Debugging AEM as a Cloud Service build and deployments

Adobe Cloud Manager facilitates the code build and deployments to AEM as a Cloud Service. Failures may occur during steps in the build process, requiring action to resolve them. This guide walks through understanding common failures in in the deployment, and how to best approach them.

Cloud manage build pipeline

Validation

The validation step simply ensures that basic Cloud Manager configurations are valid. Common validation failures include:

The environment is in an invalid state

  • Error message: The environment is in an invalid state.
    The environment is in an invalid state
  • Cause: The pipeline’s target environment is in a transitional state at which time it cannot accept new builds.
  • Resolution: Wait for the state to resolve to a running (or update available) state. If the environment is being deleted, re-create the environment, or choose a different environment to build to.

The environment associate with the pipeline cannot be found

  • Error message: The environment is marked as deleted.
    The environment is marked as deleted
  • Cause: The environment the pipeline is configured to use has been deleted.
    Even if a new environment of the same name is re-created, Cloud Manager will not automatically re-associate the pipeline to that same-named environment.
  • Resolution: Edit the pipeline configuration, and re-select the environment to deploy to.

The Git branch associated with the pipeline cannot be found

  • Error message: Invalid pipeline: XXXXXX. Reason=Branch=xxxx not found in repository.
    Invalid pipeline: XXXXXX. Reason=Branch=xxxx not found in repository
  • Cause: The Git branch the pipeline is configured to use has been deleted.
  • Resolution: Re-create the missing Git branch using the exact same name, or re-configure the pipeline to build from a different, existing branch.

Build & Unit Testing

Build and Unit Testing

The Build and Unit Testing phase performs a Maven build (mvn clean package) of the project checked out from the pipeline’s configured Git branch.

Errors identified in this phase should be re-producible building the project locally, with the following exceptions:

  • A maven dependency not available on Maven Central is used, and the Maven repository containing the dependency is either:

    • Unreachable from Cloud Manager, such as a private internal Maven repository, or the Maven repository requires authentication and the incorrect credentials have been provided.
    • Not explicitly registered in the project’s pom.xml. Note that, including Maven repositories is discouraged as it increases build times.
  • Unit tests fail due to timing issues. This may occur when unit tests are timing-sensitive. A strong indicator is relying on .sleep(..) in the test code.

  • The use of unsupported Maven plugins.

Code Scanning

Code Scanning

Code scanning performs static code analysis using a mix of Java and AEM-specific best practices.

Code scanning results in a build failure if a Critical Security vulnerabilities exist in the code. Lesser violations can be overridden, but it is recommended they are fixed. Note that code scanning is imperfect and can result in false positives.

To resolve code scanning issues, download the CSV-fomatted report provided by Cloud Manager via the Download Details button and review any entries.

For more details see AEM specific rules, see Cloud Manager documentations’ custom AEM-specific code scanning rules.

Build Images

Build Images

Build image is responsible for combining the built code artifacts created in the Build & Unit Testing step with the AEM Release, to form a single deployable artifact.

While any code build and compilation issues are found during Build & Unit Testing, there may be configuration or structural issues identified when attempting to combine the custom build artifact with the AEM release.

Duplicate OSGi configurations

When multiple OSGi configurations resolve via runmode for the target AEM environment, the Build Image step fails with the error:

[ERROR] Unable to convert content-package [/tmp/packages/enduser.all-1.0-SNAPSHOT.zip]:
Configuration 'com.example.ExampleComponent' already defined in Feature Model 'com.example.groupId:example.all:slingosgifeature:xxxxx:X.X',
set the 'mergeConfigurations' flag to 'true' if you want to merge multiple configurations with same PID

Cause 1

  • Cause: The AEM project’s all package, contains multiple code packages, and the same OSGi configuration is provided by more than one of the code packages, resulting in a conflict, resulting the Build Image step unable to decide which should be used, thus failing the build. Note this does not apply to OSGi factory configurations, as long as they have unique names.
  • Resolution: Review all code packages (including any included 3rd party code packages) being deployed as part of the AEM application, looking for duplicate OSGi configurations that resolve, via runmode, to the target environment. The error message’s guidance of “set the mergeConfigurations flag to true” is not possible in AEM as a Cloud service, and should be ignored.

Cause 2

  • Cause: The AEM project’s incorrectly includes the same code package twice, resulting in the duplication of any OSGi configuration contained in said package.
  • Resolution: Review all pom.xml’s of packages embedded in the all project, and ensure they have the filevault-package-maven-plugin configuration set to <cloudManagerTarget>none</cloudManagerTarget>.

