def test_dag_dump() -> None:
"""Test simple DAG dump to file."""
with Fun(MockServer(), options(distributed=False)) as db:
dat = put(b"bla bla")
dat2 = put(b"blaXbla")
errorstep = morph(raises, dat2)
step1 = morph(upper, dat)
step2 = shell("cat file1 file2", inp=dict(file1=step1, file2=dat))
step2b = utils.concat(step2.stdout, errorstep, strict=False)
step3 = shell("cat file1", inp=dict(file1=step2b))
step4 = shell("cat file1", inp=dict(file1=step1))
step4b = shell("cat file2", inp=dict(file2=step4.stdout))
out = utils.concat(step1, dat, step2.stdout, step3.stdout)
_dag.build_dag(db, out.hash)
execute(step2b)
execute(step4b)
wait_for(step4b, 1.0)
reset(step4)
nodes, artefacts, labels, links = _graphviz.export(
db, [out.hash, step4b.hash])
dot = _graphviz.format_dot(nodes, artefacts, labels, links,
[out.hash, step4b.hash])
assert len(dot) > 0
assert len(nodes) == 8
assert len(labels) == 8
# TODO pass through dot for testing?
with open("g.dot", "w") as f:
f.write(dot)
def test_double_execution(nworkers: int) -> None:
"""Test multiple executions of the same task."""
# This test will fail if a job is re-executed multiple times.
# external
from rq.job import get_current_job
def track_runs(inp: bytes) -> bytes:
job = get_current_job()
db: Redis[bytes] = job.connection
val = db.incrby("sentinel", 1)
time.sleep(0.5)
return str(val).encode()
with f.ManagedFun(nworkers=nworkers):
# wait_for_workers(db, nworkers)
dat = f.put(b"bla bla")
step1 = f.morph(track_runs, dat)
step1a = f.shell(
"cat file1",
inp=dict(file1=step1),
)
step1b = f.shell(
"cat file2",
inp=dict(file2=step1),
)
f.execute(step1a)
f.execute(step1b)
f.wait_for(step1a, timeout=10.0)
f.wait_for(step1b, timeout=10.0)
assert f.take(step1a.stdout) == b"1"
def test_worker_killed(nworkers: int, sig: int) -> None:
"""Test what happens when 'funsies worker' gets killed."""
# std
import os
def kill_funsies_worker(*inp: bytes) -> bytes:
pid = os.getppid()
os.kill(pid, sig)
time.sleep(1.0)
return b"what"
def cap(inp: bytes) -> bytes:
return inp.upper()
with f.ManagedFun(nworkers=nworkers) as db:
wait_for_workers(db, nworkers)
s1 = f.reduce(kill_funsies_worker,
b"bla bla",
b"bla bla",
opt=f.options(timeout=5))
s1b = f.morph(cap, s1)
f.execute(s1b)
if nworkers == 1:
# no other workers to pick up the slack
with pytest.raises(TimeoutError):
f.wait_for(s1b, timeout=1)
else:
# everything is ok
f.wait_for(s1b, timeout=5)
assert f.take(s1b) == b"WHAT"
def test_job_killed(nworkers: int, sig: int) -> None:
"""Test what happens when 'funsies worker' is ok but its job gets killed."""
# std
import os
def kill_self(*inp: bytes) -> bytes:
pid = os.getpid()
os.kill(pid, sig)
time.sleep(2.0)
return b"what"
def cap(inp: bytes) -> bytes:
return inp.upper()
with f.ManagedFun(nworkers=nworkers) as db:
wait_for_workers(db, nworkers)
s1 = f.reduce(kill_self,
b"bla bla",
b"bla bla",
opt=f.options(timeout=3))
s1b = f.morph(cap, s1)
f.execute(s1b)
# error
f.wait_for(s1b, timeout=1)
err = f.take(s1b, strict=False)
assert isinstance(err, f.errors.Error)
assert err.kind == f.errors.ErrorKind.KilledBySignal
def test_data_race(nworkers: int) -> None:
"""Test a data race when execute calls are interleaved."""
