=? AND resolved_run_id='' "
"AND reissue_count=1 AND retry_after=? AND last_reissued_run_id='9'",
(BOARD, CHILD, SOURCE_RUN, ORDINAL, CONFIRM_RUN, RETRY_AFTER),
).rowcount
if changed == 1:
return True
with supervisor_state._connect(BOARD) as connection:
row = connection.execute(
"SELECT source_run_id,source_ordinal,issued_run_id,resolved_run_id,reissue_count,retry_after,last_reissued_run_id "
"FROM publication_retries WHERE board=? AND child_task_id=?", (BOARD, CHILD)
).fetchone()
if tuple(row or ()) == (SOURCE_RUN, ORDINAL, "", "", 1, RETRY_AFTER, "9"):
return False
raise ValueError("published confirmation changed before CAS release")
if __name__ == "__main__":
import argparse
import os
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--confirm-published", action="store_true")
parser.add_argument(
"--pool-db", type=Path,
default=Path(os.environ.get("HERMES_HOME", "/opt/data")) / "execution-pool/assignments.db",
)
args = parser.parse_args()
release(args.pool_db)
print(f"released {BOARD}/{CHILD} failed_run={FAILED_RUN} source_run={SOURCE_RUN}")
if args.confirm_published:
confirm_published_release(args.pool_db)
print(f"released {BOARD}/{CHILD} published_run={CONFIRM_RUN} source_run={SOURCE_RUN}")
else:
release(args.pool_db)
print(f"released {BOARD}/{CHILD} failed_run={FAILED_RUN} source_run={SOURCE_RUN}")

89
testing/tests/test_release_soteria_publication_retry.py
View File

@ -31,10 +31,12 @@ def _receipt() -> dict:
},
"head": release.PRESERVED_HEAD,
"result_digest": release.RECEIPT_DIGEST,
"title": "Replace cache literals safely.",
"body": "Preserved validation evidence.",
}
def _pool(tmp_path: Path, *, active: bool = False, digest: str | None = None) -> tuple[Path, str]:
def _pool(tmp_path: Path, *, active: bool = False, digest: str | None = None, run_id: str = release.FAILED_RUN) -> tuple[Path, str]:
"""Create only the terminal pool columns consumed by the release guard."""
database = tmp_path / "pool.db"
connection = sqlite3.connect(database)
@ -50,7 +52,7 @@ def _pool(tmp_path: Path, *, active: bool = False, digest: str | None = None) ->
actual_digest = hashlib.sha256(encoded.encode()).hexdigest()
connection.execute(
"INSERT INTO assignments VALUES(?,?,?,?,?,?,?,?,?)",
(release.BOARD, release.CHILD, release.FAILED_RUN, json.dumps({"scm_resume": resume}), encoded,
(release.BOARD, release.CHILD, run_id, json.dumps({"scm_resume": resume}), encoded,
digest or actual_digest, "finalized", 0, 1),
)
if active:
@ -63,12 +65,12 @@ def _pool(tmp_path: Path, *, active: bool = False, digest: str | None = None) ->
return database, actual_digest
def _native(monkeypatch, *, event=None, latest=None):
def _native(monkeypatch, *, event=None, latest=None, status="blocked", block_kind="capability"):
"""Install the actual scalar contract expected from the native adapter."""
