Martin Blanchard pushed to branch jmac/remote_execution_client at BuildStream / buildstream
Commits:
-
10032dc8
by Jim MacArthur at 2018-09-04T11:08:26Z
-
a788daf5
by Martin Blanchard at 2018-09-04T11:08:28Z
-
29d79445
by Jim MacArthur at 2018-09-04T11:08:28Z
-
1e543f12
by Jim MacArthur at 2018-09-04T11:08:28Z
-
f286e267
by Jim MacArthur at 2018-09-04T11:08:28Z
-
97d821f3
by Jim MacArthur at 2018-09-04T11:08:28Z
-
f1791579
by Jim MacArthur at 2018-09-04T11:08:28Z
-
e7fdb96e
by Jim MacArthur at 2018-09-04T11:08:28Z
-
5787aeff
by Jim MacArthur at 2018-09-04T11:08:28Z
-
ef92a11f
by Jim MacArthur at 2018-09-04T11:08:28Z
-
40ad5608
by Martin Blanchard at 2018-09-04T11:08:28Z
-
b6f4cb20
by Martin Blanchard at 2018-09-04T11:08:28Z
-
111214d4
by Martin Blanchard at 2018-09-04T11:08:28Z
-
c571b9da
by Martin Blanchard at 2018-09-04T11:08:28Z
15 changed files:
- buildstream/_artifactcache/cascache.py
- buildstream/_project.py
- buildstream/buildelement.py
- buildstream/data/projectconfig.yaml
- buildstream/element.py
- buildstream/plugins/elements/autotools.py
- buildstream/plugins/elements/cmake.py
- buildstream/plugins/elements/make.py
- buildstream/plugins/elements/meson.py
- buildstream/plugins/elements/qmake.py
- buildstream/sandbox/__init__.py
- + buildstream/sandbox/_sandboxremote.py
- buildstream/sandbox/sandbox.py
- buildstream/storage/_casbaseddirectory.py
- doc/source/format_project.rst
Changes:
... | ... | @@ -19,6 +19,7 @@ |
19 | 19 |
|
20 | 20 |
import hashlib
|
21 | 21 |
import itertools
|
22 |
+import io
|
|
22 | 23 |
import multiprocessing
|
23 | 24 |
import os
|
24 | 25 |
import signal
|
... | ... | @@ -66,6 +67,10 @@ class CASCache(ArtifactCache): |
66 | 67 |
self._calculate_cache_quota()
|
67 | 68 |
|
68 | 69 |
self._enable_push = enable_push
|
70 |
+ if self._enable_push:
|
|
71 |
+ self._uuid = str(uuid.uuid4())
|
|
72 |
+ else:
|
|
73 |
+ self._uuid = None
|
|
69 | 74 |
|
70 | 75 |
# Per-project list of _CASRemote instances.
|
71 | 76 |
self._remotes = {}
|
... | ... | @@ -76,6 +81,7 @@ class CASCache(ArtifactCache): |
76 | 81 |
################################################
|
77 | 82 |
# Implementation of abstract methods #
|
78 | 83 |
################################################
|
84 |
+ |
|
79 | 85 |
def contains(self, element, key):
|
80 | 86 |
refpath = self._refpath(self.get_artifact_fullname(element, key))
|
81 | 87 |
|
... | ... | @@ -259,6 +265,25 @@ class CASCache(ArtifactCache): |
259 | 265 |
|
260 | 266 |
return False
|
261 | 267 |
|
268 |
+ def pull_tree(self, project, digest):
|
|
269 |
+ """ Pull a single Tree rather than an artifact.
|
|
270 |
+ Does not update local refs. """
|
|
271 |
+ |
|
272 |
+ for remote in self._remotes[project]:
|
|
273 |
+ try:
|
|
274 |
+ remote.init()
|
|
275 |
+ |
|
276 |
+ digest = self._fetch_tree(remote, digest)
|
|
277 |
+ |
|
278 |
+ # no need to pull from additional remotes
|
|
279 |
+ return digest
|
|
280 |
+ |
|
281 |
+ except grpc.RpcError as e:
|
|
282 |
+ if e.code() != grpc.StatusCode.NOT_FOUND:
|
|
283 |
+ raise
|
|
284 |
+ |
|
285 |
+ return None
|
|
286 |
+ |
|
262 | 287 |
def link_key(self, element, oldkey, newkey):
|
263 | 288 |
oldref = self.get_artifact_fullname(element, oldkey)
|
264 | 289 |
newref = self.get_artifact_fullname(element, newkey)
|
... | ... | @@ -267,8 +292,46 @@ class CASCache(ArtifactCache): |
267 | 292 |
|
268 | 293 |
self.set_ref(newref, tree)
|
269 | 294 |
|
295 |
+ def _push_refs_to_remote(self, refs, remote):
|
|
296 |
+ skipped_remote = True
|
|
297 |
+ try:
|
|
298 |
+ for ref in refs:
|
|
299 |
+ tree = self.resolve_ref(ref)
|
|
300 |
+ |
|
301 |
+ # Check whether ref is already on the server in which case
|
|
302 |
+ # there is no need to push the artifact
|
|
303 |
+ try:
|
|
304 |
+ request = buildstream_pb2.GetReferenceRequest()
|
|
305 |
+ request.key = ref
|
|
306 |
+ response = remote.ref_storage.GetReference(request)
|
|
307 |
+ |
|
308 |
+ if response.digest.hash == tree.hash and response.digest.size_bytes == tree.size_bytes:
|
|
309 |
+ # ref is already on the server with the same tree
|
|
310 |
+ continue
|
|
311 |
+ |
|
312 |
+ except grpc.RpcError as e:
|
|
313 |
+ if e.code() != grpc.StatusCode.NOT_FOUND:
|
|
314 |
+ # Intentionally re-raise RpcError for outer except block.
