[Notes] [Git][BuildStream/buildstream][raoul/802-refactor-artifactcache] 2 commits: artifactcache: 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:

• a81320e3
  by Raoul Hidalgo Charman at 2019-01-08T15:17:45Z

  artifactcache: 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 requests them
  if not. The request command allows batching of blobs where appropriate.
  
  Tests have been updated to ensure the correct tmpdir is set up in process
  wrappers, else invalid cross link errors happen in the CI. Additional asserts
  have been added to check that the temporary directories are cleared by the end
  of a pull.
  
  Part of #802
  

• d6cc99a5
  by Raoul Hidalgo Charman at 2019-01-08T15:17:49Z

  artifact cache: implement new push methods
  
  Part of #802
  

6 changed files:

• buildstream/_artifactcache.py
• buildstream/_cas/cascache.py
• buildstream/_cas/casremote.py
• tests/artifactcache/pull.py
• tests/artifactcache/push.py
• tests/integration/pullbuildtrees.py

Changes:

buildstream/_artifactcache.py

---

...	...	@@ -21,6 +21,7 @@ import multiprocessing
21	21	import os
22	22	import string
23	23	from collections.abc import Mapping
	24	+import grpc
24	25
25	26	from .types import _KeyStrength
26	27	from ._exceptions import ArtifactError, CASError, LoadError, LoadErrorReason
...	...	@@ -28,7 +29,7 @@ from ._message import Message, MessageType
28	29	from . import utils
29	30	from . import _yaml
30	31
31		-from ._cas.casremote import CASRemote, CASRemoteSpec
	32	+from ._cas.casremote import BlobNotFound, CASRemote, CASRemoteSpec
32	33
33	34
34	35	CACHE_SIZE_FILE = "cache_size"
...	...	@@ -607,16 +608,37 @@ class ArtifactCache():
607	608
608	609	for remote in push_remotes:
609	610	remote.init()
	611	+ pushed_remote = False
610	612	display_key = element._get_brief_display_key()
611	613	element.status("Pushing artifact {} -> {}".format(display_key, remote.spec.url))
612	614
613		- if self.cas.push(refs, remote):
614		- element.info("Pushed artifact {} -> {}".format(display_key, remote.spec.url))
	615	+ try:
	616	+ for ref in refs:
	617	+ # Check whether ref is already on the server in which case
	618	+ # there is no need to push the ref
	619	+ root_digest = self.cas.resolve_ref(ref)
	620	+ response = remote.get_reference(ref)
	621	+ if (response is not None and
	622	+ response.hash == root_digest.hash and
	623	+ response.size_bytes == root_digest.size_bytes):
	624	+ element.info("Remote ({}) already has {} cached".format(
	625	+ remote.spec.url, element._get_brief_display_key()))
	626	+ continue
	627	+
	628	+ # upload blobs
	629	+ self._send_directory(root_digest, remote)
	630	+
	631	+ remote.update_reference(ref, root_digest)
	632	+ pushed_remote = True
	633	+
	634	+ except grpc.RpcError as e:
	635	+ if e.code() != grpc.StatusCode.RESOURCE_EXHAUSTED:
	636	+ raise CASError("Failed to push ref {}: {}"
	637	+ .format(refs, e), temporary=True) from e
	638	+
	639	+ if pushed_remote is True:
615	640	pushed = True
616		- else:
617		- element.info("Remote ({}) already has {} cached".format(
618		- remote.spec.url, element._get_brief_display_key()
619		- ))
	641	+ element.info("Pushed artifact {} -> {}".format(display_key, remote.spec.url))
620	642
621	643	return pushed
622	644
...	...	@@ -644,19 +666,31 @@ class ArtifactCache():
644	666	display_key = element._get_brief_display_key()
645	667	element.status("Pulling artifact {} <- {}".format(display_key, remote.spec.url))
646	668
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))
649		- if subdir:
650		- # Attempt to extract subdir into artifact extract dir if it already exists
651		- # without containing the subdir. If the respective artifact extract dir does not
652		- # exist a complete extraction will complete.
653		- self.extract(element, key, subdir)
654		- # no need to pull from additional remotes
655		- return True
656		- else:
	669	+ root_digest = remote.get_reference(ref)
	670	+
	671	+ if not root_digest:
657	672	element.info("Remote ({}) does not have {} cached".format(
658		- remote.spec.url, element._get_brief_display_key()
659		- ))
	673	+ remote.spec.url, element._get_brief_display_key()))
	674	+ continue
	675	+
	676	+ try:
	677	+ self._fetch_directory(remote, root_digest, excluded_subdirs)
	678	+ except BlobNotFound:
	679	+ element.info("Remote ({}) is missing blobs for {}".format(
	680	+ remote.spec.url, element._get_brief_display_key()))
	681	+ continue
	682	+
	683	+ self.cas.set_ref(ref, root_digest)
	684	+
	685	+ if subdir:
	686	+ # Attempt to extract subdir into artifact extract dir if it already exists
	687	+ # without containing the subdir. If the respective artifact extract dir does not
	688	+ # exist a complete extraction will complete.
	689	+ self.extract(element, key, subdir)
	690	+
	691	+ element.info("Pulled artifact {} <- {}".format(display_key, remote.spec.url))
	692	+ # no need to pull from additional remotes
	693	+ return True
660	694
661	695	except CASError as e:
662	696	raise ArtifactError("Failed to pull artifact {}: {}".format(
...	...	@@ -671,15 +705,26 @@ class ArtifactCache():
671	705	#
672	706	# Args:
673	707	# project (Project): The current project
674		- # digest (Digest): The digest of the tree
	708	+ # tree_digest (Digest): The digest of the tree
675	709	#
676		- def pull_tree(self, project, digest):
	710	+ def pull_tree(self, project, tree_digest):
677	711	for remote in self._remotes[project]:
678		- digest = self.cas.pull_tree(remote, digest)
679		-
680		- if digest:
681		- # no need to pull from additional remotes
682		- return digest
	712	+ try:
	713	+ for blob_digest in remote.yield_tree_digests(tree_digest):
	714	+ if self.cas.check_blob(blob_digest):
	715	+ continue
	716	+ remote.request_blob(blob_digest)
	717	+ for blob_file in remote.get_blobs():
	718	+ self.cas.add_object(path=blob_file.name, link_directly=True)
	719	+
	720	+ # Get the last batch
	721	+ for blob_file in remote.get_blobs(complete_batch=True):
	722	+ self.cas.add_object(path=blob_file.name, link_directly=True)
	723	+
	724	+ except BlobNotFound:
	725	+ continue
	726	+ else:
	727	+ return tree_digest
683	728
684	729	return None
685	730
...	...	@@ -708,7 +753,7 @@ class ArtifactCache():
708	753	return
709	754
710	755	for remote in push_remotes:
711		- self.cas.push_directory(remote, directory)
	756	+ self._send_directory(directory.ref, remote)
712	757
713	758	# push_message():
714	759	#
...	...	@@ -793,6 +838,14 @@ class ArtifactCache():
793	838	with self.context.timed_activity("Initializing remote caches", silent_nested=True):
794	839	self.initialize_remotes(on_failure=remote_failed)
795	840
	841	+ def _send_directory(self, root_digest, remote):
	842	+ required_blobs = self.cas.yield_directory_digests(root_digest)
	843	+ missing_blobs = remote.find_missing_blobs(required_blobs)
	844	+ for blob in missing_blobs.values():
	845	+ blob_file = self.cas.objpath(blob)
	846	+ remote.upload_blob(blob, blob_file, final=True)
	847	+ remote.send_update_batch()
	848	+
796	849	# _write_cache_size()
797	850	#
798	851	# Writes the given size of the artifact to the cache's size file
...	...	@@ -917,6 +970,19 @@ class ArtifactCache():
917	970	stat = os.statvfs(volume)
918	971	return stat.f_bsize * stat.f_bavail, stat.f_bsize * stat.f_blocks
919	972
	973	+ def _fetch_directory(self, remote, root_digest, excluded_subdirs):
	974	+ for blob_digest in remote.yield_directory_digests(
	975	+ root_digest, excluded_subdirs=excluded_subdirs):
	976	+ if self.cas.check_blob(blob_digest):
	977	+ continue
	978	+ remote.request_blob(blob_digest)
	979	+ for blob_file in remote.get_blobs():
	980	+ self.cas.add_object(path=blob_file.name, link_directly=True)
	981	+
	982	+ # Request final CAS batch
	983	+ for blob_file in remote.get_blobs(complete_batch=True):
	984	+ self.cas.add_object(path=blob_file.name, link_directly=True)
	985	+
920	986
921	987	# _configured_remote_artifact_cache_specs():
922	988	#