with f.ManagedFun(nworkers=nworkers):
dat = f.put(b"bla bla")
step1 = f.morph(lambda x: x.decode().upper().encode(), dat)
step2 = f.shell(
"cat file1 file2; grep 'bla' file2 file1 > file3; date >> file3",
inp=dict(file1=step1, file2=dat),
out=["file2", "file3"],
)
f.execute(step1)
f.execute(step2)
f.wait_for(step1, timeout=20.0)
f.wait_for(step2, timeout=20.0)
def test_dynamic_dump() -> None:
"""Test whether a dynamic DAG gets graphed properly."""
def split(a: bytes, b: bytes) -> list[dict[str, int]]:
a = a.split()
b = b.split()
out = []
for ia, ib in zip(a, b):
out += [{
"sum": int(ia.decode()) + int(ib.decode()),
"product": int(ia.decode()) * int(ib.decode()),
}]
return out
def apply(inp: Artefact[dict[str, Any]]) -> Artefact[str]:
out = funsies.morph(lambda x: f"{x['sum']}//{x['product']}", inp)
return out
def combine(inp: Sequence[Artefact[str]]) -> Artefact[bytes]:
def enc(inp: str) -> bytes:
return inp.encode()
out = [funsies.morph(enc, x, out=Encoding.blob) for x in inp]
return funsies.utils.concat(*out)
with funsies.ManagedFun(nworkers=1) as db:
num1 = funsies.put(b"1 2 3 4 5")
num2 = funsies.put(b"11 10 11 10 11")
outputs = dynamic.sac(
split,
apply,
combine,
num1,
num2,
out=Encoding.blob,
)
outputs = funsies.morph(lambda x: x, outputs)
nodes, artefacts, labels, links = _graphviz.export(db, [outputs.hash])
assert len(artefacts) == 4 # not yet generated subdag parents
print(len(artefacts))
funsies.execute(outputs)
funsies.wait_for(outputs, timeout=1.0)
nodes, artefacts, labels, links = _graphviz.export(db, [outputs.hash])
assert len(artefacts) == 22 # generated subdag parents
assert funsies.take(outputs) == b"12//1112//2014//3314//4016//55"
def test_artefact_disk_distributed() -> None:
"""Test whether artefacts on disk works on different nodes."""
# funsies
import funsies as f
with tempfile.TemporaryDirectory() as td:
with f.ManagedFun(nworkers=1, data_url=f"file://{td}"):
dat = f.put(b"bla bla")
step1 = f.morph(lambda x: x.decode().upper().encode(), dat)
step2 = f.shell("cat file1 file2",
inp=dict(file1=step1, file2=dat))
step2b = f.shell("cat file1", inp=dict(file1=step1))
f.execute(step2)
f.wait_for(step2, 1.0)
out = f.take(step2.stdout)
assert out == b"BLA BLAbla bla"
f.execute(step2b)
f.wait_for(step2b, 1.0)
out = f.take(step2b.stdout)
assert out == b"BLA BLA"
def test_waiting_on_map_reduce() -> None:
"""Test waiting on the (linked) result of map-reduce."""
def split(a: bytes, b: bytes) -> list[dict[str, int]]:
a = a.split()
b = b.split()
out = []
for ia, ib in zip(a, b):
out += [{
"sum": int(ia.decode()) + int(ib.decode()),
"product": int(ia.decode()) * int(ib.decode()),
}]
return out
def apply(inp: Artefact) -> Artefact:
out = funsies.morph(lambda x: f"{x['sum']}//{x['product']}", inp)
return out
def combine(inp: Sequence[Artefact]) -> Artefact:
out = [
funsies.morph(lambda y: y.encode(), x, out=Encoding.blob)
for x in inp
]
return funsies.utils.concat(*out)
with funsies.ManagedFun(nworkers=1):
num1 = funsies.put(b"1 2 3 4 5")
num2 = funsies.put(b"11 10 11 10 11")
outputs = dynamic.sac(
split,
apply,
combine,
num1,
num2,
out=Encoding.blob,
)
funsies.execute(outputs)
funsies.wait_for(outputs, timeout=1.0)
assert funsies.take(outputs) == b"12//1112//2014//3314//4016//55"
def test_raising_funsie() -> None:
"""Test funsie that raises an error.