|
|
315 |
+ raise
|
|
316 |
+ |
|
317 |
+ self._send_directory(remote, tree)
|
|
318 |
+ |
|
319 |
+ request = buildstream_pb2.UpdateReferenceRequest()
|
|
320 |
+ request.keys.append(ref)
|
|
321 |
+ request.digest.hash = tree.hash
|
|
322 |
+ request.digest.size_bytes = tree.size_bytes
|
|
323 |
+ remote.ref_storage.UpdateReference(request)
|
|
324 |
+ |
|
325 |
+ skipped_remote = False
|
|
326 |
+ except grpc.RpcError as e:
|
|
327 |
+ if e.code() != grpc.StatusCode.RESOURCE_EXHAUSTED:
|
|
328 |
+ raise ArtifactError("Failed to push artifact {}: {}".format(refs, e), temporary=True) from e
|
|
329 |
+ |
|
330 |
+ return not skipped_remote
|
|
331 |
+ |
|
270 | 332 |
def push(self, element, keys):
|
271 |
- refs = [self.get_artifact_fullname(element, key) for key in keys]
|
|
333 |
+ |
|
334 |
+ refs = [self.get_artifact_fullname(element, key) for key in list(keys)]
|
|
272 | 335 |
|
273 | 336 |
project = element._get_project()
|
274 | 337 |
|
... | ... | @@ -278,95 +341,77 @@ class CASCache(ArtifactCache): |
278 | 341 |
|
279 | 342 |
for remote in push_remotes:
|
280 | 343 |
remote.init()
|
281 |
- skipped_remote = True
|
|
282 |
- element.info("Pushing {} -> {}".format(element._get_brief_display_key(), remote.spec.url))
|
|
283 |
- |
|
284 |
- try:
|
|
285 |
- for ref in refs:
|
|
286 |
- tree = self.resolve_ref(ref)
|
|
287 |
- |
|
288 |
- # Check whether ref is already on the server in which case
|
|
289 |
- # there is no need to push the artifact
|
|
290 |
- try:
|
|
291 |
- request = buildstream_pb2.GetReferenceRequest()
|
|
292 |
- request.key = ref
|
|
293 |
- response = remote.ref_storage.GetReference(request)
|
|
294 |
- |
|
295 |
- if response.digest.hash == tree.hash and response.digest.size_bytes == tree.size_bytes:
|
|
296 |
- # ref is already on the server with the same tree
|
|
297 |
- continue
|
|
298 |
- |
|
299 |
- except grpc.RpcError as e:
|
|
300 |
- if e.code() != grpc.StatusCode.NOT_FOUND:
|
|
301 |
- # Intentionally re-raise RpcError for outer except block.
|
|
302 |
- raise
|
|
303 |
- |
|
304 |
- missing_blobs = {}
|
|
305 |
- required_blobs = self._required_blobs(tree)
|
|
306 |
- |
|
307 |
- # Limit size of FindMissingBlobs request
|
|
308 |
- for required_blobs_group in _grouper(required_blobs, 512):
|
|
309 |
- request = remote_execution_pb2.FindMissingBlobsRequest()
|
|
310 |
- |
|
311 |
- for required_digest in required_blobs_group:
|
|
312 |
- d = request.blob_digests.add()
|
|
313 |
- d.hash = required_digest.hash
|
|
314 |
- d.size_bytes = required_digest.size_bytes
|
|
315 |
- |
|
316 |
- response = remote.cas.FindMissingBlobs(request)
|
|
317 |
- for digest in response.missing_blob_digests:
|
|
318 |
- d = remote_execution_pb2.Digest()
|
|
319 |
- d.hash = digest.hash
|
|
320 |
- d.size_bytes = digest.size_bytes
|
|
321 |
- missing_blobs[d.hash] = d
|
|
322 |
- |
|
323 |
- # Upload any blobs missing on the server
|
|
324 |
- skipped_remote = False
|
|
325 |
- for digest in missing_blobs.values():
|
|
326 |
- uuid_ = uuid.uuid4()
|
|
327 |
- resource_name = '/'.join(['uploads', str(uuid_), 'blobs',
|
|
328 |
- digest.hash, str(digest.size_bytes)])
|
|
329 |
- |
|
330 |
- def request_stream(resname):
|
|
331 |
- with open(self.objpath(digest), 'rb') as f:
|
|
332 |
- assert os.fstat(f.fileno()).st_size == digest.size_bytes
|
|
333 |
- offset = 0
|
|
334 |
- finished = False
|
|
335 |
- remaining = digest.size_bytes
|
|
336 |
- while not finished:
|
|
337 |
- chunk_size = min(remaining, 64 * 1024)
|
|
338 |
- remaining -= chunk_size
|
|
339 |
- |
|
340 |
- request = bytestream_pb2.WriteRequest()
|
|
341 |
- request.write_offset = offset
|
|
342 |
- # max. 64 kB chunks
|
|
343 |
- request.data = f.read(chunk_size)
|
|
344 |
- request.resource_name = resname
|
|
345 |
- request.finish_write = remaining <= 0
|
|
346 |
- yield request
|
|
347 |
- offset += chunk_size
|
|
348 |
- finished = request.finish_write
|
|
349 |
- response = remote.bytestream.Write(request_stream(resource_name))
|
|
350 |
- |
|
351 |
- request = buildstream_pb2.UpdateReferenceRequest()
|
|
352 |
- request.keys.append(ref)
|
|
353 |
- request.digest.hash = tree.hash
|
|
354 |
- request.digest.size_bytes = tree.size_bytes
|
|
355 |
- remote.ref_storage.UpdateReference(request)
|
|
356 |
- |
|
357 |
- pushed = True
|
|
358 | 344 |
|
359 |
- except grpc.RpcError as e:
|
|
360 |
- if e.code() != grpc.StatusCode.RESOURCE_EXHAUSTED:
|
|
361 |
- raise ArtifactError("Failed to push artifact {}: {}".format(refs, e), temporary=True) from e
|
|
345 |
+ element.info("Pushing {} -> {}".format(element._get_brief_display_key(), remote.spec.url))
|
|
362 | 346 |
|
363 |
- if skipped_remote:
|
|
347 |
+ if self._push_refs_to_remote(refs, remote):
|
|
348 |
+ pushed = True
|
|
349 |
+ else:
|
|
364 | 350 |
self.context.message(Message(
|
365 | 351 |
None,
|
366 | 352 |
MessageType.SKIPPED,
|
367 | 353 |
"Remote ({}) already has {} cached".format(
|
368 | 354 |
remote.spec.url, element._get_brief_display_key())
|
369 | 355 |
))
|
356 |
+ |
|
357 |
+ return pushed
|
|
358 |
+ |
|
359 |
+ def push_directory(self, project, directory):
|
|
360 |
+ |
|
361 |
+ push_remotes = [r for r in self._remotes[project] if r.spec.push]
|
|
362 |
+ |
|
363 |
+ if directory.ref is None:
|
|
364 |
+ return None
|
|
365 |
+ |
|
366 |
+ for remote in push_remotes:
|
|
367 |
+ remote.init()
|
|
368 |
+ |
|
369 |
