230 lines
13 KiB
Python
230 lines
13 KiB
Python
"""Bounded whole-suite orchestration with one shared deadline and cost allowance."""
|
|
from __future__ import annotations
|
|
|
|
import hashlib
|
|
import math
|
|
import time
|
|
|
|
import suite_backends
|
|
from suite_assignments import decode
|
|
from suite_contract import (EXECUTION_REVISION, MAX_BODY, MAX_RESULT, MODELS, PROMPT_REVISION,
|
|
PROMPT_SHA256, REVISION, Problem, digest, encoded, reasoning_selection,
|
|
validate_partition)
|
|
from suite_policy import (BASE_NAME_LIMIT, MAX_GROUP, POLICY_REVISION, invocation,
|
|
validate_natural)
|
|
from suite_sizing import cap_families, review_summary
|
|
|
|
MAX_PASSES = 5
|
|
|
|
|
|
def ordered_request(request, alternative=False):
|
|
"""Use stable independent orders while retaining every complete case record."""
|
|
def key(case):
|
|
alias = case["alias"]
|
|
return hashlib.sha256((POLICY_REVISION + ":order-b:" + alias).encode()).hexdigest() if alternative else alias
|
|
cases = sorted(request["cases"], key=key)
|
|
if alternative and len(cases) > 1 and [c["alias"] for c in cases] == sorted(c["alias"] for c in cases):
|
|
cases.reverse()
|
|
return {**request, "cases": cases}
|
|
|
|
|
|
def capacity(call, provider, count, family_count=None):
|
|
"""Check the actual complete pass, including proposal/review input and schema."""
|
|
model = MODELS[provider]
|
|
input_bytes = len(call["input"].encode()) + len(call["system"].encode()) + len(encoded(call["schema"]))
|
|
if input_bytes > MAX_BODY:
|
|
raise Problem("pass_request_too_large", 413, review_pass=call["stage"], input_bytes=input_bytes)
|
|
if family_count is None:
|
|
estimate = 1024 + 128 * count
|
|
else:
|
|
estimate = 1024 + 48 * count + 384 * family_count
|
|
reserve = estimate + (8192 if provider == "claude" else 0)
|
|
bound = input_bytes + model["overhead"]
|
|
if reserve > model["output"] or bound + model["output"] * model.get("max_turns", 1) > model["context"]:
|
|
raise Problem("pass_capacity", 422, review_pass=call["stage"], input_bytes=input_bytes,
|
|
output_reservation_tokens=reserve)
|
|
return {"input_bytes": input_bytes, "input_token_bound": bound, "input_token_count": None,
|
|
"output_reservation_tokens": reserve, "output_reservation_verified": False,
|
|
"input_count_method": "Complete UTF-8 input/system/schema byte bound plus harness overhead; not a tokenizer"}
|
|
|
|
|
|
def preflight_workflow(request):
|
|
"""Choose once, respecting permissions; recheck actual expanded inputs at every pass."""
|
|
count = len(request["cases"])
|
|
candidates = ["local"]
|
|
if request["routing"]["allow_external"]:
|
|
candidates += request["routing"]["allowed_external_providers"]
|
|
reasons = {}
|
|
for provider in candidates:
|
|
model = MODELS[provider]
|
|
if not model["enabled"]:
|
|
reasons[provider] = "unverified_output_capacity"
|
|
continue
|
|
try:
|
|
initial = capacity(invocation("proposal_a", request), provider, count)
|
|
# Even a minimum-size reconciliation must fit; its real proposals and
|
|
# every subsequent request get checked again before any model call.
|
|
capacity(invocation("reconciliation", request), provider, count, 1)
|
|
except Problem as exc:
|
|
reasons[provider] = exc.code
|
|
continue
|
|
return {"provider": provider, **model, **initial, **reasoning_selection(request, provider),
|
|
"configuration_revision": REVISION,
|
|
"prompt_revision": PROMPT_REVISION, "prompt_sha256": PROMPT_SHA256,
|
|
"execution_revision": EXECUTION_REVISION, "policy_revision": POLICY_REVISION,
|
|
"case_count": count, "source_sha256": digest(request["cases"]),
|
|
"minimum_model_passes": 3, "maximum_model_passes": MAX_PASSES,
|
|
"max_final_group_cases": MAX_GROUP, "max_final_name_characters": 64,
|
|
"later_pass_capacity_verified": False, "later_pass_checks": "before_each_invocation"}
|
|
raise Problem("capacity_or_unsupported_backend", 422, candidates=reasons)
|
|
|
|
|
|
def sum_usage(records):
|
|
"""Aggregate measured counts only; keep an unknown counter unknown."""
