- Author(s): lidizheng, pfreixes
- Approver: gnossen, rmariano
- Status: Approved
- Implemented in: Python
- Last updated: 2020-05-13
- Discussions at:
A brand new set of async API that will solve concurrency issues and performance issues for gRPC Python, which is available to Python 3.6+.
- Asynchronous processing perfectly fits IO-intensive gRPC use cases;
- Resolve a long-living design flaw of thread exhaustion problem;
- Performance is much better than the multi-threading model.
Quote from asyncio
package
documentation:
asyncio
is a library to write concurrent code using theasync
/await
syntax.
asyncio
is used as a foundation for multiple Python asynchronous frameworks that provide high-performance network and web-servers, database connection libraries, distributed task queues, etc.
In the asynchronous I/O model, the computation tasks are packed as generators (aka. coroutines) that will yield the ownership of thread while it is blocked by IO operations or wait for other tasks to complete.
The design of coroutine in Python can trace back to Python 2.5 in PEP 342 --
Coroutines via Enhanced Generators.
It introduces new generator methods like send
, throw
, and close
. With
these methods, the caller of the generator can pass values and exceptions into a
paused generator function, hence diminish the need for wiring complex callbacks.
To further simplify the usage of a coroutine, Python 3.5 introduces
async
/await
syntax that can easily transform a normal function into a
coroutine (PEP 492). And the
Twisted
style event loop abstractions is introduced in PEP
3156.
Another important piece is the asynchronous generator. It is introduced in Python 3.6 by PEP 525.
To read more about asyncio
implementation in CPython, see:
Currently, there are two futures in the standard library.
concurrent.futures.Future
added in Python 3.2asyncio.Future
added in Python 3.4
They are built for different threading models. Luckily, the CPython maintainers
are actively trying to make them compatible (they are almost compatible). By
the time this design doc is written, comparing to concurrent.futures.Future
,
asyncio.Future
has the following difference:
This class is not thread-safe.
result() and exception() do not take a timeout argument and raise an exception when the future isn't done yet.
Callbacks registered with add_done_callback() are always called via the event loop's call_soon().
This class is not compatible with the wait() and as_completed() methods in the concurrent.futures package.
gRPC Python also has its definition of
Future
interface since 2015. The concurrency story by that time is not that clear as
now. Here is the reasoning of why gRPC Python needed a dedicated Future
object, and incompatibilities:
Python doesn't have a Future interface in its standard library. In the absence of such a standard, three separate, incompatible implementations (concurrent.futures.Future, ndb.Future, and asyncio.Future) have appeared. This interface attempts to be as compatible as possible with concurrent.futures.Future. From ndb.Future it adopts a traceback-object accessor method.
Unlike the concrete and implemented Future classes listed above, the Future class defined in this module is an entirely abstract interface that anyone may implement and use.
The one known incompatibility between this interface and the interface of concurrent.futures.Future is that this interface defines its own CancelledError and TimeoutError exceptions rather than raising the implementation-private concurrent.futures._base.CancelledError and the built-in-but-only-in-3.3-and-later TimeoutError.
Although the design of Future
in Python finally settled down, it's not
recommended to expose low-level API like asyncio.Future
object. The Python
documentation suggests that we should let the application to decide which
Future
implementation they want to use, and hide the ways to operate them
directly.
The rule of thumb is to never expose Future objects in user-facing APIs, and the recommended way to create a Future object is to call loop.create_future(). This way alternative event loop implementations can inject their own optimized implementations of a Future object.
In a single-threaded application, creating a coroutine object doesn't necessarily mean it is scheduled to be executed in the event loop.
The functions defined by async def
, underlying, is a function that returns a
Python generator. If the program calls an async def
function, it will NOT be
executed. This behavior is one of the main reason why mixing async def
function with normal function is a bad idea.
There are three mechanisms to schedule coroutines:
- Await the coroutine
await asyncio.sleep(1)
; - Submit the coroutine to the event loop object
loop.call_soon(coro)
; - Turn coroutine into an
asyncio.Task
objectasyncio.ensure_future(coro)
.
This gRFC intended to fully utilize asyncio
and create a Pythonic paradigm for
gRPC Python programming in Python 3 runtime.
asyncio
is great, but not silver bullet to solve everything. It has its
limitations and unique advantages. gRPC Python as a framework should empower
tech-savvy users to use the cutting-edge features, in the same time as we allow
majority of our users to use Python in their most familiar way.
The new API will be isolated from the current API. The implementation of the new API is an entirely different stack than the current stack. On the downside, most gRPC objects and definitions can't be shared between these two sets of API. After all, it is not a good practice to introduce uncertainty (return coroutine or regular object) to our API. This decision is made to respect our contract of API since GA in 2016, including the underlying behaviors. Developers who use our current API should not be affected by this effort. We plan to support two stacks in the long-run.
For users who want to migrate to new API, the granularity migration is per
channel, per server level. For both channel-side and server-side, the adoption
of asyncio
does not only include cosmetic changes of adding async
/await
keywords throughout the code base, it also requires thinking about the change
from a potentially multi-threaded application to a single-threaded asynchronous
application. For example, the synchronization methods have to be changed to use
the asyncio
edition (see Synchronization
Primitives).
All asyncio
related API will be kept inside the grpc.aio
module. Take the
most common use case as an example, the new API will diverge from the old API
since the creation of Channel
and Server
object. New creation methods are
added to instantiate async Server
or Channel
object:
import grpc
server = grpc.aio.server(...)
channel = grpc.aio.insecure_channel(...)
channel = grpc.aio.secure_channel(...)
channel = grpc.aio.intercept_channel(...)
To reduce cognitive burden, this gRFC tries to make the new async API share the same parameters as the current API, and keep the usage similar.
Server side:
class AsyncGreeter(helloworld_pb2_grpc.GreeterServicer):
async def SayHello(self,
request: helloworld_pb2_grpc.HelloRequest,
context: grpc.aio.ServicerContext
) -> helloworld_pb2_grpc.HelloReply:
await asyncio.sleep(1)
return helloworld_pb2.HelloReply(message="Hello, %s!" % request.name)
server = grpc.aio.server()
server.add_insecure_port(":50051")
helloworld_pb2_grpc.add_GreeterServicer_to_server(AsyncGreeter(), server)
await server.start()
await server.wait_for_termination()
Client side:
async with grpc.aio.insecure_channel("localhost:50051") as channel:
stub = echo_pb2_grpc.EchoStub(channel)
response = await stub.Hi(echo_pb2.EchoRequest(message="ping"))
async for response in stub.StreamingHi(...):
process(response)
Existing streaming API has usability problem that its logic complexes user application. For client side, it requires the request iterator to be defined above the line of call, and the consumption of responses to be coded below the line of invocation. So when the application wants to send request based on received response, it would be quite challenging and usually involves multi-threading, which is not optimal in Python world.
