def test_gcs_text():
# set up
set_up()
w = BatchWriter(
inner_writer=GoogleCloudStorageWriter,
project="testing",
blob_size=1024,
format="jsonl",
dataset=f"{BUCKET_NAME}/test/gcs/dataset/text",
)
for i in range(250):
w.append({"index": i + 300})
w.finalize()
# read the files we've just written, we should be able to
# read over both paritions.
r = Reader(
inner_reader=GoogleCloudStorageReader,
project="testing",
dataset=f"{BUCKET_NAME}/test/gcs/dataset/text",
persistence=STORAGE_CLASS.MEMORY,
)
assert r.count() == 250, r
def test_index():
# step back through time
shutil.rmtree("_temp/data/tweets", ignore_errors=True)
r = Reader(inner_reader=DiskReader,
dataset="tests/data/tweets",
raw_path=True)
w = BatchWriter(inner_writer=DiskWriter,
dataset="_temp/data/tweets",
index_on=["username"])
for item in r:
w.append(item)
w.finalize()
index = glob.glob("_temp/data/tweets/**/*username.idx", recursive=True)
assert len(index) == 1, index
with open(index[0], "rb") as f:
idx = f.read()
# test the recently created index outside the reader
i = Index(io.BytesIO(idx))
assert i.search("SwiftOnSecurity") == []
assert i.search("BBCNews") == [1, 2, 4, 24, 25, 44], i.search("BBCNews")
# test the filter with an index
ri = Reader(
inner_reader=DiskReader,
dataset="_temp/data/tweets",
filters="username = '******'",
)
ri = list(ri)
assert len(ri) == 6
def test_reader_filters_single_filter():
"""ensure the reader filter is working as expected"""
r = Reader(
inner_reader=DiskReader,
dataset="tests/data/tweets/",
raw_path=True,
filters="username == 'NBCNews'",
persistence=STORAGE_CLASS.MEMORY,
)
assert r.count() == 44, r.count()
def test_reader_filters_multiple_filter():
"""ensure the reader filter is working as expected"""
r = Reader(
inner_reader=DiskReader,
dataset="tests/data/tweets/",
raw_path=True,
filters="username = '******' and timestamp >= '2020-01-12T07:11:04'",
persistence=STORAGE_CLASS.MEMORY,
)
assert r.count() == 34, r
def test_can_read_parquet():
r = Reader(
inner_reader=DiskReader,
dataset="tests/data/formats/parquet",
raw_path=True,
persistence=STORAGE_CLASS.MEMORY,
)
assert r.count() == 57581, r.count()
assert isinstance(r.first(), dict)
def test_reader_partitions_read_without_referring_to_partition():
"""
test if we reference a folder with partitions (by_) without referencing the
partition, we pick a partition and read it like it's not there
"""
DATA_DATE = datetime.date(2020, 2, 3)
records = Reader(
dataset="tests/data/partitioned",
inner_reader=DiskReader,
start_date=DATA_DATE,
end_date=DATA_DATE,
persistence=STORAGE_CLASS.MEMORY,
)
assert records.count() == 25, records.count()
def test_disk_text():
try:
w = BatchWriter(
inner_writer=DiskWriter,
blob_size=1024,
format="jsonl",
dataset=f"_temp/test/gcs/dataset/text",
)
for i in range(250):
w.append({"index": i + 300})
w.finalize()
# read the files we've just written, we should be able to
# read over both paritions.
r = Reader(
inner_reader=DiskReader,
dataset=f"_temp/test/gcs/dataset/text",
)
l = list(r)
assert len(l) == 250, len(l)
except Exception as e: # pragma: no cover
raise e
def test_gcs_parquet():
try:
# set up the stub
set_up()
w = BatchWriter(
inner_writer=GoogleCloudStorageWriter,
project="testing",
format="parquet",
dataset=f"{BUCKET_NAME}/test/gcs/dataset",
)
for i in range(100):
w.append({"$$": i * 300})
w.finalize()
