Skip to content

Latest commit

 

History

History
170 lines (145 loc) · 6.48 KB

TROUBLESHOOTING.md

File metadata and controls

170 lines (145 loc) · 6.48 KB
Raw

Troubleshooting

ALPN is not configured properly

If you see exceptions related to ALPN is not configured properly, such as:

Caused by: java.lang.IllegalArgumentException: ALPN is not configured properly. See https://github.com/grpc/grpc-java/blob/master/SECURITY.md#troubleshooting for more information.

Please use the compatibility checker to see if your environment is compatible with grpc-based clients. The incompatibility can mean that:

ClassNotFoundException, NoSuchMethodError, NoClassDefFoundError

These errors are usually caused by having multiple versions or conflicting versions of the same dependency in the classpath. Usually these dependency conflicts occur with guava or protobuf-java.

There may be multiple sources for classpath conflicts:

  • Multiple versions of the same transitive dependency in the dependency tree
  • Your runtime classpath has different versions of dependencies than what you specified in the build

For example, if you have a direct or a transitive dependency on Guava version 19.0, and google-cloud-java uses Guava version 28.1, then google-cloud-java could be using Guava methods that don't exist in Guava 19.0, and cause NoSuchMethodError.

Similarily, if your classpath has an older version of protobuf-java, but google-cloud-java requires a newer version, then you may see NoClassDefFoundError that fails to initialize google-cloud-java classes, e.g.:

java.lang.NoClassDefFoundError: Could not initialize class com.google.pubsub.v1.PubsubMessage$AttributesDefaultEntryHolder

Validate the conflict

Check the dependency tree to see if you have multiple versions of the same dependencies, e.g.:

$ mvn dependency:tree

Look for versions of potentially conflicting dependencies like guava, protobuf-java, etc.

If you experience the error only during runtime, then your runtime environment might be introducing conflicting JARs into your runtime classpath. A typical case is that Hadoop, Spark, or other server software that your application runs on has conflicting versions netty, guava, or protobuf-java JARs in the classpath.

Detecting conflicts during build

To detect dependency linkage errors at compile time, add the Linkage Checker Enforcer Rule in your pom.xml:

      <plugin>
        <groupId>org.apache.maven.plugins</groupId>
        <artifactId>maven-enforcer-plugin</artifactId>
        <version>3.0.0-M3</version>
        <dependencies>
          <dependency>
            <groupId>com.google.cloud.tools</groupId>
            <artifactId>linkage-checker-enforcer-rules</artifactId>
            <version>1.1.3</version>
          </dependency>
        </dependencies>
        <executions>
          <execution>
            <id>enforce-linkage-checker</id>
            <!-- Important! Should run after compile -->
            <phase>verify</phase>
            <goals>
              <goal>enforce</goal>
            </goals>
            <configuration>
              <rules>
                <LinkageCheckerRule
                    implementation="com.google.cloud.tools.dependencies.enforcer.LinkageCheckerRule"/>
              </rules>
            </configuration>
          </execution>
        </executions>
      </plugin>

There is no way to detect runtime classpath conflicts though. You need to be fully aware of which JARs/classes are included in the runtime classpath as every server environment is different.

Resolving the conflict

There are different strategies to resolve conflicts, but you must understand the root cause of the conflicts, e.g.:

  • If you have control over the dependency tree, you can upgrade offending dependencies (e.g., upgrading Guava version). This is the least hackish approach but it is a lot of work that can require multiple releases of multiple libraries to sync everything up.
  • If you can't modify and push new versions of your dependencies, import com.google.cloud:libraries-bom:4.0.0 (or a more recent version) and use that to select consistent dependency versions. This is the easiest route. For example, this is how you can depend on consistent versions of Guava and com.google.cloud:google-cloud-storage without explicitly setting the version of either one:
  ...
  <dependencyManagement>
    <dependencies>
      <dependency>
        <groupId>com.google.cloud</groupId>
        <artifactId>libraries-bom</artifactId>
        <version>4.1.1</version>
        <type>pom</type>
        <scope>import</scope>
       </dependency>
     </dependencies>
  </dependencyManagement>
  ...
  <dependencies>
    <dependency>
      <groupId>com.google.cloud</groupId>
      <artifactId>google-cloud-storage</artifactId>
    </dependency>
    <dependency>
      <groupId>com.google.guava</groupId>
      <artifactId>guava</artifactId>
    </dependency>
    ...
  </dependencies>
  ...

  • If changing dependency versions causes other failures, consider shading dependencies that conflict with google-cloud-java.

    For example, to shade guava and protobuf-java:

  <plugin>
    <groupId>org.apache.maven.plugins</groupId>
    <artifactId>maven-shade-plugin</artifactId>
    <version>...</version>
    <executions>
      <execution>
        <phase>package</phase>
        <goals>
          <goal>shade</goal>
        </goals>
        <configuration>
          <keepDependenciesWithProvidedScope>false</keepDependenciesWithProvidedScope>
          <relocations>
            <!-- move protobuf to a shaded package -->
            <relocation>
              <pattern>com.google.protobuf</pattern>
              <shadedPattern>myapp.shaded.com.google.protobuf</shadedPattern>
            </relocation>
            <!-- move Guava to a shaded package -->
            <relocation>
              <pattern>com.google.common</pattern>
              <shadedPattern>myapp.shaded.com.google.common</shadedPattern>
            </relocation>
          </relocations>
        </configuration>
      </execution>
    </executions>
  </plugin>