For server side, the same problem remains for non-trivial streaming servicer
handlers. The yield
syntax constraint the message sending should happen in the
handler itself, otherwise the user needs to code a dedicated iterator class. For
larger application, it would be a pain to create such iterator class for each
streaming handler.
euroelessar@ points out several great reasons for the new streaming API:
- Enable
with
statement for streaming API, which allows clean resource management; - Easier to buffer outbound messages when peer pushes back (e.g.
yield
will stop the execution of current coroutine, butcall.write
can be scheduled in event loop); - Enable application to have both pending read and pending write in one coroutine, and react on the first finisher (with iterator syntax this will be quite hard to get it right).
So, this gRFC introduces a new pattern of streaming API that reads/writes message to peer with explicit call.
### Client side
with stub.StreamingHi() as streaming_call:
request = echo_pb2.EchoRequest(message="ping")
await streaming_call.write(request)
response = await streaming_call.read()
while response != grpc.aio.EOF:
process(response)
response = await streaming_call.read()
### Server side
class AsyncGreeter(helloworld_pb2_grpc.GreeterServicer):
async def StreamingHi(self,
unused_request_iterator,
context: grpc.aio.ServicerContext
) -> None:
bootstrap_request = await context.read()
...(bootstrap_request)
while ...:
response = ...
await context.write(response)
### Client side
async def async_request_iterator() -> HelloRequest:
for request in [...]:
yield response
# No await needed, the response_iterator is grpc.aio.Call
response_iterator = stub.StreamingHi(async_request_iterator())
async for response in response_iterator:
process(response)
### Server side
class Greeter(helloworld_pb2_grpc.GreeterServicer):
async def StreamingHi(self, request_iterator, context):
async for request in request_iterator:
yield response
Existing API still has advantage over simpler use cases of gRPC, and iterator syntax is Pythonic to use. Also, for backward compatibility concern, iterator-based API needs to stay. So, both new and current streaming API need to be included in the surface API.
To keep the function signature stable, the new streaming API only requires
enhancement to the context object (grpc.aio.Call
on client-side, and
grpc.aio.ServicerContext
on the server-side).
Notably, both new and current streaming API allows read and write messages simultaneously in different thread or coroutine. However, the application should not mixing new and current streaming API at the same time. Reading or writing to the iterator and context object might have synchronization issue that the framework does not guaranteed to protect against.
Here is an example of the core part of the gRPC ProtoBuf plugin generated code:
# Client-side stub
class GreeterStub(object):
def __init__(self, channel):
self.SayHello = channel.unary_unary(...)
# Server-side servicer
def add_GreeterServicer_to_server(servicer, server):
...
Both channel
and server
object are passed in by the application. The
generated code is agnostic to which set of API the application uses, as long as
we respect the method name in our API contract.
The async API related classes and functions, including shared APIs, should be generated on the same page.
Currently, for RPC invocations, gRPC Python provides 3 options: __call__
,
with_call
, and futures
. They behaves slightly different in whether block or
not, and the number of return values (see
UnaryUnaryMultiCallable
for details). Developers have to bare in mind the subtlety between these 3
options across 4 types of RPC (not all combination are supported).
Semantically, those 3 options grant users ability to:
- Directly getting the RPC response.
- Check metadata of the RPC.
- Control the life cycle of the RPC.
- Handle the RPC failure.
It is understandable for current design to provide different options, since
Python doesn't have a consensus Future
before. The simplicity of an RPC call
will be ruined if the original designer of gRPC Python API tries to merge those
3 options.
However, thanks to the official definition of asyncio.Task
/asyncio.Future
.
They can be merged into one method by returning a grpc.aio.Call
that extends
(or composites) asyncio.Task
.
In asyncio
, it is expected to await
on asynchronous operations. Hence, it is
nature for stub calls to return an asyncio.Task
compatible object, which
representing the RPC call.
Also, the grpc.aio.Call
object provides gRPC specific semantics to manipulate
the ongoing RPC. So, all above functionality can be solved by one single merged
method.
# Usage 1: A simple call
response = await stub.Hi(...)
# Usage 2: Check metadata
call = stub.Hi(...)
if validate(await call.initial_metadata()):
response = await call
print(f'Getting response [{response}] with code [{call.code()}]')
else:
raise ValueError('Failed to validate initial metadata')
# Usage 3: Control the life cycle
call = stub.Hi(...)
await async_stuff_that_takes_time()
if call.is_active() and call.time_remaining() < REMAINING_TIME_THRESHOLD:
call.cancel()
# Usage 4: Error handling
try:
response = await stub.FailedHi(...)
except grpc.RpcError as rpc_error:
print(f'RPC failed: {rpc_error.code()}')
The new API intended to solve all concurrency issue with asynchronous I/O. However, it also introduces a migration challenge for our users. For users who want to gradually migrate from current stack to async stack, their potentially non-asyncio native logic may block the entire thread. If the thread is blocked, then the event loop will be blocked, then the whole process will end up deadlocking.
However, by supporting the executors, we can allow mixing async and sync method handlers on the server side, which further reduce the cost of migration.
Fire-and-forget is a valid use case in async programming. For the non-critical
tasks, the program may schedule the execution without checking the result. As
mentioned in "Python Coroutines in asyncio
" section, the coroutine won't be
scheduled unless we explicitly do so. A dedicated prefix or suffix of the
function should help to remind developers to await the coroutine.
### Forget to await RPC
while 1:
stub.ReportLoad(ReportRequest(
timestamp=...,
metrics=[...],
)) # <- RPC not sent
await asyncio.sleep(3)
### Await on whatever function starts with "Async"
while 1:
await stub.AsyncReportLoad(ReportRequest(
timestamp=...,
metrics=[...],
))
await asyncio.sleep(3)
CPython developers also consider this problem. Their solution is that if a
coroutine object gets deallocated without execution, the interpreter will log an
RuntimeWarning
to the standard error output.