buildstream/_cas/cascache.py

---

...	...	@@ -18,23 +18,16 @@
18	18	# Jürg Billeter <juerg billeter codethink co uk>
19	19
20	20	import hashlib
21		-import itertools
22	21	import os
23	22	import stat
24	23	import tempfile
25		-import uuid
26	24	import contextlib
27	25
28		-import grpc
29		-
30	26	from .._protos.build.bazel.remote.execution.v2 import remote_execution_pb2
31		-from .._protos.buildstream.v2 import buildstream_pb2
32	27
33	28	from .. import utils
34	29	from .._exceptions import CASError
35	30
36		-from .casremote import BlobNotFound, _CASBatchRead, _CASBatchUpdate
37		-
38	31
39	32	# A CASCache manages a CAS repository as specified in the Remote Execution API.
40	33	#
...	...	@@ -183,73 +176,6 @@ class CASCache():
183	176
184	177	return modified, removed, added
185	178
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(instance_name=remote.spec.instance_name)
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		- # pull_tree():
231		- #
232		- # Pull a single Tree rather than a ref.
233		- # Does not update local refs.
234		- #
235		- # Args:
236		- # remote (CASRemote): The remote to pull from
237		- # digest (Digest): The digest of the tree
238		- #
239		- def pull_tree(self, remote, digest):
240		- try:
241		- remote.init()
242		-
243		- digest = self._fetch_tree(remote, digest)
244		-
245		- return digest
246		-
247		- except grpc.RpcError as e:
248		- if e.code() != grpc.StatusCode.NOT_FOUND:
249		- raise
250		-
251		- return None
252		-
253	179	# link_ref():
254	180	#
255	181	# Add an alias for an existing ref.
...	...	@@ -263,73 +189,6 @@ class CASCache():
263	189
264	190	self.set_ref(newref, tree)
265	191
266		- # push():
267		- #
268		- # Push committed refs to remote repository.
269		- #
270		- # Args:
271		- # refs (list): The refs to push
272		- # remote (CASRemote): The remote to push to
273		- #
274		- # Returns:
275		- # (bool): True if any remote was updated, False if no pushes were required
276		- #
277		- # Raises:
278		- # (CASError): if there was an error
279		- #
280		- def push(self, refs, remote):
281		- skipped_remote = True
282		- try:
283		- for ref in refs:
284		- tree = self.resolve_ref(ref)
285		-
286		- # Check whether ref is already on the server in which case
287		- # there is no need to push the ref
288		- try:
289		- request = buildstream_pb2.GetReferenceRequest(instance_name=remote.spec.instance_name)
290		- request.key = ref
291		- response = remote.ref_storage.GetReference(request)
292		-
293		- if response.digest.hash == tree.hash and response.digest.size_bytes == tree.size_bytes:
294		- # ref is already on the server with the same tree
295		- continue
296		-
297		- except grpc.RpcError as e:
298		- if e.code() != grpc.StatusCode.NOT_FOUND:
299		- # Intentionally re-raise RpcError for outer except block.
300		- raise
301		-
302		- self._send_directory(remote, tree)
303		-
304		- request = buildstream_pb2.UpdateReferenceRequest(instance_name=remote.spec.instance_name)
305		- request.keys.append(ref)
306		- request.digest.hash = tree.hash
307		- request.digest.size_bytes = tree.size_bytes
308		- remote.ref_storage.UpdateReference(request)
309		-
310		- skipped_remote = False
311		- except grpc.RpcError as e:
312		- if e.code() != grpc.StatusCode.RESOURCE_EXHAUSTED:
313		- raise CASError("Failed to push ref {}: {}".format(refs, e), temporary=True) from e
314		-
315		- return not skipped_remote
316		-
317		- # push_directory():
318		- #
319		- # Push the given virtual directory to a remote.
320		- #
321		- # Args:
322		- # remote (CASRemote): The remote to push to
323		- # directory (Directory): A virtual directory object to push.
324		- #
325		- # Raises:
326		- # (CASError): if there was an error
327		- #
328		- def push_directory(self, remote, directory):
329		- remote.init()
330		-
331		- self._send_directory(remote, directory.ref)
332		-
333	192	# objpath():
334	193	#
335	194	# Return the path of an object based on its digest.
...	...	@@ -591,6 +450,37 @@ class CASCache():
591	450	reachable = set()
592	451	self._reachable_refs_dir(reachable, tree, update_mtime=True)
593	452
	453	+ # Check to see if a blob is in the local CAS
	454	+ # return None if not
	455	+ def check_blob(self, digest):
	456	+ objpath = self.objpath(digest)
	457	+ if os.path.exists(objpath):
	458	+ # already in local repository
	459	+ return objpath
	460	+ else:
	461	+ return None
	462	+
	463	+ def yield_directory_digests(self, directory_digest):
	464	+ # parse directory, and recursively add blobs
	465	+ d = remote_execution_pb2.Digest()
