gRPC Java 1.68.1 Release Notes

v1.68.0 was a mistake. This is the first release of version 1.68.x

• xds: Fix NullPointerException introduced in "Fix load reporting when pick first is used for locality-routing" (#11553). This was in 1.67.1 but not 1.68.0

• core: JSON parsing rejects duplicate keys in objects (#11575) (4be69e3f8). This is the existing behavior in C core. Duplicate keys in objects are dangerous as which value takes effect is undefined. Previously, the last value was used
• okhttp: Detect transport executors with no remaining threads (#11503) (3a6be9ca1). The transport uses two threads, but one is on-demand. If the executor provided to builder.transportExecutor() runs out of threads (e.g., it is a fixed-size thread pool), all transports can be wedged, unable to run on-demand tasks, until keepalive kills one of them. Two threads are now used when handshaking a new transport, and the transport will time out after 1 second with “Timed out waiting for second handshake thread” if two threads are unavailable
• gcp-csm-o11y: Get mesh_id value from CSM_MESH_ID environment variable, instead of getting it from bootstrap file (84d30afad)

• New grpc-context-override-opentelemetry artifact (#11523) (782a44ad6) (#11599) (e59ae5fad). This is a io.grpc.Context storage override to store its state in io.opentelemetry.context.Context. Libraries should not add a dependency on this artifact, as applications can only have one storage override in their classpath
• New grpc-s2a artifact. It is a transport that offloads the handshake similar to ALTS, but for TLS. It provides io.grpc.s2a.S2AChannelCredentials
• api: Enhance name resolver `ResolutionResult` to hold addresses or error so the single listener API onResult2 is used to convey both success and error cases for name resolution (#11330) (1ded8aff8)
• core: Handle NameResolver/LoadBalancer exceptions when panicking (b692b9d26). This expands the class of bugs that will fail RPCs with the panic error, versus some undefined behavior
• core: Use the default service config in case of initial name resolver address resolution error (#11577) (fa26a8bc5)
• core: StreamTracer.inboundMessageRead() now reports uncompressed message size when the message does not need compression (#11598) (2aae68e11). Previously it always reported -1 (unknown)
• netty: Avoid TCP_USER_TIMEOUT warning when explicitly specifying a non-epoll channel type to use (#11564) (62f409810)
• okhttp: Don't warn about missing Conscrypt (6f3542297). This is especially helpful when using TLS but not running on Android
• android: For UdsChannelBuilder, use fake IP instead of localhost (a908b5e40). This avoids an unnecessary DNS lookup
• xds: Add xDS node ID in select control plane errors to enable cross-referencing with control plane logs when debugging (f3cf7c3c7)
• xds: Enhanced how ADS stream terminations are handled, specifically addressing cases where a response has or hasn't been received (#2e9c3e19f)
• binder: Update status code documentation for Android 11's package visibility rules. (#11551) (99be6e985)
• binder: Update binderDied() error description to spell out the possibilities for those unfamiliar with Android internals. (#11628) (46c1b387f)
• example-gauth: Use application default creds instead of file argument (#11595) (94a0a0d1c)

• Updated protobuf-java to 3.25.5. This helps avoid CVE-2024-7254 (2ff837ab6)

Thanks to:
@Juneezee
@lgalfaso
@bestbeforetoday
@hlx502
@JoeCqupt

xds: Stop extending RoundRobin in WeightedRoundRobin (dc83446d9) xds: Remove ExperimentalApi for WeightedRoundRobin LB as it is already package private. (0d2ad8901)