RuntimeWarning: coroutine '...' was never awaited
For new asyncio
related behavior, we will write unit tests dedicated to the
new stack. However, currently, there are more than a hundred test cases in gRPC
Python. It's a significant amount of work to refactor each one of them for the
new API.
A straight forward solution is building a wrapper over async API to simulate current API. The wrapper itself shouldn't be too hard to implement. However, some tests are tightly coupled with the implementation detail that might break; manual labor may still be required for this approach.
We should compare the cost of manual refactoring and wrapper when we want to reuse the existing test cases.
Besides grpcio
package, currently gRPC Python also own:
grpcio-tools
(no service)grpcio-testing
grpcio-reflection
grpcio-health-checking
grpcio-channelz
grpcio-status
(no service)
Apart from grpcio-tools
and grpcio-status
, all other packages have at least
one gRPC service implementation. They will also require migration to adopt
asyncio
. This design doc propose to keep their current API untouched, and add
new sets of async APIs.
To propagate HTTP/2 flow control push back, the new async API needs to aware of the flow control mechanism. Most complex logic is handled in C-Core, except there is a single rule the wrapper layer needs to follow: there can only be one outstanding read/write operation on each call.
It means if the application fires two write in parallel, one of them have to wait until the other one finishes. Also, that rule doesn't prohibit reading and writing at the same time.
So, even if all the read/write is asynchronous in asyncio
, we will have to
either enforce the rule ourselves by adding locks in our implementation. Or we
can pass down the synchronization responsibility to our users.
contextvars
is a Python 3.7 feature that allows applications to set coroutine
local variables (just like thread local variables). In the past, gRPC Python
uses thread local to achieve two goals:
- Support distributed tracing library that work across languages;
- Support deadline propagation along the chain of RPCs.
For distributed tracing library (like OpenCensus), the tracing information is preserved as one of the metadata, and it is expected to be implicitly promoted from inbound metadata to outbound metadata. The tracing information will be used to monitor the life cycle of an request, and the length it has stayed in each services.
For deadline propagation, the deadline of upstream server will be implicitly pass down to downstream server, so downstream services can react to that information and save computation resources or perform flow control.
Acceptance criteria of this feature:
- The implementation of such feature should supported by official package, and users are not expected to directly access those metadata;
- Application logic has higher priority than the implicit propagation (e.g. setting timeout explicitly will override the propagated value);
- Application has the choice to diverse coroutine local variables;
- The exception error string should be informative (e.g. pointing out the timeout is due to upstream deadline).
Further discussion around this topic, see grpc-api-examples#2.
Typing makes a difference. This gRFC intended to introduce typing to gRPC Python as much as we can. To change generated code, we need to update the gRPC protoc plugin to generate Python 3 stubs and servicers. To date, this gRFC intended to make the new async API compatible with existing generated code. So, the Python 3 generated code will have two options:
- Add a third generation mode to protoc plugin that generates stubs / servicers with typing;
- Generating an additional file for the typed code.
By other asynchronous libraries, they mean libraries that provide their own
Future, Coroutine, Event-loops, like gevent
/Twisted
/Tornado
. In general,
it is challenging to support multiple async mechanisms inter-operate together,
and it has to be discussed case by case.
For Tornado
, they are using asyncio
underneath their abstraction of
asynchronous mechanisms. After v6.0.0, they dropped support for Python 2
entirely. So, it should work fine with our new API.
For Twisted
, with some monkey patch code to connect its Deferred
/Future
object to asyncio
's API, it should work.
For gevent
, unfortunately, it works by monkey-patching Python APIs that
including threading
and various transport library, and the work is executed in
gevent
managed event loop. gevent
and asyncio
doesn't compatible with each
other out of box.
- It would be quite confusing to mix functions that return
asyncio
coroutine and normal Python functions; - We don't want to imply that switching to use the new stack requires zero code changes.
- Also, the current contract of API constraint ourselves from changing their behavior after GA for two years.
The proposed API interface might be obsoleted, please refer to our master branch or our API reference.
# grpc.aio.Channel
class Channel:
"""Affords RPC invocation via generic methods on client-side.
Channel objects implement the Async Context Manager type, although they need
not support being entered and exited multiple times.
"""
def check_connectivity_state(self, try_to_connect: bool=False) -> grpc.ChannelConnectivity:
"""Check the connectivity state of a channel.
This is an EXPERIMENTAL API.
Args:
try_to_connect: a bool indicate whether the Channel should try to connect to peer or not.
Returns:
A ChannelConnectivity object.
"""
async def watch_connectivity_state(self,
last_observed_state: grpc.ChannelConnectivity,
timeout_seconds: float) -> Optional[grpc.ChannelConnectivity]:
"""Watch for a change in connectivity state.
This is an EXPERIMENTAL API.
Once the channel connectivity state is different from
last_observed_state, the function will return the new connectivity
state. If deadline expires BEFORE the state is changed, None will be
returned.
Args:
try_to_connect: a bool indicate whether the Channel should try to connect to peer or not.
Returns:
A ChannelConnectivity object or None.
"""
async def channel_ready(self) -> None:
"""Creates a coroutine that blocks until the Channel is READY."""
def unary_unary(self,
method: Text,
request_serializer: Optional[Callable[[Request], bytes]]=None,
response_deserializer: Optional[Callable[[bytes], Response]]=None
) -> grpc.aio.UnaryUnaryMultiCallable[Request, Response]:
"""Creates a UnaryUnaryMultiCallable for a unary call method.
Args:
method: The name of the RPC method.
request_serializer: Optional behaviour for serializing the request
message. Request goes unserialized in case None is passed.
response_deserializer: Optional behaviour for deserializing the
response message. Response goes undeserialized in case None
is passed.
Returns:
A UnaryUnaryMultiCallable value for the named unary call method.
"""
def unary_stream(self,
method: Text,
request_serializer: Optional[Callable[[Request], bytes]]=None,
response_deserializer: Optional[Callable[[bytes], Response]]=None
) -> grpc.aio.UnaryStreamMultiCallable[Request, Response]:
"""Creates a UnaryStreamMultiCallable for a server streaming method.
Args:
method: The name of the RPC method.
request_serializer: Optional behaviour for serializing the request
message. Request goes unserialized in case None is passed.
response_deserializer: Optional behaviour for deserializing the
response message. Response goes undeserialized in case None is
passed.
Returns:
A UnaryStreamMultiCallable value for the name server streaming method.