Malformed repoinit script

Repoinit scripts define baseline content, users, ACLs, etc. In AEM as a Cloud Service, repoinit scripts are applied during Build Image, however on AEM SDK’s local quickstart they are applied when the OSGi repoinit factory configuration is activated. Because of this, Repoinit scripts may quietly fail (with logging) on AEM SDK’s local quickstart, and but cause the Build Image step to fail, halting deployment.

  • Cause: A repoinit script is malformed. This may leave your repository in an incomplete state as any repoinit scripts after the failing script are not executed against the repository.
  • Resolution: Review the AEM SDK’s local quickstart when the repoinit script OSGi configuration is deployed to determine if and what the errors are.

Unsatisfied repoinit content dependency

Repoinit scripts define baseline content, users, ACLs, etc. In AEM SDK’s local quickstart, repoinit scripts are applied when the repoinit OSGi factory configuration is activated, or in other words, after the repository is active and may have incurred content changes directly or via content packages. In AEM as a Cloud Service, repoinit scripts are applied during Build Image against a repository that may not contain content the repoinit script depends on.

  • Cause: A repoinit script depends on content that is not existent.
  • Resolution: Ensure the content the repoinit script depends on exists. Often, this indicates an inadequately defined repoinit scripts that is missing directives that define these missing, but required, content structures. This can reproduced locally by deleting AEM, unpacking the Jar and adding the repoinit OSGi configuration containing the repoinit script to the install folder, and starting AEM. The error will present itself in the AEM SDK local quickstart’s error.log.

Application’s Core Components version is greater than deployed version

This issue only affects non-production environments that do NOT auto-update to the latest AEM Release.

AEM as a Cloud Service automatically includes the latest Core Components version in every AEM Release, meaning after an AEM as a Cloud Service environment is, automatically or manually updated, is had the latest version of Core Components deployed to it.

Is is possible for the Build Image step will fail when:

  • The deploying application updates the Core Components maven dependency version in the core (OSGi bundle) project
  • The deploying application is then deployed to a sandbox (non-production) AEM as a Cloud Service environment which has not been updated to use an AEM release that contains that new Core Components version.

To prevent this failure, whenever an Update of the AEM as a Cloud Service environment is available, include the update as part of the next build/deploy, and always ensure the updates are included after incrementing the Core Components version in the application code base.

  • Symptoms:
    The Build Image step fails with an ERROR reporting that com.adobe.cq.wcm.core.components... package(s) at specific version ranges could not be imported by the core project.

    code language-none
    [ERROR] Bundle com.example.core:0.0.3-SNAPSHOT is importing package(s) Package com.adobe.cq.wcm.core.components.models;version=[12.13,13) in start level 20 but no bundle is exporting these for that start level in the required version range.
    [ERROR] Analyser detected errors on feature 'com.adobe.granite:aem-ethos-app-image:slingosgifeature:aem-runtime-application-publish-dev:1.0.0-SNAPSHOT'. See log output for error messages.
    [INFO] ------------------------------------------------------------------------
    [INFO] BUILD FAILURE
    [INFO] ------------------------------------------------------------------------
    
  • Cause: The application’s OSGi bundle (defined in the core project) imports Java classes from Core Components core dependency, at a different version level than is what’s deployed to AEM as a Cloud Service.

  • Resolution:

    • Using Git, revert to a working commit that exists prior to Core Component version increment. Push this commit to a Cloud Manager Git branch and perform an Update of the environment from this branch. This will upgrade AEM as a Cloud Service to the latest AEM Release, which will include the later Core Components version. Once the AEM as a Cloud Service is updated to the latest AEM Release, which will have the latest Core Components version, redeploy the originally failing code.
    • To reproduce this issue locally, ensure the AEM SDK version is the same AEM release version the AEM as a Cloud Service environment is using.

Create an Adobe Support case

If the above troubleshooting approaches do not resolve the issue, please create an Adobe Support case, via:

  • Adobe Admin Console > Support Tab > Create Case

    If you are a member of multiple Adobe Orgs, ensure the Adobe Org that has failing pipeline is selected in the Adobe Orgs switcher prior to creating the case.

Deploy to

The Deploy to step is responsible for taking the code artifact generated in Build Image, starts up new AEM Author and Publish services using it, and upon success, removes any old AEM Author and Publish services. Mutable content packages and indexes are installed and updated in this step as well.

Familiarize yourself with AEM as a Cloud Service logs prior to debugging the Deploy to step. The aemerror log contains information around the start up and shutdown of pods which may be pertinent to Deploy to issues. Note that the log available via the Download Log button in the Cloud Manager’s Deploy to step is not the aemerror log, and does not contain detailed information pertaining to your applications start up.

