Jim MacArthur pushed to branch jmac/remote_execution_rebase at BuildStream / buildstream
Commits:
-
f562a458
by Jim MacArthur at 2018-08-03T14:11:29Z
-
ea823782
by Jim MacArthur at 2018-08-03T14:11:29Z
-
51f4f007
by Jim MacArthur at 2018-08-03T14:11:29Z
-
8a6e9160
by Jim MacArthur at 2018-08-03T14:11:29Z
-
2f088cdd
by Jim MacArthur at 2018-08-03T14:11:29Z
-
3726a4f2
by Jim MacArthur at 2018-08-03T14:11:51Z
-
0d2ec2ba
by Jim MacArthur at 2018-08-03T14:11:51Z
-
bbad0b88
by Jim MacArthur at 2018-08-03T14:11:51Z
-
8a920f76
by Jim MacArthur at 2018-08-03T14:11:51Z
-
6d560c88
by Jim MacArthur at 2018-08-03T14:11:51Z
-
2237b816
by Jim MacArthur at 2018-08-03T14:11:51Z
5 changed files:
- buildstream/_artifactcache/cascache.py
- buildstream/buildelement.py
- buildstream/element.py
- + buildstream/sandbox/_sandboxremote.py
- buildstream/sandbox/sandbox.py
Changes:
| ... | ... | @@ -213,7 +213,6 @@ class CASCache(ArtifactCache): |
| 213 | 213 |
remotes_for_project = self._remotes[element._get_project()]
|
| 214 | 214 |
return any(remote.spec.push for remote in remotes_for_project)
|
| 215 | 215 |
|
| 216 |
- |
|
| 217 | 216 |
def pull_key(self, key, size_bytes, project):
|
| 218 | 217 |
""" Pull a single key rather than an artifact.
|
| 219 | 218 |
Does not update local refs. """
|
| ... | ... | @@ -362,9 +361,9 @@ class CASCache(ArtifactCache): |
| 362 | 361 |
refs = [self.get_artifact_fullname(element, key) for key in keys]
|
| 363 | 362 |
|
| 364 | 363 |
project = element._get_project()
|
| 365 |
- return self.push_refs(element, refs, project)
|
|
| 364 |
+ return self.push_refs(refs, project, element=element)
|
|
| 366 | 365 |
|
| 367 |
- def push_refs(self, element, refs, project, may_have_dependencies=True):
|
|
| 366 |
+ def push_refs(self, refs, project, may_have_dependencies=True, element=None):
|
|
| 368 | 367 |
|
| 369 | 368 |
push_remotes = [r for r in self._remotes[project] if r.spec.push]
|
| 370 | 369 |
|
| ... | ... | @@ -374,7 +373,7 @@ class CASCache(ArtifactCache): |
| 374 | 373 |
remote.init()
|
| 375 | 374 |
if self._push_refs_to_remote(refs, remote, may_have_dependencies):
|
| 376 | 375 |
pushed = True
|
| 377 |
- else:
|
|
| 376 |
+ elif element:
|
|
| 378 | 377 |
self.context.message(Message(
|
| 379 | 378 |
None,
|
| 380 | 379 |
MessageType.SKIPPED,
|
| ... | ... | @@ -399,7 +398,6 @@ class CASCache(ArtifactCache): |
| 399 | 398 |
return pushed
|
| 400 | 399 |
|
| 401 | 400 |
def _verify_ref_on_remote(self, ref, remote):
|
| 402 |
- pushed = False
|
|
| 403 | 401 |
tree = self.resolve_ref(ref)
|
| 404 | 402 |
|
| 405 | 403 |
# Check whether ref is already on the server in which case
|
| ... | ... | @@ -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 |
|
| ... | ... | @@ -1537,6 +1537,8 @@ class Element(Plugin): |
| 1537 | 1537 |
finally:
|
| 1538 | 1538 |
if collect is not None:
|
| 1539 | 1539 |
try:
|
| 1540 |
+ # Sandbox will probably have replaced its virtual directory, so get it again
|
|
| 1541 |
+ sandbox_vroot = sandbox.get_virtual_directory()
|
|
| 1540 | 1542 |
collectvdir = sandbox_vroot.descend(collect.lstrip(os.sep).split(os.sep))
|
| 1541 | 1543 |
except VirtualDirectoryError:
|
| 1542 | 1544 |
# No collect directory existed
|
| 1 |
+#!/usr/bin/env python3
|
|
| 2 |
+#
|
|
| 3 |
+# Copyright (C) 2018 Codethink Limited
|
|
| 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 time
|
|
| 23 |
+ |
|
| 24 |
+import grpc
|
|
| 25 |
+ |
|
| 26 |
+from . import Sandbox
|
|
| 27 |
+from ..storage._filebaseddirectory import FileBasedDirectory
|
|
| 28 |
+from ..storage._casbaseddirectory import CasBasedDirectory
|
|
| 29 |
+from .._protos.build.bazel.remote.execution.v2 import remote_execution_pb2, remote_execution_pb2_grpc
|
|
| 30 |
+from .._protos.google.longrunning import operations_pb2, operations_pb2_grpc
|
|
| 31 |
+ |
|
| 32 |
+from .._artifactcache.cascache import CASCache
|
|
| 33 |
+ |
|
| 34 |
+ |
|
| 35 |
+class SandboxError(Exception):
|
|
| 36 |
+ pass
|
|
| 37 |
+ |
|
| 38 |
+ |
|
| 39 |
+# SandboxRemote()
|
|
| 40 |
+#
|
|
| 41 |
+# This isn't really a sandbox, it's a stub which sends all the source
|
|
| 42 |
+# to a remote server and retrieves the results from it.
