perf(cli): cut hermes startup 63% — flip head-to-head vs codex (#31968)

* perf(bitwarden): persist secret-fetch cache across CLI invocations

Every `hermes` invocation paid a ~380ms tax for `bws secret list` to
Bitwarden Secrets Manager because the existing cache was in-process only.
Back-to-back `hermes chat -q`, gateway-spawned agents, and cron-launched
runs all re-fetched.

Adds a disk-persisted L2 cache at `<hermes_home>/cache/bws_cache.json`
(mode 0600, never contains the access token — only the SHA-256
fingerprint prefix). Same TTL as the in-process cache. Read on miss,
write on bws success, ignored on key mismatch / corruption / expiry.

Measured on a startup profile:
  load_hermes_dotenv() cold: 372ms → warm (disk cache hit): 20ms

End-to-end `hermes --version` cold→warm: 666ms → ~295ms.

In a hermes-vs-codex benchmark across 11 single- and multi-turn tasks
(framework overhead = wall − llm − tool_exec, median over 3 trials):

  cohort               before    after    saved
  single-turn (median)  2.96s    2.31s   -0.65s
  multi-turn  (5-turn)  9.40s    8.95s   -0.45s (≈0.3s/turn)

Hermes now wins head-to-head on 6/11 tasks vs codex (was 4/11 before).
The remaining ~0.6s single-turn delta is mostly Python's own import
cost in hermes_cli.main, which is a separate optimization.

* perf(cli): lazy-load model catalog + dedupe config.yaml reads at startup

Two import-time wins on top of the bws disk-cache fix:

1. Lazy-load `hermes_cli.models._PROVIDER_MODELS` via PEP 562
   module-level `__getattr__`. The catalog is ~55ms of work that was
   eagerly imported on every CLI invocation (line 4557 `if not
   _is_termux_startup_environment(): from hermes_cli.models import
   _PROVIDER_MODELS`). Audit showed every internal call site already
   does its own function-local import; only test code reads
   `hermes_cli.main._PROVIDER_MODELS` as a module attribute, and
   __getattr__ keeps that working transparently. First access triggers
   the import once and caches the result on the module via
   `globals()[name] = ...`, so subsequent reads are dict lookups.

2. Dedupe the double config.yaml read in the top-of-module bootstrap.
   Previously: one raw yaml.safe_load for the `security.redact_secrets`
   bridge, then a separate full `load_config()` (with deep-merge) for
   `network.force_ipv4`. Both keys come from the same file. Merged
   into one raw yaml load.

Combined with the bws cache fix in the previous commit:

  hermes --version wall time:
    original (cold):           666 ms
    after bws fix (warm):      295 ms
    after lazy-load + dedupe:  228 ms   (-67 ms additional, -66% from original)