"""
def stream_unary(self,
method: Text,
request_serializer: Optional[Callable[[Request], bytes]]=None,
response_deserializer: Optional[Callable[[bytes], Response]]=None
) -> grpc.aio.StreamUnaryMultiCallable[Request, Response]:
"""Creates a StreamUnaryMultiCallable for a client streaming method.
Args:
method: The name of the RPC method.
request_serializer: Optional behaviour for serializing the request
message. Request goes unserialized in case None is passed.
response_deserializer: Optional behaviour for deserializing the
response message. Response goes undeserialized in case None is
passed.
Returns:
A StreamUnaryMultiCallable value for the named client streaming method.
"""
def stream_stream(self,
method: Text,
request_serializer: Optional[Callable[[Request], bytes]]=None,
response_deserializer: Optional[Callable[[bytes], Response]]=None
) -> grpc.aio.StreamStreamMultiCallable[Request, Response]:
"""Creates a StreamStreamMultiCallable for a bi-directional streaming method.
Args:
method: The name of the RPC method.
request_serializer: Optional behaviour for serializing the request
message. Request goes unserialized in case None is passed.
response_deserializer: Optional behaviour for deserializing the
response message. Response goes undeserialized in case None
is passed.
Returns:
A StreamStreamMultiCallable value for the named bi-directional streaming method.
"""
def close(self) -> None:
"""Closes this Channel and releases all resources held by it.
Closing the Channel will immediately terminate all RPCs active with the
Channel and it is not valid to invoke new RPCs with the Channel.
This method is idempotent.
"""
# grpc.aio.UnaryUnaryMultiCallable
class UnaryUnaryMultiCallable(Generic[Request, Response]):
"""Affords invoking an async unary RPC from client-side."""
def __call__(self,
request: Request,
timeout: Optional[int]=None,
metadata: Optional[Sequence[Tuple[Text, AnyStr]]]=None,
credentials: Optional[grpc.CallCredentials]=None,
wait_for_ready: Optional[bool]=None,
compression: Optional[grpc.Compression]=None
) -> grpc.aio.Call[Response]:
"""Schedules the underlying RPC.
Args:
request: The request value for the RPC.
timeout: An optional duration of time in seconds to allow
for the RPC.
metadata: Optional :term:`metadata` to be transmitted to the
service-side of the RPC.
credentials: An optional CallCredentials for the RPC. Only valid for
secure Channel.
wait_for_ready: This is an EXPERIMENTAL argument. An optional
flag to enable wait for ready mechanism
compression: An element of grpc.compression, e.g.
grpc.compression.Gzip. This is an EXPERIMENTAL option.
Returns:
An awaitable object grpc.aio.Call that returns the response value.
Raises:
RpcError: Indicating that the RPC terminated with non-OK status. The
raised RpcError will also be a Call for the RPC affording the RPC's
metadata, status code, and details.
"""
# grpc.aio.UnaryStreamMultiCallable
class UnaryStreamMultiCallable(Generic[Request, Response]):
"""Affords invoking an async server streaming RPC from client-side."""
def __call__(self,
request: Request,
timeout: Optional[int]=None,
metadata: Optional[Sequence[Tuple[Text, AnyStr]]]=None,
credentials: Optional[grpc.CallCredentials]=None,
wait_for_ready: Optional[bool]=None,
compression: Optional[grpc.Compression]=None
) -> grpc.aio.Call[AsyncIterable[Response]]:
"""Schedules the underlying RPC.
Args:
request: The request value for the RPC.
timeout: An optional duration of time in seconds to allow
for the RPC.
metadata: Optional :term:`metadata` to be transmitted to the
service-side of the RPC.
credentials: An optional CallCredentials for the RPC. Only valid for
secure Channel.
wait_for_ready: This is an EXPERIMENTAL argument. An optional
flag to enable wait for ready mechanism
compression: An element of grpc.compression, e.g.
grpc.compression.Gzip. This is an EXPERIMENTAL option.
Returns:
An awaitable object grpc.aio.Call that returns the async iterator of response values.
Raises:
RpcError: Indicating that the RPC terminated with non-OK status. The
raised RpcError will also be a Call for the RPC affording the RPC's
metadata, status code, and details.
"""
# grpc.aio.StreamUnaryMultiCallable
class StreamUnaryMultiCallable(Generic[Request, Response]):
"""Affords invoking an async client streaming RPC from client-side."""
def __call__(self,
request_iterator: Optional[AsyncIterable[Request]]=None,
timeout: Optional[int]=None,
metadata: Optional[Sequence[Tuple[Text, AnyStr]]]=None,
credentials: Optional[grpc.CallCredentials]=None,
wait_for_ready: Optional[bool]=None,
compression: Optional[grpc.Compression]=None
) -> grpc.aio.Call[Response]:
"""Schedules the underlying RPC.
Args:
request_iterator: An async iterator that yields request values for
the RPC.
timeout: An optional duration of time in seconds to allow
for the RPC.
metadata: Optional :term:`metadata` to be transmitted to the
service-side of the RPC.
credentials: An optional CallCredentials for the RPC. Only valid for
secure Channel.
wait_for_ready: This is an EXPERIMENTAL argument. An optional
flag to enable wait for ready mechanism
compression: An element of grpc.compression, e.g.
grpc.compression.Gzip. This is an EXPERIMENTAL option.
Returns:
An awaitable object grpc.aio.Call presents the RPC.
Raises:
RpcError: Indicating that the RPC terminated with non-OK status. The
raised RpcError will also be a Call for the RPC affording the RPC's
metadata, status code, and details.
"""
# grpc.aio.StreamStreamMultiCallable
class StreamStreamMultiCallable(Generic[Request, Response]):
"""Affords invoking an async bi-directional RPC from client-side."""
def __call__(self,
request_iterator: Optional[AsyncIterable[Request]]=None,
timeout: Optional[int]=None,
metadata: Optional[Sequence[Tuple[Text, AnyStr]]]=None,
credentials: Optional[grpc.CallCredentials]=None,
wait_for_ready: Optional[bool]=None,
compression: Optional[grpc.Compression]=None
) -> grpc.aio.Call[AsyncIterable[Response]]:
"""Schedules the underlying RPC.
