Add althing seat monitor-behavior test artifacts
Disposable harnesses from the wake-RFI thread (01M2WT3F): exit1-behavior-test.py covers the monitor exit-code/gate path including a sustained-failure/no-backoff assertion; hash_gate_repro.py covers the detection-time hash persistence crash-before-read boundary with consumption/unchanged/changed controls. Both fail nonzero on mismatch. Committing so seat artifact provenance rides git rather than file mtimes.
This commit is contained in:
+94
@@ -0,0 +1,94 @@
|
||||
"""Disposable behavioral test: monitor source exit!=0 -> ok=False, nothing persisted.
|
||||
|
||||
Calls cron.monitor.check_monitor and cron.scheduler._apply_monitor_gate directly
|
||||
against fabricated job dicts and throwaway monitor scripts. No live job touched,
|
||||
no agent woken. Exit codes: 0 = all assertions held, 1 = mismatch.
|
||||
"""
|
||||
import os, sys, json, tempfile
|
||||
sys.path.insert(0, os.path.expanduser("~/.hermes/hermes-agent"))
|
||||
os.environ.setdefault("HERMES_HOME", os.path.expanduser("~/.hermes"))
|
||||
|
||||
from cron.monitor import check_monitor
|
||||
from cron.scheduler import _apply_monitor_gate
|
||||
|
||||
SDIR = os.path.expanduser("~/.hermes/scripts")
|
||||
results = []
|
||||
|
||||
def mkscript(name, body):
|
||||
p = os.path.join(SDIR, name)
|
||||
with open(p, "w") as f:
|
||||
f.write(body)
|
||||
os.chmod(p, 0o755)
|
||||
return os.path.relpath(p, SDIR)
|
||||
|
||||
# Script S1: emits the sentinel then exit 1 (proposed repaired failure path)
|
||||
s1 = mkscript("_t_exit1.sh",
|
||||
'#!/usr/bin/env bash\necho "PEEK-FAILED rc=2"\nexit 1\n')
|
||||
# Script S2: emits the sentinel then exit 0 (current production behaviour)
|
||||
s2 = mkscript("_t_exit0.sh",
|
||||
'#!/usr/bin/env bash\necho "PEEK-FAILED rc=2"\nexit 0\n')
|
||||
|
||||
def fake_job(jid, script, prior_hash=None):
|
||||
st = {"last_output_hash": prior_hash} if prior_hash else None
|
||||
return {"id": jid, "name": jid, "monitor_script": script,
|
||||
"monitor_state": st, "schedule": {"kind": "interval", "minutes": 5}}
|
||||
|
||||
# --- T1: exit 1 -> check_monitor ok=False, no state persisted
|
||||
job = fake_job("_t_exit1_job", s1)
|
||||
out = check_monitor(job)
|
||||
results.append(("T1 exit1 => ok=False", out.ok is False))
|
||||
results.append(("T1 error carries sentinel", "PEEK-FAILED" in (out.error or "")))
|
||||
from cron.jobs import get_job
|
||||
persisted = get_job("_t_exit1_job")
|
||||
results.append(("T1 nothing persisted (no monitor_state on a real store)",
|
||||
persisted is None or not (persisted.get("monitor_state") or {}).get("last_output_hash")))
|
||||
|
||||
# --- T2: exit 0 -> ok=True changed=True (sentinel becomes a persisted hash)
|
||||
job0 = fake_job("_t_exit0_job", s2)
|
||||
out0 = check_monitor(job0)
|
||||
results.append(("T2 exit0 => ok=True changed=True", out0.ok is True and out0.changed is True))
|
||||
# simulate what a real store would now hold (job not in the store, so update_job
|
||||
# has nothing to write; fabricate the persisted hash the same way the monitor did)
|
||||
from cron.monitor import hash_monitor_output
|
||||
persisted_hash = hash_monitor_output("PEEK-FAILED rc=2")
|
||||
suppressed = check_monitor(fake_job("_t_exit0_job", s2, prior_hash=persisted_hash))
|
||||
results.append(("T2 re-run with persisted hash suppressed (outage wakes once per transition)",
|
||||
suppressed.ok is True and suppressed.changed is False))
|
||||
|
||||
# --- T3: _apply_monitor_gate on the exit-1 job returns the error early-result
|
||||
early, prompt, ctx = _apply_monitor_gate(fake_job("_t_exit1_job", s1), "_t_exit1_job", "t", None)
|
||||
ok_gate = (early is not None and early[0] is False and "source failed" in early[1])
|
||||
results.append(("T3 gate early-returns failure alert, no agent run", ok_gate))
|
||||
# repeated failing tick: gate must alert AGAIN (nothing persisted to suppress against)
|
||||
early2, _, _ = _apply_monitor_gate(fake_job("_t_exit1_job", s1), "_t_exit1_job", "t", None)
|
||||
results.append(("T3 second failing tick alerts again (no dedup state)",
|
||||
early2 is not None and early2[0] is False))