This test is specifically designed to catch the bug fixed in fa9af6a4
where funsies that raised did not release their locks, leading to a race
condition.
"""
def raising_fun(*inp: str) -> bytes:
raise RuntimeError("this funsie raises.")
with f.ManagedFun(nworkers=2):
s0a = f.morph(lambda x: x, "bla blabla")
s0b = f.morph(lambda x: x, "blala")
s1 = f.reduce(raising_fun, "bla bla", s0a, s0b, strict=True)
f.execute(s1)
f.wait_for(s1, timeout=2)
with pytest.raises(UnwrapError):
_ = f.take(s1)
s2 = f.morph(lambda x: x, s1)
f.execute(s2)
f.wait_for(s2, timeout=0.5)
def test_timeout_deadlock() -> None:
"""Test funsies that time out.
Here we explicitly check if dependents are still enqueued or if the whole
thing deadlocks.
"""
def timeout_fun(*inp: str) -> bytes:
time.sleep(3.0)
return b"what"
def cap(inp: bytes) -> bytes:
return inp.capitalize()
with f.ManagedFun(nworkers=2):
# Test when python function times out
s1 = f.reduce(timeout_fun,
"bla bla",
"bla bla",
opt=f.options(timeout=1))
s1b = f.morph(cap, s1)
# Test when shell function times out
s2 = f.shell("sleep 20", "echo 'bla bla'", opt=f.options(timeout=1))
s2b = f.morph(cap, s2.stdouts[1])
f.execute(s1b, s2b)
# Check err for reduce
f.wait_for(s1b, timeout=1.5)
err = f.take(s1b, strict=False)
assert isinstance(err, f.errors.Error)
assert err.kind == f.errors.ErrorKind.JobTimedOut
assert err.source == s1.parent
# Check err for shell
f.wait_for(s2b, timeout=1.5)
err = f.take(s2b, strict=False)
assert isinstance(err, f.errors.Error)
assert err.kind == f.errors.ErrorKind.JobTimedOut
assert err.source == s2.hash
def test_nested_map_reduce(nworkers: int) -> None:
"""Test nested map-reduce."""
# ------------------------------------------------------------------------
# Inner
def sum_inputs(*inp: int) -> int:
out = 0
for el in inp:
out += el
return out
def split_inner(inp: str) -> list[int]:
a = inp.split(" ")
return [int(el) for el in a]
def apply_inner(inp: Artefact) -> Artefact:
return funsies.reduce(sum_inputs, inp, 1)
def combine_inner(inp: Sequence[Artefact]) -> Artefact:
return funsies.reduce(sum_inputs, *inp)
# ------------------------------------------------------------------------
# outer
def split_outer(inp: list[str], fac: int) -> list[str]:
out = [x + f" {fac}" for x in inp]
return out
def apply_outer(inp: Artefact) -> Artefact:
outputs = dynamic.sac(
split_inner,
apply_inner,
combine_inner,
inp,
out=Encoding.json,
)
return outputs
def combine_outer(inp: Sequence[Artefact]) -> Artefact:
out = [
funsies.morph(lambda y: f"{y}".encode(), x, out=Encoding.blob)
for x in inp
]
return funsies.utils.concat(*out, join=b",,")
with funsies.ManagedFun(nworkers=nworkers):
num1 = funsies.put("1 2 3 4 5")
outputs = dynamic.sac(split_inner,
apply_inner,
combine_inner,
num1,
out=Encoding.json)
funsies.execute(outputs)
funsies.wait_for(outputs, timeout=30.0)
assert funsies.take(outputs) == 20
# Now try the nested one
num = funsies.put(["1 2", "3 4 7", "10 12", "1"])
factor = funsies.put(-2)
# split -> 1 2 -2|3 4 7 -2|10 12 -2| 1 -2
# apply -> split2 -> 1, 2,-2 | 3,4,7,-2|10,12,-2|1,-2
# apply2 -> 2, 3,-1 | 4,5,8,-1|11,13,-1|2,-1
# combine2 -> 4|16|23|1
# combine -> 4,,16,,23,,1
ans = b"4,,16,,23,,1"
outputs = dynamic.sac(
split_outer,
apply_outer,
combine_outer,
num,
factor,
out=Encoding.blob,
)
funsies.execute(outputs)
funsies.wait_for(outputs, timeout=30.0)
assert funsies.take(outputs) == ans
# abort if molecule is empty
struct = not_empty(gen3d.out["struct.mol"])