	466	+ d.hash = directory_digest.hash
	467	+ d.size_bytes = directory_digest.size_bytes
	468	+ yield d
	469	+
	470	+ directory = remote_execution_pb2.Directory()
	471	+
	472	+ with open(self.objpath(directory_digest), 'rb') as f:
	473	+ directory.ParseFromString(f.read())
	474	+
	475	+ for filenode in directory.files:
	476	+ d = remote_execution_pb2.Digest()
	477	+ d.hash = filenode.digest.hash
	478	+ d.size_bytes = filenode.digest.size_bytes
	479	+ yield d
	480	+
	481	+ for dirnode in directory.directories:
	482	+ yield from self.yield_directory_digests(dirnode.digest)
	483	+
594	484	################################################
595	485	# Local Private Methods #
596	486	################################################
...	...	@@ -779,202 +669,3 @@ class CASCache():
779	669
780	670	for dirnode in directory.directories:
781	671	yield from self._required_blobs(dirnode.digest)
782		-
783		- # _ensure_blob():
784		- #
785		- # Fetch and add blob if it's not already local.
786		- #
787		- # Args:
788		- # remote (Remote): The remote to use.
789		- # digest (Digest): Digest object for the blob to fetch.
790		- #
791		- # Returns:
792		- # (str): The path of the object
793		- #
794		- def _ensure_blob(self, remote, digest):
795		- objpath = self.objpath(digest)
796		- if os.path.exists(objpath):
797		- # already in local repository
798		- return objpath
799		-
800		- with tempfile.NamedTemporaryFile(dir=self.tmpdir) as f:
801		- remote._fetch_blob(digest, f)
802		-
803		- added_digest = self.add_object(path=f.name, link_directly=True)
804		- assert added_digest.hash == digest.hash
805		-
806		- return objpath
807		-
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		- def _fetch_tree(self, remote, digest):
906		- # download but do not store the Tree object
907		- with tempfile.NamedTemporaryFile(dir=self.tmpdir) as out:
908		- remote._fetch_blob(digest, out)
909		-
910		- tree = remote_execution_pb2.Tree()
911		-
912		- with open(out.name, 'rb') as f:
913		- tree.ParseFromString(f.read())
914		-
915		- tree.children.extend([tree.root])
916		- for directory in tree.children:
917		- for filenode in directory.files:
918		- self._ensure_blob(remote, filenode.digest)
919		-
920		- # place directory blob only in final location when we've downloaded
921		- # all referenced blobs to avoid dangling references in the repository
922		- dirbuffer = directory.SerializeToString()
923		- dirdigest = self.add_object(buffer=dirbuffer)
924		- assert dirdigest.size_bytes == len(dirbuffer)
925		-
926		- return dirdigest
927		-
928		- def _send_directory(self, remote, digest, u_uid=uuid.uuid4()):
929		- required_blobs = self._required_blobs(digest)
930		-
931		- missing_blobs = dict()
932		- # Limit size of FindMissingBlobs request
933		- for required_blobs_group in _grouper(required_blobs, 512):
934		- request = remote_execution_pb2.FindMissingBlobsRequest(instance_name=remote.spec.instance_name)
935		-
936		- for required_digest in required_blobs_group:
937		- d = request.blob_digests.add()
938		- d.hash = required_digest.hash
939		- d.size_bytes = required_digest.size_bytes
940		-
941		- response = remote.cas.FindMissingBlobs(request)
942		- for missing_digest in response.missing_blob_digests:
943		- d = remote_execution_pb2.Digest()
944		- d.hash = missing_digest.hash
945		- d.size_bytes = missing_digest.size_bytes
946		- missing_blobs[d.hash] = d
947		-
948		- # Upload any blobs missing on the server
949		- self._send_blobs(remote, missing_blobs.values(), u_uid)
950		-
951		- def _send_blobs(self, remote, digests, u_uid=uuid.uuid4()):
952		- batch = _CASBatchUpdate(remote)
953		-
954		- for digest in digests:
955		- with open(self.objpath(digest), 'rb') as f:
956		- assert os.fstat(f.fileno()).st_size == digest.size_bytes
957		-
958		- if (digest.size_bytes >= remote.max_batch_total_size_bytes or
959		- not remote.batch_update_supported):
960		- # Too large for batch request, upload in independent request.
961		- remote._send_blob(digest, f, u_uid=u_uid)
962		- else:
963		- if not batch.add(digest, f):
964		- # Not enough space left in batch request.
965		- # Complete pending batch first.
966		- batch.send()
967		- batch = _CASBatchUpdate(remote)
968		- batch.add(digest, f)
969		-
970		- # Send final batch
971		- batch.send()
972		-
973		-
974		-def _grouper(iterable, n):
975		- while True:
976		- try:
977		- current = next(iterable)
978		- except StopIteration:
979		- return
980		- yield itertools.chain([current], itertools.islice(iterable, n - 1))

