[Notes] [Git][BuildStream/buildstream][raoul/802-refactor-artifactcache] 2 commits: _cas: Move pull logic into CASRemote

[Raoul Hidalgo Charman <gitlab mg gitlab com>]

Title: GitLab

Raoul Hidalgo Charman pushed to branch raoul/802-refactor-artifactcache at BuildStream / buildstream

Commits:

• 510744fa
  by Raoul Hidalgo Charman at 2018-12-17T16:49:02Z

  _cas: Move pull logic into CASRemote
  
  Seperates the pull logic into a remote/local API , so that artifact cache
  iterates over blob digests checks whether it has them, and then downloads them
  if not. Currently doesn't use batching.
  
  Part of #802
  

• 1bfb661d
  by Raoul Hidalgo Charman at 2018-12-17T17:00:33Z

  _casremote.py: Add batching to pull command
  
  Part of #802
  

3 changed files:

• buildstream/_artifactcache.py
• buildstream/_cas/cascache.py
• buildstream/_cas/casremote.py

Changes:

buildstream/_artifactcache.py

---

...	...	@@ -28,7 +28,7 @@ from ._message import Message, MessageType
28	28	from . import utils
29	29	from . import _yaml
30	30
31		-from ._cas.casremote import CASRemote, CASRemoteSpec
	31	+from ._cas.casremote import BlobNotFound, CASRemote, CASRemoteSpec
32	32
33	33
34	34	CACHE_SIZE_FILE = "cache_size"
...	...	@@ -634,7 +634,7 @@ class ArtifactCache():
634	634	# Returns:
635	635	# (bool): True if pull was successful, False if artifact was not available
636	636	#
637		- def pull(self, element, key, *, progress=None, subdir=None, excluded_subdirs=None):
	637	+ def pull(self, element, key, *, progress=None, subdir=None, excluded_subdirs=[]):
638	638	ref = self.get_artifact_fullname(element, key)
639	639
640	640	project = element._get_project()
...	...	@@ -644,13 +644,34 @@ class ArtifactCache():
644	644	display_key = element._get_brief_display_key()
645	645	element.status("Pulling artifact {} <- {}".format(display_key, remote.spec.url))
646	646
647		- if self.cas.pull(ref, remote, progress=progress, subdir=subdir, excluded_subdirs=excluded_subdirs):
648		- element.info("Pulled artifact {} <- {}".format(display_key, remote.spec.url))
	647	+ root_digest = remote.get_reference(ref)
	648	+
	649	+ if root_digest:
	650	+ try:
	651	+ for blob_digest in remote.yield_blob_digests(
	652	+ root_digest, progress=progress, subdir=subdir,
	653	+ excluded_subdirs=excluded_subdirs):
	654	+ if self.cas.check_blob(blob_digest):
	655	+ continue
	656	+ remote.request_blob(blob_digest)
	657	+ for blob_file in remote.get_blobs():
	658	+ self.cas.add_object(path=blob_file.name, link_directly=True)
	659	+
	660	+ for blob_file in remote.get_blobs(request_batch=True):
	661	+ self.cas.add_object(path=blob_file.name, link_directly=True)
	662	+ self.cas.set_ref(ref, root_digest)
	663	+ except BlobNotFound:
	664	+ element.info("Remote ({}) is missing blobs for {}".format(
	665	+ remote.spec.url, element._get_brief_display_key()))
	666	+ continue
	667	+
649	668	if subdir:
650	669	# Attempt to extract subdir into artifact extract dir if it already exists
651	670	# without containing the subdir. If the respective artifact extract dir does not
652	671	# exist a complete extraction will complete.
653	672	self.extract(element, key, subdir)
	673	+
	674	+ element.info("Pulled artifact {} <- {}".format(display_key, remote.spec.url))
654	675	# no need to pull from additional remotes
655	676	return True
656	677	else:

buildstream/_cas/cascache.py

---

...	...	@@ -33,7 +33,7 @@ from .._protos.buildstream.v2 import buildstream_pb2
33	33	from .. import utils
34	34	from .._exceptions import CASError
35	35
36		-from .casremote import BlobNotFound, _CASBatchRead, _CASBatchUpdate
	36	+from .casremote import _CASBatchUpdate
37	37
38	38
39	39	# A CASCache manages a CAS repository as specified in the Remote Execution API.
...	...	@@ -183,50 +183,6 @@ class CASCache():
183	183
184	184	return modified, removed, added
185	185
186		- # pull():
187		- #
188		- # Pull a ref from a remote repository.
189		- #
190		- # Args:
191		- # ref (str): The ref to pull
192		- # remote (CASRemote): The remote repository to pull from
193		- # progress (callable): The progress callback, if any
194		- # subdir (str): The optional specific subdir to pull
195		- # excluded_subdirs (list): The optional list of subdirs to not pull
196		- #
197		- # Returns:
198		- # (bool): True if pull was successful, False if ref was not available
199		- #
200		- def pull(self, ref, remote, *, progress=None, subdir=None, excluded_subdirs=None):
201		- try:
202		- remote.init()
203		-
204		- request = buildstream_pb2.GetReferenceRequest()
205		- request.key = ref
206		- response = remote.ref_storage.GetReference(request)
207		-
208		- tree = remote_execution_pb2.Digest()
209		- tree.hash = response.digest.hash
210		- tree.size_bytes = response.digest.size_bytes
211		-
212		- # Check if the element artifact is present, if so just fetch the subdir.
213		- if subdir and os.path.exists(self.objpath(tree)):
214		- self._fetch_subdir(remote, tree, subdir)
215		- else:
216		- # Fetch artifact, excluded_subdirs determined in pullqueue
217		- self._fetch_directory(remote, tree, excluded_subdirs=excluded_subdirs)
218		-
219		- self.set_ref(ref, tree)
220		-
221		- return True
222		- except grpc.RpcError as e:
223		- if e.code() != grpc.StatusCode.NOT_FOUND:
224		- raise CASError("Failed to pull ref {}: {}".format(ref, e)) from e
225		- else:
226		- return False
227		- except BlobNotFound as e:
228		- return False
229		-
230	186	# pull_tree():
231	187	#
232	188	# Pull a single Tree rather than a ref.
...	...	@@ -591,6 +547,16 @@ class CASCache():
591	547	reachable = set()
592	548	self._reachable_refs_dir(reachable, tree, update_mtime=True)
593	549
	550	+ # Check to see if a blob is in the local CAS
	551	+ # return None if not
	552	+ def check_blob(self, digest):
	553	+ objpath = self.objpath(digest)
	554	+ if os.path.exists(objpath):
	555	+ # already in local repository
	556	+ return objpath
	557	+ else:
	558	+ return None
	559	+
594	560	################################################
595	561	# Local Private Methods #
596	562	################################################
...	...	@@ -805,103 +771,6 @@ class CASCache():
805	771
806	772	return objpath
807	773
808		- def _batch_download_complete(self, batch):
809		- for digest, data in batch.send():
810		- with tempfile.NamedTemporaryFile(dir=self.tmpdir) as f:
811		- f.write(data)
812		- f.flush()
813		-
814		- added_digest = self.add_object(path=f.name, link_directly=True)
815		- assert added_digest.hash == digest.hash
816		-
817		- # Helper function for _fetch_directory().
818		- def _fetch_directory_batch(self, remote, batch, fetch_queue, fetch_next_queue):
819		- self._batch_download_complete(batch)
820		-
821		- # All previously scheduled directories are now locally available,
822		- # move them to the processing queue.
823		- fetch_queue.extend(fetch_next_queue)
824		- fetch_next_queue.clear()
825		- return _CASBatchRead(remote)
826		-
827		- # Helper function for _fetch_directory().
828		- def _fetch_directory_node(self, remote, digest, batch, fetch_queue, fetch_next_queue, *, recursive=False):
829		- in_local_cache = os.path.exists(self.objpath(digest))
830		-
831		- if in_local_cache:
832		- # Skip download, already in local cache.
833		- pass
834		- elif (digest.size_bytes >= remote.max_batch_total_size_bytes or
835		- not remote.batch_read_supported):
836		- # Too large for batch request, download in independent request.
837		- self._ensure_blob(remote, digest)
838		- in_local_cache = True
839		- else:
840		- if not batch.add(digest):
841		- # Not enough space left in batch request.
842		- # Complete pending batch first.
843		- batch = self._fetch_directory_batch(remote, batch, fetch_queue, fetch_next_queue)
844		- batch.add(digest)
845		-
846		- if recursive:
847		- if in_local_cache:
848		- # Add directory to processing queue.
849		- fetch_queue.append(digest)
850		- else:
851		- # Directory will be available after completing pending batch.
852		- # Add directory to deferred processing queue.
853		- fetch_next_queue.append(digest)
854		-
855		- return batch
856		-
857		- # _fetch_directory():
858		- #
859		- # Fetches remote directory and adds it to content addressable store.
860		- #
861		- # Fetches files, symbolic links and recursively other directories in
862		- # the remote directory and adds them to the content addressable
863		- # store.
864		- #
865		- # Args:
866		- # remote (Remote): The remote to use.
867		- # dir_digest (Digest): Digest object for the directory to fetch.
868		- # excluded_subdirs (list): The optional list of subdirs to not fetch
869		- #
870		- def _fetch_directory(self, remote, dir_digest, *, excluded_subdirs=None):
871		- fetch_queue = [dir_digest]
872		- fetch_next_queue = []
873		- batch = _CASBatchRead(remote)
874		- if not excluded_subdirs:
875		- excluded_subdirs = []
876		-
877		- while len(fetch_queue) + len(fetch_next_queue) > 0:
878		- if not fetch_queue:
879		- batch = self._fetch_directory_batch(remote, batch, fetch_queue, fetch_next_queue)
880		-
881		- dir_digest = fetch_queue.pop(0)
882		-
883		- objpath = self._ensure_blob(remote, dir_digest)
884		-
885		- directory = remote_execution_pb2.Directory()
886		- with open(objpath, 'rb') as f:
887		- directory.ParseFromString(f.read())
888		-
889		- for dirnode in directory.directories:
890		- if dirnode.name not in excluded_subdirs:
891		- batch = self._fetch_directory_node(remote, dirnode.digest, batch,
892		- fetch_queue, fetch_next_queue, recursive=True)
893		-
894		- for filenode in directory.files:
895		- batch = self._fetch_directory_node(remote, filenode.digest, batch,
896		- fetch_queue, fetch_next_queue)
897		-
898		- # Fetch final batch
899		- self._fetch_directory_batch(remote, batch, fetch_queue, fetch_next_queue)
900		-
901		- def _fetch_subdir(self, remote, tree, subdir):
902		- subdirdigest = self._get_subdir(tree, subdir)
903		- self._fetch_directory(remote, subdirdigest)
904		-
905	774	def _fetch_tree(self, remote, digest):
906	775	# download but do not store the Tree object
907	776	with tempfile.NamedTemporaryFile(dir=self.tmpdir) as out:

