def setUp(self):
self.pkg = Package()
self.entry_with_hash = PackageEntry(
PhysicalKey('test-bucket', 'with-hash', 'with-hash'),
42,
{
'type': 'SHA256',
'value': '0' * 64
},
{},
)
self.entry_without_hash = PackageEntry(
PhysicalKey('test-bucket', 'without-hash', 'without-hash'),
42,
None,
{},
)
self.pkg.set('with-hash', self.entry_with_hash)
self.pkg.set('without-hash', self.entry_without_hash)
def prepare_pkg(self, *, copy_data):
expected_pkg = Package()
pkg_entries = self.entries.items()
if copy_data:
pkg_entries = [(
lk,
e.with_physical_key(
PhysicalKey(self.dst_bucket, f'{self.dst_pkg_name}/{lk}',
'dst_' + e.physical_key.version_id)),
) for lk, e in pkg_entries]
for lk, entry in pkg_entries:
expected_pkg.set(lk, entry)
expected_pkg._set_commit_message(None)
return expected_pkg
class HashCalculationTest(unittest.TestCase):
def setUp(self):
self.pkg = Package()
self.entry_with_hash = PackageEntry(
PhysicalKey('test-bucket', 'with-hash', 'with-hash'),
42,
{
'type': 'SHA256',
'value': '0' * 64
},
{},
)
self.entry_without_hash = PackageEntry(
PhysicalKey('test-bucket', 'without-hash', 'without-hash'),
42,
None,
{},
)
self.pkg.set('with-hash', self.entry_with_hash)
self.pkg.set('without-hash', self.entry_without_hash)
def test_calculate_pkg_hashes(self):
boto_session = mock.MagicMock()
with mock.patch.object(
t4_lambda_pkgpush,
'calculate_pkg_entry_hash') as calculate_pkg_entry_hash_mock:
t4_lambda_pkgpush.calculate_pkg_hashes(boto_session, self.pkg)
calculate_pkg_entry_hash_mock.assert_called_once_with(
mock.ANY, self.entry_without_hash)
@mock.patch.object(t4_lambda_pkgpush, 'S3_HASH_LAMBDA_MAX_FILE_SIZE_BYTES',
1)
def test_calculate_pkg_hashes_too_large_file_error(self):
s3_client = mock.MagicMock()
with pytest.raises(t4_lambda_pkgpush.FileTooLargeForHashing):
t4_lambda_pkgpush.calculate_pkg_hashes(s3_client, self.pkg)
def test_calculate_pkg_entry_hash(self):
get_s3_client_mock = mock.MagicMock()
s3_client_mock = get_s3_client_mock.return_value
s3_client_mock.generate_presigned_url.return_value = 'https://example.com'
with mock.patch("t4_lambda_pkgpush.invoke_hash_lambda",
return_value='0' * 64) as invoke_hash_lambda_mock:
t4_lambda_pkgpush.calculate_pkg_entry_hash(get_s3_client_mock,
self.entry_without_hash)
get_s3_client_mock.assert_called_once_with(
self.entry_without_hash.physical_key.bucket)
invoke_hash_lambda_mock.assert_called_once_with(
s3_client_mock.generate_presigned_url.return_value)
s3_client_mock.generate_presigned_url.assert_called_once_with(
ClientMethod='get_object',
ExpiresIn=t4_lambda_pkgpush.
S3_HASH_LAMBDA_SIGNED_URL_EXPIRES_IN_SECONDS,
Params={
'Bucket': self.entry_without_hash.physical_key.bucket,
'Key': self.entry_without_hash.physical_key.path,
'VersionId': self.entry_without_hash.physical_key.version_id,
},
)
assert self.entry_without_hash.hash == {
'type': 'SHA256',
'value': invoke_hash_lambda_mock.return_value,
}
def test_invoke_hash_lambda(self):
lambda_client_stubber = Stubber(t4_lambda_pkgpush.lambda_)
lambda_client_stubber.activate()
self.addCleanup(lambda_client_stubber.deactivate)
test_hash = '0' * 64
test_url = 'https://example.com'
lambda_client_stubber.add_response(
'invoke',
service_response={
'Payload': io.BytesIO(b'"%s"' % test_hash.encode()),
},
expected_params={
'FunctionName': t4_lambda_pkgpush.S3_HASH_LAMBDA,
'Payload': '"%s"' % test_url,
},
)
assert t4_lambda_pkgpush.invoke_hash_lambda(test_url) == test_hash
lambda_client_stubber.assert_no_pending_responses()
def test_invoke_hash_lambda_error(self):
