Files
Gökay Aydoğan 75a4ce6324 Platform pack (#76)
* feat: v2.1.0 — fal platform utilities (async fan-out, cost tools, recovery)

Six new nodes under FAL/Platform, built on fal's queue/request-id/pricing
primitives:

- Fal Submit + Fal Collect: queue jobs without blocking and collect
  results later — N generations run in parallel inside one graph
- Fal Result by Request ID: re-fetch any past generation from fal's
  queue without re-generating or re-billing
- Fal Cost Estimator: per-run cost report parsed from registry pricing
  before anything is queued
- Fal Session Costs: in-memory ledger of every fal call this session
  (endpoint, duration, request id, est. cost) with running total
- Fal Save Media from URL: persist expiring CDN results into the
  ComfyUI output directory

Core: utils/pricing.py (pricing-text parser + registry lookup),
utils/ledger.py (thread-safe session ledger), api.py gains submit_only,
result_from_request_id, request-id capture via on_enqueue, and
automatic ledger recording; any_endpoint helpers extracted for reuse.

16 new tests (61 total).

* feat: v2.2.0 — persistent result cache, spend guard, URL passthrough

Three fal-platform features on top of the 2.1 platform pack:

- Persistent result cache (sqlite, [cache] config): identical fal calls
  are served free from disk across ComfyUI restarts; uploads are
  deduplicated by content hash so identical inputs upload once. Bypass
  per node via force_rerun. Cache hits are never blocked by the spend
  guard and never recorded as spend.
- Spend Guard + billing ([spend_guard] config): preflight refuses to
  submit once the session's estimated spend reaches session_budget_usd
  or the account balance (fal Billing API) falls below min_balance_usd.
  New Fal Account Balance node; ledger reconciliation against the
  aggregated usage API (fal exposes no per-request billed amounts).
- fal-CDN URL passthrough: every dynamic node's media inputs gain
  optional *_direct_url twins (URL sent directly, no download/re-upload;
  1,405 twins across 992 nodes) and image-kind nodes output image_urls
  alongside IMAGE — fal→fal chains with zero local I/O.

Review fixes: path traversal blocked in FalSaveMediaURL (confined to the
output dir, atomic O_EXCL filename claim), FalAnyEndpoint force_rerun now
bypasses the cache, cache lookup ordered before spend-guard preflight,
result recovery no longer counts as new session spend.

14 new tests (76 total).
2026-07-02 14:54:07 +03:00

57 lines
1.4 KiB
Python

"""Unit tests for the session cost ledger."""
from __future__ import annotations
import importlib
import threading
import pytest
from conftest import PKG, _load_package
@pytest.fixture()
def ledger(pricing_mod=None):
_load_package()
mod = importlib.import_module(f"{PKG}.nodes.utils.ledger")
instance = mod.SessionLedger()
instance.reset()
yield instance
instance.reset()
def test_record_and_totals(ledger):
ledger.record("fal-ai/a", "req-1", 2.5, 0.10)
ledger.record("fal-ai/b", "req-2", 1.0, None)
entries = ledger.entries()
assert len(entries) == 2
assert ledger.total_cost() == pytest.approx(0.10)
assert ledger.unknown_cost_count() == 1
def test_report_mentions_calls(ledger):
ledger.record("fal-ai/kling-video/v3/pro/image-to-video", "abc123", 12.4, 0.35)
report = ledger.report()
assert "kling" in report
assert "abc123" in report
def test_reset(ledger):
ledger.record("fal-ai/a", None, 1.0, 0.5)
ledger.reset()
assert ledger.entries() == []
assert ledger.total_cost() == 0.0
def test_thread_safety(ledger):
def worker():
for _ in range(100):
ledger.record("fal-ai/t", None, 0.1, 0.01)
threads = [threading.Thread(target=worker) for _ in range(10)]
for t in threads:
t.start()
for t in threads:
t.join()
assert len(ledger.entries()) == 1000
assert ledger.total_cost() == pytest.approx(10.0)