buildstream/_cas/casremote.py

---

1	1	from collections import namedtuple
2	2	import io
	3	+import itertools
3	4	import os
4	5	import multiprocessing
5	6	import signal
	7	+import tempfile
6	8	from urllib.parse import urlparse
7	9	import uuid
8	10
...	...	@@ -96,6 +98,11 @@ class CASRemote():
96	98	self.tmpdir = str(tmpdir)
97	99	os.makedirs(self.tmpdir, exist_ok=True)
98	100
	101	+ self.__tmp_downloads = [] # files in the tmpdir waiting to be added to local caches
	102	+
	103	+ self.__batch_read = None
	104	+ self.__batch_update = None
	105	+
99	106	def init(self):
100	107	if not self._initialized:
101	108	url = urlparse(self.spec.url)
...	...	@@ -153,6 +160,7 @@ class CASRemote():
153	160	request = remote_execution_pb2.BatchReadBlobsRequest()
154	161	response = self.cas.BatchReadBlobs(request)
155	162	self.batch_read_supported = True
	163	+ self.__batch_read = _CASBatchRead(self)
156	164	except grpc.RpcError as e:
157	165	if e.code() != grpc.StatusCode.UNIMPLEMENTED:
158	166	raise
...	...	@@ -163,6 +171,7 @@ class CASRemote():
163	171	request = remote_execution_pb2.BatchUpdateBlobsRequest()
164	172	response = self.cas.BatchUpdateBlobs(request)
165	173	self.batch_update_supported = True
	174	+ self.__batch_update = _CASBatchUpdate(self)
166	175	except grpc.RpcError as e:
167	176	if (e.code() != grpc.StatusCode.UNIMPLEMENTED and
168	177	e.code() != grpc.StatusCode.PERMISSION_DENIED):
...	...	@@ -259,6 +268,210 @@ class CASRemote():
259	268
260	269	return message_digest
261	270
	271	+ # get_reference():
	272	+ #
	273	+ # Args:
	274	+ # ref (str): The ref to request
	275	+ #
	276	+ # Returns:
	277	+ # (digest): digest of ref, None if not found
	278	+ #
	279	+ def get_reference(self, ref):
	280	+ try:
	281	+ self.init()
	282	+
	283	+ request = buildstream_pb2.GetReferenceRequest()
	284	+ request.key = ref
	285	+ return self.ref_storage.GetReference(request).digest
	286	+ except grpc.RpcError as e:
	287	+ if e.code() != grpc.StatusCode.NOT_FOUND:
	288	+ raise CASError("Failed to find ref {}: {}".format(ref, e)) from e
	289	+ else:
	290	+ return None
	291	+
	292	+ # update_reference():
	293	+ #
	294	+ # Args:
	295	+ # ref (str): Reference to update
	296	+ # digest (Digest): New digest to update ref with
	297	+ def update_reference(self, ref, digest):
	298	+ request = buildstream_pb2.UpdateReferenceRequest()
	299	+ request.keys.append(ref)
	300	+ request.digest.hash = digest.hash
	301	+ request.digest.size_bytes = digest.size_bytes
	302	+ self.ref_storage.UpdateReference(request)
	303	+
	304	+ def get_tree_blob(self, tree_digest):
	305	+ self.init()
	306	+ f = tempfile.NamedTemporaryFile(dir=self.tmpdir)
	307	+ self._fetch_blob(tree_digest, f)
	308	+
	309	+ tree = remote_execution_pb2.Tree()
	310	+ with open(f.name, 'rb') as tmp:
	311	+ tree.ParseFromString(tmp.read())
	312	+
	313	+ return tree
	314	+
	315	+ # yield_directory_digests():
	316	+ #
	317	+ # Recursively iterates over digests for files, symbolic links and other
	318	+ # directories starting from a root digest
	319	+ #
	320	+ # Args:
	321	+ # root_digest (digest): The root_digest to get a tree of
	322	+ # progress (callable): The progress callback, if any
	323	+ # subdir (str): The optional specific subdir to pull
	324	+ # excluded_subdirs (list): The optional list of subdirs to not pull
	325	+ #
	326	+ # Returns:
	327	+ # (iter digests): recursively iterates over digests contained in root directory
	328	+ #
	329	+ def yield_directory_digests(self, root_digest, *, progress=None,
	330	+ subdir=None, excluded_subdirs=None):
	331	+ self.init()
	332	+
	333	+ # Fetch artifact, excluded_subdirs determined in pullqueue
	334	+ if excluded_subdirs is None:
	335	+ excluded_subdirs = []
	336	+
	337	+ # get directory blob
	338	+ f = tempfile.NamedTemporaryFile(dir=self.tmpdir)
	339	+ self._fetch_blob(root_digest, f)
	340	+
	341	+ directory = remote_execution_pb2.Directory()
	342	+ with open(f.name, 'rb') as tmp:
	343	+ directory.ParseFromString(tmp.read())
	344	+
	345	+ yield root_digest
	346	+ for filenode in directory.files:
	347	+ yield filenode.digest
	348	+
	349	+ for dirnode in directory.directories:
	350	+ if dirnode.name not in excluded_subdirs:
	351	+ yield from self.yield_directory_digests(dirnode.digest)
	352	+
	353	+ # yield_tree_digests():
	354	+ #
	355	+ # Fetches a tree file from digests and then iterates over child digests
	356	+ #
	357	+ # Args:
	358	+ # tree_digest (digest): tree digest
	359	+ #
	360	+ # Returns:
	361	+ # (iter digests): iterates over digests in tree message
	362	+ def yield_tree_digests(self, tree_digest):
	363	+ self.init()
	364	+
	365	+ # get tree file
	366	+ f = tempfile.NamedTemporaryFile(dir=self.tmpdir)