|
|
| 43 |
+#
|
|
| 44 |
+class SandboxRemote(Sandbox):
|
|
| 45 |
+ |
|
| 46 |
+ def __init__(self, *args, **kwargs):
|
|
| 47 |
+ super().__init__(*args, **kwargs)
|
|
| 48 |
+ self.user_ns_available = kwargs['user_ns_available']
|
|
| 49 |
+ self.die_with_parent_available = kwargs['die_with_parent_available']
|
|
| 50 |
+ self.cascache = None
|
|
| 51 |
+ |
|
| 52 |
+ def _get_cascache(self):
|
|
| 53 |
+ if self.cascache is None:
|
|
| 54 |
+ self.cascache = CASCache(self._get_context())
|
|
| 55 |
+ self.cascache.setup_remotes(use_config=True)
|
|
| 56 |
+ return self.cascache
|
|
| 57 |
+ |
|
| 58 |
+ def __run_remote_command(self, cascache, command, input_root_digest, environment):
|
|
| 59 |
+ environment_variables = []
|
|
| 60 |
+ for(k, v) in environment.items():
|
|
| 61 |
+ environment_variables.append(remote_execution_pb2.Command.EnvironmentVariable(name=k, value=v))
|
|
| 62 |
+ remote_command = remote_execution_pb2.Command(arguments=command, environment_variables=environment_variables)
|
|
| 63 |
+ |
|
| 64 |
+ # Serialise this into the cascache...
|
|
| 65 |
+ command_digest = cascache.add_object(buffer=remote_command.SerializeToString())
|
|
| 66 |
+ |
|
| 67 |
+ command_ref = 'worker-command/{}'.format(command_digest.hash)
|
|
| 68 |
+ cascache.set_ref(command_ref, command_digest)
|
|
| 69 |
+ |
|
| 70 |
+ command_push_successful = cascache.push_refs([command_ref], self._get_project(), may_have_dependencies=False)
|
|
| 71 |
+ if command_push_successful or cascache.verify_key_pushed(command_ref, self._get_project()):
|
|
| 72 |
+ # Next, try to create a communication channel
|
|
| 73 |
+ port = 50051
|
|
| 74 |
+ channel = grpc.insecure_channel('dekatron.office.codethink.co.uk:{}'.format(port))
|
|
| 75 |
+ stub = remote_execution_pb2_grpc.ExecutionStub(channel)
|
|
| 76 |
+ ops_stub = operations_pb2_grpc.OperationsStub(channel)
|
|
| 77 |
+ |
|
| 78 |
+ # Having done that, create and send the action.
|
|
| 79 |
+ |
|
| 80 |
+ action = remote_execution_pb2.Action(command_digest=command_digest,
|
|
| 81 |
+ input_root_digest=input_root_digest,
|
|
| 82 |
+ output_files=[],
|
|
| 83 |
+ output_directories=[self._output_directory],
|
|
| 84 |
+ platform=None,
|
|
| 85 |
+ timeout=None,
|
|
| 86 |
+ do_not_cache=True)
|
|
| 87 |
+ |
|
| 88 |
+ request = remote_execution_pb2.ExecuteRequest(instance_name='default',
|
|
| 89 |
+ action=action,
|
|
| 90 |
+ skip_cache_lookup=True)
|
|
| 91 |
+ |
|
| 92 |
+ operation = stub.Execute(request) # Returns Operation
|
|
| 93 |
+ job_name = operation.name
|
|
| 94 |
+ else:
|
|
| 95 |
+ # Source push failed
|
|
| 96 |
+ return None
|
|
| 97 |
+ while True:
|
|
| 98 |
+ # TODO: Timeout
|
|
| 99 |
+ # Refresh the operation data periodically using the name
|
|
| 100 |
+ request = operations_pb2.GetOperationRequest(name=job_name)
|
|
| 101 |
+ operation = ops_stub.GetOperation(request)
|
|
| 102 |
+ time.sleep(1)
|
|
| 103 |
+ if operation.done:
|
|
| 104 |
+ break
|
|
| 105 |
+ return operation
|
|
| 106 |
+ |
|
| 107 |
+ def process_job_output(self, output_directories, output_files):
|
|
| 108 |
+ # output_directories is an array of OutputDirectory objects.