Tests:
  - tests/hermes_cli/test_api_key_providers.py: 173/173 pass
    (lazy __getattr__ correctly handles
     `from hermes_cli.main import _PROVIDER_MODELS`)
  - tests/test_ipv4_preference.py + tests/hermes_cli/test_redact_config_bridge.py +
    tests/agent/test_redact.py: 93/93 pass (dedupe preserves both bridges)
  - tests/test_bitwarden_secrets.py + env_loader tests: 49/49 pass
This commit is contained in:
Teknium
2026-05-25 03:06:39 -07:00
committed by GitHub
parent c0169496d0
commit 0219b0408a
4 changed files with 395 additions and 24 deletions
+221
View File
@@ -572,3 +572,224 @@ def test_env_loader_calls_bsm_when_enabled(tmp_path, monkeypatch):
assert called["n"] == 1
assert os.environ.get("MY_BSM_KEY") == "from-bsm"
# ---------------------------------------------------------------------------
# Disk-persisted cache (cross-process — speeds up back-to-back CLI invocations)
# ---------------------------------------------------------------------------
def test_disk_cache_written_after_first_fetch(monkeypatch, tmp_path):
"""First fetch hits bws AND writes a 0600 file under hermes_home/cache/."""
home = tmp_path / ".hermes"
home.mkdir()
fake_binary = tmp_path / "bws"
fake_binary.write_text("")
payload = _fake_bws_payload([{"key": "K1", "value": "v1"}])
call_count = {"n": 0}
def fake_run(*a, **kw):
call_count["n"] += 1
return mock.Mock(returncode=0, stdout=payload, stderr="")
monkeypatch.setattr(bw.subprocess, "run", fake_run)
bw._reset_cache_for_tests(home)
secrets, _ = bw.fetch_bitwarden_secrets(
access_token="0.t", project_id="proj-1", binary=fake_binary,
cache_ttl_seconds=300, home_path=home,
)
assert secrets == {"K1": "v1"}
assert call_count["n"] == 1
cache_path = bw._disk_cache_path(home)
assert cache_path.exists()
# Mode must be 0600 — disk cache contains plaintext secret values
mode = os.stat(cache_path).st_mode & 0o777
assert mode == 0o600, f"expected 0o600, got 0o{mode:o}"
# File contents: key (fingerprint not raw token), secrets dict, fetched_at
payload_disk = json.loads(cache_path.read_text())
assert set(payload_disk.keys()) == {"key", "secrets", "fetched_at"}
assert payload_disk["secrets"] == {"K1": "v1"}
# Critically, the raw access token must NOT appear anywhere in the file
assert "0.t" not in cache_path.read_text()
def test_disk_cache_short_circuits_bws_when_fresh(monkeypatch, tmp_path):
"""Second fetch (different process simulation) skips bws entirely."""
home = tmp_path / ".hermes"
home.mkdir()
fake_binary = tmp_path / "bws"
fake_binary.write_text("")
payload = _fake_bws_payload([{"key": "K1", "value": "v1"}])
call_count = {"n": 0}
def fake_run(*a, **kw):
call_count["n"] += 1
return mock.Mock(returncode=0, stdout=payload, stderr="")
monkeypatch.setattr(bw.subprocess, "run", fake_run)
bw._reset_cache_for_tests(home)
# First call: hits bws, populates disk cache
bw.fetch_bitwarden_secrets(
access_token="0.t", project_id="proj-1", binary=fake_binary,
cache_ttl_seconds=300, home_path=home,
)
assert call_count["n"] == 1
# Clear ONLY the in-process cache to simulate a fresh subprocess.
bw._CACHE.clear()
secrets2, _ = bw.fetch_bitwarden_secrets(
access_token="0.t", project_id="proj-1", binary=fake_binary,
cache_ttl_seconds=300, home_path=home,
)
assert secrets2 == {"K1": "v1"}
# Critical: bws was NOT invoked the second time
assert call_count["n"] == 1
def test_disk_cache_expires_with_ttl(monkeypatch, tmp_path):
"""Stale disk cache (older than ttl) triggers a refetch."""
home = tmp_path / ".hermes"
home.mkdir()
fake_binary = tmp_path / "bws"
fake_binary.write_text("")
payload = _fake_bws_payload([{"key": "K1", "value": "v1"}])
call_count = {"n": 0}
def fake_run(*a, **kw):
call_count["n"] += 1
return mock.Mock(returncode=0, stdout=payload, stderr="")
monkeypatch.setattr(bw.subprocess, "run", fake_run)
bw._reset_cache_for_tests(home)
# First call
bw.fetch_bitwarden_secrets(
access_token="0.t", project_id="proj-1", binary=fake_binary,