buildstream/_cas/casremote.py

---

...	...	@@ -3,6 +3,7 @@ import io
3	3	import os
4	4	import multiprocessing
5	5	import signal
	6	+import tempfile
6	7	from urllib.parse import urlparse
7	8	import uuid
8	9
...	...	@@ -92,6 +93,11 @@ class CASRemote():
92	93	self.tmpdir = tmpdir
93	94	os.makedirs(tmpdir, exist_ok=True)
94	95
	96	+ self.__tmp_downloads = [] # files in the tmpdir waiting to be added to local caches
	97	+
	98	+ self.__batch_read = None
	99	+ self.__batch_update = None
	100	+
95	101	def init(self):
96	102	if not self._initialized:
97	103	url = urlparse(self.spec.url)
...	...	@@ -149,6 +155,7 @@ class CASRemote():
149	155	request = remote_execution_pb2.BatchReadBlobsRequest()
150	156	response = self.cas.BatchReadBlobs(request)
151	157	self.batch_read_supported = True
	158	+ self.__batch_read = _CASBatchRead(self)
152	159	except grpc.RpcError as e:
153	160	if e.code() != grpc.StatusCode.UNIMPLEMENTED:
154	161	raise
...	...	@@ -159,6 +166,7 @@ class CASRemote():
159	166	request = remote_execution_pb2.BatchUpdateBlobsRequest()
160	167	response = self.cas.BatchUpdateBlobs(request)
161	168	self.batch_update_supported = True
	169	+ self.__batch_update = _CASBatchUpdate(self)
162	170	except grpc.RpcError as e:
163	171	if (e.code() != grpc.StatusCode.UNIMPLEMENTED and
164	172	e.code() != grpc.StatusCode.PERMISSION_DENIED):
...	...	@@ -255,6 +263,78 @@ class CASRemote():
255	263
256	264	return message_digest
257	265
	266	+ # get_reference():
	267	+ #
	268	+ # Args:
	269	+ # ref (str): The ref to request
	270	+ #
	271	+ def get_reference(self, ref):
	272	+ try:
	273	+ self.init()
	274	+
	275	+ request = buildstream_pb2.GetReferenceRequest()
	276	+ request.key = ref
	277	+ return self.ref_storage.GetReference(request).digest
	278	+ except grpc.RpcError as e:
	279	+ if e.code() != grpc.StatusCode.NOT_FOUND:
	280	+ raise CASError("Failed to find ref {}: {}".format(ref, e)) from e
	281	+ else:
	282	+ return None
	283	+
	284	+ # yield_blob_digests():
	285	+ #
	286	+ # Iterate over blobs digests from a reference
	287	+ #
	288	+ # Args:
	289	+ # root_digest (str): The root_digest to get a tree of
	290	+ # progress (callable): The progress callback, if any
	291	+ # subdir (str): The optional specific subdir to pull
	292	+ # excluded_subdirs (list): The optional list of subdirs to not pull
	293	+ #
	294	+ # Returns:
	295	+ # (iter): True if pull was successful, False if ref was not available
	296	+ #
	297	+ def yield_blob_digests(self, root_digest, *, progress=None, subdir=None, excluded_subdirs=[]):
	298	+ self.init()
	299	+
	300	+ # TODO add subdir stuff
	301	+ # Fetch artifact, excluded_subdirs determined in pullqueue
	302	+ yield from self._yield_directory_digests(root_digest, excluded_subdirs=excluded_subdirs)
	303	+
	304	+ # request_blob():
	305	+ #
	306	+ # Request blob, triggering download depending via bytestream or cas
	307	+ # BatchReadBlobs depending on size.
	308	+ #
	309	+ # Args:
	310	+ # digest (Digest): digest of the requested blob
	311	+ #
	312	+ def request_blob(self, digest):
	313	+ if (not self.batch_read_supported or
