gRPC performance improvement

[richieyan]

Motivation: Armeria's the gRPC request (de)serialization is in IO thread. If the deserialization process takes too long, the IO task may experience significant delays. Move the deserialization from the IO thread to BlockingExecutor thread will improve performance significantly.

Modifications:

• Move the execution of gRPC (de)serialization in com.linecorp.armeria.internal.server.grpc.AbstractServerCall#onRequestMessage from the IO thread to the BlockingExecutor.

Result:

• Closes #5152
• Large and complex gRPC request processing performance will improve significantly.

[github-actions[bot]]

🔍 Build Scan® (commit: b4219be4851cff276a1137e76c5aef2758a4a48d)

Job name	Status	Build Scan®
build-windows-latest-jdk-19	✅	https://ge.armeria.dev/s/7bib2eb34wke2
build-self-hosted-jdk-8	✅	https://ge.armeria.dev/s/qrgjjlxdbk35s
build-self-hosted-jdk-19-snapshot-blockhound	✅	https://ge.armeria.dev/s/d3ejp7n7jktas
build-self-hosted-jdk-17-min-java-17-coverage	✅	https://ge.armeria.dev/s/yf77hg3knx3wa
build-self-hosted-jdk-17-min-java-11	✅	https://ge.armeria.dev/s/2rco3tciidqss
build-self-hosted-jdk-17-leak	✅	https://ge.armeria.dev/s/opvaqdx2uwn3w
build-self-hosted-jdk-11	✅	https://ge.armeria.dev/s/nrycqwzo4icni
build-macos-12-jdk-19	❌ (failure)	https://ge.armeria.dev/s/nu7ee4yvh4onk

[jrhee17]

command:

./gradlew --no-daemon :benchmarks:jmh:clean :benchmarks:jmh:jmh -Pjmh.includes=grpc.downstream.DownstreamSimpleBenchmark.simple$ -Pjmh.params="clientType=NORMAL;useBlockingTaskExecutor=true,false"

Note: main branch: https://ge.armeria.dev/s/wrjaj3csesbe6

Benchmark                         (clientType)  (useBlockingTaskExecutor)   Mode  Cnt     Score    Error  Units
DownstreamSimpleBenchmark.simple        NORMAL                       true  thrpt    5  1425.960 ±  3.838  ops/s
DownstreamSimpleBenchmark.simple        NORMAL                      false  thrpt    5  1504.965 ± 41.929  ops/s

grpc-performance-improv: https://ge.armeria.dev/s/4gsvjzwljnaee

Benchmark                         (clientType)  (useBlockingTaskExecutor)   Mode  Cnt     Score    Error  Units
DownstreamSimpleBenchmark.simple        NORMAL                       true  thrpt    5  1391.880 ± 34.705  ops/s
DownstreamSimpleBenchmark.simple        NORMAL                      false  thrpt    5  1496.237 ± 55.994  ops/s