Args:
request_iterator: An async iterator that yields request values for
the RPC.
timeout: An optional duration of time in seconds to allow
for the RPC.
metadata: Optional :term:`metadata` to be transmitted to the
service-side of the RPC.
credentials: An optional CallCredentials for the RPC. Only valid for
secure Channel.
wait_for_ready: This is an EXPERIMENTAL argument. An optional
flag to enable wait for ready mechanism
compression: An element of grpc.compression, e.g.
grpc.compression.Gzip. This is an EXPERIMENTAL option.
Returns:
An awaitable object grpc.aio.Call that returns the async iterator of response values.
Raises:
RpcError: Indicating that the RPC terminated with non-OK status. The
raised RpcError will also be a Call for the RPC affording the RPC's
metadata, status code, and details.
"""
# grpc.aio.RpcContext
class RpcContext:
"""Provides RPC-related information and action."""
def cancelled(self) -> bool:
"""Return True if the RPC is cancelled.
The RPC is cancelled when the cancellation was requested with cancel().
Returns:
A bool indicates whether the RPC is cancelled or not.
"""
def done(self) -> bool:
"""Return True if the RPC is done.
An RPC is done if the RPC is completed, cancelled or aborted.
Returns:
A bool indicates if the RPC is done.
"""
def time_remaining(self) -> Optional[float]:
"""Describes the length of allowed time remaining for the RPC.
Returns:
A nonnegative float indicating the length of allowed time in seconds
remaining for the RPC to complete before it is considered to have
timed out, or None if no deadline was specified for the RPC.
"""
def cancel(self) -> bool:
"""Cancels the RPC.
Idempotent and has no effect if the RPC has already terminated.
Returns:
A bool indicates if the cancellation is performed or not.
"""
def add_done_callback(self, callback: DoneCallbackType) -> None:
"""Registers a callback to be called on RPC termination.
Args:
callback: A callable object will be called with the call object as
its only argument.
"""
# grpc.aio.Call
class Call(grpc.aio.RpcContext):
"""The abstract base class of an RPC on the client-side."""
async def initial_metadata(self) -> MetadataType:
"""Accesses the initial metadata sent by the server.
Returns:
The initial :term:`metadata`.
"""
async def trailing_metadata(self) -> MetadataType:
"""Accesses the trailing metadata sent by the server.
Returns:
The trailing :term:`metadata`.
"""
async def code(self) -> grpc.StatusCode:
"""Accesses the status code sent by the server.
Returns:
The StatusCode value for the RPC.
"""
async def details(self) -> Text:
"""Accesses the details sent by the server.
Returns:
The details string of the RPC.
"""
# grpc.aio.UnaryUnaryCall
class UnaryUnaryCall(Generic[RequestType, ResponseType], Call):
"""The abstract base class of an unary-unary RPC on the client-side."""
def __await__(self) -> Awaitable[ResponseType]:
"""Await the response message to be ready.
Returns:
The response message of the RPC.
"""
# grpc.aio.UnaryStreamCall
class UnaryStreamCall(Generic[RequestType, ResponseType], Call):
def __aiter__(self) -> AsyncIterable[ResponseType]:
"""Returns the async iterable representation that yields messages.
Under the hood, it is calling the "read" method.
Returns:
An async iterable object that yields messages.
"""
async def read(self) -> Union[EOFType, ResponseType]:
"""Reads one message from the stream.
Read operations must be serialized when called from multiple
coroutines.
Returns:
A response message, or an `grpc.aio.EOF` to indicate the end of the
stream.
"""
# grpc.aio.StreamUnaryCall
class StreamUnaryCall(Generic[RequestType, ResponseType], Call):
async def write(self, request: RequestType) -> None:
"""Writes one message to the stream.
Raises:
An RpcError exception if the write failed.
"""
async def done_writing(self) -> None:
"""Notifies server that the client is done sending messages.
After done_writing is called, any additional invocation to the write
function will fail. This function is idempotent.
"""
def __await__(self) -> Awaitable[ResponseType]:
"""Await the response message to be ready.
Returns:
The response message of the stream.
"""
# grpc.aio.StreamStreamCall
class StreamStreamCall(Generic[RequestType, ResponseType], Call):
def __aiter__(self) -> AsyncIterable[ResponseType]:
"""Returns the async iterable representation that yields messages.
Under the hood, it is calling the "read" method.
Returns:
An async iterable object that yields messages.
"""
async def read(self) -> Union[EOFType, ResponseType]:
"""Reads one message from the stream.
Read operations must be serialized when called from multiple
coroutines.
Returns:
A response message, or an `grpc.aio.EOF` to indicate the end of the
stream.
"""
async def write(self, request: RequestType) -> None:
"""Writes one message to the stream.
Raises:
An RpcError exception if the write failed.
"""
async def done_writing(self) -> None:
"""Notifies server that the client is done sending messages.
After done_writing is called, any additional invocation to the write
function will fail. This function is idempotent.
"""
# grpc.aio.Server
class Server:
"""Serves RPCs."""
def add_generic_rpc_handlers(
self,
generic_rpc_handlers: Iterable[grpc.aio.GenericRpcHandlers]
) -> None:
"""Registers GenericRpcHandlers with this Server.
This method is only safe to call before the server is started.
Args:
generic_rpc_handlers: An iterable of GenericRpcHandlers that will be
used to service RPCs.
"""
def add_insecure_port(self, address: Text) -> int:
"""Opens an insecure port for accepting RPCs.
This method may only be called before starting the server.
Args:
address: The address for which to open a port. If the port is 0,
or not specified in the address, then gRPC runtime will choose a port.
Returns:
An integer port on which server will accept RPC requests.
"""
def add_secure_port(self,
address: Text,
server_credentials: grpc.ServerCredentials) -> int:
"""Opens a secure port for accepting RPCs.
This method may only be called before starting the server.
Args:
address: The address for which to open a port.
if the port is 0, or not specified in the address, then gRPC
runtime will choose a port.
server_credentials: A ServerCredentials object.
Returns:
An integer port on which server will accept RPC requests.
"""
def start(self) -> None:
"""Starts this Server.
This method may only be called once. (i.e. it is not idempotent).
"""
def stop(self, grace: Optional[float]) -> asyncio.Event:
"""Stops this Server.
This method immediately stop service of new RPCs in all cases.
If a grace period is specified, this method returns immediately
and all RPCs active at the end of the grace period are aborted.
If a grace period is not specified (by passing None for `grace`),
all existing RPCs are aborted immediately and this method
blocks until the last RPC handler terminates.
This method is idempotent and may be called at any time.