+ self._send_directory(remote, directory.ref)
|
|
370 |
+ |
|
371 |
+ return directory.ref
|
|
372 |
+ |
|
373 |
+ def push_message(self, project, message):
|
|
374 |
+ |
|
375 |
+ push_remotes = [r for r in self._remotes[project] if r.spec.push]
|
|
376 |
+ |
|
377 |
+ message_buffer = message.SerializeToString()
|
|
378 |
+ message_sha = hashlib.sha256(message_buffer)
|
|
379 |
+ message_digest = remote_execution_pb2.Digest()
|
|
380 |
+ message_digest.hash = message_sha.hexdigest()
|
|
381 |
+ message_digest.size_bytes = len(message_buffer)
|
|
382 |
+ |
|
383 |
+ for remote in push_remotes:
|
|
384 |
+ remote.init()
|
|
385 |
+ |
|
386 |
+ with io.BytesIO(message_buffer) as b:
|
|
387 |
+ self._send_blob(remote, message_digest, b)
|
|
388 |
+ |
|
389 |
+ return message_digest
|
|
390 |
+ |
|
391 |
+ def _verify_digest_on_remote(self, remote, digest):
|
|
392 |
+ # Check whether ref is already on the server in which case
|
|
393 |
+ # there is no need to push the artifact
|
|
394 |
+ request = remote_execution_pb2.FindMissingBlobsRequest()
|
|
395 |
+ request.blob_digests.extend([digest])
|
|
396 |
+ |
|
397 |
+ response = remote.cas.FindMissingBlobs(request)
|
|
398 |
+ if digest in response.missing_blob_digests:
|
|
399 |
+ return False
|
|
400 |
+ |
|
401 |
+ return True
|
|
402 |
+ |
|
403 |
+ def verify_key_pushed(self, project, digest):
|
|
404 |
+ |
|
405 |
+ push_remotes = [r for r in self._remotes[project] if r.spec.push]
|
|
406 |
+ |
|
407 |
+ pushed = False
|
|
408 |
+ |
|
409 |
+ for remote in push_remotes:
|
|
410 |
+ remote.init()
|
|
411 |
+ |
|
412 |
+ if self._verify_digest_on_remote(remote, digest):
|
|
413 |
+ pushed = True
|
|
414 |
+ |
|
370 | 415 |
return pushed
|
371 | 416 |
|
372 | 417 |
################################################
|
... | ... | @@ -599,6 +644,7 @@ class CASCache(ArtifactCache): |
599 | 644 |
################################################
|
600 | 645 |
# Local Private Methods #
|
601 | 646 |
################################################
|
647 |
+ |
|
602 | 648 |
def _checkout(self, dest, tree):
|
603 | 649 |
os.makedirs(dest, exist_ok=True)
|
604 | 650 |
|
... | ... | @@ -782,16 +828,16 @@ class CASCache(ArtifactCache): |
782 | 828 |
for dirnode in directory.directories:
|
783 | 829 |
yield from self._required_blobs(dirnode.digest)
|
784 | 830 |
|
785 |
- def _fetch_blob(self, remote, digest, out):
|
|
831 |
+ def _fetch_blob(self, remote, digest, stream):
|
|
786 | 832 |
resource_name = '/'.join(['blobs', digest.hash, str(digest.size_bytes)])
|
787 | 833 |
request = bytestream_pb2.ReadRequest()
|
788 | 834 |
request.resource_name = resource_name
|
789 | 835 |
request.read_offset = 0
|
790 | 836 |
for response in remote.bytestream.Read(request):
|
791 |
- out.write(response.data)
|
|
837 |
+ stream.write(response.data)
|
|
838 |
+ stream.flush()
|
|
792 | 839 |
|
793 |
- out.flush()
|
|
794 |
- assert digest.size_bytes == os.fstat(out.fileno()).st_size
|
|
840 |
+ assert digest.size_bytes == os.fstat(stream.fileno()).st_size
|
|
795 | 841 |
|
796 | 842 |
def _fetch_directory(self, remote, tree):
|
797 | 843 |
objpath = self.objpath(tree)
|
... | ... | @@ -827,6 +873,92 @@ class CASCache(ArtifactCache): |
827 | 873 |
digest = self.add_object(path=out.name)
|
828 | 874 |
assert digest.hash == tree.hash
|
829 | 875 |
|
876 |
+ def _fetch_tree(self, remote, digest):
|
|
877 |
+ # download but do not store the Tree object
|
|
878 |
+ with tempfile.NamedTemporaryFile(dir=self.tmpdir) as out:
|
|
879 |
+ self._fetch_blob(remote, digest, out)
|
|
880 |
+ |
|
881 |
+ tree = remote_execution_pb2.Tree()
|
|
882 |
+ |
|
883 |
+ with open(out.name, 'rb') as f:
|
|
884 |
+ tree.ParseFromString(f.read())
|
|
885 |
+ |
|
886 |
+ tree.children.extend([tree.root])
|
|
887 |
+ for directory in tree.children:
|
|
888 |
+ for filenode in directory.files:
|
|
889 |
+ fileobjpath = self.objpath(filenode.digest)
|
|
890 |
+ if os.path.exists(fileobjpath):
|
|
891 |
+ # already in local cache
|
|
892 |
+ continue
|
|
893 |
+ |
|
894 |
+ with tempfile.NamedTemporaryFile(dir=self.tmpdir) as f:
|
|
895 |
+ self._fetch_blob(remote, filenode.digest, f)
|
|
896 |
+ |
|
897 |
+ added_digest = self.add_object(path=f.name)
|
|
898 |
+ assert added_digest.hash == filenode.digest.hash
|
|
899 |
+ |
|
900 |
+ # place directory blob only in final location when we've downloaded
|
|
901 |
+ # all referenced blobs to avoid dangling references in the repository
|
|
902 |
+ dirbuffer = directory.SerializeToString()
|
|
903 |
+ dirdigest = self.add_object(buffer=dirbuffer)
|
|
904 |
+ assert dirdigest.size_bytes == len(dirbuffer)
|
|
905 |
+ |
|
906 |
+ return dirdigest
|
|
907 |
+ |
|
908 |
+ def _send_blob(self, remote, digest, stream):
|
|
909 |
+ resource_name = '/'.join(['uploads', self._uuid, 'blobs',
|
|
910 |
+ digest.hash, str(digest.size_bytes)])
|
|
911 |
+ |
|
912 |
+ def request_stream(resname, instream):
|
|
913 |
+ offset = 0
|
|
914 |
+ finished = False
|
|
915 |
+ remaining = digest.size_bytes
|
|
916 |
+ while not finished:
|
|
917 |
+ chunk_size = min(remaining, 64 * 1024)
|
|
918 |
+ remaining -= chunk_size
|
|
919 |
+ |
|
920 |
+ request = bytestream_pb2.WriteRequest()
|
|
921 |
+ request.write_offset = offset
|
|
922 |
+ # max. 64 kB chunks
|
|
923 |
+ request.data = instream.read(chunk_size)
|
|
924 |
+ request.resource_name = resname
|
|
925 |