|
||||
|
||||
# --- T4: sustained-failure then recovery: no backoff, no auto-pause, job keeps firing.
|
||||
# Source claim (svos-dev): the monitor-failure branch early-returns only;
|
||||
# _block_and_pause_job is unreachable from it. Behavioral close: N consecutive
|
||||
# failing ticks each alert (no dedup/pause), a recovery tick passes the gate
|
||||
# open (early=None), and the job is never left paused/absent-of-schedule.
|
||||
fail_alerts = 0
|
||||
for _ in range(5):
|
||||
early_t, _, ctx_t = _apply_monitor_gate(fake_job("_t_exit1_job", s1), "_t_exit1_job", "t", None)
|
||||
if early_t is not None and early_t[0] is False:
|
||||
fail_alerts += 1
|
||||
results.append(("T4 5 consecutive failing ticks each alert (no backoff/disable)", fail_alerts == 5))
|
||||
after = get_job("_t_exit1_job")
|
||||
results.append(("T4 job not paused/disabled by failures (absent or state!=paused)",
|
||||
after is None or after.get("state") not in ("paused", "blocked")))
|
||||
# recovery: a script that succeeds with a NEW output passes the gate open
|
||||
s3 = mkscript("_t_recover.sh", '#!/usr/bin/env bash\necho "[9001]"\nexit 0\n')
|
||||
early_r, _, ctx_r = _apply_monitor_gate(fake_job("_t_rec_job", s3), "_t_rec_job", "t", None)
|
||||
results.append(("T4 recovery tick passes gate open (early=None, monitor ctx present)",
|
||||
early_r is None and ctx_r is not None))
|
||||
os.remove(os.path.join(SDIR, s3))
|
||||
|
||||
# cleanup
|
||||
for f in (s1, s2):
|
||||
os.remove(os.path.join(SDIR, f))
|
||||
|
||||
for name, ok in results:
|
||||
print(("PASS" if ok else "FAIL"), "-", name)
|
||||
sys.exit(0 if all(ok for _, ok in results) else 1)
|
||||
@@ -0,0 +1,87 @@
|
||||
"""Behavioral repro: cron monitor hash gate persists at DETECTION time, so a
|
||||
run that dies before consuming mail is suppressed on the next tick if the
|
||||
observed output is unchanged. Disposable job dict + in-memory persist; the
|
||||
live job and store are untouched.
|
||||
|
||||
Controls per forseti (2740):
|
||||
- positive control: changed ID set DOES admit another run
|
||||
- suppression control: unchanged output after a SUCCESSFUL run suppresses
|
||||
- crash case: persist detection, simulate death before 'read', reload state
|
||||
as after restart, rerun with SAME id set -> suppressed (the finding)
|
||||
Each condition repeated 3x.
