* Reintroduce the EventEngine default factory
An application can provide an EventEngine factory function that allows
gRPC internals to create EventEngines as needed. This factory would be
used when no EventEngine is provided for some given channel or server,
and where an EventEngine otherwise could not be provided by the
application. Note that there currently is no API to provide an
EventEngine per channel or per server.
I've also deleted some previous iterations on global EventEngine and
EventEngine factory ideas. This new code lives in a public API, and
coexists with iomgr instead of being isolated to an EventEngine-specific
iomgr implementation.
* add proper namespaces, and fix description
* put factory functions in their own file (for replaceability)
* add synchronization
* generate_projects.sh
* extract event_engine_base and event_engine_factory targets
Also separate iomgr/event_engine files in the BUILD, with comments
* gpr_platform
* move all EE factory declarations to event_engine_base
Makes internal hackery easier.
* add missing deps
* reorder dep alphabetically
* comment style change
* Assert Android API >= v21
This precedes a change that would otherwise break on older Android APIs,
but in a more obvious way.
* error if __ANDROID_API__ is not defined
* update all Andriod minSdkVersions to 21
* csharp experimental: android-19 to 21
Due to limitation of JVM, when user want to create binder channel in
threads created in unmanaged native code, they will need to call this
new API first to make sure Java helper classes can correctly be found.
Unused code in jni_utils.cc are also cleaned up in this commit
* [BinderTransport] Avoid depending on NdkBinder at compile time
We would like to make it possible to use BinderTransport in a APK that
has min sdk version lower than 29 (NdkBinder was introduced at 29)
We copies constants and type definitions from Ndk headers, creates a
same name wrapper for every NdkBinder API we use in
grpc_binder::ndk_util namespace.
We will try to load libbinder_ndk.so and resolve the symbol when the
NdkBinder API wrappers are invoked.
* regenerate projects
* Add GRPC_NO_BINDER guard
* Expose experimental binder transport API
New headers are added
`grpcpp/create_channel_binder.h `: interfaces for creating client
channel
`grpcpp/security/binder_credentials.h`: interfaces for binder server
credentials
`grpcpp/security/binder_security_policy.h`: interfaces for binder
security policy, which is used by both server and client. Individual
security policies are merged into this single header.
Users can now depend on the `grpc++_binder` target to use the headers
listed above.
* Regenerate projects
This introduces the new `Closure` type, and new cancellation semantics.
I've also fixed a few bugs in the EventEngine iomgr implementation that
had gone unnoticed.
* Fix OOM issues in qps tests
* Add more verbose logging.
* Fix clang error
* Fix race between IsCancelled and Read
* Fix build errors from using bool in C code
This commit adds a new symbol GPR_SUPPORT_BINDER_TRANSPORT to
port_platform.h
This will help avoid surprising compilation failure when compiled with
old NDK or low API level
It was determined that an explicit Shutdown method is not necessary, and further, it can be challenging to implement efficiently in some cases. Instead, EventEngines are expected to clean themselves up upon destruction. Anything that relies on an EventEngine's existence must coordinate itself to ensure the engine remains alive for as long as it's needed.
Internal bug b/170934515
In some cases, calling CQ::Next will return true without passing up a
new tag value, which is very much illegal. My expectation is that this
is due to messing up between SHUTDOWN and TIMEOUT in lower layer code
that doesn't particularly matter much to most callers, but was being
erroneously checked here.
It is not possible for such a function to be implemented in a way that
is understood by annotalysis. Mark it deprecated and replace instances
of its use with direct mutex/condvar usage.
Add a bunch of missing thread safety annotations while I'm here.
* Enable retries by default, but add a separate arg for hedging.
* don't need to explicitly enable retries in xDS config selector
* clang-format
* don't need retry_enabled bit anymore
* fix HTTP client filter to restore the send_message op in the batch
* fix retry cancellation when a batch fails on call attempt
* fix clang-tidy
* fix client channel to pass down batches even after cancellation
* fix retry code to pass transport stats back up to the surface
* add some missing payload propagation
* fix retry handling of callbacks for pending batches
* avoid scheduling the same callback twice
* fix some trace messages
* don't avoid starting recv_initial_metadata or recv_message due to recv_trailing_metadata already being started internally
* avoid restarting recv_trailing_metadata after commit if we've already started it internally
* use fast path when retries are not configured
Craig Tiller