+ request.finish_write = remaining <= 0
|
|
926 |
+ |
|
927 |
+ yield request
|
|
928 |
+ |
|
929 |
+ offset += chunk_size
|
|
930 |
+ finished = request.finish_write
|
|
931 |
+ |
|
932 |
+ response = remote.bytestream.Write(request_stream(resource_name, stream))
|
|
933 |
+ |
|
934 |
+ assert response.committed_size == digest.size_bytes
|
|
935 |
+ |
|
936 |
+ def _send_directory(self, remote, digest):
|
|
937 |
+ required_blobs = self._required_blobs(digest)
|
|
938 |
+ |
|
939 |
+ missing_blobs = dict()
|
|
940 |
+ # Limit size of FindMissingBlobs request
|
|
941 |
+ for required_blobs_group in _grouper(required_blobs, 512):
|
|
942 |
+ request = remote_execution_pb2.FindMissingBlobsRequest()
|
|
943 |
+ |
|
944 |
+ for required_digest in required_blobs_group:
|
|
945 |
+ d = request.blob_digests.add()
|
|
946 |
+ d.hash = required_digest.hash
|
|
947 |
+ d.size_bytes = required_digest.size_bytes
|
|
948 |
+ |
|
949 |
+ response = remote.cas.FindMissingBlobs(request)
|
|
950 |
+ for missing_digest in response.missing_blob_digests:
|
|
951 |
+ d = remote_execution_pb2.Digest()
|
|
952 |
+ d.hash = missing_digest.hash
|
|
953 |
+ d.size_bytes = missing_digest.size_bytes
|
|
954 |
+ missing_blobs[d.hash] = d
|
|
955 |
+ |
|
956 |
+ # Upload any blobs missing on the server
|
|
957 |
+ for blob_digest in missing_blobs.values():
|
|
958 |
+ with open(self.objpath(blob_digest), 'rb') as f:
|
|
959 |
+ assert os.fstat(f.fileno()).st_size == blob_digest.size_bytes
|
|
960 |
+ self._send_blob(remote, blob_digest, f)
|
|
961 |
+ |
|
830 | 962 |
|
831 | 963 |
# Represents a single remote CAS cache.
|
832 | 964 |
#
|
... | ... | @@ -129,6 +129,7 @@ class Project(): |
129 | 129 |
|
130 | 130 |
self.artifact_cache_specs = None
|
131 | 131 |
self._sandbox = None
|
132 |
+ self._remote_execution = None
|
|
132 | 133 |
self._splits = None
|
133 | 134 |
|
134 | 135 |
self._context.add_project(self)
|
... | ... | @@ -471,7 +472,7 @@ class Project(): |
471 | 472 |
'aliases', 'name',
|
472 | 473 |
'artifacts', 'options',
|
473 | 474 |
'fail-on-overlap', 'shell', 'fatal-warnings',
|
474 |
- 'ref-storage', 'sandbox', 'mirrors'
|
|
475 |
+ 'ref-storage', 'sandbox', 'mirrors', 'remote-execution'
|
|
475 | 476 |
])
|
476 | 477 |
|
477 | 478 |
#
|
... | ... | @@ -489,6 +490,9 @@ class Project(): |
489 | 490 |
# Load sandbox configuration
|
490 | 491 |
self._sandbox = _yaml.node_get(config, Mapping, 'sandbox')
|
491 | 492 |
|
493 |
+ # Load remote execution configuration
|
|
494 |
+ self._remote_execution = _yaml.node_get(config, Mapping, 'remote-execution')
|
|
495 |
+ |
|
492 | 496 |
# Load project split rules
|
493 | 497 |
self._splits = _yaml.node_get(config, Mapping, 'split-rules')
|
494 | 498 |
|
... | ... | @@ -155,6 +155,9 @@ class BuildElement(Element): |
155 | 155 |
command_dir = build_root
|
156 | 156 |
sandbox.set_work_directory(command_dir)
|
157 | 157 |
|
158 |
+ # Tell sandbox which directory is preserved in the finished artifact
|
|
159 |
+ sandbox.set_output_directory(install_root)
|
|
160 |
+ |
|
158 | 161 |
# Setup environment
|
159 | 162 |
sandbox.set_environment(self.get_environment())
|
160 | 163 |
|
... | ... | @@ -204,3 +204,6 @@ shell: |
204 | 204 |
# Command to run when `bst shell` does not provide a command
|
205 | 205 |
#
|
206 | 206 |
command: [ 'sh', '-i' ]
|
207 |
+ |
|
208 |
+remote-execution:
|
|
209 |
+ url: ""
|
|
\ No newline at end of file |
... | ... | @@ -95,6 +95,7 @@ from . import _site |
95 | 95 |
from ._platform import Platform
|
96 | 96 |
from .plugin import CoreWarnings
|
97 | 97 |
from .sandbox._config import SandboxConfig
|
98 |
+from .sandbox._sandboxremote import SandboxRemote
|
|
98 | 99 |
|
99 | 100 |
from .storage.directory import Directory
|
100 | 101 |
from .storage._filebaseddirectory import FileBasedDirectory
|
... | ... | @@ -250,6 +251,9 @@ class Element(Plugin): |
250 | 251 |
# Extract Sandbox config
|
251 | 252 |
self.__sandbox_config = self.__extract_sandbox_config(meta)
|
252 | 253 |
|
254 |
+ # Extract remote execution URL
|
|
255 |
+ self.__remote_execution_url = self.__extract_remote_execution_config(meta)
|
|
256 |
+ |
|
253 | 257 |
def __lt__(self, other):
|
254 | 258 |
return self.name < other.name
|
255 | 259 |
|
... | ... | @@ -1570,6 +1574,8 @@ class Element(Plugin): |
1570 | 1574 |
finally:
|
1571 | 1575 |
if collect is not None:
|
1572 | 1576 |
try:
|
1577 |
+ # Sandbox will probably have replaced its virtual directory, so get it again
|
|
1578 |
+ sandbox_vroot = sandbox.get_virtual_directory()
|
|
1573 | 1579 |
collectvdir = sandbox_vroot.descend(collect.lstrip(os.sep).split(os.sep))
|
1574 | 1580 |
except VirtualDirectoryError:
|
1575 | 1581 |
# No collect directory existed
|
... | ... | @@ -2146,7 +2152,32 @@ class Element(Plugin): |
2146 | 2152 |
project = self._get_project()
|
2147 | 2153 |
platform = Platform.get_platform()
|
2148 | 2154 |
|
2149 |
- if directory is not None and os.path.exists(directory):
|
|
2155 |
+ if self.__remote_execution_url and self.BST_VIRTUAL_DIRECTORY:
|
|
2156 |
+ if not self.__artifacts.has_push_remotes(element=self):
|
|
2157 |
+ # Give an early warning if remote execution will not work
|
|
2158 |
+ raise ElementError("Artifact {} is configured to use remote execution but has no push remotes. "
|
|
2159 |
+ .format(self.name) +
|
|
2160 |
+ "The remote artifact server(s) may not be correctly configured or contactable.")
