Overwatch was built to enable Databricks' customers, employees, and partners to quickly / easily understand operations within Databricks deployments. As enterprise adoption increases there’s an ever-growing need for strong governance. Overwatch means to enable users to quickly answer questions and then drill down to make effective operational changes. Common examples of operational activities Overwatch assists with are:
Overwatch is currently a Databricks Labs project and has no direct cost associated with it other than the costs incurred with running the daily jobs. Audit logging is required and as such a Azure Databricks Premium SKU OR the equivalent AWS Premium Plan or above. As a reference, a cost analysis was performed at a large Databricks customer. This customer had >1000 named users with >400 daily active users with a contract price with Databricks over $2MM/year. Overwatch costs them approximately $8 USD / day to run which includes storage, compute, and DBUs. $8 / day == approximately $3,000 / year or 0.15% of Databricks contract price. This is just a single reference customer but cost monitoring for your Overwatch job should be the same as any other job you run. At the time of this writing Overwatch has only been deployed as a pilot and has already proven several million dollars in value. The savings customers found through these efficiency improvements allowed them to accelerate their backlogs, permanently lower the total cost of ownership, and even take on additional exploratory efforts.
Overwatch has two primary modes, Historical & Realtime
Overwatch amalgamates and unifies all the logs produced by Spark and Databricks via a periodic job run (typically 1x/day). The Overwatch job then enriches this data through various API calls to the Databricks platform and, in some cases, the cloud provider. Finally, all the data is pulled together into a data model that enables simplified queries for powerful insights and governance. Below are a few example reports produced by Overwatch. The batch / historical data is typically used to identify the proper boundaries that can later be used to set up monitoring / alerts in the real-time modules.
|Hot Notebooks||Heavy User Breakdown|
Overwatch does have a configurable, near-realtime component that allows users to retrieve metrics at a configurable frequency (i.e. 5s/120s/etc). This is enabled through DropWizard (a default spark monitoring platform) and a time-series database (TSDB) called “whisper” which is wrapped by Graphite. Graphite offers many exposed APIs upon which any front-end TSDB reporting tool can be used, such as Grafana. Graphite and Grafana are used by default as they do not require a license for basic use and Grafana is very common in the industry. Very powerful monitoring dashboards along with configured alerting can be enabled through Grafana.
The realtime architecture has been created but 1-click delivery has not yet been enabled. Currently, realtime integration is up to the customer to integrate. For guidance / assistance in configuring the Overwatch real-time, please reach out to your Databricks account team to set up an discovery meeting.
|Simple Spark Dashboard||IO Dashboard||Advanced Dashboard|