	367	+ self._fetch_blob(tree_digest, f)
	368	+ tree = remote_execution_pb2.Tree()
	369	+ tree.ParseFromString(f.read())
	370	+
	371	+ tree.children.extend([tree.root])
	372	+ for directory in tree.children:
	373	+ for filenode in directory.files:
	374	+ yield filenode.digest
	375	+
	376	+ # add the directory to downloaded tmp files to be added
	377	+ f2 = tempfile.NamedTemporaryFile(dir=self.tmpdir)
	378	+ f2.write(directory.SerializeToString())
	379	+ self.__tmp_downloads.append(f2)
	380	+
	381	+ # Add the tree directory to downloads right at the end
	382	+ self.__tmp_downloads.append(f)
	383	+
	384	+ # request_blob():
	385	+ #
	386	+ # Request blob, triggering download depending via bytestream or cas
	387	+ # BatchReadBlobs depending on size.
	388	+ #
	389	+ # Args:
	390	+ # digest (Digest): digest of the requested blob
	391	+ #
	392	+ def request_blob(self, digest):
	393	+ if (not self.batch_read_supported or
	394	+ digest.size_bytes > self.max_batch_total_size_bytes):
	395	+ f = tempfile.NamedTemporaryFile(dir=self.tmpdir)
	396	+ self._fetch_blob(digest, f)
	397	+ self.__tmp_downloads.append(f)
	398	+ elif self.__batch_read.add(digest) is False:
	399	+ self._download_batch()
	400	+ self.__batch_read.add(digest)
	401	+
	402	+ # get_blobs():
	403	+ #
	404	+ # Yield over downloaded blobs in the tmp file locations, causing the files
	405	+ # to be deleted once they go out of scope.
	406	+ #
	407	+ # Args:
	408	+ # complete_batch (bool): download any outstanding batch read request
	409	+ #
	410	+ # Returns:
	411	+ # iterator over NamedTemporaryFile
	412	+ def get_blobs(self, complete_batch=False):
	413	+ # Send read batch request and download
	414	+ if (complete_batch is True and
	415	+ self.batch_read_supported is True):
	416	+ self._download_batch()
	417	+
	418	+ while self.__tmp_downloads:
	419	+ yield self.__tmp_downloads.pop()
	420	+
	421	+ # upload_blob():
	422	+ #
	423	+ # Push blobs given an iterator over blob files
	424	+ #
	425	+ def upload_blob(self, digest, blob_file, u_uid=uuid.uuid4(), final=False):
	426	+ with open(blob_file, 'rb') as f:
	427	+ assert os.fstat(f.fileno()).st_size == digest.size_bytes
	428	+
	429	+ if (digest.size_bytes >= self.max_batch_total_size_bytes or
	430	+ not self.batch_update_supported):
	431	+ # Too large for batch request, upload in independent request.
	432	+ self._send_blob(digest, f, u_uid=u_uid)
	433	+ else:
	434	+ if self.__batch_update.add(digest, f) is False:
	435	+ self.__batch_update.send()
	436	+ self.__batch_update = _CASBatchUpdate(self)
	437	+ self.__batch_update.add(digest, f)
	438	+
	439	+ def send_update_batch(self):
	440	+ # make sure everything is sent
	441	+ self.__batch_update.send()
	442	+ self.__batch_update = _CASBatchUpdate(self)
	443	+
	444	+ # find_missing_blobs()
	445	+ #
	446	+ # Does FindMissingBlobs request to remote
	447	+ #
	448	+ # Args:
	449	+ # required_blobs ([Digest]): list of blobs required
	450	+ # u_uid (str): uuid4
	451	+ #
	452	+ # Returns:
	453	+ # (Dict(Digest)): missing blobs
	454	+ def find_missing_blobs(self, required_blobs, u_uid=uuid.uuid4()):
	455	+ self.init()
	456	+ missing_blobs = dict()
	457	+ # Limit size of FindMissingBlobs request
	458	+ for required_blobs_group in _grouper(required_blobs, 512):
	459	+ request = remote_execution_pb2.FindMissingBlobsRequest()
	460	+
	461	+ for required_digest in required_blobs_group:
	462	+ d = request.blob_digests.add()
	463	+ d.hash = required_digest.hash
	464	+ d.size_bytes = required_digest.size_bytes
	465	+
	466	+ response = self.cas.FindMissingBlobs(request)
	467	+ for missing_digest in response.missing_blob_digests:
	468	+ d = remote_execution_pb2.Digest()
	469	+ d.hash = missing_digest.hash
	470	+ d.size_bytes = missing_digest.size_bytes
	471	+ missing_blobs[d.hash] = d
	472	+
	473	+ return missing_blobs
	474	+
262	475	################################################
263	476	# Local Private Methods #
264	477	################################################
...	...	@@ -301,6 +514,24 @@ class CASRemote():
301	514
302	515	assert response.committed_size == digest.size_bytes
303	516
	517	+ def _download_batch(self):
	518	+ for _, data in self.__batch_read.send():
	519	+ f = tempfile.NamedTemporaryFile(dir=self.tmpdir)
	520	+ f.write(data)
	521	+ f.flush()
	522	+ self.__tmp_downloads.append(f)
	523	+
	524	+ self.__batch_read = _CASBatchRead(self)
	525	+
	526	+
	527	+def _grouper(iterable, n):
	528	+ while True:
	529	+ try:
	530	+ current = next(iterable)
	531	+ except StopIteration:
	532	+ return
	533	+ yield itertools.chain([current], itertools.islice(iterable, n - 1))
	534	+
304	535
305	536	# Represents a batch of blobs queued for fetching.
306	537	#