Deploy to

The three primary reasons why the Deploy to step may fail:

The Cloud Manager pipeline holds an old AEM version

  • Cause: A Cloud Manager pipeline holds an older version of AEM than what is deployed to the target environment. This may happens when a pipeline is re-used and pointed at a new environment that is running a later version of AEM. This can be identified by checking to see if the environment’s AEM version is greater than than pipeline’s AEM version.
    The Cloud Manager pipeline holds an old AEM version

  • Resolution:

    • If the target environment has an Update Available, select Update from the environment’s actions, and then re-run the build.
    • If the target environment does not have an Update Available, this means it is running the latest version of AEM. To resolve this, delete the pipeline and re-create it.

Cloud Manager times out

Code running during the start up of the newly deployed AEM service takes so long that Cloud Manager times out before the deploy can complete. In these cases, the deployment may eventually succeed, even thought Cloud Manager status reported Failed.

  • Cause: Custom code may execute operations, such as large queries or content traversals, triggered early on in OSGi bundle or Component life-cycles significantly delaying the start up time of AEM.
  • Resolution: Review the implementation for code that runs early in the OSGi Bundle’s lifecycle, and review the aemerror logs for AEM Author and Publish services around the time of the failure (log time in GMT) as shown by the Cloud Manager, and look for log messages indicating any custom log running processes.

Incompatible code or configuration

Most code and configuration violations are caught in earlier in the build, however it is possible for custom code or configuration to be incompatible with the AEM as a Cloud Service and go undetected until it executes in the container.

  • Cause: Custom code may invoke lengthy operations, such as large queries or content traversals, triggered early on in OSGi bundle or Component life-cycles significantly delaying the start up time of AEM.

  • Resolution: Review the aemerror logs for AEM Author and Publish services around the time (log time in GMT) of the failure as shown by the Cloud Manager.

    1. Review the logs for any ERRORS thrown by the Java classes provided by the custom application. If any issues are found, resolve, push the fixed code, and re-build the pipeline.
    2. Review the logs for any ERRORS reported by aspects of AEM that you are extending/interacting with in your custom application, and investigate those; these ERRORs may not be directly attributed to Java classes. If any issues are found, resolve, push the fixed code, and re-build the pipeline.

Including /var in content package

/var is mutable containing a variety of transient, runtime content. Including /var in a content packages (ex. ui.content) deployed via Cloud Manager may cause the Deploy to step to fail.

This issue difficult to identify as it does not result in a failure on the initial deployment, only on subsequent deployments. Noticeable symptoms include:

  • The initial deployment succeeds, however new or changed mutable content, that is part of the deployment, does not appear to exists on AEM Publish service.
  • Activation/De-activation of content in AEM Author is blocked
  • Subsequent deployments fail in the Deploy to step, with the Deploy to step failing after approximately 60 minutes.

To validate this issue is the cause of the failing behavior:

  1. Determining that at least one content package that is part of the deployment, writes to /var.

  2. Verify the primary (bolded) distribution queue is blocked at:

    • AEM Author > Tools > Deployment > Distribution
      Blocked distribution queue
  3. On failing subsequent deployment, download Cloud Manager’s “Deploy to” logs using the Download Log button:

    Download deploy to logs

    … and verify there is approximately 60 minutes between the log statements:

    code language-none
    2020-01-01T01:01:02+0000 Begin deployment in aem-program-x-env-y-dev [CorrelationId: 1234]
    

    … and …

    code language-none
    2020-01-01T02:04:10+0000 Failed deployment in aem-program-x-env-y-dev
    

    Note that this log will not contain these indicators on the initial deployments that reports as successful, rather only on subsequent failing deployments.

  • Cause: AEM’s replication service user used to deploy content packages to the AEM Publish service cannot write to /var on AEM Publish. This results in the deployment of the content package to the AEM Publish service to fail.

  • Resolution: The following ways to resolve this issues are listed in the order of preference:

    1. If the /var resources are not necessary remove any resources under /var from content packages that are deployed as part of your application.
    2. If the /var resources are necessary, define the node structures using repoinit. Repoinit scripts can be targeted to AEM Author, AEM Publish or both, via OSGi runmodes.
    3. If the /var resources are only required on AEM author and cannot be reasonably modeled using repoinit, move them to a discrete content package, that is only installed on AEM Author by embedding it in the all package in an AEM Author runmode folder (<target>/apps/example-packages/content/install.author</target>).
    4. Provide appropriate ACLs to the sling-distribution-importer service user as described in this Adobe KB.

Create an Adobe Support case

If the above troubleshooting approaches do not resolve the issue, please create an Adobe Support case, via:

  • Adobe Admin Console > Support Tab > Create Case

    If you are a member of multiple Adobe Orgs, ensure the Adobe Org that has failing pipeline is selected in the Adobe Orgs switcher prior to creating the case.

recommendation-more-help
4859a77c-7971-4ac9-8f5c-4260823c6f69