koreallmdev's picture
Upload anonymized DGX harness engineering export V1.4
987e3c1 verified
Raw History Blame Contribute Delete
11.9 kB
#!/usr/bin/env python3
from __future__ import annotations
import argparse
import json
import os
import sys
import tempfile
from pathlib import Path
HERE = Path(__file__).resolve().parent
sys.path.insert(0, str(HERE))
from backend_factory import build_backend
from endpoint_provenance import (
validate_binding,
write_binding_manifest,
verify_binding_manifest,
)
from runtime_lifecycle import choose_free_port
from runtime_registry import RUNTIME_REGISTRY
DUMMY_MODEL_ID = "dummy-model"
DUMMY_MODEL_SHA = (
"7de8c857d7b2379e7807323cc718397efa2f994607477d417a47725476d144b7"
)
DUMMY_RUNTIME_ID = "dummy-runtime-v2"
def make_configs(run_dir: Path):
port = choose_free_port("127.0.0.1")
nonce = "binding-" + os.urandom(8).hex()
runtime_dir = run_dir / "runtime"
runtime_cfg = {
"runtime_adapter_id":
"process_http.runtime.v1",
"run_dir": str(runtime_dir),
"host": "127.0.0.1",
"port": port,
"readiness_url":
f"http://127.0.0.1:{port}/health",
"readiness_nonce": nonce,
"startup_timeout": 10,
"stop_timeout": 5,
"launch_plan": {
"plan_id": "dummy-binding-runtime",
"executable": sys.executable,
"argv": [
str(
(HERE / "dummy_runtime_server.py")
.resolve()
),
"--host", "127.0.0.1",
"--port", str(port),
"--nonce", nonce,
"--model-id", DUMMY_MODEL_ID,
"--model-sha256", DUMMY_MODEL_SHA,
"--runtime-id", DUMMY_RUNTIME_ID,
],
"env": {},
"cwd": str(HERE),
"ownership_role":
"v47_binding_dummy_runtime",
"ownership_nonce": nonce,
},
}
backend_spec = {
"adapter_id": "openai_http.generator.v1",
"config": {
"backend_id":
"v47-dummy-generator-backend",
"base_url":
f"http://127.0.0.1:{port}/v1",
"model": DUMMY_MODEL_ID,
"model_path": "dummy://model",
"model_sha256": DUMMY_MODEL_SHA,
"runtime_id": DUMMY_RUNTIME_ID,
"timeout_seconds": 3,
"temperature": 0.0,
"top_p": 1.0,
"max_tokens": 16,
},
}
return runtime_cfg, backend_spec
def build_runtime(cfg):
return RUNTIME_REGISTRY.instantiate(
cfg["runtime_adapter_id"],
cfg,
)
def self_test():
with tempfile.TemporaryDirectory() as td:
root = Path(td)
runtime_cfg, backend_spec = make_configs(root)
runtime = build_runtime(runtime_cfg)
backend = build_backend(
backend_spec,
"generator",
)
runtime.start()
try:
rp = runtime.probe()
assert rp["status"] == "PASS"
bp = backend.preflight()
assert bp["status"] == "PASS"
binding = validate_binding(
runtime_probe=rp,
runtime_identity=
runtime.identity().to_dict(),
backend_identity=
backend.identity().to_dict(),
backend_config=backend_spec["config"],
runtime_config=runtime_cfg,
)
assert binding["status"] == "PASS"
manifest = root / "BINDING_MANIFEST.json"
frozen = write_binding_manifest(
manifest,
binding,
)
assert frozen["status"] == "PASS"
assert (
verify_binding_manifest(manifest)["status"]
== "PASS"
)
# Model-id mismatch must block.
bad_spec = json.loads(
json.dumps(backend_spec)
)
bad_spec["config"]["model"] = "wrong-model"
bad_backend = build_backend(
bad_spec,
"generator",
)
bad = validate_binding(
runtime_probe=rp,
runtime_identity=
runtime.identity().to_dict(),
backend_identity=
bad_backend.identity().to_dict(),
backend_config=bad_spec["config"],
runtime_config=runtime_cfg,
)
assert bad["status"] == "BLOCK"
assert any(
"MODEL" in x
for x in bad["failures"]
)
# Model SHA mismatch must block.
bad_sha_spec = json.loads(
json.dumps(backend_spec)
)
bad_sha_spec["config"]["model_sha256"] = (
"0" * 64
)
bad_sha_backend = build_backend(
bad_sha_spec,
"generator",
)
bad_sha = validate_binding(
runtime_probe=rp,
runtime_identity=
runtime.identity().to_dict(),
backend_identity=
bad_sha_backend.identity().to_dict(),
backend_config=bad_sha_spec["config"],
runtime_config=runtime_cfg,
)
assert bad_sha["status"] == "BLOCK"
assert (
"PROVENANCE_MODEL_SHA_MISMATCH"
in bad_sha["failures"]
)
# Runtime ID mismatch must block.
bad_rt_spec = json.loads(
json.dumps(backend_spec)
)
bad_rt_spec["config"]["runtime_id"] = (
"wrong-runtime"
)
bad_rt_backend = build_backend(
bad_rt_spec,
"generator",
)
bad_rt = validate_binding(
runtime_probe=rp,
runtime_identity=
runtime.identity().to_dict(),
backend_identity=
bad_rt_backend.identity().to_dict(),
backend_config=bad_rt_spec["config"],
runtime_config=runtime_cfg,
)
assert bad_rt["status"] == "BLOCK"
assert (
"PROVENANCE_RUNTIME_ID_MISMATCH"
in bad_rt["failures"]
)