|
|
2161 |
+ |
|
2162 |
+ self.info("Using a remote sandbox for artifact {}".format(self.name))
|
|
2163 |
+ |
|
2164 |
+ sandbox = SandboxRemote(context, project,
|
|
2165 |
+ directory,
|
|
2166 |
+ stdout=stdout,
|
|
2167 |
+ stderr=stderr,
|
|
2168 |
+ config=config,
|
|
2169 |
+ server_url=self.__remote_execution_url,
|
|
2170 |
+ allow_real_directory=False)
|
|
2171 |
+ yield sandbox
|
|
2172 |
+ |
|
2173 |
+ elif directory is not None and os.path.exists(directory):
|
|
2174 |
+ if self.__remote_execution_url:
|
|
2175 |
+ self.warn("Artifact {} is configured to use remote execution but element plugin does not support it."
|
|
2176 |
+ .format(self.name), detail="Element plugin '{kind}' does not support virtual directories."
|
|
2177 |
+ .format(kind=self.get_kind()), warning_token="remote-failure")
|
|
2178 |
+ |
|
2179 |
+ self.info("Falling back to local sandbox for artifact {}".format(self.name))
|
|
2180 |
+ |
|
2150 | 2181 |
sandbox = platform.create_sandbox(context, project,
|
2151 | 2182 |
directory,
|
2152 | 2183 |
stdout=stdout,
|
... | ... | @@ -2318,6 +2349,20 @@ class Element(Plugin): |
2318 | 2349 |
return SandboxConfig(self.node_get_member(sandbox_config, int, 'build-uid'),
|
2319 | 2350 |
self.node_get_member(sandbox_config, int, 'build-gid'))
|
2320 | 2351 |
|
2352 |
+ # Remote execution configuration data (server URL), to be used by the remote sandbox.
|
|
2353 |
+ #
|
|
2354 |
+ def __extract_remote_execution_config(self, meta):
|
|
2355 |
+ if self.__is_junction:
|
|
2356 |
+ return None
|
|
2357 |
+ else:
|
|
2358 |
+ project = self._get_project()
|
|
2359 |
+ project.ensure_fully_loaded()
|
|
2360 |
+ if project._remote_execution:
|
|
2361 |
+ rexec_config = _yaml.node_chain_copy(project._remote_execution)
|
|
2362 |
+ return self.node_get_member(rexec_config, str, 'url')
|
|
2363 |
+ else:
|
|
2364 |
+ return None
|
|
2365 |
+ |
|
2321 | 2366 |
# This makes a special exception for the split rules, which
|
2322 | 2367 |
# elements may extend but whos defaults are defined in the project.
|
2323 | 2368 |
#
|
... | ... | @@ -57,7 +57,8 @@ from buildstream import BuildElement |
57 | 57 |
|
58 | 58 |
# Element implementation for the 'autotools' kind.
|
59 | 59 |
class AutotoolsElement(BuildElement):
|
60 |
- pass
|
|
60 |
+ # Supports virtual directories (required for remote execution)
|
|
61 |
+ BST_VIRTUAL_DIRECTORY = True
|
|
61 | 62 |
|
62 | 63 |
|
63 | 64 |
# Plugin entry point
|
... | ... | @@ -56,7 +56,8 @@ from buildstream import BuildElement |
56 | 56 |
|
57 | 57 |
# Element implementation for the 'cmake' kind.
|
58 | 58 |
class CMakeElement(BuildElement):
|
59 |
- pass
|
|
59 |
+ # Supports virtual directories (required for remote execution)
|
|
60 |
+ BST_VIRTUAL_DIRECTORY = True
|
|
60 | 61 |
|
61 | 62 |
|
62 | 63 |
# Plugin entry point
|
... | ... | @@ -38,7 +38,8 @@ from buildstream import BuildElement |
38 | 38 |
|
39 | 39 |
# Element implementation for the 'make' kind.
|
40 | 40 |
class MakeElement(BuildElement):
|
41 |
- pass
|
|
41 |
+ # Supports virtual directories (required for remote execution)