|
|
| 109 |
+ # output_files is an array of OutputFile objects.
|
|
| 110 |
+ #
|
|
| 111 |
+ # We only specify one output_directory, so it's an error
|
|
| 112 |
+ # for there to be any output files or more than one directory at the moment.
|
|
| 113 |
+ |
|
| 114 |
+ if output_files:
|
|
| 115 |
+ raise SandboxError("Output files were returned when we didn't request any.")
|
|
| 116 |
+ elif len(output_directories) > 1:
|
|
| 117 |
+ error_text = "More than one output directory was returned from the build server: {}"
|
|
| 118 |
+ raise SandboxError(error_text.format(output_directories))
|
|
| 119 |
+ |
|
| 120 |
+ digest = output_directories[0].tree_digest
|
|
| 121 |
+ if digest is None or digest.hash is None or digest.hash == "":
|
|
| 122 |
+ raise SandboxError("Output directory structure had no digest attached.")
|
|
| 123 |
+ |
|
| 124 |
+ # Now do a pull to ensure we have the necessary parts.
|
|
| 125 |
+ cascache = self._get_cascache()
|
|
| 126 |
+ cascache.pull_key(digest.hash, digest.size_bytes, self._get_project())
|
|
| 127 |
+ path_components = os.path.split(self._output_directory)
|
|
| 128 |
+ |
|
| 129 |
+ # Now what we have is a digest for the output. Once we return, the calling process will
|
|
| 130 |
+ # attempt to descend into our directory and find that directory, so we need to overwrite
|
|
| 131 |
+ # that.
|
|
| 132 |
+ |
|
| 133 |
+ if not path_components:
|
|
| 134 |
+ # The artifact wants the whole directory; we could just return the returned hash in its
|
|
| 135 |
+ # place, but we don't have a means to do that yet.
|
|
| 136 |
+ raise SandboxError("Unimplemented: Output directory is empty or equal to the sandbox root.")
|
|
| 137 |
+ |
|
| 138 |
+ # At the moment, we will get the whole directory back in the first directory argument and we need
|
|
| 139 |
+ # to replace the sandbox's virtual directory with that. Creating a new virtual directory object
|
|
| 140 |
+ # from another hash will be interesting, though...
|
|
| 141 |
+ |
|
| 142 |
+ new_dir = CasBasedDirectory(self._get_context(), ref=digest)
|
|
| 143 |
+ self.set_virtual_directory(new_dir)
|
|
| 144 |
+ |
|
| 145 |
+ def run(self, command, flags, *, cwd=None, env=None):
|
|
| 146 |
+ # Upload sources
|
|
| 147 |
+ upload_vdir = self.get_virtual_directory()
|
|
| 148 |
+ if isinstance(upload_vdir, FileBasedDirectory):
|
|
| 149 |
+ # Make a new temporary directory to put source in
|
|
| 150 |
+ upload_vdir = CasBasedDirectory(self._get_context(), ref=None)
|
|
| 151 |
+ upload_vdir.import_files(self.get_virtual_directory().get_underlying_directory())
|
|
| 152 |
+ |
|
| 153 |
+ # Now, push that key (without necessarily needing a ref) to the remote.
|
|
| 154 |
+ cascache = self._get_cascache()
|
|
| 155 |
+ |
|
| 156 |
+ ref = 'worker-source/{}'.format(upload_vdir.ref.hash)
|
|
| 157 |
+ upload_vdir._save(ref)
|
|
| 158 |
+ source_push_successful = cascache.push_refs([ref], self._get_project())
|
|
| 159 |
+ # Fallback to the sandbox default settings for
|
|
| 160 |
+ # the cwd and environment.