Passing a smaller grace value in a subsequent call will have
the effect of stopping the Server sooner (passing None will
have the effect of stopping the server immediately). Passing
a larger grace value in a subsequent call *will not* have the
effect of stopping the server later (i.e. the most restrictive
grace value is used).
Args:
grace: A duration of time in seconds or None.
Returns:
A threading.Event that will be set when this Server has completely
stopped, i.e. when running RPCs either complete or are aborted and
all handlers have terminated.
"""
async def wait_for_termination(self, timeout: Optional[float]=None) -> bool:
"""Block current thread until the server stops.
This is an EXPERIMENTAL API.
The wait will not consume computational resources during blocking, and
it will block until one of the two following conditions are met:
1) The server is stopped or terminated;
2) A timeout occurs if timeout is not `None`.
The timeout argument works in the same way as `threading.Event.wait()`.
https://docs.python.org/3/library/threading.html#threading.Event.wait
Args:
timeout: A floating point number specifying a timeout for the
operation in seconds.
Returns:
A bool indicates if the operation times out.
"""
# grpc.aio.GenericRpcHandler
class GenericRpcHandler:
"""An implementation of arbitrarily many RPC methods."""
def service(self,
handler_call_details: grpc.HandlerCallDetails
) -> Union[grpc.RpcMethodHandler, grpc.aio.RpcMethodHandler, None]:
"""Returns the handler for servicing the RPC.
Args:
handler_call_details: A HandlerCallDetails describing the RPC.
Returns:
An grpc.RpcMethodHandler if the handler is a normal Python function;
or an grpc.aio.RpcMethodHandler if the handler is a coroutine;
otherwise, None.
"""
# grpc.aio.RpcMethodHandler
class RpcMethodHandler(Generic[Request, Response]):
"""An implementation of a single RPC method.
Attributes:
request_streaming: Whether the RPC supports exactly one request message
or any arbitrary number of request messages.
response_streaming: Whether the RPC supports exactly one response message
or any arbitrary number of response messages.
request_deserializer: A callable behavior that accepts a byte string and
returns an object suitable to be passed to this object's business
logic, or None to indicate that this object's business logic should be
passed the raw request bytes.
response_serializer: A callable behavior that accepts an object produced
by this object's business logic and returns a byte string, or None to
indicate that the byte strings produced by this object's business logic
should be transmitted on the wire as they are.
unary_unary: This object's application-specific business logic as a
callable value that takes a request value and a ServicerContext object
and returns a response value. Only non-None if both request_streaming
and response_streaming are False.
unary_stream: This object's application-specific business logic as a
callable value that takes a request value and a ServicerContext object
and returns an iterator of response values. Only non-None if
request_streaming is False and response_streaming is True.
stream_unary: This object's application-specific business logic as a
callable value that takes an iterator of request values and a
ServicerContext object and returns a response value. Only non-None if
request_streaming is True and response_streaming is False.
stream_stream: This object's application-specific business logic as a
callable value that takes an iterator of request values and a
ServicerContext object and returns an iterator of response values.
Only non-None if request_streaming and response_streaming are both
True.
"""
request_streaming: bool
response_streaming: bool
request_deserializer: Optional[Callable[[bytes], Request]]
response_serializer: Optional[Callable[[Response], bytes]]
unary_unary: Optional[Callable[[Request, grpc.aio.ServicerContext], Response]]
unary_stream: Optional[Callable[[Request, grpc.aio.ServicerContext], Optional[AsyncIterable[Response]]]]
stream_unary: Optional[Callable[[AsyncIterable[Request], grpc.aio.ServicerContext], Response]]
stream_stream: Optional[Callable[[AsyncIterable[Request], grpc.aio.ServicerContext], Optional[AsyncIterable[Response]]]]
# grpc.aio.ServicerContext
class ServicerContext(Generic[Request, Response], grpc.RpcContext):
"""A context object passed to method implementations."""
def invocation_metadata(self) -> Optional[Sequence[Tuple[Text, AnyStr]]]:
"""Accesses the metadata from the sent by the client.
Returns:
The invocation :term:`metadata`.
"""
def peer(self) -> Text:
"""Identifies the peer that invoked the RPC being serviced.
Returns:
A string identifying the peer that invoked the RPC being serviced.
The string format is determined by gRPC runtime.
"""
def peer_identities(self) -> Iterable[Text]:
"""Gets one or more peer identity(s).
Equivalent to
servicer_context.auth_context().get(servicer_context.peer_identity_key())
Returns:
An iterable of the identities, or None if the call is not
authenticated. Each identity is returned as a raw bytes type.
"""
def peer_identity_key(self) -> Text:
"""The auth property used to identify the peer.
For example, "x509_common_name" or "x509_subject_alternative_name" are
used to identify an SSL peer.
Returns:
The auth property (string) that indicates the
peer identity, or None if the call is not authenticated.
"""
def auth_context(self) -> Mapping[Text, Iterable[bytes]]:
"""Gets the auth context for the call.
Returns:
A map of strings to an iterable of bytes for each auth property.
"""
def set_compression(self, compression: grpc.Compression) -> None:
"""Set the compression algorithm to be used for the entire call.
This is an EXPERIMENTAL method.
Args:
compression: An element of grpc.compression, e.g.
grpc.compression.Gzip.
"""
async def send_initial_metadata(self, initial_metadata: Sequence[Tuple[Text, AnyStr]]) -> None:
"""Sends the initial metadata value to the client.
This method need not be called by implementations if they have no
metadata to add to what the gRPC runtime will transmit.
Args:
initial_metadata: The initial :term:`metadata`.
"""
async def set_trailing_metadata(self, trailing_metadata: Sequence[Tuple[Text, AnyStr]]) -> None:
"""Sends the trailing metadata for the RPC.
This method need not be called by implementations if they have no
metadata to add to what the gRPC runtime will transmit.
Args:
trailing_metadata: The trailing :term:`metadata`.
"""
async def abort(self, code: grpc.StatusCode, details: Text) -> NoReturn:
"""Raises an exception to terminate the RPC with a non-OK status.
The code and details passed as arguments will supercede any existing
ones.
Args:
code: A StatusCode object to be sent to the client.
It must not be StatusCode.OK.
details: A UTF-8-encodable string to be sent to the client upon
termination of the RPC.
Raises:
Exception: An exception is always raised to signal the abortion the
RPC to the gRPC runtime.
"""
async def abort_with_status(self, status: grpc.Status) -> NoReturn:
"""Raises an exception to terminate the RPC with a non-OK status.
