Optimize OkHttp transport's memory use by getting rid of queuing writes in
AsyncFrameWriter. If any write is pending due to connection issue or by flow
control, AsyncFrameWriter can use at least 8K per each task (task includes
buffer) even if the actual payload is very small. To merge pending writes,
Async mechanism is moved from AsyncFrameWriter to AsyncSink (AsyncSink
is used by okio's FrameWriter). AsyncSink is still relying on okio's buffer to
decide merging writes or not.
Resolves#4860
Currently only protos in alts, grpclb, and services are synced. Once
some java_package options are "upstreamed", benchmarks and
interop-testing could also be synced.
the fundamental issue is that the sha-1 sums for the github and maven
jars for re2j are different. (see
https://github.com/google/re2j/issues/83 for discussion.) the sha used
at grpc-java head corresponds to the github jar, but the target itself
refers to the maven jar.
the reason no one has noticed is that @com_google_re2j//jar is not
actually used as a dependency in any bazel target in this repo. this
will change once the services/ directory becomes a bazel package. (the
source files in services/ do depend on re2j.)
Since the Resource shared the executor service between invocations, but
didn't null it out on shutdown, it could bring up a new channel with a
terminated event loop. The channel would then proceed to panic on usage.
I noticed this problem while looking into what was necessary for #4755.
* makes Census tracing factories at the end of the user added ones
* makes more vars in AbstractServerImplBuilder package private
* annotates methods in ASIB to be clearer
* simplifies several of the setters to be single line
* Makes the generics on the Tracer factories proper
All files other than the following are generated by `import.sh`.
```
settings.gradle
xds/build.gradle
xds/third_party/envoy/import.sh
xds/third_party/protoc-gen-validate/import.sh
```
This also reduces the time of the bandwidth tests from ~6 seconds to ~2,
each (they have about 1s of overhead).
Fixes#2951. The flake is very frequent in #4980.
Currently ManagedChannelImpl will interpret empty address list from
NameResolver as an error, and LoadBalancer will NEVER receive an empty
list from handleResolvedAddressGroups(). There is a case in the
request-routing design where a LoadBalancer only receives service
config with which it constructs children NameResolver/LoadBalancer
pairs, thus no address is expected at this level.
canHandleEmptyAddressListFromNameResolution() is a signal to Channel
(and to the parent LoadBalancer in case of hierachical LoadBalancers)
about whether it accepts empty address lists. The default is false,
which is the current behavior.
The logic is currently duplicated in ManagedChannelImpl and
AutoConfiguredLoadBalancer. The one in ManagedChannelImpl will be
removed once we delete ManagedChannelBuilder.loadBalancerFactory() as
AutoConfiguredLoadBalancer will always be the top-level LoadBalancer
by then.
Previously, overwriting an existing Key would cause the original CallOptions instance to also be mutated.
See #5142
Also adds a regression test for this issue.
This allows calling build() multiple times as well as prevents other
interceptors from being able to detect that we've implemented
CallCredential attachment via an interceptor. (Previously they could
have set their own CallCredentials which would have overridden the
default creds.)
ManagedChannelBuilder.loadBalancingFactory() overrides the proper
policy selection logic implemented in AutoConfiguredLoadBalancer, thus
has problems in cases where NameResolver returns balancer addresses,
because custom LoadBalancers normally don't differentiate between
normal server addresses with balancer addresses. The policy selection
logic will filter out balancer addresses.
When service owner turns on grpclb through service config, it
shouldn't break existing clients that don't have grpclb in their
classpath.
Resolves#4602
ZHANG Dapeng