다음을 통해 공유


Troubleshooting & diagnostics for Machine Learning Server

Important

This content is being retired and may not be updated in the future. The support for Machine Learning Server will end on July 1, 2022. For more information, see What's happening to Machine Learning Server?

Applies to: Machine Learning Server, Microsoft R Server 9.1

You can assess the health of your web and compute node environment using the diagnostic tests in the Administration Utility. This utility is installed by default with Machine Learning Server (and R Server).

Armed with this information, you can identify unresponsive components, execution problems, and access the log files.

The set of diagnostic tests includes:

  • A general health check of the configuration
  • A trace of an execution (R code, Python code, or a web service)

Additional troubleshooting topics are also covered.

Test your configuration

  1. Launch the diagnostic tests:

    • On Machine Learning Server 9.3 and later (or any other machine that has access to Machine Learning Server), launch a command-line window or terminal with administrator (Windows) or root/sudo (Linux) privileges and run this command.

      # With elevated privileges, run the following commands.
      # Authenticate with the CLI if you haven't already
      az login --mls
      
      # If running from another machine, specify the MLS endpoint
      az ml admin --mls --endpoint <Endpoint>
      
      # Run diagnostics
      az ml admin diagnostic run
      
    • On earlier versions (9.0 - 9.2):

      1. Launch the administration utility with administrator privileges (Windows) or root/sudo privileges (Linux).

      2. From menus, choose Run Diagnostic Tests, then Test configuration for a 'health report' of the configuration including a code execution test.

        If you have not authenticated yet, provide your username and password.

  2. Review the test results. If any issues arise, a raw report appears. You can also investigate the log files and attempt to resolve the issues.

  3. After making your corrections, restart the component in question. It may take a few minutes for a component to restart.

  4. Rerun the diagnostic test to make sure all is running smoothly now.

You can also get a health report directly using the status API call.

Trace a code execution

To go through the execution of a specific line of code and retrieve request IDs for debugging purposes, run a trace.

  1. Launch the code trace tests:

    • On Machine Learning Server 9.3 and later, launch a command-line window or terminal with administrator (Windows) or root/sudo (Linux) privileges and run one of these commands to trace the execution of a code block or script:

      # Authenticate with the CLI if you haven't already
      az login --mls
      
      # If running tests from another machine, specify the MLS endpoint
      az ml admin --mls --endpoint <Endpoint>
      
      # Trace execution of an inline R code block
      az ml admin diagnostic code --runtime R --block "x<-5;x"
      # Trace execution of an R script
      az ml admin diagnostic code --runtime R --file <path/to/script.R>
      # Trace execution of an inline Python code block
      az ml admin diagnostic code --runtime python --block "x=5;print(x)"
      # Trace execution of a Python script
      az ml admin diagnostic code --runtime python --file <path/to/script.py>
      
    • On earlier versions (9.0 - 9.2):

      1. Launch the administration utility with administrator privileges (Windows) or root/sudo privileges (Linux).

      2. From menus, choose Run Diagnostic Tests, then either Trace R code execution or Trace Python code execution depending on the language you are using.

      3. When prompted, enter the code you want to trace.

      4. To start the trace, press the Enter key (carriage return).

  2. Review the trace output.

Trace a web service execution

To go through the execution of a specific web service and retrieve request IDs for debugging purposes, run a trace.

  1. Launch the web service execution tests:

    • On Machine Learning Server 9.3 and later, launch a command-line window or terminal with administrator (Windows) or root/sudo (Linux) privileges and run this command specifying the web service name and version:

      # Authenticate with the CLI if you haven't already
      az login --mls
      
      # If running tests from another machine, specify the MLS endpoint
      az ml admin --mls --endpoint <Endpoint>
      
      az ml admin diagnostic service --name <service-name> --version <version>
      
    • On earlier versions (9.0 - 9.2):

      1. Launch the administration utility with administrator privileges (Windows) or root/sudo privileges (Linux).

      2. From menus, choose Run Diagnostic Tests, then Trace service execution.

        If you have not authenticated yet, you must provide your username and password.

      3. Enter the service name and version after the syntax '<service-name>/<version>' such as my-service/1.1.

      4. To start the trace, press the Enter key (carriage return).

  2. Review the trace output to better understand how the execution is running or failing.

Log files and levels

Review the log and configuration files for any component that was identified as experiencing issues. You can find the logs in the <node-install-path>\logs folder under your web and compute node installation paths. (Locate the install path for your version.)

If there are any issues, you must solve them before continuing. For extra help, consult or post questions to our forum or contact technical support.

By default, the logging level is set to Warning so as not to slow performance. However, whenever you encounter an issue that you want to share with technical support or a forum, you can change the logging level to capture more information. The following logging levels are available:

Level Description
Verbose The most detailed comprehensive logging level of all activity, which is rarely (if ever) enabled in production environments
Debug Logs robust details including internal system events, which are not necessarily observable
Information Logs system events that correspond to its responsibilities and functions
Warning Logs only when service is degraded, endangered, or may be behaving outside of its expected parameters. (Default level)
Error Logs only errors (functionality is unavailable or expectations broken)
Critical Logs only fatal events that crash the application

To update the logging level:

  1. On each compute node AND each web node, open the configuration file, <node-install-path>/appsettings.json. (Find the install path for your version.)