	314	+ digest.size_bytes > self.max_batch_total_size_bytes):
	315	+ f = tempfile.NamedTemporaryFile(dir=self.tmpdir)
	316	+ self._fetch_blob(digest, f)
	317	+ self.__tmp_downloads.append(f)
	318	+ elif self.__batch_read.add(digest) is False:
	319	+ self._download_batch()
	320	+ self.__batch_read.add(digest)
	321	+
	322	+ # get_blobs():
	323	+ #
	324	+ # Yield over downloaded blobs in the tmp file locations, causing the files
	325	+ # to be deleted once they go out of scope.
	326	+ #
	327	+ # Returns:
	328	+ # iterator over NamedTemporaryFile
	329	+ def get_blobs(self, request_batch=False):
	330	+ # Send read batch request and download
	331	+ if (request_batch is True and
	332	+ self.batch_read_supported is True):
	333	+ self._download_batch()
	334	+
	335	+ while self.__tmp_downloads:
	336	+ yield self.__tmp_downloads.pop()
	337	+
258	338	################################################
259	339	# Local Private Methods #
260	340	################################################
...	...	@@ -269,6 +349,36 @@ class CASRemote():
269	349
270	350	assert digest.size_bytes == os.fstat(stream.fileno()).st_size
271	351
	352	+ # _yield_directory_digests():
	353	+ #
	354	+ # Fetches remote directory and adds it to content addressable store.
	355	+ #
	356	+ # Fetches files, symbolic links and recursively other directories in
	357	+ # the remote directory and adds them to the content addressable
	358	+ # store.
	359	+ #
	360	+ # Args:
	361	+ # dir_digest (Digest): Digest object for the directory to fetch.
	362	+ # excluded_subdirs (list): The optional list of subdirs to not fetch
	363	+ #
	364	+ def _yield_directory_digests(self, dir_digest, *, excluded_subdirs=[]):
	365	+ # get directory blob
	366	+ f = tempfile.NamedTemporaryFile(dir=self.tmpdir)
	367	+ self._fetch_blob(dir_digest, f)
	368	+ self.__tmp_downloads.append(f)
	369	+
	370	+ # need to read in directory structure to iterate over it
	371	+ directory = remote_execution_pb2.Directory()
	372	+ with open(f.name, 'rb') as tmp:
	373	+ directory.ParseFromString(tmp.read())
	374	+
	375	+ yield dir_digest
	376	+ for filenode in directory.files:
	377	+ yield filenode.digest
	378	+ for dirnode in directory.directories:
	379	+ if dirnode.name not in excluded_subdirs:
	380	+ yield from self._yield_directory_digests(dirnode.digest)
	381	+
272	382	def _send_blob(self, digest, stream, u_uid=uuid.uuid4()):
273	383	resource_name = '/'.join(['uploads', str(u_uid), 'blobs',
274	384	digest.hash, str(digest.size_bytes)])
...	...	@@ -297,6 +407,15 @@ class CASRemote():
297	407
298	408	assert response.committed_size == digest.size_bytes
299	409
	410	+ def _download_batch(self):
	411	+ for _, data in self.__batch_read.send():
	412	+ f = tempfile.NamedTemporaryFile(dir=self.tmpdir)
	413	+ f.write(data)
	414	+ f.flush()
	415	+ self.__tmp_downloads.append(f)
	416	+
	417	+ self.__batch_read = _CASBatchRead(self)
	418	+
300	419
301	420	# Represents a batch of blobs queued for fetching.
302	421	#

—
View it on GitLab.
You're receiving this email because of your account on gitlab.com. If you'd like to receive fewer emails, you can adjust your notification settings.