# read the files we've just written, we should be able to
# read over both paritions.
r = Reader(
inner_reader=GoogleCloudStorageReader,
project="testing",
dataset=f"{BUCKET_NAME}/test/gcs/dataset",
)
l = list(r)
assert isinstance(l[0], dict)
assert len(l) == 100, len(l)
except Exception as e: # pragma: no cover
raise e
def test_reader_writer_format_parquet():
do_writer_compressed("parquet")
g = glob.glob("_temp/**/*.parquet", recursive=True)
assert len(g) > 0, g
r = Reader(inner_reader=DiskReader, dataset="_temp")
l = len(list(r))
shutil.rmtree("_temp", ignore_errors=True)
assert l == 200000, l
def test_reader_writer():
do_writer()
r = Reader(inner_reader=DiskReader, dataset="_temp")
l = len(list(r))
shutil.rmtree("_temp", ignore_errors=True)
assert l == 200000, l
def test_reader_filters_no_filter():
"""ensure the reader filter is working as expected"""
r = Reader(inner_reader=DiskReader,
dataset="tests/data/tweets/",
raw_path=True)
for index, item in enumerate(r):
pass
assert index == 49, index
def test_reader_writer_format_default():
do_writer_default()
g = glob.glob("_temp/**/*.zstd", recursive=True)
assert len(g) > 0, g
r = Reader(inner_reader=DiskReader, dataset="_temp")
l = len(list(r))
shutil.rmtree("_temp", ignore_errors=True)
assert l == 200000, l
def test_cursor():
"""
Test that when we break a read in two, we read the right amount of records.
"""
import json
test_counter = 0
number_of_records = get_records()
lim = number_of_records // 4 * 3
hashes = []
reader = Reader(inner_reader=DiskReader,
dataset="tests/data/tweets/",
partitions=[])
for row in reader["tweet"].take(lim):
hashes.append(hash(json.dumps(row)))
test_counter += 1
cursor = reader.cursor
print(cursor)
assert cursor["location"] == (
(lim - 1) % 25), f"{cursor['location']}, {lim}, {(lim % 25)}"
assert cursor["partition"] == 5122091051124077700, cursor["partition"]
reader = Reader(
inner_reader=DiskReader,
dataset="tests/data/tweets/",
partitions=[],
cursor=cursor,
)
for i, row in enumerate(reader["tweet"].take(100)):
hashes.append(hash(json.dumps(row)))
test_counter += 1
# we should have read the number of expected records
assert (number_of_records == test_counter
), f"{number_of_records} - {test_counter}, {i}"
# we shouldn't have captured any duplicates
assert len(hashes) == len(
set(hashes)), f"{len(hashes)} == {len(set(hashes))}"
def time_it(dataset, username):
start = time.perf_counter_ns()
reader = Reader(
inner_reader=DiskReader,
dataset=dataset,
raw_path=True,
filters=("user_name", "==", username),
)
res = [r for r in reader]
print(len(res))
return (time.perf_counter_ns() - start) / 1e9
def _inner(p, pt):
# step back through time
r = Reader(
inner_reader=DiskReader,
dataset=p,
partitions=pt,
start_date=datetime.date(2021, 1, 1),
end_date=datetime.date(2021, 1, 1),
freshness_limit="30d",
)
assert len(list(r)) == 50
def _inner(p, pt):
# step back through time
r = Reader(
inner_reader=DiskReader,
dataset=p,
partitions=pt,
start_date=datetime.date(2021, 1, 1),
end_date=datetime.date(2021, 1, 1),
freshness_limit="5d",
)
with pytest.raises(DataNotFoundError):
assert len(list(r)) == 0
def test_ignore_flag():
"""
test we ignore invalidated frames
"""
DATA_DATE = datetime.date(2021, 3, 29)
records = Reader(
dataset="tests/data/framed",
inner_reader=DiskReader,
start_date=DATA_DATE,
end_date=DATA_DATE,
)
print(next(records))
assert next(records).get("test") == 1
def test_ignore_flag_step_back_days():
"""
test that we step back a day if all of the frames have been invalidated
"""
DATA_DATE = datetime.date(2021, 3, 30)
records = Reader(
dataset="tests/data/framed",
inner_reader=DiskReader,
start_date=DATA_DATE,
end_date=DATA_DATE,
freshness_limit="24h",
)
print(next(records))
def test_ignore_non_complete_frames():
"""
test we ignore frames without the complete flag
"""
DATA_DATE = datetime.date(2021, 3, 28)
records = Reader(
dataset="tests/data/framed",
inner_reader=DiskReader,
start_date=DATA_DATE,
end_date=DATA_DATE,
)
print(next(records))
assert next(records).get("test") == 1
def test_reader_writer_format_jsonl():
do_writer_compressed("jsonl")
g = glob.glob("_temp/**/*.jsonl", recursive=True)
assert len(g) > 0, g