lambda_client_stubber = Stubber(t4_lambda_pkgpush.lambda_)
lambda_client_stubber.activate()
self.addCleanup(lambda_client_stubber.deactivate)
test_url = 'https://example.com'
lambda_client_stubber.add_response(
'invoke',
service_response={
'FunctionError': 'Unhandled',
'Payload': io.BytesIO(b'some error info'),
},
expected_params={
'FunctionName': t4_lambda_pkgpush.S3_HASH_LAMBDA,
'Payload': '"%s"' % test_url,
},
)
with pytest.raises(t4_lambda_pkgpush.S3HashLambdaUnhandledError):
t4_lambda_pkgpush.invoke_hash_lambda(test_url)
lambda_client_stubber.assert_no_pending_responses()
def setUpClass(cls):
super().setUpClass()
pkg = Package()
pkg._set_commit_message(cls.parent_commit_message)
pkg._workflow = {
'config':
f's3://{cls.parent_bucket}/.quilt/workflows/config.yml?versionId=configVersion',
'id': 'gamma',
'schemas': {
'top-secret':
f's3://{cls.parent_bucket}/top-secret.schema.json?versionId=schemaVersion'
},
}
pkg.set_meta({'meta': 'old meta'})
cls.entries = cls.get_pkg_entries()
for lk, entry in cls.entries.items():
pkg.set(lk, entry)
manifest_buf = io.BytesIO()
pkg._dump(manifest_buf)
cls.parent_manifest = manifest_buf.getvalue()
cls.parent_top_hash = pkg.top_hash
cls.src_params = {
'parent': {
'registry': cls.src_registry,
'name': cls.parent_pkg_name,
'top_hash': cls.parent_top_hash,
},
}
def _mock_package_build(self,
entries,
*,
message=...,
expected_workflow=...):
if message is ...:
message = self.dst_commit_message
# Use a test package to verify manifest entries
test_pkg = Package()
test_pkg.set_meta(self.meta)
# Mock hashing package objects
for entry in entries:
pkey = PhysicalKey.from_url(entry['physical_key'])
hash_obj = {'type': 'SHA256', 'value': entry['hash']}
test_entry = PackageEntry(pkey, entry['size'], hash_obj,
entry.get('meta'))
test_pkg.set(entry['logical_key'], entry=test_entry)
mocked_workflow_data = 'some-workflow-data'
test_pkg._workflow = mocked_workflow_data
# build the manifest from the test_package
test_pkg._set_commit_message(message)
manifest = io.BytesIO()
test_pkg.dump(manifest)
manifest.seek(0)
self.s3_stubber.add_response(
'put_object',
service_response={},
expected_params={
'Body': manifest.read(),
'Bucket': self.dst_bucket,
'Key': f'.quilt/packages/{test_pkg.top_hash}',
},
)
self.s3_stubber.add_response(
'put_object',
service_response={},
expected_params={
'Body':
str.encode(test_pkg.top_hash),
'Bucket':
self.dst_bucket,
'Key':
f'.quilt/named_packages/{self.dst_pkg_name}/{str(int(self.mock_timestamp))}',
},
)
self.s3_stubber.add_response(
'put_object',
service_response={},
expected_params={
'Body': str.encode(test_pkg.top_hash),
'Bucket': self.dst_bucket,
'Key': f'.quilt/named_packages/{self.dst_pkg_name}/latest',
},
)
with mock.patch(
'quilt3.workflows.validate',
return_value=mocked_workflow_data) as workflow_validate_mock:
yield
workflow_validate_mock.assert_called_once_with(
registry=get_package_registry(self.dst_registry),
workflow=expected_workflow,
name=self.dst_pkg_name,
pkg=mock.ANY, # TODO: probably this should be more specific.
message=message,
)
def create_package(
manifest: pd.DataFrame,
step_pkg_root: Path,
filepath_columns: List[str] = ["filepath"],
metadata_columns: List[str] = [],
) -> Tuple[Package, pd.DataFrame]:
# Make a copy
relative_manifest = manifest.copy(deep=True)
# Create empty package
pkg = Package()
# Create associate mappings: List[Dict[str, str]]
# This list is in index order. Meaning that as the column values are descended we
# can simply add a new associate to the already existing associate map at that list
# index.
associates = []
