4

i'm aware of the existence of this thread:
where are the individual dataproc spark logs?

However if i ssh connect to a worker node vm and navigate to the /tmp folder this is all i see: ls /tmp

Is anyone able to pinpoint me to the exact location?

also for some reason i can't navigate directly from UI to stdout/stderr of the single task as it says that i'm unable to reach the site whenever i try to access the logs from the link in the UI

Dagang
  • 24,586
  • 26
  • 88
  • 133
nonoDa
  • 413
  • 2
  • 16
  • Are you looking for Spark driver log or executor logs? How did you configure logs when creating the cluster? Did you use the default config? – Dagang Sep 15 '21 at 23:01
  • i'm talking about specific executor logs, i used the default config – nonoDa Sep 16 '21 at 07:32

2 Answers2

3

The previous answer looks to be outdated.

If you are talking about the container logs, then:

  1. On clusters with a 1.5 or newer image, Yarn log aggregation is enabled by default and the remote log directory is set to be the temp bucket for the cluster. You can look the location up under /etc/hadoop/conf/yarn-site.xml, and the configuration is yarn.nodemanager.remote-app-log-dir.
  2. On clusters with a 1.4 or older image, log aggregation is not enabled by default, so the container logs will be under /var/log/hadoop-yarn/userlogs on the worker nodes where the containers were run.
cyxxy
  • 608
  • 3
  • 8
2

In 1.5 or newer versions, dataproc:yarn.log-aggregation.enabled is set to true by default. Under the hood, the yarn.log-aggregation-enable property in /etc/hadoop/conf/yarn-site.xml is set to true, and the container logs are controlled by the yarn.nodemanager.remote-app-log-dir property which is set to gs://<cluster-tmp-bucket>/<cluster-uuid>/yarn-logs by default. Check this doc for more details on Dataproc tmp bucket.

In addition to dumping the logs at the location, there are several other ways to view the logs:

  1. YARN CLI: If the cluster has not been deleted, SSH into the master node, then run yarn logs -applicationId <app-id>. If you are not sure about the app ID, run yarn application -list -appStates ALL to list all apps. This method works only when log aggregation is enabled.

  2. YARN Application Timeline server: If you enabled Component Gateway and the cluster has not been deleted, open the cluster's "YARN Application Timeline" link in the "WEB INTERFACES" tab of the cluster's web UI, find the application attemp and its containers, click the "Logs" link. This method works only when log aggregation is enabled.

  3. Cloud Logging: YARN container logs are available in Cloud Logging even after the cluster is deleted.

    3.1) When dataproc:dataproc.logging.stackdriver.job.yarn.container.enable if false (which is the default) or the job is submitted through CLI e.g., spark-submit instead of Dataproc jobs API , it is under the projects/<project-id>/logs/yarn-userlogs log name of the cluster resource:

    resource.type="cloud_dataproc_cluster"
resource.labels.cluster_name=<cluster-name>
resource.labels.cluster_uuid=<cluster-uuid>
log_name="projects/<project-id>/logs/yarn-userlogs"

    3.2) When dataproc:dataproc.logging.stackdriver.job.yarn.container.enable if true, it is under the projects/<project-id>/logs/dataproc.job.yarn.container log name of the job resource:

    resource.type="cloud_dataproc_job"
resource.labels.job_id=<job_id>
resource.labels.job_uuid=<job_uuid>
log_name="projects/<project-id>/logs/dataproc.job.yarn.container"

In Dataproc 1.4 (deprecated) or older versions, the yarn.log-aggregation-enable property in /etc/hadoop/conf/yarn-site.xml is set to fasle by default, and the container logs are controlled by the yarn.nodemanager.log-dirs property which is set to /var/log/hadoop-yarn/userlogs by default.

Dagang
  • 24,586
  • 26
  • 88
  • 133