RoundRobinLoadBalancerFactory creates a new Picker instance every time the set of provided address groups changes or the connection state of subchannels associated with existing address groups changes. In certain scenarios, such as deployment/replacement of the target service cluster, this can lead to high churn of Picker objects. Given that each new Picker's subchannel index is initialized to zero, in these scenarios requests can end up getting disproportionately routed through subchannels (and hence server nodes) which are earlier in the list of address groups.
At Netflix we have measured that some service nodes end up taking 3-4x the load that of other nodes during deployment.
This commit randomizes the start index of the RoundRobinLoadBalancerFactory.Picker which eliminates this behavior.
Passing a promise to WriteQueue was only misused to add a listener on
the promise before issuing the write. Although in this case the listener
ordering will be "random" because listeners are being added from two
different threads, in general we always want to add a listener after the
write returns to let any lower-level listeners be registered first.
Future work can resolve the "random" listener order by passing the
listener to the WriteQueue and adding the listener from the event loop.
We previously passed a custom variable to the checkstyle configuration.
In Gradle 4.0 config_loc was added for the same purpose. The default
configDir is $projectDir/config/checkstyle which is different for each
project; we need to override it to always point to the root project.
This opens up the ability of dependency locking and the now-stable Maven
Publish Plugin. Also failOnVersionConflict no longer needs to be
commented out for the dependency insight report.
Truth 0.42 brings in some Java 8 bytecode, but they are only in
annotations. So we remove them for gae-java7, otherwise they cause the
build to fail with messages like:
> Unable to stage app: Class file is Java 8 but max supported is Java 7: com/google/auto/value/extension/memoized/Memoized.class in /usr/local/google/home/ejona/clients/grpc-java/gae-interop-testing/gae-jdk7/build/exploded-grpc-gae-interop-testing-jdk7/WEB-INF/lib/auto-value-annotations-1.6.2.jar
> Unable to stage app: Class file is Java 8 but max supported is Java 7: org/checkerframework/dataflow/qual/Pure$Kind.class in /tmpfs/src/github/grpc-java/gae-interop-testing/gae-jdk7/build/exploded-grpc-gae-interop-testing-jdk7/WEB-INF/lib/checker-qual-2.5.3.jar
I manually tested the interop client with Java 7 and it ran without
issue.
This fixes the issues experienced with Truth 0.42 before in #4664.
Related: google/truth#479
Prevent multiple effective close calls either by successful completion
of a cancel or complete notification, or through successive exceptions
handled within a single call.
If the server sends all response message but does not send the trailers, client throws
StatusRuntimeException: INTERNAL: Received unexpected EOS on DATA frame from server
DoNotMock was removed from error_prone_annotations in 2.1.3, because
there was no enforcement mechanism (which is in google/error-prone#572).
Guava and Trust also depend on error_prone_annotations and are beginning
to use newer versions, so our usage of DoNotMock is causing diamond
dependency problems. This allows us to update to 2.2.0.
The annotations were useful internally; we're solving that in cl/205294089.
This greatly reduces the amount of content. The removed content was
already deteriorating. For example, it references grpc-java-deploy which
is no longer present nor necessary. It also didn't suggest using
build_docker.sh for building protoc-artifacts.
While the docs could have been updated, without actually running them
they are unlikely to be functional. Removing the docs seems better than
spending effort of maintaining them, especially since they add confusion
to the process and are a scary amount of content for new devs.
The buildscript was updated to pull in a fix to protoc-artifacts to
support TLS 1.2 (google/protobuf#4879), which was the only remaining
reason to have our own container.
Docker container building was split out into a separate build_docker.sh
so that people can call it blindly and get the container necessary for
run_in_docker.sh.
We always want to use a consistent version of protobuf; avoid the need
for the caller (which may be a person running the script) to specify the
version.
Tom Leach