|
|
values = [r.get("usage") for r in records]
|
|
if not values or any(not isinstance(v, dict) for v in values):
|
|
return None
|
|
keys = set().union(*(v.keys() for v in values))
|
|
result = {}
|
|
for key in keys:
|
|
counts = [v.get(key) for v in values]
|
|
if all(type(n) in (int, float) for n in counts):
|
|
result[key] = sum(counts)
|
|
elif all(isinstance(n, dict) for n in counts):
|
|
result[key] = sum_usage([{"usage": n} for n in counts])
|
|
else:
|
|
result[key] = None
|
|
return result
|
|
|
|
|
|
class Workflow:
|
|
"""Keep content in memory and allow only one pinned provider across model passes."""
|
|
|
|
def __init__(self, request, selected, cancel, client_ip, progress=None):
|
|
self.request, self.provider, self.cancel = request, selected["provider"], cancel
|
|
self.reasoning = selected.get("reasoning", MODELS[self.provider]["reasoning"])
|
|
self.client_ip, self.progress = client_ip, progress or (lambda _: None)
|
|
self.started = time.monotonic()
|
|
self.deadline = self.started + request["execution"]["max_seconds"]
|
|
self.cost_limit, self.spent = request["execution"]["max_cost_usd"], 0.0
|
|
self.records = []
|
|
self.last_metadata = {}
|
|
self.stage = "proposal_a"
|
|
|
|
def checkpoint(self):
|
|
"""Fail closed on cancellation or exhaustion of the shared job budgets."""
|
|
if self.cancel.is_set():
|
|
raise Problem("cancelled", 409)
|
|
if time.monotonic() >= self.deadline:
|
|
raise Problem("job_time_budget_exhausted", 504)
|
|
if self.spent > self.cost_limit:
|
|
raise Problem("job_cost_budget_exhausted", 502)
|
|
|
|
def call(self, stage, source, context=None, family_count=None):
|
|
"""Run one fresh complete-input invocation, retaining metadata before validation."""
|
|
self.stage = stage
|
|
self.checkpoint()
|
|
if len(self.records) >= MAX_PASSES:
|
|
raise Problem("model_pass_limit", 502)
|
|
call = invocation(stage, source, context)
|
|
# Keep the preflight choice across independent ordering and larger review prompts.
|
|
call["reasoning"] = self.reasoning
|
|
limits = capacity(call, self.provider, len(source["cases"]), family_count)
|
|
remaining_seconds = self.deadline - time.monotonic()
|
|
remaining_cost = self.cost_limit - self.spent
|
|
if remaining_cost <= 0:
|
|
raise Problem("job_cost_budget_exhausted", 502)
|
|
effective = {**source, "execution": {**source["execution"],
|
|
"max_seconds": remaining_seconds, "max_cost_usd": remaining_cost}}
|
|
started = time.monotonic()
|
|
def report(activity=None):
|
|
now = time.monotonic()
|
|
self.progress({"current_pass": stage, "completed_model_passes": len(self.records),
|
|
"maximum_model_passes": MAX_PASSES, "passes": self.records,
|
|
"heartbeat_at": time.time(), "pass_elapsed_seconds": round(now - started, 1),
|
|
"job_elapsed_seconds": round(now - self.started, 1),
|
|
"job_remaining_seconds": round(max(0, self.deadline - now), 1),
|
|
"cost_used_usd_estimate": round(self.spent, 8),
|
|
"cost_limit_usd_estimate": self.cost_limit, **(activity or {})})
|
|
report()
|
|
if self.provider == "claude":
|
|
value, metadata = suite_backends.claude_generate(effective, self.cancel, invocation=call, progress=report)
|
|
elif self.provider == "local":
|
|
value, metadata = suite_backends.local_generate(effective, self.cancel, self.client_ip, invocation=call)
|
|
else:
|
|
raise Problem("unsupported_backend", 422)
|
|
cost = metadata.get("cost_usd_estimate") if self.provider == "claude" else 0.0
|
|
record = {"stage": stage, "provider": self.provider, "model": MODELS[self.provider]["model"],
|
|
"reasoning": self.reasoning,
|
|
"wall_seconds": round(time.monotonic() - started, 3), **limits,
|
|
"system_sha256": hashlib.sha256(call["system"].encode()).hexdigest(),
|
|
"schema_sha256": digest(call["schema"]),
|
|
"case_order_sha256": digest([c["alias"] for c in source["cases"]]),
|
|
"allocated_seconds": remaining_seconds, "allocated_cost_usd": remaining_cost,
|
|
"usage": metadata.get("usage"), "cli_turns": metadata.get("turns"),
|
|
"duration_api_ms": metadata.get("duration_api_ms"),
|
|
"cost_usd_estimate": cost, "cli_diagnostics": metadata.get("cli_diagnostics")}
|
|
self.records.append(record)
|
|
self.last_metadata = metadata
|
|
if type(cost) not in (int, float) or not math.isfinite(cost) or cost < 0:
|
|
raise Problem("budget_accounting_unavailable", 502)
|
|
self.spent += cost
|
|
self.checkpoint()
|
|
report({"cli_running": False})
|
|
return decode(value, call, source)
|
|
|
|
def execute(self):
|
|
"""Perform independent discovery, reconciliation, and bounded semantic reviews."""