|
|
42 |
+ BST_VIRTUAL_DIRECTORY = True
|
|
42 | 43 |
|
43 | 44 |
|
44 | 45 |
# Plugin entry point
|
... | ... | @@ -53,7 +53,8 @@ from buildstream import BuildElement |
53 | 53 |
|
54 | 54 |
# Element implementation for the 'meson' kind.
|
55 | 55 |
class MesonElement(BuildElement):
|
56 |
- pass
|
|
56 |
+ # Supports virtual directories (required for remote execution)
|
|
57 |
+ BST_VIRTUAL_DIRECTORY = True
|
|
57 | 58 |
|
58 | 59 |
|
59 | 60 |
# Plugin entry point
|
... | ... | @@ -33,7 +33,8 @@ from buildstream import BuildElement |
33 | 33 |
|
34 | 34 |
# Element implementation for the 'qmake' kind.
|
35 | 35 |
class QMakeElement(BuildElement):
|
36 |
- pass
|
|
36 |
+ # Supports virtual directories (required for remote execution)
|
|
37 |
+ BST_VIRTUAL_DIRECTORY = True
|
|
37 | 38 |
|
38 | 39 |
|
39 | 40 |
# Plugin entry point
|
... | ... | @@ -20,3 +20,4 @@ |
20 | 20 |
from .sandbox import Sandbox, SandboxFlags
|
21 | 21 |
from ._sandboxchroot import SandboxChroot
|
22 | 22 |
from ._sandboxbwrap import SandboxBwrap
|
23 |
+from ._sandboxremote import SandboxRemote
|
1 |
+#!/usr/bin/env python3
|
|
2 |
+#
|
|
3 |
+# Copyright (C) 2018 Bloomberg LP
|
|
4 |
+#
|
|
5 |
+# This program is free software; you can redistribute it and/or
|
|
6 |
+# modify it under the terms of the GNU Lesser General Public
|
|
7 |
+# License as published by the Free Software Foundation; either
|
|
8 |
+# version 2 of the License, or (at your option) any later version.
|
|
9 |
+#
|
|
10 |
+# This library is distributed in the hope that it will be useful,
|
|
11 |
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
12 |
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
13 |
+# Lesser General Public License for more details.
|
|
14 |
+#
|
|
15 |
+# You should have received a copy of the GNU Lesser General Public
|
|
16 |
+# License along with this library. If not, see <http://www.gnu.org/licenses/>.
|
|
17 |
+#
|
|
18 |
+# Authors:
|
|
19 |
+# Jim MacArthur <jim macarthur codethink co uk>
|
|
20 |
+ |
|
21 |
+import os
|
|
22 |
+import re
|
|
23 |
+from urllib.parse import urlparse
|
|
24 |
+ |
|
25 |
+import grpc
|
|
26 |
+ |
|
27 |
+from . import Sandbox
|
|
28 |
+from ..storage._filebaseddirectory import FileBasedDirectory
|
|
29 |
+from ..storage._casbaseddirectory import CasBasedDirectory
|
|
30 |
+from .._protos.build.bazel.remote.execution.v2 import remote_execution_pb2, remote_execution_pb2_grpc
|
|
31 |
+from .._artifactcache.cascache import CASCache
|
|
32 |
+ |
|
33 |
+ |
|
34 |
+class SandboxError(Exception):
|
|
35 |
+ pass
|
|
36 |
+ |
|
37 |
+ |
|
38 |
+# SandboxRemote()
|
|
39 |
+#
|
|
40 |
+# This isn't really a sandbox, it's a stub which sends all the sources and build
|
|
41 |
+# commands to a remote server and retrieves the results from it.
|
|
42 |
+#
|
|
43 |
+class SandboxRemote(Sandbox):
|
|
44 |
+ |
|
45 |
+ def __init__(self, *args, **kwargs):
|
|
46 |
+ super().__init__(*args, **kwargs)
|
|
47 |
+ self.cascache = None
|
|
48 |
+ |
|
49 |
+ url = urlparse(kwargs['server_url'])
|
|
50 |
+ if not url.scheme or not url.hostname or not url.port:
|
|
51 |
+ raise SandboxError("Configured remote URL '{}' does not match the expected layout. "
|
|
52 |
+ .format(self.server_url) +
|
|
53 |
+ "It should be of the form <protocol>://<domain name>:<port>.")
|
|
54 |
+ elif url.scheme != 'http':
|
|
55 |
+ raise SandboxError("Configured remote '{}' uses an unsupported protocol. "
|
|
56 |
+ "Only plain HTTP is currenlty supported (no HTTPS).")
|
|
57 |
+ |
|
58 |
+ self.server_url = '{}:{}'.format(url.hostname, url.port)
|
|
59 |
+ |
|
60 |
+ def _get_cascache(self):
|
|
61 |
+ if self.cascache is None:
|
|
62 |
+ self.cascache = CASCache(self._get_context())
|
|
63 |
+ self.cascache.setup_remotes(use_config=True)
|
|
64 |
+ return self.cascache
|
|
65 |
+ |
|
66 |
+ def run_remote_command(self, command, input_root_digest, working_directory, environment):
|
|
67 |
+ # Sends an execution request to the remote execution server.
|
|
68 |
+ #
|
|
69 |
+ # This function blocks until it gets a response from the server.
|
|
70 |
+ #
|
|
71 |
+ environment_variables = [remote_execution_pb2.Command.
|
|
72 |
+ EnvironmentVariable(name=k, value=v)
|
|
73 |
+ for (k, v) in environment.items()]
|
|
74 |
+ |
|
75 |
+ # Create and send the Command object.
|
|
76 |
+ remote_command = remote_execution_pb2.Command(arguments=command,
|
|
77 |
+ working_directory=working_directory,
|
|
78 |
+ environment_variables=environment_variables,
|
|
79 |
+ output_files=[],
|
|
80 |
+ output_directories=[self._output_directory],
|
|
81 |
+ platform=None)
|
|
82 |
+ |
|
83 |
+ cascache = self._get_cascache()
|
|
84 |
+ # Upload the Command message to the remote CAS server
|
|
85 |
+ command_digest = cascache.push_message(self._get_project(), remote_command)
|
|
86 |
+ if not command_digest or not cascache.verify_key_pushed(self._get_project(), command_digest):
|
|
87 |
+ # Command push failed
|
|
88 |
+ return None
|
|
89 |
+ |
|
90 |
+ # Create and send the action.
