* extract self-defined Locality into XdsLocality class
* separate out functionalities for recording client load from lrsClient, xds load balancer will directly interact with XdsLoadStatsStore to set up locality counters
* added GRPC to constant TRAFFICDIRECTOR_HOSTNAME_FIELD name to better match that in XdsComms
* fixed bug of using the wrong cluster name in load report's ClusterStats, it should be GSLB service name, which is responsed by load report response (same as that in EDS response).
* added a new line to the end of files.
* Revert "fixed bug of using the wrong cluster name in load report's ClusterStats, it should be GSLB service name, which is responsed by load report response (same as that in EDS response)."
This reverts commit 6097dd4066.
* rephrase interface comment for StatsStore
* added equality and hashCode test for XdsLocality
The test case `XdsLbStateTest.handleSubchannelState()` was introduced before `LocalityStore` refactored out of `XdsLbState`. After `LocalityStore` refactored out, the test case should not be in `XdsLbStateTest` anymore. The test case is already covered in `LocalityStoreTest`.
During migration, the name resolver may not know when the client has been upgraded to xds, so it may still send grpclb v1 addresses with a list of policies including both grpclb v1 and xds.
HealthCheckingLoadBalancer.shutdown() calls
hcState.onSubchannelState(SHUTDOWN) which removes that hcState from
helper.hcStates. Therefore, if more than one Subchannels are present,
ConcurrentModificationException will be thrown.
Since HealthCheckingLoadBalancer.shutdown() will clear the hcStates
set after the loop, it's unnecessary to do the deletion within the
loop. However, when a Subchannel is shutdown by LoadBalancer, its
HcState still needs to be removed. To do that, change moves the
deletion to Subchannel.shutdown().
This add perfmark annotations in some key places, notably on transport/application boundaries, and thread hop locations. Perfmark records to a thread-local buffer the events that happen in each thread. Perfmark is disabled by default, and will compile to a noop unless Perfmark.setEnabled is invoked. This should make it free when disable, and pretty fast when it is enabled.
It is important that started tasks are ended, so several places in our code are moved to either try-finally blocks, or moved into a private method. I realize this is ugly, but I think it is manageable. In the future, we can look at making an agent or compiler plugin that simplifies the recording.
Linking between threads is done with a Link object, which is created on the "outbound" task, and used on the "inbound" task. This is slightly more verbose, and does has a small amount of runtime overhead, even when disabled. (for null checks, slightly higher memory usage, etc.) I think this is okay to, because it makes other optimizations much easier.
Depending on jdk:toolchain causes java_grpc_library to always use the
_default_ toolchain, even if the user tried to override it. Changing to
:current_java_toolchain allows the rule to use the user-selected
toolchain when overridden.
Tested by adding to BUILD:
load("@bazel_tools//tools/jdk:default_java_toolchain.bzl", "default_java_toolchain")
default_java_toolchain(
name = "mychain",
misc = ["-Amy=flag"],
visibility = ["//visibility:public"],
)
And then verifying -Amy=flag is in the output of:
bazel aquery --java_toolchain=:mychain services:_reflection_java_grpc
Fixes#5841
com.github.dcendents:android-maven-gradle-plugin is incompatible with
Gradle 5 and the project hasn't seen any activity in over a year, so it
seems unlikely to get fixed.
We want to use maven-publish anyway, since that's what we use elsewhere.
Previously, AutoConfiguredLoadBalancer was also handling it, but it
doesn't trigger retries. By returning true for
canHandleEmptyAddressListFromNameResolution(),
AutoConfiguredLoadBalancer effectively by-passed the empty-result
handling logic in ManagedChannelImpl, thus resolution retries were
never triggered.
This change requires AutoConfiguredLoadBalancer to stop being a
LoadBalancer, for its tryHandleResolvedAddresses(). It doesn't
cause any trouble because AutoConfiguredLoadBalancer has become
less and less like a LoadBalancer during the service config changes.
The former is deprecated and replaced by the latter in Mockito 2.
However, there is a functional difference: ArgumentMatchers will reject
`null` and check the type if the matcher specified a type (e.g.
`any(Class)` or `anyInt()`). `any()` will remain to accept anything.
As mentioned in 5188[1], the default used with the enableKeepAlive API
conflicted with the default server enforcement. Instead of fixing it,
remove it. These APIs were deprecated in v1.3.0 in April 2017.
1. https://github.com/grpc/grpc-java/issues/5188#issuecomment-482269303
- Cancel fallback timer and/or exit fallback mode once receiving an EDS response indicating 100% drop.
- Also update balancing state once receiving the first EDS response with drop information when the channel is at the initial IDLE state.
* implemented utility methods to create ClientStreamTracer.Factory with OrcaReportListener installed for retrieving per-request ORCA data
* added unit tests
* use delegatesTo instead of spy
* implemented OrcaReportingHelper delegating to some original Helper for load balancing policies accessing OOB metric reports
* added unit tests for out-of-band ORCA metric accessing API in a separate test class
* rebase to master, resolve the breaking change of StreamInfo class being final with builder
* trashed hashCode/equal for OrcaReportingConfig
* changed log level and channel trace event level to ERROR as required by design doc
* added OrcaReportingHelperWrapper layer to allow updating report interval at any time
* reverse the naming of parent/child helper, child helper is the outer-most helper in the wrapping structure
* changed orca listener interface to use separate listener interfaces for per-request and out-of-band cases
* added more comprehensive unit tests
* added test case for per-request reporting that parent creates its own stream tracer
* fixed bug of directly assign reporting config, which would cause it be mutated later
* separate test cases for updating reporting config at different time
* fixed lint style error
* polish comments
* minor polish in unit tests
* refactor OrcaUtil class into OrcaOobUtil and OrcaPerRequestUtil and get rid of static methods for easier user testing
* hide BackoffPolicyProvider and Stopwatch supplier in OrcaOobUtil's public API
* add javadoc for getInstance() methods
* ensure the same Subchannel instance created by the helper that has corresponding OrcaOobReportListener registered are passed to the listener callback
* removed costNames foe OrcaReportingConfig
* removed redundant checks
* reformated the OrcaOobUtilTest class to put helper methods in the bottom
* fixed impl with changes made on Subchannel (SubchannelStateListener now ties with Subchannel)
* fixed comments
* added usage examples in javadoc for OrcaUtils
* add method comments for OrcaUtil's listener API threading
* make fields in OrcaReportingConfig final
* fixed OrcaOobUtilTest for calling setOrcaReportingConfig inside syncContext
* added ExperimentalApi annotation for Orca utils
The `method` and `status` shouldn't be propagated in the first place,
but in previous OpenCensus implementation all tags are propagating by
default. Now with the TagMetadata it may make sense to change them to
local tags.
This will be a breaking change to users who depend on the behavior that
these tags propagate through process boundaries.
The old createSubchannel() doesn't require being called from
sync-context, thus its implementation schedules start() on
sync-context because start() has such requirement. However, if a
LoadBalancer call createSubchannel() in sync-context, start() is
queued and only called after the control exits the sync-context. If
createSubchannel() is immediately followed by other Subchannel methods
that requires start(), such as requestConnection(), they will fail.
This will be a very common issue as most LoadBalancers call
createSubchannel() in the sync-context.
This fix splits out the real work of start() into internalStart() which
is called by the old createSubchannel().
Chengyuan Zhang
ZHANG Dapeng
Carl Mastrangelo
Kun Zhang
Eric Anderson
Jihun Cho
Ran
Manuel Kollus
Ryan Michela
Tim van der Lippe
Yang Song