The status passed as argument will supercede any existing status code,
status message and trailing metadata.
This is an EXPERIMENTAL API.
Args:
status: A grpc.Status object. The status code in it must not be
StatusCode.OK.
Raises:
Exception: An exception is always raised to signal the abortion the
RPC to the gRPC runtime.
"""
def set_code(self, code: grpc.StatusCode) -> None:
"""Sets the value to be used as status code upon RPC completion.
This method need not be called by method implementations if they wish
the gRPC runtime to determine the status code of the RPC.
Args:
code: A StatusCode object to be sent to the client.
"""
def set_details(self, details: Text) -> None:
"""Sets the value to be used as detail string upon RPC completion.
This method need not be called by method implementations if they have
no details to transmit.
Args:
details: A UTF-8-encodable string to be sent to the client upon
termination of the RPC.
"""
def disable_next_message_compression(self) -> None:
"""Disables compression for the next response message.
This is an EXPERIMENTAL method.
This method will override any compression configuration set during
server creation or set on the call.
"""
async def read(self) -> Request:
"""Reads one message from the RPC.
Only one read operation is allowed simultaneously. Mixing new streaming API and old
streaming API will resulted in undefined behavior.
Returns:
A response message of the RPC.
Raises:
An RpcError exception if the read failed.
"""
async def write(self, message: Response) -> None:
"""Writes one message to the RPC.
Only one write operation is allowed simultaneously. Mixing new streaming API and old
streaming API will resulted in undefined behavior.
Raises:
An RpcError exception if the write failed.
"""
def add_done_callback(self, callback: Callable[[ServicerContext], None]) -> None:
"""Registers a callback to be called on RPC termination.
Args:
callback: A callable that accepts the grpc.aio.ServicerContext itself.
"""
# grpc.aio.unary_unary_rpc_method_handler
def unary_unary_rpc_method_handler(behavior: Callable[[Request, grpc.aio.ServicerContext], Response],
request_deserializer: Optional[Callable[[bytes], Request]]=None,
response_serializer: Optional[Callable[[Response], bytes]]=None
) -> grpc.aio.RpcMethodHandler[Request, Response]:
"""Creates an RpcMethodHandler for a unary-unary RPC method.
Args:
behavior: The implementation of an RPC that accepts one request
and returns one response.
request_deserializer: An optional behavior for request deserialization.
response_serializer: An optional behavior for response serialization.
Returns:
An RpcMethodHandler object that is typically used by grpc.Server.
"""
# grpc.aio.unary_stream_rpc_method_handler
def unary_stream_rpc_method_handler(behavior: Callable[[Request, grpc.aio.ServicerContext], Optional[AsyncIterable[Response]]],
request_deserializer: Optional[Callable[[bytes], Request]]=None,
response_serializer: Optional[Callable[[Response], bytes]]=None
) -> grpc.aio.RpcMethodHandler[Request, Response]:
"""Creates an RpcMethodHandler for a unary-stream RPC method.
Args:
behavior: The implementation of an RPC that accepts one request
and returns an iterator of response values.
request_deserializer: An optional behavior for request deserialization.
response_serializer: An optional behavior for response serialization.
Returns:
An RpcMethodHandler object that is typically used by grpc.Server.
"""
# grpc.aio.stream_unary_rpc_method_handler
def stream_unary_rpc_method_handler(behavior: Callable[[AsyncIterable[Request], grpc.aio.ServicerContext], Response],
request_deserializer: Optional[Callable[[bytes], Request]]=None,
response_serializer: Optional[Callable[[Response], bytes]]=None
) -> grpc.aio.RpcMethodHandler[Reqeust, Response]:
"""Creates an RpcMethodHandler for a stream-unary RPC method.
Args:
behavior: The implementation of an RPC that accepts an iterator of
request values and returns a single response value.
request_deserializer: An optional behavior for request deserialization.
response_serializer: An optional behavior for response serialization.
Returns:
An RpcMethodHandler object that is typically used by grpc.Server.
"""
# grpc.aio.stream_stream_rpc_method_handler
def stream_stream_rpc_method_handler(behavior: Callable[[AsyncIterable[Request], grpc.aio.ServicerContext], Optional[AsyncIterable[Response]]],
request_deserializer: Optional[Callable[[bytes], Request]]=None,
response_serializer: Optional[Callable[[Response], bytes]]=None
) -> grpc.aio.RpcMethodHandler[Request, Response]:
"""Creates an RpcMethodHandler for a stream-stream RPC method.
Args:
behavior: The implementation of an RPC that accepts an iterator of
request values and returns an iterator of response values.
request_deserializer: An optional behavior for request deserialization.
response_serializer: An optional behavior for response serialization.
Returns:
An RpcMethodHandler object that is typically used by grpc.Server.
"""
# grpc.aio.method_handlers_generic_handler
def method_handlers_generic_handler(service: Text,
method_handlers: Mapping[Text, grpc.aio.RpcMethodHandler]
) -> grpc.aio.GenericRpcHandler:
"""Creates a GenericRpcHandler from RpcMethodHandlers.
Args:
service: The name of the service that is implemented by the
method_handlers.
method_handlers: A dictionary that maps method names to corresponding
RpcMethodHandler.
Returns:
A GenericRpcHandler. This is typically added to the grpc.Server object
with add_generic_rpc_handlers() before starting the server.
"""
# grpc.aio.ServerInterceptor
class ServerInterceptor:
"""Affords intercepting incoming RPCs on the service-side.
This is an EXPERIMENTAL API.
"""
async def intercept_service(self,
continuation: Callable[[grpc.HandlerCallDetails], grpc.aio.RpcMethodHandler],
handler_call_details: grpc.HandlerCallDetails
) -> grpc.aio.RpcMethodHandler:
"""Intercepts incoming RPCs before handing them over to a handler.
Args:
continuation: A function that takes a HandlerCallDetails and
proceeds to invoke the next interceptor in the chain, if any,
or the RPC handler lookup logic, with the call details passed
as an argument, and returns an RpcMethodHandler instance if
the RPC is considered serviced, or None otherwise.
handler_call_details: A HandlerCallDetails describing the RPC.
Returns:
An RpcMethodHandler with which the RPC may be serviced if the
interceptor chooses to service this RPC, or None otherwise.