# Create metadata reduction map
# This will be used to clean up and standardize the metadata access after object
# construction. Metadata column name to boolean value for should or should not
# reduce metadata values. This will be used during the "clean up the package
# metadata step". If we have multiple files each with the same keys for the
# metadata, but for one reason or another, one packaged file's value for a certain
# key is a list while another's is a single string, this leads to a confusing mixed
# return value API for the same _type_ of object. Example:
# fov/
# obj1/
# {example_key: "hello"}
# obj2/
# {example_key: ["hello", "world"]}
# Commonly this happens when a manifest has rows of unique instances of a child
# object but retains a reference to a parent object, example: rows of information
# about unique cells that were all generated using the same algorithm, whose
# information is stored in a column, for each cell information row. This could
# result in some files (which only have one cell) being a single string while other
# files (which have more than one cell) being a list of the same string over and
# over again. "Why spend all this time to reduce/ collapse the metadata anyway?",
# besides making it so that users won't have to call `obj2.meta["example_key"][0]`
# every time they want the value, and besides the fact that it standardizes the
# metadata api, the biggest reason is that S3 objects can only have 2KB of metadata,
# without this reduction/ collapse step, manifests are more likely to hit that limit
# and cause a package distribution error.
metadata_reduction_map = {
index_col: True
for index_col in metadata_columns
}
# Set all files
with tqdm(
total=len(filepath_columns) * len(relative_manifest),
desc="Constructing package",
) as pbar:
for col in filepath_columns:
# Update values to the logical key as they are set
for i, val in enumerate(relative_manifest[col].values):
# Fully resolve the path
physical_key = Path(val).expanduser().resolve()
# Try creating a logical key from the relative of step
# local staging to the filepath
#
# Ex:
# step_pkg_root = "local_staging/raw"
# physical_key = "local_staging/raw/images/some_file.tiff"
# produced logical_key = "images/some_file.tiff"
try:
logical_key = str(
file_utils._filepath_rel2abs(physical_key).relative_to(
file_utils._filepath_rel2abs(step_pkg_root)))
except ValueError:
# Create logical key from merging column and filename
# Also remove any obvious "path" type words from column name
#
# Ex:
# physical_key = "/some/abs/path/some_file.tiff"
# column = "SourceReadPath"
# produced logical_key = "source/some_file.tiff"
stripped_col = col.lower().replace("read",
"").replace("path", "")
logical_key = f"{stripped_col}/{physical_key.name}"
if physical_key.is_file():
relative_manifest[col].values[i] = logical_key
# Create metadata dictionary to attach to object
meta = {}
for meta_col in metadata_columns:
# Short reference to current metadata value
v = relative_manifest[meta_col].values[i]
# Enforce simple JSON serializable type
# First check if value is a numpy value
# It likely is because pandas relies on numpy
# All numpy types have the "dtype" attribute and can be cast to
# python type by using the `item` function, details here:
# https://docs.scipy.org/doc/numpy/reference/generated/numpy.ndarray.item.html
if hasattr(v, "dtype"):
v = v.item()
# Cast to JSON serializable type
v = file_utils.make_json_serializable(
v, f"Value from column: {meta_col}, index: {i}")
# Update metadata with value
meta[meta_col] = [v]
# Check if object already exists
if logical_key in pkg:
# Join the two meta dictionaries
joined_meta = {}
for meta_col, curr_v in pkg[logical_key].meta.items():
# Join the values for the current iteration of the metadata
joined_values = [*curr_v, *meta[meta_col]]
# Only check if the metadata at this index can be reduced
# if currently is still being decided. We know if the
# metadata value at this index is still be decided if:
# the boolean value in the metadata reduction map is True,
# as in, this index can be reduced or collapsed.
# The other reason to make this check is so that we don't
# override an earlier False reduction value. In the case
# where early on we encounter an instance of the metadata
# that should not be reduced but then later on we say it
# can be, this check prevents that. As we want all metadata
# access across the dataset to be uniform.
if metadata_reduction_map[meta_col]:
# Update the metadata reduction map
# For the current column being checked, as long as it
# is still being determined that the column can be
# reduced (aka we have entered this if block) check if
# we can still reduce the metadata after the recent
# addition. "We can reduce the metadata if the count of
# the first value (or any value) is the same as the
# length of the entire list of values".
# This runs quickly for small lists as seen here:
# https://stackoverflow.com/questions/3844801/check-if-all-elements-in-a-list-are-identical
metadata_reduction_map[
meta_col] = joined_values.count(
joined_values[0]) == len(
joined_values) # noqa F501
# Attached the joined values to the joined metadata
joined_meta[meta_col] = joined_values
# Update meta
pkg[logical_key].set_meta(joined_meta)
# Object didn't already exist, simply set it
else:
pkg.set(logical_key, physical_key, meta)
# Update associates
try:
associates[i][col] = logical_key
except IndexError:
associates.append({col: logical_key})
else:
relative_manifest[col].values[i] = logical_key
pkg.set_dir(logical_key, physical_key)
# Update progress bar
pbar.update()
# Clean up package metadata
pkg = _recursive_clean(pkg, metadata_reduction_map)
# Attach associates
for i, associate_mapping in tqdm(
enumerate(associates),
desc="Creating associate metadata blocks"):
for col, lk in associate_mapping.items():
# Having dictionary expansion in this order means that associates will
# override a prior existing `associates` key, this is assumed safe
# because attach_associates was set to True.
pkg[lk].set_meta({
**pkg[lk].meta,
**{
"associates": associate_mapping
}
})
return pkg, relative_manifest