Introduce onResult2 in NameResolver Listener2 that returns Status (90d0fabb1) xds: Add test that uses real DnsNR with ClusterResolverLB (9bc1a93f6) android-interop-testing: Enable -Xlint:deprecation (b8e3ae9a4) api: Move ClientStreamTracerTest from core to api (780e4ba08) core: Don't reuse channels in PickFirstLeafLB test (e567b4427) core: In PickFirst LB, pass around SubchannelData instead of Subchannel (15456f8f0) CONTRIBUTING.md: Update commit guidelines (18d73a368) util: SocketAddress.toString() cannot be used for equality (f866c805c) util: Delete unused MultiChildLB.ChildLbState.getSubchannels() (2f4f7f0ec) xds: Replace WrrHelper with a per-child Helper (d1dcfb045) xds: WRRPicker must not access unsynchronized data in ChildLbState (0d47f5bd1) netty: Use DefaultELG with LocalChannel in test because LocalChannel is not guaranteed to be compatible with NioEventLoopGroup (4ab34229f) util: Remove implicit requestConnection() on IDLE from MultiChildLB (a6f8ebf33) util: MultiChildLb children should always start with a NoResult picker (b5989a540) xds: Delegate more RingHashLB address updates to MultiChildLB (fd8734f34) cronet: Add internal API to specify Network (c2eccca3b) interop-testing: Add dualstack interop test kokoro config (#11422) (941a1c37a) example: Dualstack example (#11451) (72a977bf7) xds: Separate xds clients for each authority, each with its own connection to an xds server. (#11484) (d034a56cb) rls: Fix log statements incorrectly referring to "LRS" (#11497) (c63e35488) xds: Envoy proto sync to 2024-07-06 (#11401) (96a788a34) xds: cncf/xds proto sync to 2024-07-24 (#11417) (0017c98f6) xds: Import RLQS protos (#11418) (c29763d88) core: Make once-set fields of AbstractClientStream as final(#11389) (40e2b165b) opentelemetry: add binary format, grpcTraceBinContextPropagator (#11409) (043ba556b) opentelemetry: add OpenTelemetryTracingModule (#11477) (421e2371e) example: delete duplicate and unused code in KeepAliveClient.java (6a9bc3ba1) util: Remove minor convenience functions from MultiChildLB (909c4bc38) core: PickFirstLeafLoadBalancer Index.size() should be number of addresses (c120e364d) util: Remove MultiChildLB.getImmutableChildMap() (778a00b62) core: Never have null PickFirstLeafLoadBalancer Index (8bd97953a) core: Remove useless NPE check for syncContext in PF (33687d398) core: Avoid exception handling in PF for invalid index (4914ffc59) core: PF.requestConnection() is possible when READY (2c93791c9) core: In PF, remove useless requestConnection for CONNECTING subchannel (82a8d5739) core: In PF, remove extraneous index.reset() (9762945f8) interop-testing: Move interop tests only used by test client to the test client (6c9f92a72) util: Replace RoundRobinLoadBalancer.EmptyPicker with FixedResultPicker (f20167d60) core: In PF, disjoint update while READY should transition to IDLE (ee3ffef3e) xds: ClusterManagerLB must update child configuration (10d6002cb) util: Remove child policy config from MultiChildLB state (01389774d) util: MultiChildLB children know if they are active (4cb646519) util: Focus MultiChildLB updates around ResolvedAddresses of children (cfecc4754) core: touch() buffer when detach()ing (e821d5e15) xds: XdsClient should unsubscribe on last resource (#11264) (448ec4f37) stub: Add newAttachMetadataServerInterceptor() MetadataUtil (#11458) (6dbd1b9d5)

xds: Fix load reporting when pick first is used for locality-routing. (#11495) (1dae144f0) core: Revert "Enable new PickFirst LB (#11348)" (#11425) (cc1cbe987)

.github/workflows: Bump action major versions from Node16 to Node20 (#11476) (66075eab8) interop-testing: Remove unused implementation deps (1f9d50226) Revert "Netty upgrade to 4.1.110 in grpc-java (#11273)" (9bed655c5) Netty upgrade to 4.1.110 in grpc-java (#11273) (f9b072cfe) Upgrade Netty to 4.1.110 and tcnative to 2.0.65 (#11444) (70ae83288) Migrate from the deprecated Charsets constants (in Guava) to the StandardCharsets constants (in the JDK). (00136096e) Migrate from the deprecated Charsets constants (in Guava) to the StandardCharsets constants (in the JDK) (06135a074) Migrate from Charsets to StandardCharsets. (#11482) (2fe1a13cd) examples: Upgrade Maven plugin versions (75012a5be) Remove direct dependency on j2objc (ff8e41376)