|
|
| 161 |
+ |
|
| 162 |
+ if env is None:
|
|
| 163 |
+ env = self._get_environment()
|
|
| 164 |
+ |
|
| 165 |
+ # We want command args as a list of strings
|
|
| 166 |
+ if isinstance(command, str):
|
|
| 167 |
+ command = [command]
|
|
| 168 |
+ |
|
| 169 |
+ # Now transmit the command to execute
|
|
| 170 |
+ if source_push_successful or cascache.verify_key_pushed(ref, self._get_project()):
|
|
| 171 |
+ response = self.__run_remote_command(cascache, command, upload_vdir.ref, env)
|
|
| 172 |
+ |
|
| 173 |
+ if response is None or response.HasField("error"):
|
|
| 174 |
+ # Build failed, so return a failure code
|
|
| 175 |
+ return 1
|
|
| 176 |
+ else:
|
|
| 177 |
+ |
|
| 178 |
+ # At the moment, response can either be an
|
|
| 179 |
+ # ExecutionResponse containing an ActionResult, or an
|
|
| 180 |
+ # ActionResult directly.
|
|
| 181 |
+ executeResponse = remote_execution_pb2.ExecuteResponse()
|
|
| 182 |
+ if response.response.Is(executeResponse.DESCRIPTOR):
|
|
| 183 |
+ # Unpack ExecuteResponse and set response to its response
|
|
| 184 |
+ response.response.Unpack(executeResponse)
|
|
| 185 |
+ response = executeResponse
|
|
| 186 |
+ |
|
| 187 |
+ actionResult = remote_execution_pb2.ActionResult()
|
|
| 188 |
+ if response.response.Is(actionResult.DESCRIPTOR):
|
|
| 189 |
+ response.response.Unpack(actionResult)
|
|
| 190 |
+ self.process_job_output(actionResult.output_directories, actionResult.output_files)
|
|
| 191 |
+ else:
|
|
| 192 |
+ raise SandboxError("Received unknown message from server (expected ExecutionResponse).")
|
|
| 193 |
+ else:
|
|
| 194 |
+ raise SandboxError("Failed to verify that source has been pushed to the remote artifact cache.")
|
|
| 195 |
+ 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]:
|
| ... | ... | @@ -142,11 +144,29 @@ class Sandbox(): |
| 142 | 144 |
self._vdir = FileBasedDirectory(self._root)
|
| 143 | 145 |
return self._vdir
|
| 144 | 146 |
|
| 147 |
+ def set_virtual_directory(self, vdir):
|
|
| 148 |
+ """ Sets virtual directory. Useful after remote execution
|
|
| 149 |
+ has rewritten the working directory. """
|
|
| 150 |
+ self._vdir = vdir
|
|
| 151 |
+ |
|
| 152 |
+ def get_virtual_toplevel_directory(self):
|
|
| 153 |
+ """Fetches the sandbox's toplevel directory
|
|
| 154 |
+ |
|
| 155 |
+ The toplevel directory contains 'root', 'scratch' and later
|
|
| 156 |
+ 'artifact' where output is copied to.
|
|
| 157 |
+ |
|
| 158 |
+ Returns:
|
|
| 159 |
+ (str): The sandbox toplevel directory
|
|
| 160 |
+ |
|
| 161 |
+ """
|
|
| 162 |
+ # For now, just create a new Directory every time we're asked
|
|
| 163 |
+ return FileBasedDirectory(self.__directory)
|
|
| 164 |
+ |
|
| 145 | 165 |
def set_environment(self, environment):
|
| 146 | 166 |
"""Sets the environment variables for the sandbox
|
| 147 | 167 |
|
| 148 | 168 |
Args:
|
| 149 |
- directory (dict): The environment variables to use in the sandbox
|
|
| 169 |
+ environment (dict): The environment variables to use in the sandbox
|
|
| 150 | 170 |
"""
|
| 151 | 171 |
self.__env = environment
|
| 152 | 172 |
|
| ... | ... | @@ -158,6 +178,15 @@ class Sandbox(): |
| 158 | 178 |
"""
|
| 159 | 179 |
self.__cwd = directory
|
| 160 | 180 |
|
| 181 |
+ def set_output_directory(self, directory):
|
|
| 182 |
+ """Sets the output directory - the directory which is preserved
|
|
| 183 |
+ as an artifact after assembly.
|
|
| 184 |
+ |
|
| 185 |
+ Args:
|
|
| 186 |
+ directory (str): An absolute path within the sandbox
|
|
| 187 |
+ """
|
|
| 188 |
+ self._output_directory = directory
|
|
| 189 |
+ |
|
| 161 | 190 |
def mark_directory(self, directory, *, artifact=False):
|
| 162 | 191 |
"""Marks a sandbox directory and ensures it will exist
|
| 163 | 192 |
|
