Files
app/backend/tests/flow/test_resources.py
T
stroblmeandClaude Opus 5 058f16ec1d Close eight open SDK tasks: the pidfile, the log, cards, names and a live curve
Each was a loose end recorded under `### SDK` in the notepad.

`serve` takes its own pidfile down on SIGTERM. uvicorn restores the handler it
found and re-raises the signal it stopped on, so the default handler ended the
process without unwinding and the `finally` never ran — which is what a stop
sends, and what left `serve.pid` behind.

`serve.log` is cut back past 5 MB by the engine rather than by the screen that
started it, so an adopted engine is bounded too. Gated on its own stdout being
an appended regular file, which is what makes the cut safe: the kernel then
puts the next write at the new end.

Cards are counted from `/dev/nvidia[0-9]*`, so `FLOW_GPUS`/`--gpus` of 0 means
"work it out" the way `FLOW_CPUS` always has. The engine counts, not the
accountant — a remote worker builds one of those from its own inventory, and
detecting there would hand it the engine host's cards. The worker counts last:
what a batch job says it was granted still wins.

`GET /runs/metrics/names` is the distinct over a selection that `--list` and
the terminal's metric picker were approximating by reading the newest run that
had measured anything, which missed a name only an older run ever wrote.

`MetricSink` announces each batch it has written (`run_metric`, carrying the
names). Not a per-point event: one covers up to 500 points or two seconds of
them, and the rows stay the record. The terminal comparison fills in as the
first readings land instead of staying blank until reopened, and the browser
refetches the run and any comparison rather than the list behind them.

`retry --group` pages the list route by `before` instead of stopping at 500.

The terminal dashboard takes the terminal's colours (`ansi-dark`), and the web
UI can re-pair from Settings without disconnecting first.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PRQ9bmTvCbqCwXo9mxZzzV
2026-09-02 16:40:51 +02:00