Eng Zer Jun engzerjun@gmail.com Jiajing LU lujiajing1126@gmail.com Petr Portnov | PROgrm_JARvis mrJARVIScraft@gmail.com sunpe sunpeng.dev@gmail.com

This was supposed to be v1.67.0, but there was a mistake during the release process. This has everything in v1.67.1, except for:

• xds: Fix NullPointerException introduced in "Fix load reporting when pick first is used for locality-routing" (https://github.com/grpc/grpc-java/pull/11553)

• stub: Support setting onReadyThreshold through AbstractStub. (#11320) (25a8b7c50)
• util: Stabilize AdvancedTlsX509TrustManager, an X509ExtendedTrustManager that allows users to configure advanced TLS features, such as root certificate reloading and peer cert custom verification. (658cbf6cf)
• util: Align AdvancedTlsX509{Key and Trust}Manager. (#11385)
• util: Add GracefulSwitchLoadBalancer config (ebed04798) and mark switchTo() deprecated. (85e0a01ec). GracefulSwitchLoadBalancer now receives its configuration like a regular load balancer.
• binder: Introduce AllowSecurityPolicy to allow calling code to not have to wait on async/slow implementations. BinderTransport now submits async implementations to an executor. (#11272) (7fee6a3fe)
• api: Add convenience method in ServerBuilder for adding a list of service implementations to the handler registry together. (#11285) (85ed05300)

• examples: Improve example Bazel WORKSPACE to demonstrate referencing grpc-xds. (5ec0187e2)
• examples: Include Bazel bzlmod configuration (36e687f9d). There are now examples for both non-bzlmod and bzlmod.
• core: Fixes to PickFirstLeafLoadBalancer
  • Eliminate NPE after recovering from a temporary name resolution failure. (#11298)
  • Deduplicate addresses. (#11342, #11345)
• core: Change default to use the new pick first load balancer (PickFirstLeafLoadBalancer). (#11348)
• core: Use retryThrottling from defaultServiceConfig when the name resolver config doesn't provide this config. (#11274) (062ebb4d7)
• netty: Enable use of Netty 4.1.111 by avoiding the optimization provided by NettyAdaptiveCumulator if Netty is on version 4.1.111 or later. (#11367)
• binder: Set a default connect timeout of 60 seconds. (#11359) (21dec3092)
• binder: Make BinderServer own ServerAuthInterceptor's executor that helps avoid leaks. (#11293) (15ad9f546)
• services:: Added ProtoReflectionServiceV1 for the v1 reflection protocol. The preexisting ProtoReflectionService implements the v1alpha reflection protocol. (#11237) (0aa976c4e)

• binder: Add missing synchronization to prevent races when calling awaitTermination(). (#11277) (14fd81f59)
• util: Fix AdvancedTlsX509TrustManager validation on servers when using SSLSocket. Previously it would try to use a null SSLEngine . (dcb1c018c)