tests/artifactcache/pull.py

---

...	...	@@ -110,7 +110,7 @@ def test_pull(cli, tmpdir, datafiles):
110	110	# See https://github.com/grpc/grpc/blob/master/doc/fork_support.md for details
111	111	process = multiprocessing.Process(target=_queue_wrapper,
112	112	args=(_test_pull, queue, user_config_file, project_dir,
113		- artifact_dir, 'target.bst', element_key))
	113	+ artifact_dir, tmpdir, 'target.bst', element_key))
114	114
115	115	try:
116	116	# Keep SIGINT blocked in the child process
...	...	@@ -126,14 +126,18 @@ def test_pull(cli, tmpdir, datafiles):
126	126	assert not error
127	127	assert cas.contains(element, element_key)
128	128
	129	+ # Check that the tmp dir is cleared out
	130	+ assert os.listdir(os.path.join(str(tmpdir), 'cache', 'tmp')) == []
129	131
130		-def _test_pull(user_config_file, project_dir, artifact_dir,
	132	+
	133	+def _test_pull(user_config_file, project_dir, artifact_dir, tmpdir,
131	134	element_name, element_key, queue):
132	135	# Fake minimal context
133	136	context = Context()
134	137	context.load(config=user_config_file)
135	138	context.artifactdir = artifact_dir
136	139	context.set_message_handler(message_handler)
	140	+ context.tmpdir = os.path.join(str(tmpdir), 'cache', 'tmp')
137	141
138	142	# Load the project manually
139	143	project = Project(project_dir, context)
...	...	@@ -218,7 +222,7 @@ def test_pull_tree(cli, tmpdir, datafiles):
218	222	# See https://github.com/grpc/grpc/blob/master/doc/fork_support.md for details
219	223	process = multiprocessing.Process(target=_queue_wrapper,
220	224	args=(_test_push_tree, queue, user_config_file, project_dir,
221		- artifact_dir, artifact_digest))
	225	+ artifact_dir, tmpdir, artifact_digest))
222	226
223	227	try:
224	228	# Keep SIGINT blocked in the child process
...	...	@@ -239,6 +243,9 @@ def test_pull_tree(cli, tmpdir, datafiles):
239	243	# Assert that we are not cached locally anymore
240	244	assert cli.get_element_state(project_dir, 'target.bst') != 'cached'
241	245
	246	+ # Check that the tmp dir is cleared out
	247	+ assert os.listdir(os.path.join(str(tmpdir), 'cache', 'tmp')) == []
	248	+
242	249	tree_digest = remote_execution_pb2.Digest(hash=tree_hash,
243	250	size_bytes=tree_size)
244	251
...	...	@@ -246,7 +253,7 @@ def test_pull_tree(cli, tmpdir, datafiles):
246	253	# Use subprocess to avoid creation of gRPC threads in main BuildStream process
247	254	process = multiprocessing.Process(target=_queue_wrapper,
248	255	args=(_test_pull_tree, queue, user_config_file, project_dir,
249		- artifact_dir, tree_digest))
	256	+ artifact_dir, tmpdir, tree_digest))
250	257
251	258	try:
252	259	# Keep SIGINT blocked in the child process
...	...	@@ -267,13 +274,18 @@ def test_pull_tree(cli, tmpdir, datafiles):
267	274	# Ensure the entire Tree stucture has been pulled
268	275	assert os.path.exists(cas.objpath(directory_digest))
269	276
	277	+ # Check that the tmp dir is cleared out
	278	+ assert os.listdir(os.path.join(str(tmpdir), 'cache', 'tmp')) == []
	279	+
270	280
271		-def _test_push_tree(user_config_file, project_dir, artifact_dir, artifact_digest, queue):
	281	+def _test_push_tree(user_config_file, project_dir, artifact_dir, tmpdir,
	282	+ artifact_digest, queue):
272	283	# Fake minimal context
273	284	context = Context()
274	285	context.load(config=user_config_file)
275	286	context.artifactdir = artifact_dir
276	287	context.set_message_handler(message_handler)
	288	+ context.tmpdir = os.path.join(str(tmpdir), 'cache', 'tmp')
277	289
278	290	# Load the project manually
279	291	project = Project(project_dir, context)
...	...	@@ -304,12 +316,14 @@ def _test_push_tree(user_config_file, project_dir, artifact_dir, artifact_digest
304	316	queue.put("No remote configured")
305	317
306	318
307		-def _test_pull_tree(user_config_file, project_dir, artifact_dir, artifact_digest, queue):
	319	+def _test_pull_tree(user_config_file, project_dir, artifact_dir, tmpdir,
	320	+ artifact_digest, queue):
308	321	# Fake minimal context
309	322	context = Context()
310	323	context.load(config=user_config_file)
311	324	context.artifactdir = artifact_dir
312	325	context.set_message_handler(message_handler)
	326	+ context.tmpdir = os.path.join(str(tmpdir), 'cache', 'tmp')
313	327
314	328	# Load the project manually
315	329	project = Project(project_dir, context)