"""
# grpc.aio.UnaryUnaryClientInterceptor
class UnaryUnaryClientInterceptor:
"""Affords intercepting unary-unary invocations."""
async def intercept_unary_unary(self,
continuation: Callable[[grpc.ClientCallDetails, Request], grpc.aio.Call[Response]],
client_call_details: grpc.ClientCallDetails,
request: Request
) -> grpc.aio.Call[Response]:
"""Intercepts a unary-unary invocation asynchronously.
Args:
continuation: A function that proceeds with the invocation by
executing the next interceptor in chain or invoking the
actual RPC on the underlying Channel. It is the interceptor's
responsibility to call it if it decides to move the RPC forward.
The interceptor can use
`response_future = continuation(client_call_details, request)`
to continue with the RPC. `continuation` returns an object that is
both a Call for the RPC and a Future. In the event of RPC
completion, the return Call-Future's result value will be
the response message of the RPC. Should the event terminate
with non-OK status, the returned Call-Future's exception value
will be an RpcError.
client_call_details: A ClientCallDetails object describing the
outgoing RPC.
request: The request value for the RPC.
Returns:
An object that is both a Call for the RPC and a Future.
In the event of RPC completion, the return Call-Future's
result value will be the response message of the RPC.
Should the event terminate with non-OK status, the returned
Call-Future's exception value will be an RpcError.
"""
# grpc.aio.UnaryStreamClientInterceptor
class UnaryStreamClientInterceptor:
"""Affords intercepting unary-stream invocations."""
async def intercept_unary_stream(self,
continuation: Callable[[grpc.ClientCallDetails, Request], grpc.aio.Call[AsyncIterable[Response]]],
client_call_details: grpc.ClientCallDetails,
request: Request
) -> grpc.aio.Call[AsyncIterable[Response]]:
"""Intercepts a unary-stream invocation.
Args:
continuation: A function that proceeds with the invocation by
executing the next interceptor in chain or invoking the
actual RPC on the underlying Channel. It is the interceptor's
responsibility to call it if it decides to move the RPC forward.
The interceptor can use
`response_iterator = continuation(client_call_details, request)`
to continue with the RPC. `continuation` returns an object that is
both a Call for the RPC and an iterator for response values.
Drawing response values from the returned Call-iterator may
raise RpcError indicating termination of the RPC with non-OK
status.
client_call_details: A ClientCallDetails object describing the
outgoing RPC.
request: The request value for the RPC.
Returns:
An object that is both a Call for the RPC and an iterator of
response values. Drawing response values from the returned
Call-iterator may raise RpcError indicating termination of
the RPC with non-OK status.
"""
# grpc.aio.StreamUnaryClientInterceptor
class StreamUnaryClientInterceptor:
"""Affords intercepting stream-unary invocations."""
async def intercept_stream_unary(self,
continuation: Callable[[grpc.ClientCallDetails, Optional[AsyncIterable[Request]]], grpc.aio.Call[Response]],
client_call_details: grpc.ClientCallDetails,
request_iterator: AsyncIterable[Request]
) -> grpc.aio.Call[Response]:
"""Intercepts a stream-unary invocation asynchronously.
Args:
continuation: A function that proceeds with the invocation by
executing the next interceptor in chain or invoking the
actual RPC on the underlying Channel. It is the interceptor's
responsibility to call it if it decides to move the RPC forward.
The interceptor can use
`response_future = continuation(client_call_details, request_iterator)`
to continue with the RPC. `continuation` returns an object that is
both a Call for the RPC and a Future. In the event of RPC completion,
the return Call-Future's result value will be the response message
of the RPC. Should the event terminate with non-OK status, the
returned Call-Future's exception value will be an RpcError.
client_call_details: A ClientCallDetails object describing the
outgoing RPC.
request_iterator: An iterator that yields request values for the RPC.
Returns:
An object that is both a Call for the RPC and a Future.
In the event of RPC completion, the return Call-Future's
result value will be the response message of the RPC.
Should the event terminate with non-OK status, the returned
Call-Future's exception value will be an RpcError.
"""
# grpc.aio.StreamStreamClientInterceptor
class StreamStreamClientInterceptor:
"""Affords intercepting stream-stream invocations."""
async def intercept_stream_stream(self,
continuation: Callable[[grpc.ClientCallDetails, Optional[AsyncIterable[Request]]], grpc.aio.Call[AsyncIterable[Response]]],
client_call_details: grpc.ClientCallDetails,
request_iterator: AsyncIterable[Request]
) -> grpc.aio.Call[AsyncIterable[Response]]:
"""Intercepts a stream-stream invocation.
Args:
continuation: A function that proceeds with the invocation by
executing the next interceptor in chain or invoking the
actual RPC on the underlying Channel. It is the interceptor's
responsibility to call it if it decides to move the RPC forward.
The interceptor can use
`response_iterator = continuation(client_call_details, request_iterator)`
to continue with the RPC. `continuation` returns an object that is
both a Call for the RPC and an iterator for response values.
Drawing response values from the returned Call-iterator may
raise RpcError indicating termination of the RPC with non-OK
status.
client_call_details: A ClientCallDetails object describing the
outgoing RPC.
request_iterator: An iterator that yields request values for the RPC.
Returns:
An object that is both a Call for the RPC and an iterator of
response values. Drawing response values from the returned
Call-iterator may raise RpcError indicating termination of
the RPC with non-OK status.
"""
# grpc.aio.EOF is a unique object per process that evaluates to False
class _EOF:
def __bool__(self):
return False
EOF = _EOF()
# grpc.BaseError
class BaseError(Exception):
"""The base class for all exceptions generated by gRPC framework."""
# grpc.AbortError
class AbortError(BaseError):
"""Raised when calling abort in servicer methods.
This exception should not be suppressed. Applications may catch it to
perform certain clean-up logic, and then re-raise it.
"""
# grpc.UsageError
class UsageError(BaseError):
"""Raised when the usage might lead to undefined behavior."""
APIs in the following categories remain in top level:
- Credentials related classes including for channel, call, and server;
- Channel connectivity Enum class;
- Status code Enum class;
- Compression method Enum class;
grpc.RpcError
exception class;grpc.RpcContext
class;grpc.ClientCallDetails
class;grpc.HandlerCallDetails
class;
Reviewers has thrown many valuable proposals. This design doc may not be the ideal place for those discussions.
- Design a easy way to use channel arguments grpc#19734.
- Integrate type annotation into generated code grpc#20479.
- Project Dashboard: gRPC Async API
- API reference available since v1.28.0