• compiler: Upgrade from CentOS 7 to AlmaLinux 8 for the pre-compiled Linux protoc-gen-grpc-java (71eb5fb9f). This adds a runtime dependency on libstdc++
• Upgrade animal-sniffer-annotations to 1.24 (a97738518)
• Upgrade error_prone_annotations to 2.28.0 (a97738518)
• Upgrade proto-google-common-protos to 2.41.0 (a97738518)
• Upgrade google-auth-library to 1.23.0 (a97738518)
• Upgrade gson to 2.11.0 (a97738518)
• Upgrade guava to 33.2.1 (a97738518)
• Upgrade opentelemetry to 1.40.0 (a97738518)
• Upgrade perfmark-api to 0.27.0 (a97738518)
• Upgrade protobuf-java to 3.25.3 (a97738518)
• xds: Remove unused opencensus-proto dependency (e7c3803b5)
• bazel: Replace @com_github_cncf_udpa usage with preexisting @com_github_cncf_xds; delete @com_github_cncf_udpa repo alias for xds (6dd6ca9f9)
• bazel: Upgrade envoyproxy/data-plane-api to 1611a730 (c540993aa). The version used by Gradle had been updated in 1.62.0 (68334a01), but the bazel version had not
• bazel: Use com_google_protobuf instead of com_google_protobuf_javalite (7a25e6895). Bazel’s protobuf rules no longer use the old com_google_protobuf_javalite repository name
• bazel: Don't require protobuf to be in maven_install (d3c2f5a2d). Protobuf’s targets are generally just used directly; this fixed the only place that used maven’s artifact() syntax

@hlx502 @erm-g @jdcormie @JoaoVitorStein @cfredri4

• okhttp: Workaround SSLSocket not noticing socket is closed
• netty: skip using NettyAdaptiveCumulator when Netty is on version 4.1.111 or later

• netty: Restore old behavior of NettyAdaptiveCumulator, but avoid using that class if Netty is on version 4.1.111 or later

• netty: Restore old behavior of NettyAdaptiveCumulator, but avoid using that class if Netty is on version 4.1.111 or later

• netty:Fix Netty composite buffer merging to be compatible with Netty 4.1.111 (1.64.x backport) by @larry-safran in https://github.com/grpc/grpc-java/pull/11303

grpc-netty in this release is compatible with Netty 4.1.111; it fixes the incompatibility that caused data corruption. grpc-netty-shaded is still using Netty 4.1.100.

• New module grpc-gcp-csm-observability (df8cfe9dd)

• api: Add ClientStreamTracer.inboundHeaders(Metadata) (960012d76). This is the same as the existing inboundHeaders(), but is provided the Metadata
• api: Fix various typos in the documentation (#11144) (6ec744f2a)
• core: When queuing RPCs, don’t request picks from the LB twice (8844cf7b8). This could be viewed as a small performance optimization, but mainly reduces the amount of race-handling code
• util: Improve AdvancedTlsX509KeyManager’s documentation, verification, and testing. (#11139) (781b4c457) This change shows @ExperimentalApi being removed, but it was re-added in 3c97245 before the release
• examples: Fix broken command in reflection readme (#11131) (c31dbf48a)
• binder: Add a connection timeout (#11255) (791f894e2)

• core: Exit idle mode when delayed transport is in use (fea577c80). This was a long-standing race that could cause RPCs to hang, but was very unlikely to be hit. Avoiding the double-picking (8844cf7b8) made the race more visible
• netty: Fix Netty composite buffer merging to be compatible with Netty 4.1.111 (#11294) (0fea7dd). The previous behavior easily caused data corruption
• okhttp: Workaround SSLSocket not noticing socket is closed (a28357e19). Previously, shutting down when a new connection was being established could result in the server never becoming terminated
• inprocess: Fix listener race if transport is shutdown while starting (e4e7f3a06). This issue was unlikely to be hit outside of specialized tests
• services: restore //services:binarylog bazel target (#11292) (d57f271). This fixes a regression introduced in 1.62.2
• binder: Wait for all server transports to terminate before returning the security policy executor to the object pool (#11240) (34ee600dc)
• binder: Reject further SETUP_TRANSPORT requests post-BinderServer shutdown (#11260) (1670e97f7)
• bazel: Include missing com_google_protobuf_javalite in MODULE.bazel (#11147) (f995c121e)

@hakusai22 @firov @mateusazis @Mir3605 @niloc132

• netty: Release SendGrpcFrameCommand when stream is missing (#11116) (fb9a10809)
• Change defaults to use the older PickFirstLoadBalancer and disable Happy Eyeballs. This disables a performance optimization added in v1.63.0 (#11120) We have had a report that the new implementation can trigger a NullPointerException