  2. Search for the section starting with "Logging": {

  3. Set the logging level for "Default", which captures Machine Learning Server default events. For debugging support, use the 'Debug' level.

  4. Set the logging level for "System", which captures Machine Learning Server .NET core events. For debugging support, use the 'Debug' level. Use the same value as for "Default".

  5. Save the file.

  6. Restart the node services.

  7. Repeat these changes on every compute node and every web node.
    Each node should have the same appsettings.json properties.

  8. Repeat the same operation(s) that were running when the error(s) occurred.

  9. Collect the log files from each node for debugging.

Trace user actions

Using Information Level logging, any Core API or Service Consumption API call can be logged. Additionally, the UserPrincipalName of the responsible user is also recorded in the logs. The user session is given a unique ID called LoginSessionId on successful login, which is included in subsequent log entries detailing actions (REST APIs) performed by the user during that session. LoginSessionId allows a more fine-grained association of user actions to a particular user session.

To enable information logging, update the "LogLevel" for "Default" to "Information" on the web node, using the instructions provided above.

"LogLevel": {
     "Default": "Information"
}

Now consider a user action flow in which a user logs in, creates a session, and deletes that session. Corresponding log entries for these actions might look as follows:

2018-01-23 22:21:21.770 +00:00 [Information] {"CorrelationId":"d6587885-e729-4b12-a5aa-3352b4500b0d","Subject":{"Operation":"Login","UserPrincipalName":"azureuser","RemoteIpAddress":"x.x.x.x","LoginSessionId":"A580CF7A1ED5587BDFD2E63E26103A672DE53C6AF9929F17E6311C4405950F1408F53E9451476B7B370C621FF7F6CE5E622183B4463C2CFEEA3A9838938A5EA2"}}

2018-01-23 22:24:29.812 +00:00 [Information] {"CorrelationId":"06d3f05d-5819-4c06-a366-a74d36a1c33c","Subject":{"Operation":"REQUEST POST /sessions","UserPrincipalName":"azureuser","RemoteIpAddress":"x.x.x.x","LoginSessionId":"A580CF7A1ED5587BDFD2E63E26103A672DE53C6AF9929F17E6311C4405950F1408F53E9451476B7B370C621FF7F6CE5E622183B4463C2CFEEA3A9838938A5EA2"}}
2018-01-23 22:24:29.960 +00:00 [Information] {"CorrelationId":"06d3f05d-5819-4c06-a366-a74d36a1c33c","Subject":{"Operation":"RESPONSE POST /sessions","UserPrincipalName":"azureuser","RemoteIpAddress":"x.x.x.x","StatusCode":201}}

2018-01-23 22:28:33.661 +00:00 [Information] {"CorrelationId":"47e20e55-e5ca-4414-bd84-e3e0dd7b01cc","Subject":{"Operation":"REQUEST DELETE /sessions/fc3222d7-09bd-4a89-a959-380f1e639340/force","UserPrincipalName":"azureuser","RemoteIpAddress":"x.x.x.x","LoginSessionId":"A580CF7A1ED5587BDFD2E63E26103A672DE53C6AF9929F17E6311C4405950F1408F53E9451476B7B370C621FF7F6CE5E622183B4463C2CFEEA3A9838938A5EA2"}}
2018-01-23 22:28:34.818 +00:00 [Information] {"CorrelationId":null,"Subject":{"Operation":"RESPONSE DELETE /sessions/fc3222d7-09bd-4a89-a959-380f1e639340/force","UserPrincipalName":"azureuser","RemoteIpAddress":"x.x.x.x","StatusCode":200}}

Correlating the above logs using LoginSessionId value, you can determine that the user "azureuser" logged in, created a session, and then deleted that session during the time period range from 2018-01-23 22:21 to 2018-01-23 22:28. We can also obtain other information like the machine IP address from which "azureuser" performed these actions (RemoteIpAddress) and whether the requests succeeded or failed (StatusCode). In the second entry, notice that the Request and Response for each user action can be correlated using the CorrelationId.

Troubleshooting

This section contains pointers to help you troubleshoot some problems that can occur.

Important

  1. In addition to the info below, review the issues listed in the Known Issues article as well.
  2. If this section does not help solve your issue, file a ticket with technical support or post in our forum.

"BackEndConfiguration is missing URI" Error

If you get an BackEndConfiguration is missing URIs error when trying to install a web node, then ensure your compute nodes are installed and declared prior to installing web nodes.

Unhandled Exception: System.Reflection.TargetInvocationException: Exception has been thrown by the target of an invocation. ---> Microsoft.DeployR.Server.App.Common.Exceptions.ConfigurationException: BackEndConfiguration is missing URIs
   at Microsoft.DeployR.Server.WebAPI.Extensions.ServicesExtensions.AddDomainServices(IServiceCollection serviceCollection, IHostingEnvironment env, IConfigurationRoot configurationRoot, String LogPath)
   --- End of inner exception stack trace ---

“Cannot establish connection with the web node” Error

If you get the Cannot establish connection with the web node error, then the client is unable to establish a connection with the web node in order to log in. Perform the following steps:

If the issue persists, verify you can post to the login API using curl, fiddler, or something similar. Then, share this information with technical support or post it in our forum.