|
|
91 |
+ action = remote_execution_pb2.Action(command_digest=command_digest,
|
|
92 |
+ input_root_digest=input_root_digest,
|
|
93 |
+ timeout=None,
|
|
94 |
+ do_not_cache=True)
|
|
95 |
+ |
|
96 |
+ # Upload the Action message to the remote CAS server
|
|
97 |
+ action_digest = cascache.push_message(self._get_project(), action)
|
|
98 |
+ if not action_digest or not cascache.verify_key_pushed(self._get_project(), action_digest):
|
|
99 |
+ # Action push failed
|
|
100 |
+ return None
|
|
101 |
+ |
|
102 |
+ # Next, try to create a communication channel to the BuildGrid server.
|
|
103 |
+ channel = grpc.insecure_channel(self.server_url)
|
|
104 |
+ stub = remote_execution_pb2_grpc.ExecutionStub(channel)
|
|
105 |
+ request = remote_execution_pb2.ExecuteRequest(action_digest=action_digest,
|
|
106 |
+ skip_cache_lookup=True)
|
|
107 |
+ try:
|
|
108 |
+ operation_iterator = stub.Execute(request)
|
|
109 |
+ except grpc.RpcError:
|
|
110 |
+ return None
|
|
111 |
+ |
|
112 |
+ operation = None
|
|
113 |
+ with self._get_context().timed_activity("Waiting for the remote build to complete"):
|
|
114 |
+ # It is advantageous to check operation_iterator.code() is grpc.StatusCode.OK here,
|
|
115 |
+ # which will check the server is actually contactable. However, calling it when the
|
|
116 |
+ # server is available seems to cause .code() to hang forever.
|
|
117 |
+ for operation in operation_iterator:
|
|
118 |
+ if operation.done:
|
|
119 |
+ break
|
|
120 |
+ |
|
121 |
+ return operation
|
|
122 |
+ |
|
123 |
+ def process_job_output(self, output_directories, output_files):
|
|
124 |
+ # Reads the remote execution server response to an execution request.
|
|
125 |
+ #
|
|
126 |
+ # output_directories is an array of OutputDirectory objects.
|
|
127 |
+ # output_files is an array of OutputFile objects.
|
|
128 |
+ #
|
|
129 |
+ # We only specify one output_directory, so it's an error
|
|
130 |
+ # for there to be any output files or more than one directory at the moment.
|
|
131 |
+ #
|
|
132 |
+ if output_files:
|
|
133 |
+ raise SandboxError("Output files were returned when we didn't request any.")
|
|
134 |
+ elif not output_directories:
|
|
135 |
+ error_text = "No output directory was returned from the build server."
|
|
136 |
+ raise SandboxError(error_text)
|
|
137 |
+ elif len(output_directories) > 1:
|
|
138 |
+ error_text = "More than one output directory was returned from the build server: {}."
|
|
139 |
+ raise SandboxError(error_text.format(output_directories))
|
|
140 |
+ |
|
141 |
+ tree_digest = output_directories[0].tree_digest
|
|
142 |
+ if tree_digest is None or not tree_digest.hash:
|
|
143 |
+ raise SandboxError("Output directory structure had no digest attached.")
|
|
144 |
+ |
|
145 |
+ cascache = self._get_cascache()
|
|
146 |
+ # Now do a pull to ensure we have the necessary parts.
|
|
147 |
+ dir_digest = cascache.pull_tree(self._get_project(), tree_digest)
|
|
148 |
+ if dir_digest is None or not dir_digest.hash or not dir_digest.size_bytes:
|
|
149 |
+ raise SandboxError("Output directory structure pulling from remote failed.")
|
|
150 |
+ |
|
151 |
+ path_components = os.path.split(self._output_directory)
|
|
152 |
+ |
|
153 |
+ # Now what we have is a digest for the output. Once we return, the calling process will
|
|
154 |
+ # attempt to descend into our directory and find that directory, so we need to overwrite
|
|
155 |
+ # that.
|
|
156 |
+ |
|
157 |
+ if not path_components:
|
|
158 |
+ # The artifact wants the whole directory; we could just return the returned hash in its
|
|
159 |
+ # place, but we don't have a means to do that yet.
|
|
160 |
+ raise SandboxError("Unimplemented: Output directory is empty or equal to the sandbox root.")
|
|
161 |
+ |
|
162 |
+ # At the moment, we will get the whole directory back in the first directory argument and we need
|
|
163 |
+ # to replace the sandbox's virtual directory with that. Creating a new virtual directory object
|
|
164 |
+ # from another hash will be interesting, though...
|
|
165 |
+ |
|
166 |
+ new_dir = CasBasedDirectory(self._get_context(), ref=dir_digest)
|
|
167 |
+ self._set_virtual_directory(new_dir)
|
|
168 |
+ |
|
169 |
+ def run(self, command, flags, *, cwd=None, env=None):
|
|
170 |
+ # Upload sources
|
|
171 |
+ upload_vdir = self.get_virtual_directory()
|
|
172 |
+ |
|
173 |
+ if isinstance(upload_vdir, FileBasedDirectory):
|
|
174 |
+ # Make a new temporary directory to put source in
|
|
175 |
+ upload_vdir = CasBasedDirectory(self._get_context(), ref=None)
|
|
176 |
+ upload_vdir.import_files(self.get_virtual_directory()._get_underlying_directory())
|
|
177 |
+ |
|
178 |
+ upload_vdir.recalculate_hash()
|
|
179 |
+ |
|
180 |
+ cascache = self._get_cascache()
|
|
181 |
+ # Now, push that key (without necessarily needing a ref) to the remote.
|
|
182 |
+ vdir_digest = cascache.push_directory(self._get_project(), upload_vdir)
|
|
183 |
+ if not vdir_digest or not cascache.verify_key_pushed(self._get_project(), vdir_digest):
|
|
184 |
+ raise SandboxError("Failed to verify that source has been pushed to the remote artifact cache.")