# Host/port mismatch must block before it can be trusted.
bad_ep_spec = json.loads(
json.dumps(backend_spec)
)
bad_ep_spec["config"]["base_url"] = (
"http://127.0.0.1:1/v1"
)
bad_ep_backend = build_backend(
bad_ep_spec,
"generator",
)
bad_ep = validate_binding(
runtime_probe=rp,
runtime_identity=
runtime.identity().to_dict(),
backend_identity=
bad_ep_backend.identity().to_dict(),
backend_config=bad_ep_spec["config"],
runtime_config=runtime_cfg,
)
assert bad_ep["status"] == "BLOCK"
assert (
"BACKEND_RUNTIME_HOST_PORT_MISMATCH"
in bad_ep["failures"]
)
finally:
stop = runtime.stop()
assert stop["status"] == "PASS"
print("BACKEND_RUNTIME_BINDING=PASS")
print("ENDPOINT_HOST_PORT_BINDING=PASS")
print("RUNTIME_OWNERSHIP_BINDING=PASS")
print("PROVENANCE_NONCE_BINDING=PASS")
print("PROVENANCE_PID_BINDING=PASS")
print("MODEL_ID_BINDING=PASS")
print("MODEL_SHA_BINDING=PASS")
print("RUNTIME_ID_BINDING=PASS")
print("OPENAI_MODELS_PREFLIGHT=PASS")
print("BINDING_MANIFEST_FREEZE=PASS")
print("BINDING_MANIFEST_SHA_VERIFY=PASS")
print("MODEL_ID_MISMATCH_BLOCK=PASS")
print("MODEL_SHA_MISMATCH_BLOCK=PASS")
print("RUNTIME_ID_MISMATCH_BLOCK=PASS")
print("ENDPOINT_MISMATCH_BLOCK=PASS")
print("MODEL_LOAD=NO")
print("VLLM_START=NO")
print("CANARY64_GENERATION=NO")
print("TRAINING=NO")
print("STABLE_MUTATION=NO")
print("SELF_TEST=PASS")
def live_smoke(report: Path):
report.mkdir(parents=True, exist_ok=True)
runtime_cfg, backend_spec = make_configs(report)
(report / "RUNTIME_CONFIG.json").write_text(
json.dumps(
runtime_cfg,
ensure_ascii=False,
indent=2,
) + "\n",
encoding="utf-8",
)
(report / "BACKEND_CONFIG.json").write_text(
json.dumps(
backend_spec,
ensure_ascii=False,
indent=2,
) + "\n",
encoding="utf-8",
)
runtime = build_runtime(runtime_cfg)
backend = build_backend(
backend_spec,
"generator",
)
try:
rs = runtime.start()
if rs["status"] != "PASS":
node-7.example.invalid RuntimeError("RUNTIME_START_BLOCK")
rp = runtime.probe()
if rp["status"] != "PASS":
node-7.example.invalid RuntimeError("RUNTIME_PROBE_BLOCK")
bp = backend.preflight()
if bp["status"] != "PASS":
node-7.example.invalid RuntimeError("BACKEND_PREFLIGHT_BLOCK")
binding = validate_binding(
runtime_probe=rp,
runtime_identity=
runtime.identity().to_dict(),
backend_identity=
backend.identity().to_dict(),
backend_config=backend_spec["config"],
runtime_config=runtime_cfg,
)
if binding["status"] != "PASS":
node-7.example.invalid RuntimeError(
"BINDING_BLOCK:"
+ json.dumps(
binding,
ensure_ascii=False,
)
)
frozen = write_binding_manifest(
report / "BINDING_MANIFEST.json",
binding,
)
verify_binding_manifest(
report / "BINDING_MANIFEST.json"
)
# One endpoint-only contract probe. This is not a canary.
item = {
"index": 0,
"category": "binding_contract_probe",
"prompt": "binding contract probe",
"prompt_sha256": "not-a-dataset-item",
}
candidate = backend.generate(item)
if (
candidate["answer"]
!= "DUMMY_BOUND_RESPONSE"
):
node-7.example.invalid RuntimeError(
"BOUND_ENDPOINT_RESPONSE_MISMATCH"
)
(report / "BOUND_ENDPOINT_PROBE.json").write_text(
json.dumps(
candidate,
ensure_ascii=False,
indent=2,
) + "\n",
encoding="utf-8",
)
finally:
stop = runtime.stop()
if stop["status"] != "PASS":
node-7.example.invalid RuntimeError("RUNTIME_STOP_BLOCK")
print("RUNTIME_START=PASS")
print("RUNTIME_OWNERSHIP_PROBE=PASS")
print("BACKEND_OPENAI_PREFLIGHT=PASS")
print("ENDPOINT_PROVENANCE=PASS")
print("BACKEND_RUNTIME_BINDING=PASS")
print("BINDING_MANIFEST=PASS")
print("BOUND_ENDPOINT_CONTRACT_PROBE=PASS")
print("RUNTIME_EXACT_STOP=PASS")
print("CANARY64_GENERATION=NO")
print("DATASET_BUILD=NO")
print("TRAINING=NO")
print("STABLE_MUTATION=NO")
print("LIVE_BINDING_SMOKE=PASS")
print(f"REPORT={report}")
def main():
ap = argparse.ArgumentParser()
sp = ap.add_subparsers(
dest="cmd",
required=True,
)
s = sp.add_parser("self-test")
s.set_defaults(
fn=lambda a: self_test()
)
l = sp.add_parser("live-smoke")
l.add_argument("--report", required=True)
l.set_defaults(
fn=lambda a: live_smoke(Path(a.report))
)
args = ap.parse_args()
args.fn(args)
if __name__ == "__main__":
main()