# Generate conformers.
confab = f.shell(
"obabel input.mol -O conformers.xyz --confab --verbose",
inp={"input.mol": struct},
out=["conformers.xyz"],
)
# Optimize conformer ensemble using xtb.
optimized1 = f.dynamic.sac(
# split the xyz file
split_xyz,
# optimize the conformers
optimize_conformer_xtb,
# join the structures
make_data_output,
# input data:
confab.out["conformers.xyz"],
# output is json
out=Encoding.json,
)
# execute the workflow
f.execute(optimized1)
# block until its over
f.wait_for(optimized1)
# dump results for analysis
f.takeout(optimized1, "conformers.json")
def test_integration(reference: str, nworkers: int) -> None:
"""Test full integration."""
# make a temp file and copy reference database
dir = tempfile.mkdtemp()
if not make_reference:
shutil.copy(os.path.join(ref_dir, reference, "appendonly.aof"), dir)
shutil.copytree(os.path.join(ref_dir, reference, "data"),
os.path.join(dir, "data"))
shutil.copy(os.path.join(ref_dir, "redis.conf"), dir)
# data url
datadir = f"file://{os.path.join(dir, 'data')}"
# Dictionary for test data
test_data: dict[str, Any] = {}
def update_data(a: dict[int, int], b: list[int]) -> dict[int, int]:
for i in b:
a[i] = a.get(i, 0) + 1
return a
def sum_data(x: dict[int, int]) -> int:
return sum([int(k) * v for k, v in x.items()])
def make_secret(x: int) -> str:
return secrets.token_hex(x)
# Start funsie script
with ManagedFun(
nworkers=nworkers,
directory=dir,
data_url=datadir,
redis_args=["redis.conf"],
) as db:
integers = put([5, 4, 8, 9, 9, 10, 1, 3])
init_data = put({100: 9})
test_data["init_data"] = init_data
nbytes = put(4)
s1 = reduce(update_data, init_data, integers)
num = morph(sum_data, s1)
date = shell("date").stdout
test_data["date"] = date
rand = morph(make_secret, nbytes)
s4 = template(
"date:{{date}}\n" + "some random bytes:{{random}}\n" +
"a number: {{num}}\n" + "a string: {{string}}\n",
{
"date": date,
"random": rand,
"num": num,
"string": "wazza"
},
name="a template",
)
test_data["s4"] = s4
execute(s4)
wait_for(s4, 5)
# check that the db doesn't itself include data
for k in db.keys():
assert b"data" not in k
if make_reference:
folder = os.path.join(ref_dir, reference)
os.makedirs(folder, exist_ok=True)
for name, artefact in test_data.items():
with open(os.path.join(folder, name), "wb") as f:
execute(artefact)
wait_for(artefact, 10.0)
out = take(artefact)
data2 = _serdes.encode(artefact.kind, out)
assert isinstance(data2, bytes)
f.write(data2)
shutil.copy(
os.path.join(dir, "appendonly.aof"),
os.path.join(folder, "appendonly.aof"),
)
shutil.copytree(
os.path.join(dir, "data"),
os.path.join(folder, "data"),
)
else:
# Test against reference dbs
for name, artefact in test_data.items():
execute(artefact)
wait_for(artefact, 10.0)
with open(os.path.join(ref_dir, reference, name), "rb") as f:
data = f.read()
out = take(artefact)
data_ref = _serdes.encode(artefact.kind, out)
assert isinstance(data_ref, bytes)
assert data == data_ref
# delete tempdir
shutil.rmtree(dir)
