[Notes] [Git][BuildStream/buildstream][mablanch/630-remote-execution-reconn] _sandboxremote.py: Try to reopen operation steam on failure

[Martin Blanchard <gitlab mg gitlab com>]

Title: GitLab

Martin Blanchard pushed to branch mablanch/630-remote-execution-reconn at BuildStream / buildstream

Commits:

• 4e6037a0
  by Martin Blanchard at 2018-09-24T10:49:09Z

  _sandboxremote.py: Try to reopen operation steam on failure
  
  The REAPI allows a client to reconnect to an ongoing operation stream by
  providing a WaitExecution(). If implemented on server side, BuildStream
  will try to recover from connection errors using it.
  
  https://gitlab.com/BuildStream/buildstream/issues/630
  

1 changed file:

• buildstream/sandbox/_sandboxremote.py

Changes:

buildstream/sandbox/_sandboxremote.py

---

...	...	@@ -76,8 +76,7 @@ class SandboxRemote(Sandbox):
76	76	# Upload the Command message to the remote CAS server
77	77	command_digest = cascache.push_message(self._get_project(), remote_command)
78	78	if not command_digest or not cascache.verify_digest_pushed(self._get_project(), command_digest):
79		- # Command push failed
80		- return None
	79	+ raise SandboxError("Failed pushing build command to remote CAS.")
81	80
82	81	# Create and send the action.
83	82	action = remote_execution_pb2.Action(command_digest=command_digest,
...	...	@@ -88,27 +87,57 @@ class SandboxRemote(Sandbox):
88	87	# Upload the Action message to the remote CAS server
89	88	action_digest = cascache.push_message(self._get_project(), action)
90	89	if not action_digest or not cascache.verify_digest_pushed(self._get_project(), action_digest):
91		- # Action push failed
92		- return None
	90	+ raise SandboxError("Failed pushing build action to remote CAS.")
93	91
94	92	# Next, try to create a communication channel to the BuildGrid server.
95	93	channel = grpc.insecure_channel(self.server_url)
96	94	stub = remote_execution_pb2_grpc.ExecutionStub(channel)
97	95	request = remote_execution_pb2.ExecuteRequest(action_digest=action_digest,
98	96	skip_cache_lookup=False)
99		- try:
100		- operation_iterator = stub.Execute(request)
101		- except grpc.RpcError:
102		- return None
	97	+
	98	+ def __run_remote_command(stub, execute_request=None, running_operation=None):
	99	+ try:
	100	+ last_operation = None
	101	+ if execute_request is not None:
	102	+ operation_iterator = stub.Execute(execute_request)
	103	+ else:
	104	+ request = remote_execution_pb2.WaitExecutionRequest(name=running_operation.name)
	105	+ operation_iterator = stub.WaitExecution(request)
	106	+
	107	+ for operation in operation_iterator:
	108	+ if operation.done:
	109	+ return operation
	110	+ else:
	111	+ last_operation = operation
	112	+ except grpc.RpcError as e:
	113	+ status_code = e.code()
	114	+ if status_code == grpc.StatusCode.UNAVAILABLE:
	115	+ raise SandboxError("Failed contacting remote execution server at {}."
	116	+ .format(self.server_url))
	117	+
	118	+ elif status_code in (grpc.StatusCode.INVALID_ARGUMENT,
	119	+ grpc.StatusCode.FAILED_PRECONDITION,
	120	+ grpc.StatusCode.RESOURCE_EXHAUSTED,
	121	+ grpc.StatusCode.INTERNAL,
	122	+ grpc.StatusCode.DEADLINE_EXCEEDED):
	123	+ raise SandboxError("{} ({}).".format(e.details(), status_code.name))
	124	+
	125	+ elif running_operation and status_code == grpc.StatusCode.UNIMPLEMENTED:
	126	+ raise SandboxError("Failed trying to recover from connection loss: "
	127	+ "server does not support operation status polling recovery.")
	128	+
	129	+ return last_operation
103	130
104	131	operation = None
105	132	with self._get_context().timed_activity("Waiting for the remote build to complete"):
106		- # It is advantageous to check operation_iterator.code() is grpc.StatusCode.OK here,
107		- # which will check the server is actually contactable. However, calling it when the
108		- # server is available seems to cause .code() to hang forever.
109		- for operation in operation_iterator:
110		- if operation.done:
111		- break
	133	+ operation = __run_remote_command(stub, execute_request=request)
	134	+ if operation is None:
	135	+ return None
	136	+ elif operation.done:
	137	+ return operation
	138	+
	139	+ while operation is not None and not operation.done:
	140	+ operation = __run_remote_command(stub, running_operation=operation)
112	141
113	142	return operation
114	143
...	...	@@ -196,7 +225,6 @@ class SandboxRemote(Sandbox):
196	225
197	226	if operation is None:
198	227	# Failure of remote execution, usually due to an error in BuildStream
199		- # NB This error could be raised in __run_remote_command
200	228	raise SandboxError("No response returned from server")
201	229
202	230	assert not operation.HasField('error') and operation.HasField('response')

—
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.