tests/artifactcache/push.py

---

...	...	@@ -89,7 +89,7 @@ def test_push(cli, tmpdir, datafiles):
89	89	# See https://github.com/grpc/grpc/blob/master/doc/fork_support.md for details
90	90	process = multiprocessing.Process(target=_queue_wrapper,
91	91	args=(_test_push, queue, user_config_file, project_dir,
92		- artifact_dir, 'target.bst', element_key))
	92	+ artifact_dir, tmpdir, 'target.bst', element_key))
93	93
94	94	try:
95	95	# Keep SIGINT blocked in the child process
...	...	@@ -105,14 +105,18 @@ def test_push(cli, tmpdir, datafiles):
105	105	assert not error
106	106	assert share.has_artifact('test', 'target.bst', element_key)
107	107
	108	+ # Check tmpdir for downloads is cleared
	109	+ assert os.listdir(os.path.join(str(tmpdir), 'cache', 'tmp')) == []
108	110
109		-def _test_push(user_config_file, project_dir, artifact_dir,
	111	+
	112	+def _test_push(user_config_file, project_dir, artifact_dir, tmpdir,
110	113	element_name, element_key, queue):
111	114	# Fake minimal context
112	115	context = Context()
113	116	context.load(config=user_config_file)
114	117	context.artifactdir = artifact_dir
115	118	context.set_message_handler(message_handler)
	119	+ context.tmpdir = os.path.join(str(tmpdir), 'cache', 'tmp')
116	120
117	121	# Load the project manually
118	122	project = Project(project_dir, context)
...	...	@@ -196,9 +200,10 @@ def test_push_directory(cli, tmpdir, datafiles):
196	200	queue = multiprocessing.Queue()
197	201	# Use subprocess to avoid creation of gRPC threads in main BuildStream process
198	202	# See https://github.com/grpc/grpc/blob/master/doc/fork_support.md for details
199		- process = multiprocessing.Process(target=_queue_wrapper,
200		- args=(_test_push_directory, queue, user_config_file,
201		- project_dir, artifact_dir, artifact_digest))
	203	+ process = multiprocessing.Process(
	204	+ target=_queue_wrapper,
	205	+ args=(_test_push_directory, queue, user_config_file, project_dir,
	206	+ artifact_dir, tmpdir, artifact_digest))
202	207
203	208	try:
204	209	# Keep SIGINT blocked in the child process
...	...	@@ -215,13 +220,17 @@ def test_push_directory(cli, tmpdir, datafiles):
215	220	assert artifact_digest.hash == directory_hash
216	221	assert share.has_object(artifact_digest)
217	222
	223	+ assert os.listdir(os.path.join(str(tmpdir), 'cache', 'tmp')) == []
218	224
219		-def _test_push_directory(user_config_file, project_dir, artifact_dir, artifact_digest, queue):
	225	+
	226	+def _test_push_directory(user_config_file, project_dir, artifact_dir, tmpdir,
	227	+ artifact_digest, queue):
220	228	# Fake minimal context
221	229	context = Context()
222	230	context.load(config=user_config_file)
223	231	context.artifactdir = artifact_dir
224	232	context.set_message_handler(message_handler)
	233	+ context.tmpdir = os.path.join(str(tmpdir), 'cache', 'tmp')
225	234
226	235	# Load the project manually
227	236	project = Project(project_dir, context)
...	...	@@ -273,7 +282,7 @@ def test_push_message(cli, tmpdir, datafiles):
273	282	# See https://github.com/grpc/grpc/blob/master/doc/fork_support.md for details
274	283	process = multiprocessing.Process(target=_queue_wrapper,
275	284	args=(_test_push_message, queue, user_config_file,
276		- project_dir, artifact_dir))
	285	+ project_dir, artifact_dir, tmpdir))
277	286
278	287	try:
279	288	# Keep SIGINT blocked in the child process
...	...	@@ -291,13 +300,16 @@ def test_push_message(cli, tmpdir, datafiles):
291	300	size_bytes=message_size)
292	301	assert share.has_object(message_digest)
293	302
	303	+ assert os.listdir(os.path.join(str(tmpdir), 'cache', 'tmp')) == []
	304	+
294	305
295		-def _test_push_message(user_config_file, project_dir, artifact_dir, queue):
	306	+def _test_push_message(user_config_file, project_dir, artifact_dir, tmpdir, queue):
296	307	# Fake minimal context
297	308	context = Context()
298	309	context.load(config=user_config_file)
299	310	context.artifactdir = artifact_dir
300	311	context.set_message_handler(message_handler)
	312	+ context.tmpdir = os.path.join(str(tmpdir), 'cache', 'tmp')
301	313
302	314	# Load the project manually
303	315	project = Project(project_dir, context)

tests/integration/pullbuildtrees.py

---

...	...	@@ -79,6 +79,9 @@ def test_pullbuildtrees(cli, tmpdir, datafiles, integration_cache):
79	79	assert os.path.isdir(buildtreedir)
80	80	default_state(cli, tmpdir, share1)
81	81
	82	+ # Check tmpdir for downloads is cleared
	83	+ assert os.listdir(os.path.join(integration_cache, 'tmp')) == []
	84	+
82	85	# Pull artifact with pullbuildtrees set in user config, then assert
83	86	# that pulling with the same user config doesn't creates a pull job,
84	87	# or when buildtrees cli flag is set.
...	...	@@ -91,6 +94,9 @@ def test_pullbuildtrees(cli, tmpdir, datafiles, integration_cache):
91	94	assert element_name not in result.get_pulled_elements()
92	95	default_state(cli, tmpdir, share1)
93	96
	97	+ # Check tmpdir for downloads is cleared
	98	+ assert os.listdir(os.path.join(integration_cache, 'tmp')) == []
	99	+
94	100	# Pull artifact with default config and buildtrees cli flag set, then assert
95	101	# that pulling with pullbuildtrees set in user config doesn't create a pull
96	102	# job.
...	...	@@ -101,6 +107,9 @@ def test_pullbuildtrees(cli, tmpdir, datafiles, integration_cache):
101	107	assert element_name not in result.get_pulled_elements()
102	108	default_state(cli, tmpdir, share1)
103	109
	110	+ # Check tmpdir for downloads is cleared
	111	+ assert os.listdir(os.path.join(integration_cache, 'tmp')) == []
	112	+
104	113	# Assert that a partial build element (not containing a populated buildtree dir)
105	114	# can't be pushed to an artifact share, then assert that a complete build element
106	115	# can be. This will attempt a partial pull from share1 and then a partial push

—
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.