|
|
185 |
+ |
|
186 |
+ # Set up environment and working directory
|
|
187 |
+ if cwd is None:
|
|
188 |
+ cwd = self._get_work_directory()
|
|
189 |
+ |
|
190 |
+ if cwd is None:
|
|
191 |
+ cwd = '/'
|
|
192 |
+ |
|
193 |
+ if env is None:
|
|
194 |
+ env = self._get_environment()
|
|
195 |
+ |
|
196 |
+ # We want command args as a list of strings
|
|
197 |
+ if isinstance(command, str):
|
|
198 |
+ command = [command]
|
|
199 |
+ |
|
200 |
+ # Now transmit the command to execute
|
|
201 |
+ operation = self.run_remote_command(command, upload_vdir.ref, cwd, env)
|
|
202 |
+ |
|
203 |
+ if operation is None:
|
|
204 |
+ # Failure of remote execution, usually due to an error in BuildStream
|
|
205 |
+ # NB This error could be raised in __run_remote_command
|
|
206 |
+ raise SandboxError("No response returned from server")
|
|
207 |
+ |
|
208 |
+ assert(not operation.HasField('error') and operation.HasField('response'))
|
|
209 |
+ |
|
210 |
+ execution_response = remote_execution_pb2.ExecuteResponse()
|
|
211 |
+ # The response is expected to be an ExecutionResponse message
|
|
212 |
+ assert(operation.response.Is(execution_response.DESCRIPTOR))
|
|
213 |
+ |
|
214 |
+ operation.response.Unpack(execution_response)
|
|
215 |
+ |
|
216 |
+ if execution_response.status.code != 0:
|
|
217 |
+ # A normal error during the build: the remote execution system
|
|
218 |
+ # has worked correctly but the command failed.
|
|
219 |
+ # execution_response.error also contains 'message' (str) and
|
|
220 |
+ # 'details' (iterator of Any) which we ignore at the moment.
|
|
221 |
+ return execution_response.status.code
|
|
222 |
+ |
|
223 |
+ action_result = execution_response.result
|
|
224 |
+ |
|
225 |
+ self.process_job_output(action_result.output_directories, action_result.output_files)
|
|
226 |
+ |
|
227 |
+ return 0
|
... | ... | @@ -99,9 +99,11 @@ class Sandbox(): |
99 | 99 |
self.__stdout = kwargs['stdout']
|
100 | 100 |
self.__stderr = kwargs['stderr']
|
101 | 101 |
|
102 |
- # Setup the directories. Root should be available to subclasses, hence
|
|
103 |
- # being single-underscore. The others are private to this class.
|
|
102 |
+ # Setup the directories. Root and output_directory should be
|
|
103 |
+ # available to subclasses, hence being single-underscore. The
|
|
104 |
+ # others are private to this class.
|
|
104 | 105 |
self._root = os.path.join(directory, 'root')
|
106 |
+ self._output_directory = None
|
|
105 | 107 |
self.__directory = directory
|
106 | 108 |
self.__scratch = os.path.join(self.__directory, 'scratch')
|
107 | 109 |
for directory_ in [self._root, self.__scratch]:
|
... | ... | @@ -144,11 +146,17 @@ class Sandbox(): |
144 | 146 |
self._vdir = FileBasedDirectory(self._root)
|
145 | 147 |
return self._vdir
|
146 | 148 |
|
149 |
+ def _set_virtual_directory(self, virtual_directory):
|
|
150 |
+ """ Sets virtual directory. Useful after remote execution
|
|
151 |
+ has rewritten the working directory.
|
|
152 |
+ """
|
|
153 |
+ self._vdir = virtual_directory
|
|
154 |
+ |
|
147 | 155 |
def set_environment(self, environment):
|
148 | 156 |
"""Sets the environment variables for the sandbox
|
149 | 157 |
|
150 | 158 |
Args:
|
151 |
- directory (dict): The environment variables to use in the sandbox
|
|
159 |
+ environment (dict): The environment variables to use in the sandbox
|
|
152 | 160 |
"""
|
153 | 161 |
self.__env = environment
|
154 | 162 |
|
... | ... | @@ -160,6 +168,15 @@ class Sandbox(): |
160 | 168 |
"""
|
161 | 169 |
self.__cwd = directory
|
162 | 170 |
|
171 |
+ def set_output_directory(self, directory):
|
|
172 |
+ """Sets the output directory - the directory which is preserved
|
|
173 |
+ as an artifact after assembly.
|
|
174 |
+ |
|
175 |
+ Args:
|
|
176 |
+ directory (str): An absolute path within the sandbox
|
|
177 |
+ """
|
|
178 |
+ self._output_directory = directory
|
|
179 |
+ |
|
163 | 180 |
def mark_directory(self, directory, *, artifact=False):
|
164 | 181 |
"""Marks a sandbox directory and ensures it will exist
|
165 | 182 |
|
... | ... | @@ -543,6 +543,15 @@ class CasBasedDirectory(Directory): |
543 | 543 |
filelist.append(k)
|
544 | 544 |
return filelist
|
545 | 545 |
|
546 |
+ def recalculate_hash(self):
|
|
547 |
+ """ Recalcuates the hash for this directory and store the results in
|
|
548 |
+ the cache. If this directory has a parent, tell it to
|
|
549 |
+ recalculate (since changing this directory changes an entry in
|
|
550 |
+ the parent). Hashes for subdirectories also get recalculated.
|
|
551 |
+ """
|
|
552 |
+ self._recalculate_recursing_up()
|
|
553 |
+ self._recalculate_recursing_down()
|
|
554 |
+ |
|
546 | 555 |
def _get_identifier(self):
|
547 | 556 |
path = ""
|
548 | 557 |
if self.parent:
|
... | ... | @@ -204,6 +204,24 @@ with an artifact share. |
204 | 204 |
You can also specify a list of caches here; earlier entries in the list
|
205 | 205 |
will have higher priority than later ones.
|
206 | 206 |
|
207 |
+Remote execution
|
|
208 |
+~~~~~~~~~~~~~~~~
|
|
209 |
+BuildStream supports remote execution using the Google Remote Execution API
|
|
210 |
+(REAPI). A description of how remote execution works is beyond the scope
|
|
211 |
+of this document, but you can specify a remote server complying with the REAPI
|
|
212 |
+using the `remote-execution` option:
|
|
213 |
+ |
|
214 |
+.. code:: yaml
|
|
215 |
+ |
|
216 |
+ remote-execution:
|
|
217 |
+ |
|
218 |
+ # A url defining a remote execution server
|
|
219 |
+ url: http://buildserver.example.com:50051
|
|
220 |
+ |
|
221 |
+The url should contain a hostname and port separated by ':'. Only plain HTTP is
|
|
222 |
+currently suported (no HTTPS).
|
|
223 |
+ |
|
224 |
+The Remote Execution API can be found via https://github.com/bazelbuild/remote-apis.
|
|
207 | 225 |
|
208 | 226 |
.. _project_essentials_mirrors:
|
209 | 227 |
|