Cloud specific logging#

This page describes how to look at various logs in different cloud providers.

Google Cloud Platform#

On GCP, by default, all logs produced by all containers and other components are sent to Google Cloud Logging. These logs are kept for 30 days, and are searchable.

Accessing the Logs Explorer#

  1. Go to the log explorer on your browser.

  2. Make sure you are in the right project, by looking at the project selector dropdown in the top bar, next to the ‘Google Cloud’ logo. If you are not in the correct project, switch to it before continuing

  3. There is a query input textbox where you can write queries using the Google Cloud Logging query language, and get output. There is also a way to explore logs by resource type, as well as time sliders. However, for most of our logs, the ‘log levels’ (error, warning, etc) are not parsed correctly, and hence are useless.

  4. Google provides a query library set of sample queries that you can access via the Library tab in Logs Explorer.

Common queries#

Kubernetes autoscaler logs#

You can find scale up or scale down events by looking for decision events

logName="projects/<project-name>/logs/" severity>=DEFAULT
jsonPayload.decision: *

Kubernetes node event logs for a particular node#


Look at hub logs#

The JupyterHub pod’s logs can be fetched with the following query:


This gives logs of all containers in the hub pod in all namespaces in the cluster. You can narrow it down to a particular namespace with:


Look at a specific user’s logs#

You can look at all user pod logs from a given namespace with:


To look at a specific user’s pod logs:


Note that you need the escaped username, rather than just the username. You can either spot it by taking a quick look at all the logs and finding out, or by using the following python code snippet:

import escapism
import string

username = "<your-username>"
escaped_username = escapism.escape(
    username, safe=set(string.ascii_lowercase + string.digits), escape_char="-"


Another super-quick shortcut is to replace any - in the username with -2d, any . with -2e and any @ with -40. If your username contains more special characters, highly recommend using the script instead - escaping errors can be frustrating!

Look at dask-gateway logs#

The following query will show logs from all the components of dask-gateway - the controller, API, proxy, individual schedulers and workers.


To look at just the scheduler and worker logs of a specific user, you can try:


You can also look by container name such as the below query for dask-worker and distributed.nanny.

resource.labels.container_name="dask-worker" OR resource.labels.container_name="distributed.nanny"

You must pass the escaped username to the query.

Full-text search across logs#

If you are looking for a specific string across all logs, you can use textPayload or jsonPayload as a field to search for depending on the log.


This is most useful when combined with any of the other queries here. For example, the following query will search across all user notebook pod logs:


Microsoft Azure#

On Azure, the logs produced by all containers and other components are sent to Microsoft Azure Monitoring, if the service is configured.

Accessing the Azure Monitoring#

Go to the Azure Monitoring, container Insights section on your browser.

Check if your cluster is in the list of monitored clusters.

If yes, then you can go the the logs section and run queries on this data.

Otherwise, it means that Azure Monitoring was not setup for your cluster and you are not able to access container logs from the portal.


Azure Monitoring is not configured on any of the 2i2c Azure clusters (i.e. the utoronto hub-cluster).

Amazon AWS#

On Amazon, the logs produced by all containers and other components are sent to Amazon Cloud Watch, if the service is configured.

Accessing the CloudWatch#

Go to the Amazon CloudWatch section on your browser.

Check if Application Insights was configured for the desired aws account and cluster.

If yes, then you can go the the logs section and run queries on the data generated by the cluster.

Otherwise, it means that Cloud Watch was not setup for your cluster and you are not able to access the logs from the console.


Amazon CloudWatch is not configured on the 2i2c Amazon clusters.