208 lines
7.4 KiB
Python

"""One machine's books: what it holds, and what is free of it.
The failure this exists for is not subtle — five concurrent nodes each sizing a
thread pool to every core starved the engine's own event loop, and three GPU
processes each preallocating most of the card deadlocked at zero utilisation.
Both come down to arithmetic nobody was doing, so the arithmetic is what is
checked here. Which machine a node goes to, and the waiting, is
``test_placement.py``.
"""
import os
import pytest
from fluksio.flow.resources import (
THREAD_VARS,
Allocation,
ResourceAccountant,
derive_env,
fair_share_env,
machine_gpus,
)
from fluksio.flow.schemas import NodeDef, Resources
def test_the_cards_are_counted_from_the_devices_the_driver_makes(monkeypatch):
"""A box with cards and an engine told nothing used to have none of them.
NVIDIA's own device nodes, counted — no vendor tool, so the one dependency
does not become two. The control files beside them are not cards.
"""
from fluksio.flow import resources
monkeypatch.setattr(
resources.glob,
"glob",
lambda pattern: (
["/dev/nvidia0", "/dev/nvidia1"] if pattern == "/dev/nvidia[0-9]*" else []
),
)
assert machine_gpus() == 2
monkeypatch.setattr(resources.glob, "glob", lambda pattern: [])
assert machine_gpus() == 0
def test_what_is_free_is_what_was_handed_out():
accountant = ResourceAccountant(cpus=4, gpus=0)
held = accountant.try_take(cpus=3, gpus=0)
assert accountant.snapshot()["cpus"] == {"total": 4, "free": 1}
accountant.give_back(held)
assert accountant.snapshot()["cpus"] == {"total": 4, "free": 4}
def test_asking_for_more_than_is_free_is_answered_not_waited_on():
"""The books never block: the placer has other machines to try first."""
accountant = ResourceAccountant(cpus=2)
assert accountant.try_take(cpus=2, gpus=0) is not None
assert accountant.try_take(cpus=2, gpus=0) is None
def test_a_gpu_is_held_by_one_node_at_a_time():
"""The deadlock was three processes each preallocating most of one card."""
accountant = ResourceAccountant(cpus=8, gpus=2)
first = accountant.try_take(cpus=1, gpus=1)
second = accountant.try_take(cpus=1, gpus=1)
assert set(first.gpus) & set(second.gpus) == set()
assert accountant.snapshot()["gpus"] == {"total": 2, "free": 0}
assert accountant.try_take(cpus=1, gpus=1) is None
accountant.give_back(first)
accountant.give_back(second)
assert accountant.snapshot()["gpus"] == {"total": 2, "free": 2}
def test_what_a_machine_could_ever_grant_is_a_different_question():
"""Busy is worth queueing for; too small is not, and reads the same."""
accountant = ResourceAccountant(cpus=4, gpus=1)
assert accountant.fits(cpus=4, gpus=1)
assert not accountant.fits(cpus=8, gpus=0)
assert not accountant.fits(cpus=1, gpus=2)
accountant.try_take(cpus=4, gpus=1)
# Still true with nothing free: it is about the machine, not the moment.
assert accountant.fits(cpus=4, gpus=1)
def test_a_machine_that_said_nothing_about_memory_is_not_a_machine_with_none():
said = ResourceAccountant(cpus=4, ram_mb=2048)
assert said.fits(cpus=1, gpus=0, ram_mb=2048)
assert not said.fits(cpus=1, gpus=0, ram_mb=4096)
held = said.try_take(cpus=1, gpus=0, ram_mb=1536)
assert said.try_take(cpus=1, gpus=0, ram_mb=1024) is None
said.give_back(held)
assert said.snapshot()["ram_mb"] == {"total": 2048, "free": 2048}
quiet = ResourceAccountant(cpus=4)
assert quiet.fits(cpus=1, gpus=0, ram_mb=999_999)
assert quiet.try_take(cpus=1, gpus=0, ram_mb=999_999) is not None
assert quiet.snapshot()["ram_mb"] is None
def test_a_release_says_so_once_it_has_let_go():
"""The placer is told outside the lock, which is what keeps the order one-way."""
seen: list[dict] = []
accountant = ResourceAccountant(cpus=2)
accountant.on_release = lambda: seen.append(accountant.snapshot()["cpus"])
accountant.give_back(accountant.try_take(cpus=2, gpus=0))
assert seen == [{"total": 2, "free": 2}]
# -----------------------------------------------------------------------------
# What the worker is started with
# -----------------------------------------------------------------------------
def test_the_share_becomes_the_thread_limit():
env = derive_env(Resources(cpus=3), Allocation(cpus=3))
assert all(env[var] == "3" for var in THREAD_VARS)
assert "CUDA_VISIBLE_DEVICES" not in env
def test_a_gpu_node_is_told_which_card_is_its():
env = derive_env(Resources(cpus=1, gpus=2), Allocation(cpus=1, gpus=(1, 3)))
assert env["CUDA_VISIBLE_DEVICES"] == "1,3"
def test_what_the_node_asked_for_wins():
"""A declaration outranks what the allocation would imply — it is deliberate."""
wanted = Resources(
cpus=4, env={"OMP_NUM_THREADS": "1", "XLA_FLAGS": "--xla_cpu_x=false"}
)
env = derive_env(wanted, Allocation(cpus=4))
assert env["OMP_NUM_THREADS"] == "1"
assert env["XLA_FLAGS"] == "--xla_cpu_x=false"
assert env["MKL_NUM_THREADS"] == "4"
def test_the_shared_pool_divides_what_it_has():
env = fair_share_env(cpus=8, workers=4)
assert all(env[var] == "2" for var in THREAD_VARS)
# Never zero, however many workers there are.
assert fair_share_env(cpus=2, workers=8)["OMP_NUM_THREADS"] == "1"
def test_an_operator_who_set_one_keeps_it(monkeypatch):
"""An explicit value in the engine's environment is an answer, not a default."""
monkeypatch.setitem(os.environ, "OMP_NUM_THREADS", "2")
env = fair_share_env(cpus=16, workers=2)
assert "OMP_NUM_THREADS" not in env
assert env["MKL_NUM_THREADS"] == "8"
# -----------------------------------------------------------------------------
# Declaration
# -----------------------------------------------------------------------------
def test_declaring_nothing_stays_exactly_as_it_was():
"""Every flow that exists today parses unchanged and is not accounted for."""
assert NodeDef(id="poll").resources is None
def test_a_misspelled_resource_is_refused():
with pytest.raises(ValueError):
Resources(cpu=4)
# -----------------------------------------------------------------------------
# Sizes with names, and sizes written out
# -----------------------------------------------------------------------------
def test_a_size_can_be_written_the_way_people_write_sizes():
assert Resources(ram="2G").ram == 2048
assert Resources(ram="512M").ram == 512
assert Resources(ram="512").ram == 512
assert Resources(ram=512).ram == 512
with pytest.raises(ValueError, match="not a size"):
Resources(ram="a lot")
def test_a_duration_is_one_unit_and_says_so_when_it_is_not():
assert Resources(duration_s="2h").duration_s == 7200
assert Resources(duration_s="30m").duration_s == 1800
assert Resources(duration_s="90").duration_s == 90
with pytest.raises(ValueError, match="not a duration"):
Resources(duration_s="1h30m")
def test_a_flavor_and_a_number_for_the_same_thing_is_two_answers():
with pytest.raises(ValueError, match="already says how much"):
Resources(flavor="gpu-small", cpus=4)
# The defaults are not an answer: an editor writing the whole object back
# sends them, and that must survive the round trip.
assert Resources(flavor="gpu-small", cpus=1, gpus=0).flavor == "gpu-small"
# Neither is a duration, which a flavor says nothing about.
assert Resources(flavor="gpu-small", duration_s="2h").duration_s == 7200