|
|
source = ordered_request(self.request)
|
|
a = self.call("proposal_a", source)
|
|
validate_partition(a, source, name_limit=BASE_NAME_LIMIT, unique_names=False)
|
|
b = self.call("proposal_b", ordered_request(source, True))
|
|
validate_partition(b, source, name_limit=BASE_NAME_LIMIT, unique_names=False)
|
|
count = max(len(a["groups"]), len(b["groups"]))
|
|
reconciled = self.call("reconciliation", source, {"proposal_a": a, "proposal_b": b}, count)
|
|
validate_natural(reconciled, source)
|
|
large = [g for g in reconciled["groups"] if len(g["members"]) > MAX_GROUP]
|
|
reviewed = audited = None
|
|
if large:
|
|
reviewed = self.call("large_family_review", source,
|
|
{"natural_partition": reconciled, "oversized_families": [g["members"] for g in large]},
|
|
len(reconciled["groups"]))
|
|
validate_natural(reviewed, source, reconciled["groups"])
|
|
audited = self.call("decision_audit", source,
|
|
{"original_partition": reconciled, "reviewed_partition": reviewed,
|
|
"oversized_families": [g["members"] for g in large]},
|
|
len(reviewed["groups"]))
|
|
validate_natural(audited, source, reconciled["groups"])
|
|
natural = audited or reconciled
|
|
final, divisions = cap_families(natural, source)
|
|
review = review_summary(a, b, reconciled, reviewed, audited, final, divisions)
|
|
self.checkpoint()
|
|
metadata = {**self.last_metadata, "passes": self.records,
|
|
"reasoning": self.reasoning,
|
|
"model_pass_count": len(self.records), "usage": sum_usage(self.records),
|
|
"cli_diagnostics_scope": "last_model_pass",
|
|
"duration_api_ms": sum(r["duration_api_ms"] for r in self.records) if all(type(r["duration_api_ms"]) in (int, float) for r in self.records) else None,
|
|
"cost_usd_estimate": round(self.spent, 8),
|
|
"turns": sum(r["cli_turns"] for r in self.records) if all(type(r["cli_turns"]) is int for r in self.records) else None,
|
|
"natural_family_count": len(natural["groups"]), "final_task_count": len(final["groups"]),
|
|
"singleton_statistics": {k: v for k, v in review["counts"].items() if "singleton" in k},
|
|
"policy_revision": POLICY_REVISION, "review_summary": review}
|
|
if len(encoded({"result": final, **metadata})) > MAX_RESULT - 16384:
|
|
raise Problem("response_too_large", 502)
|
|
return final, metadata
|
|
|
|
|
|
def generate(request, selected, cancel, client_ip, progress=None):
|
|
"""Never expose partial partitions as completion or broaden a failed route."""
|
|
workflow = Workflow(request, selected, cancel, client_ip, progress)
|
|
try:
|
|
return workflow.execute()
|
|
except Problem as exc:
|
|
details = {"failure_stage": "multi_pass_orchestration", **exc.details, "review_pass": workflow.stage,
|
|
"completed_model_passes": len(workflow.records), "passes": workflow.records,
|
|
"aggregate_usage": sum_usage(workflow.records + ([{"usage": exc.details["usage"]}] if isinstance(exc.details.get("usage"), dict) else [])),
|
|
"aggregate_cost_usd_estimate": None if exc.code == "budget_accounting_unavailable" else round(workflow.spent, 8)}
|
|
if exc.details.get("cost_usd_estimate") is not None:
|
|
details["aggregate_cost_usd_estimate"] += exc.details["cost_usd_estimate"]
|
|
raise Problem(exc.code, exc.status, **details) from None
|