Long delays when consuming web service on Spark

If you encounter long delays when trying to consume a web service created with mrsdeploy functions in a Spark compute context, you may need to add some missing folders. The Spark application belongs to a user called 'rserve2' whenever it is invoked from a web service using mrsdeploy functions.

To work around this issue, create these required folders for user 'rserve2' in local and hdfs:

hadoop fs -mkdir /user/RevoShare/rserve2
hadoop fs -chmod 777 /user/RevoShare/rserve2
 
mkdir /var/RevoShare/rserve2
chmod 777 /var/RevoShare/rserve2

Next, create a new Spark compute context:

rxSparkConnect(reset = TRUE)

When 'reset = TRUE', all cached Spark Data Frames are freed and all existing Spark applications belonging to the current user are shut down.

Compute Node Failed / HTTP status 503 on APIs (9.0.1 - Linux Only)

If you get an HTTP status 503 (Service Unavailable) response when using the Rest APIs or encounter a failure for the compute node during diagnostic testing, then one or more of the symlinks needed by deployr-rserve are missing. deployr-rserve is the R execution component for the compute node,

  1. Launch a command window with administrator privileges with root/sudo privileges.

  2. Run a diagnostic test of the system on the machine hosting the compute node.

  3. If the test reveals that the compute node test has failed, type pgrep -u rserve2 at a command prompt.

  4. If no process ID is returned, then the R execution component is not running and we need to check which symlinks are missing.

  5. At the command prompt, type tail -f /opt/deployr/9.0.1/rserve/R/log. The symlinks are revealed.

  6. Compare these symlinks to the symlinks listed in the configuration article.

  7. Add a few symlinks using the commands in the configuration article.

  8. Restart the compute node services.

  9. Run the diagnostic test or try the APIs again.

Unauthorized / HTTP status 401

If you configured Machine Learning Server to authenticate against LDAP/AD, and you experience connection issues or a 401 error, verify the LDAP settings you declared in appsettings.json. Use the ldp.exe tool to search the correct LDAP settings and compare them to what you have declared. You can also consult with any Active Directory experts in your organization to identify the correct parameters.

Configuration did not restore after upgrade

If you followed the upgrade instructions but your configuration did not persist, then put the backup of the appsettings.json file under the following directories and reinstall Machine Learning Server again:

  • On Windows: C:\Users\Default\AppData\Local\DeployR\current

  • On Linux: /etc/deployr/current

Alphanumeric error message when consuming service

If you get an error similar to Message: b55088c4-e563-459a-8c41-dd2c625e891d when consuming a web service, search compute node's log file for the alphanumeric error code to read the full error message.

Failed code execution with “ServiceBusyException” in the log

If a code execution fails and returns a ServiceBusyException error in the Web node log file, then a proxy issue may be blocking the execution.

The workaround is to:

  1. Open the R initialization file <install folder>\R_SERVER\etc\Rprofile.site.

  2. Add the following code as a new line in Rprofile.site:

    utils::setInternet2(TRUE)
    
  3. Save the file and restart Machine Learning Server.

  4. Repeat on every machine on which Machine Learning Server is installed.

  5. Run the diagnostic test or code execution again.

Error: "Microsoft.AspNetCore.Server.Kestrel.Internal.Networking.UvException: Error -97 EAFNOSUPPORT address family not supported"

Applies to: Machine Learning Server

If the following error occurs when attempting to start a web node "Microsoft.AspNetCore.Server.Kestrel.Internal.Networking.UvException: Error -97 EAFNOSUPPORT address family not supported", then disable IPv6 on your operating system.

az: error: argument command_package: invalid choice: ml

Applies to: Machine Learning Server 9.3 or later

If for some reason your azure-ml-admin-cli extension is not available or has been removed you will be met with the following error:

> az ml admin --help

az: error: argument _command_package: invalid choice: ml
usage: az [-h] [--verbose] [--debug] [--output {tsv,table,json,jsonc}]
          [--query JMESPATH]
          {aks,backup,redis,network,cosmosdb,batch,iot,dla,group,webapp,acr,dls,
           storage,mysql,vm,reservations,account,keyvault,sql,vmss,eventgrid,
           managedapp,ad,advisor,postgres,container,policy,lab,batchai,
           functionapp,identity,role,cognitiveservices,monitor,sf,resource,cdn,
           tag,feedback,snapshot,disk,extension,acs,provider,cloud,lock,image,
           find,billing,appservice,login,consumption,feature,logout,configure,
           interactive}

If you encounter this error, you can re-add the extension as such:

Windows:

> az extension add --source 'C:\Program Files\Microsoft\ML Server\Setup\azure_ml_admin_cli-0.0.1-py2.py3-none-any.whl' --yes

Linux:

> az extension add --source /opt/microsoft/mlserver/9.4.7/o16n/azure_ml_admin_cli-0.0.1-py2.py3-none-any.whl --yes