|
||||
"""
|
||||
import sys, copy
|
||||
sys.path.insert(0, "/home/lkraven/.hermes/hermes-agent")
|
||||
from cron import monitor
|
||||
|
||||
STATE = {} # in-memory stand-in for jobs.json monitor_state
|
||||
|
||||
def fake_persist(job_id, new_hash, output):
|
||||
STATE[job_id] = {"last_output_hash": new_hash, "snapshot": output}
|
||||
|
||||
monitor._persist_monitor_state = fake_persist
|
||||
monitor._read_last_output = lambda job_id: STATE.get(job_id, {}).get("snapshot", "")
|
||||
monitor._write_last_output = lambda job_id, output: STATE[job_id].__setitem__("snapshot", output)
|
||||
|
||||
OUTPUT = [""] # current observed monitor output, set per tick
|
||||
|
||||
def fake_source(job):
|
||||
return True, OUTPUT[0]
|
||||
|
||||
monitor._run_monitor_source = fake_source
|
||||
|
||||
def tick(ids):
|
||||
"""One monitor tick with the given unread id set. Returns (ok, changed)."""
|
||||
OUTPUT[0] = "\n".join(f"[{i}]" for i in ids) + "\n"
|
||||
job = {"id": "disposable-test-job", "monitor_script": "stub"}
|
||||
if "disposable-test-job" in STATE: # simulate fresh load from store
|
||||
job["monitor_state"] = copy.deepcopy(STATE["disposable-test-job"])
|
||||
out = monitor.check_monitor(job)
|
||||
return out.ok, getattr(out, "changed", None)
|
||||
|
||||
results = []
|
||||
for rep in range(1, 4):
|
||||
# --- consumption control: a SUCCESSFUL run consumes -> output genuinely changes.
|
||||
# (Was: same ids after success, which only proved unchanged-input suppression.
|
||||
# Fix: after the run consumes, the mailbox is empty, so the next observation
|
||||
# is the idle sentinel; that transition admits once, stable idle suppresses.)
|
||||
STATE.clear()
|
||||
r1 = tick(["100"]) # first observation -> wake
|
||||
r2 = tick([]) # agent consumed -> monitor now says idle
|
||||
r3 = tick([]) # still idle -> suppressed
|
||||
results.append((f"rep{rep} control-consume: first={r1} after-consume={r2} stable-idle={r3}",
|
||||
r1 == (True, True) and r2 == (True, True) and r3 == (True, False)))
|
||||
|
||||
# --- suppression control (renamed honestly): unchanged output suppresses ---
|
||||
STATE.clear()
|
||||
r1 = tick(["100"])
|
||||
r2 = tick(["100"]) # no consumption, same set next tick
|
||||
results.append((f"rep{rep} control-unchanged-suppresses: first={r1} unchanged={r2}",
|
||||
r1 == (True, True) and r2 == (True, False)))
|
||||
|
||||
# --- positive control: changed ID set admits another run ---
|
||||
STATE.clear()
|
||||
r1 = tick(["100"])
|
||||
r2 = tick(["100", "101"]) # new mail -> changed -> wake
|
||||
results.append((f"rep{rep} control-change: first={r1} changed={r2}",
|
||||
r1 == (True, True) and r2 == (True, True)))
|
||||
|
||||
# --- crash case: detection persisted, agent died BEFORE consuming mail ---
|
||||
STATE.clear()
|
||||
r1 = tick(["200", "201"]) # detection: hash persisted pre-agent
|
||||
# (agent session would run `postbox read` here; simulate it dying first)
|
||||
r2 = tick(["200", "201"]) # next tick, same unread set, post-restart reload
|
||||
r3 = tick(["200", "201"]) # and again: unbounded silence
|
||||
results.append((f"rep{rep} crash-before-read: detect={r1} retry1={r2} retry2={r3}",
|
||||
r1 == (True, True) and r2 == (True, False) and r3 == (True, False)))
|
||||
|
||||
# --- recovery path: stranded mail + a NEW message eventually wakes ---
|
||||
r4 = tick(["200", "201", "202"]) # a genuinely new message re-wakes,
|
||||
results.append((f"rep{rep} recovery-on-new-mail: {r4}", r4 == (True, True)))
|
||||
|
||||
ok = True
|
||||
for line, passed in results:
|
||||
print(("PASS " if passed else "FAIL ") + line)
|
||||
ok = ok and passed
|
||||
print("ALL PASS" if ok else "SOME FAILED")
|
||||
sys.exit(0 if ok else 1)
|
||||
Reference in New Issue
Block a user