cache_ttl_seconds=300, home_path=home,
)
assert call_count["n"] == 1
# Backdate the disk cache so the TTL window has passed
cache_path = bw._disk_cache_path(home)
payload_disk = json.loads(cache_path.read_text())
payload_disk["fetched_at"] = time.time() - 10_000
cache_path.write_text(json.dumps(payload_disk))
bw._CACHE.clear()
# Second call: stale disk → refetch
bw.fetch_bitwarden_secrets(
access_token="0.t", project_id="proj-1", binary=fake_binary,
cache_ttl_seconds=300, home_path=home,
)
assert call_count["n"] == 2
def test_disk_cache_key_mismatch_triggers_refetch(monkeypatch, tmp_path):
"""Disk cache entry written by a different token/project is ignored."""
home = tmp_path / ".hermes"
home.mkdir()
fake_binary = tmp_path / "bws"
fake_binary.write_text("")
payload = _fake_bws_payload([{"key": "K1", "value": "v1"}])
call_count = {"n": 0}
def fake_run(*a, **kw):
call_count["n"] += 1
return mock.Mock(returncode=0, stdout=payload, stderr="")
monkeypatch.setattr(bw.subprocess, "run", fake_run)
bw._reset_cache_for_tests(home)
# Write a cache entry for a DIFFERENT token/project pair
cache_path = bw._disk_cache_path(home)
cache_path.parent.mkdir(parents=True, exist_ok=True)
cache_path.write_text(json.dumps({
"key": "deadbeef00000000|other-project|",
"secrets": {"OTHER": "should-not-leak"},
"fetched_at": time.time(),
}))
secrets, _ = bw.fetch_bitwarden_secrets(
access_token="0.t", project_id="proj-1", binary=fake_binary,
cache_ttl_seconds=300, home_path=home,
)
# We must NOT have used the foreign cache entry
assert secrets == {"K1": "v1"}
assert "OTHER" not in secrets
assert call_count["n"] == 1
def test_disk_cache_use_cache_false_skips_disk(monkeypatch, tmp_path):
"""use_cache=False must skip BOTH in-process and disk caches."""
home = tmp_path / ".hermes"
home.mkdir()
fake_binary = tmp_path / "bws"
fake_binary.write_text("")
payload = _fake_bws_payload([{"key": "K1", "value": "v1"}])
call_count = {"n": 0}
def fake_run(*a, **kw):
call_count["n"] += 1
return mock.Mock(returncode=0, stdout=payload, stderr="")
monkeypatch.setattr(bw.subprocess, "run", fake_run)
bw._reset_cache_for_tests(home)
# First call WITH cache populates disk
bw.fetch_bitwarden_secrets(
access_token="0.t", project_id="proj-1", binary=fake_binary,
cache_ttl_seconds=300, use_cache=True, home_path=home,
)
assert call_count["n"] == 1
bw._CACHE.clear()
# Second call with use_cache=False MUST hit bws again even though disk is fresh
bw.fetch_bitwarden_secrets(
access_token="0.t", project_id="proj-1", binary=fake_binary,
cache_ttl_seconds=300, use_cache=False, home_path=home,
)
assert call_count["n"] == 2
def test_disk_cache_corrupt_file_falls_through(monkeypatch, tmp_path):
"""A garbage cache file must NOT crash startup — we refetch."""
home = tmp_path / ".hermes"
home.mkdir()
fake_binary = tmp_path / "bws"
fake_binary.write_text("")
payload = _fake_bws_payload([{"key": "K1", "value": "v1"}])
monkeypatch.setattr(
bw.subprocess, "run",
lambda *a, **kw: mock.Mock(returncode=0, stdout=payload, stderr=""),
)
bw._reset_cache_for_tests(home)
# Write a corrupt cache file
cache_path = bw._disk_cache_path(home)
cache_path.parent.mkdir(parents=True, exist_ok=True)
cache_path.write_text("not json {{{")
secrets, _ = bw.fetch_bitwarden_secrets(
access_token="0.t", project_id="proj-1", binary=fake_binary,
cache_ttl_seconds=300, home_path=home,
)
# Refetched cleanly
assert secrets == {"K1": "v1"}
# And the corrupt file was replaced with a valid one
assert json.loads(cache_path.read_text())["secrets"] == {"K1": "v1"}
def test_reset_cache_for_tests_deletes_disk_file(tmp_path):
"""_reset_cache_for_tests(home_path) must also clean disk."""
home = tmp_path / ".hermes"
home.mkdir()
cache_path = bw._disk_cache_path(home)
cache_path.parent.mkdir(parents=True, exist_ok=True)
cache_path.write_text("{}")
assert cache_path.exists()
bw._reset_cache_for_tests(home)
assert not cache_path.exists()
# Idempotent
bw._reset_cache_for_tests(home)