c = glob.glob("_temp/**/*.complete", recursive=True)
len(c) == 0, c
r = Reader(inner_reader=DiskReader, dataset="_temp")
l = len(list(r))
shutil.rmtree("_temp", ignore_errors=True)
assert l == 200000, l
def test_dataset_prefix_validator():
with pytest.raises(AccessDenied):
reader = Reader(dataset="dataset", valid_dataset_prefixes=["drive/"])
with pytest.raises(AccessDenied):
reader = Reader(dataset="dataset",
valid_dataset_prefixes=["list", "of", "items"])
# no whitelist - allow all
reader = Reader(project="", dataset="dataset")
# a list of one
reader = Reader(project="",
dataset="dataset",
valid_dataset_prefixes=["dataset"])
# a list of many
reader = Reader(
project="",
dataset="dataset",
valid_dataset_prefixes=["on", "the", "list", "dataset"],
)
def test_reader_all_good():
failed = False
try:
reader = Reader(
project="",
select="a, b",
dataset="",
start_date=datetime.datetime.now(),
end_date=datetime.datetime.now(),
)
except InvalidReaderConfigError:
failed = True
assert not failed
def test_cursor_as_text():
offsets = [1, 6, 8, 13, 22]
for offset in offsets:
reader = Reader(
inner_reader=DiskReader,
dataset="tests/data/tweets/",
partitions=[],
cursor='{"partition": 5122091051124077700, "location": ' +
str(offset) + ', "map":"80" }',
)
l = list(reader)
# 24 because we count from zero (the first row in the file is the 0th record)
assert len(
l) + offset == 24, f"{len(l) + offset} == {24}, {reader.cursor}"
def test_base():
reader = Reader(inner_reader=DiskReader,
dataset="tests/data/tweets/",
partitions=[])
# we read 50 records
for i in range(50):
next(reader)
assert reader.cursor["location"] == (i % 25), reader.cursor
# we can't read 51
with pytest.raises(StopIteration):
next(reader)
# range 50 actually is 50
assert len(range(50)) == 50
def test_move_to_cursor():
offsets = [1, 6, 8, 13, 22]
for offset in offsets:
reader = Reader(inner_reader=DiskReader,
dataset="tests/data/tweets/",
partitions=[])
next(reader)
steps = 1
while reader.cursor["location"] < offset:
steps += 1
next(reader)
assert offset == reader.cursor["location"]
l = len(list(reader))
# we stepped offset number of records and then read l more, this should be 50
assert steps + l == 50
def test_writer_backout():
if Path(TEST_FOLDER).exists(): # pragma: no cover
shutil.rmtree(TEST_FOLDER)
w = StreamWriter(
dataset=TEST_FOLDER,
inner_writer=DiskWriter,
schema=Schema(SCHEMA),
idle_timeout_seconds=1,
)
for record in DATA_SET:
w.append(record)
time.sleep(4)
r = Reader(dataset=TEST_FOLDER, inner_reader=DiskReader)
assert len(list(r)) == 8
def get_data():
r = Reader(inner_reader=DiskReader,
dataset="tests/data/tweets",
raw_path=True)
return r
)
res = [r for r in reader]
print(len(res))
return (time.perf_counter_ns() - start) / 1e9
import os, sys
sys.path.insert(1, os.path.join(sys.path[0], "../.."))
from mabel.data import Reader
from mabel.adapters.disk import DiskReader
from mabel.logging import get_logger
get_logger().setLevel(100)
user_name = "Verizon Support"
print("indexed\t:", time_it("tests/data/index/is", user_name))
print("not indexed\t:", time_it("tests/data/index/not", user_name))
print("indexed\t:", time_it("tests/data/index/is", user_name + "bb"))
print("not indexed\t:", time_it("tests/data/index/not", user_name + "bb"))
reader = Reader(
inner_reader=DiskReader,
dataset="tests/data/index/is",
raw_path=True,
)
idx = Index.build_index(reader, "user_name")
idx.dump("tests/data/index/is/tweets.jsonl.user_name.idx")
augmented_chunk = carry_forward + chunk
lines = augmented_chunk.split(delimiter)
carry_forward = lines.pop()
yield from lines
if carry_forward:
yield carry_forward
schema = Schema(schema_definition)
lines = read_jsonl("tests/data/index/not/tweets.jsonl")
writer = BatchWriter(
inner_writer=DiskWriter,
dataset="_temp/idx",
# schema=schema,
indexes=["user_name"],
)
for record in lines:
writer.append(record)
writer.finalize()
reader = Reader(inner_reader=DiskReader,
dataset="_temp/idx",
filters=("user_name", "==", "Remy"))
i = 0
for i, r in enumerate(reader):
print(i, r)
print(i)
def get_records():
r = Reader(inner_reader=DiskReader,
dataset="tests/data/tweets/",
partitions=[])
return len(list(r))