# Recursive application of merge sort
# split -> generates two list or raises
# recurse(x) for each values of split
# merge(left, right)
split,
lambda element: funsies_mergesort(element),
lambda lr: f.reduce(merge, lr[0], lr[1]), # type:ignore
art,
out=Encoding.json,
)
return f.reduce(
# if the subdag fails, it's because split raised. In this case, we
# just forward the arguments
ignore_error,
result,
art,
strict=False,
out=Encoding.json,
)
# run the workflow
to_be_sorted = [random.randint(0, 99) for _ in range(120)]
with f.Fun():
inp = f.put(to_be_sorted)
out = funsies_mergesort(inp)
print("output:", out.hash)
f.execute(out)
f.wait_for(out)
print(f.take(out))
def test_integration(reference: str, nworkers: int) -> None:
"""Test full integration."""
# make a temp file and copy reference database
dir = tempfile.mkdtemp()
if not make_reference:
shutil.copy(os.path.join(ref_dir, reference, "appendonly.aof"), dir)
shutil.copy(os.path.join(ref_dir, "redis.conf"), dir)
# Dictionary for test data
test_data = {}
# Start funsie script
with ManagedFun(nworkers=nworkers,
directory=dir,
redis_args=["redis.conf"]):
dat = put(b"bla bla")
step1 = morph(lambda x: x.decode().upper().encode(), dat)
step2 = shell(
"cat file1 file2; grep 'bla' file2 file1 > file3; date >> file3",
inp=dict(file1=step1, file2=dat),
out=["file2", "file3"],
)
echo = shell("sleep 1", "date")
merge = reduce(
join_bytes,
step2.out["file3"],
echo.stdouts[1],
name="merger",
)
def tolist(x: bytes, y: bytes) -> Dict[int, str]:
return {1: x.decode(), 8: y.decode()}
A = py(tolist, merge, echo.stdouts[1])
test_data["test1"] = A
def raises(inp: bytes) -> bytes:
raise RuntimeError("an error was raised")
def error_count(*inp: Result[bytes]) -> bytes:
out = utils.match_results(inp, lambda x: 0, lambda x: 1)
return str(sum(out)).encode()
err = morph(raises, dat)
count = reduce(error_count,
dat,
dat,
err,
dat,
err,
err,
echo.stdouts[0],
strict=False)
cat = utils.concat(merge,
dat,
err,
count,
echo.stdouts[1],
strict=False)
test_data["test2"] = cat
execute(step1)
wait_for(step1, timeout=10.0)
execute(step2)
wait_for(step2, timeout=10.0)
assert take(step1) == b"BLA BLA"
assert take(step2.stdout) == b"BLA BLAbla bla"
if make_reference:
folder = os.path.join(ref_dir, reference)
os.makedirs(folder, exist_ok=True)
for name, artefact in test_data.items():
with open(os.path.join(folder, name), "wb") as f:
execute(artefact)
wait_for(artefact, 10.0)
out = take(artefact)
data2 = _serdes.encode(artefact.kind, out)
assert isinstance(data2, bytes)
f.write(data2)
shutil.copy(
os.path.join(dir, "appendonly.aof"),
os.path.join(folder, "appendonly.aof"),
)
else:
# Test against reference dbs
for name, artefact in test_data.items():
execute(artefact)
wait_for(artefact, 10.0)
with open(os.path.join(ref_dir, reference, name), "rb") as f:
data = f.read()
out = take(artefact)
data_ref = _serdes.encode(artefact.kind, out)
assert isinstance(data_ref, bytes)
assert data == data_ref
shutil.rmtree(dir)