event = event or (release.FAILED_EVENT_ID, int(release.FAILED_RUN), "blocked")
latest = latest or (int(release.FAILED_RUN), "blocked", "blocked")
state = SimpleNamespace(
task=SimpleNamespace(status="blocked", current_run_id=None, block_kind="capability"),
task=SimpleNamespace(status=status, current_run_id=None, block_kind=block_kind),
event=event, latest=latest,
)
@ -118,9 +120,9 @@ def test_exact_historical_fixture_passes_all_release_guards(tmp_path, monkeypatc
_receipt_state(monkeypatch, _receipt())
_remote(monkeypatch)
release._pool_guard(pool)
release._native_guard()
release._receipt_guard()
release._pool_guard(pool, release.FAILED_RUN)
release._native_guard(release.FAILED_RUN, release.FAILED_EVENT_ID, "blocked", "blocked")
release._receipt_guard(release.FAILED_RUN)
release._remote_guard()
@ -138,12 +140,12 @@ def test_conflicting_evidence_fails_closed_without_a_release(tmp_path, monkeypat
with pytest.raises(ValueError):
if change in {"terminal_digest", "active"}:
release._pool_guard(pool)
release._pool_guard(pool, release.FAILED_RUN)
elif change == "later_event":
assert state.event[0] == 210
release._native_guard()
release._native_guard(release.FAILED_RUN, release.FAILED_EVENT_ID, "blocked", "blocked")
elif change == "changed_receipt":
release._receipt_guard()
release._receipt_guard(release.FAILED_RUN)
else:
release._remote_guard()
@ -189,3 +191,70 @@ def test_exact_cas_release_is_repeatable_and_preserves_budget_and_provenance(tmp
connection.close()
assert retry == ("", 1, release.RETRY_AFTER, "9")
assert provenance[0] == release.SOURCE_RAW_SHA
def test_published_confirmation_requires_run11_triage_and_exact_pr_prose(tmp_path, monkeypatch):
"""The post-push recovery accepts only the unchanged run-11 confirmation."""
pool, digest = _pool(tmp_path, run_id=release.CONFIRM_RUN)
monkeypatch.setattr(release, "FAILED_RESULT_SHA", digest)
_native(
monkeypatch,
event=(release.CONFIRM_EVENT_ID, int(release.CONFIRM_RUN), "block_loop_detected"),
latest=(int(release.CONFIRM_RUN), "blocked", "blocked"),
status="triage",
)
receipt = _receipt()
_receipt_state(monkeypatch, receipt, issued=release.CONFIRM_RUN)
pull = {
"state": "open", "merged": False,
"head": {"ref": release.BRANCH, "sha": release.PRESERVED_HEAD},
"base": {"ref": "main"},
"title": release._draft_title(receipt["title"]), "body": receipt["body"],
}
monkeypatch.setattr(release.scm_broker_client, "read", lambda _path: json.dumps(pull).encode())
release._pool_guard(pool, release.CONFIRM_RUN)
release._native_guard(release.CONFIRM_RUN, release.CONFIRM_EVENT_ID, "triage", "block_loop_detected")
release._published_remote_guard(release._receipt_guard(release.CONFIRM_RUN))
pull["body"] = "changed"
with pytest.raises(ValueError, match="published pull"):
release._published_remote_guard(receipt)
def test_published_confirmation_cas_is_repeatable_without_spending_budget(tmp_path, monkeypatch):
"""A crash before native triage specification leaves one harmless rerun path."""
state_db = tmp_path / "state.db"
connection = sqlite3.connect(state_db)
connection.execute(
"CREATE TABLE publication_retries(board,child_task_id,source_run_id,source_ordinal,issued_run_id,"
"resolved_run_id,reissue_count,retry_after,last_reissued_run_id)"
)
connection.execute(
"INSERT INTO publication_retries VALUES(?,?,?,?,?,?,?,?,?)",
(release.BOARD, release.CHILD, release.SOURCE_RUN, 0, release.CONFIRM_RUN, "", 1, release.RETRY_AFTER, "9"),
)
connection.commit()
connection.close()
@contextmanager
def connect(_board):
connection = sqlite3.connect(state_db)
try:
yield connection
connection.commit()
finally:
connection.close()
monkeypatch.setattr(release, "_pool_guard", lambda *_args: None)
monkeypatch.setattr(release, "_native_guard", lambda *_args: None)
monkeypatch.setattr(release, "_receipt_guard", lambda *_args: {})
monkeypatch.setattr(release, "_published_remote_guard", lambda *_args: None)
monkeypatch.setattr(release.supervisor_state, "_connect", connect)
assert release.confirm_published_release(tmp_path / "unused.db") is True
assert release.confirm_published_release(tmp_path / "unused.db") is False
connection = sqlite3.connect(state_db)
row = connection.execute("SELECT issued_run_id,reissue_count,retry_after,last_reissued_run_id FROM publication_retries").fetchone()
connection.close()
assert row == ("", 1, release.RETRY_AFTER, "9")