DevOps pros rethink cloud cost with continuous delivery tool

DevOps puts cloud cost on dev to-do list

Tyler Technologies has used Harness pipelines since 2017 to continuously deploy and automatically roll back greenfield applications that run on Kubernetes clusters in the AWS cloud. The full lifecycle of these applications is managed by developers, who previously didn't have direct visibility into how their apps used cloud resources, or experience with cloud cost management. CE bridged that gap without requiring developers to manage a separate tool or manually tag resources for tracking.

This has already prompted developers at Tyler Technologies to focus more on cost efficiencies as they plan applications, Green said.

"That wasn't something they really thought about before," he said. "Until very recently, we followed a more traditional model where we had dedicated operations people that ran our data centers, and they were the ones that were responsible for optimizing and tuning."

While developer visibility into apps can be helpful, a tool such as CE doesn't replace other cloud cost management platforms used by company executives and corporate finance departments.

"It's good for developers to be cognizant of costs and not feel like they're being blindsided by impossible mandates from a perspective they don't understand," said Charles Betz, analyst at Forrester Research. "But in large enterprises, there will still be dedicated folks managing cloud costs at scale."

The Harness CD tool deploys delegates, or software agents, to each Kubernetes cluster to carry out and monitor app deployments. CE can use those agents to identify the resources that specific apps and microservices use and compare this information to resource allocations in developers' Kubernetes manifests, identifying idle and unallocated resources.

If users don't have the Harness CD tool, CE draws on information from Kubernetes autoscaling data and associates it with specific microservices and applications. In either case, developers don't have to manually tag resources, which many other cloud cost tools require.

This was a plus for Tyler Technologies, but Betz also expressed concern about the reliability of auto-discovery. 

"There's no way to map objective tech resources to subjective business concepts without some false negatives or positives that could result in the wrong executive being charged for the wrong workload," Betz said. "Tagging is a discipline that organizations ultimately can't really get away from."

Harness roadmap includes cloud cost guidance

Tyler Technologies plans to add the CE product to Harness when it renews its license this year but hasn't yet received a specific pricing quote for the tool. Harness officials declined to disclose specific pricing numbers but said that CE will have a tiered model that charges between 1% and 5% of customers' overall cloud spending, depending on whether the cloud infrastructure is clustered or non-clustered.

"It's not quite free money -- there is a charge for this service," Green said. "But it will allow us to save costs we wouldn't even be aware of otherwise."

It will allow us to save costs we wouldn't even be aware of otherwise.

Harness plans to add recommendation features to CE in a late July release, which will give developer teams hints about how to improve cloud cost efficiency. In its initial release, developers must correct inefficiencies themselves, which Tyler's Green said would be easier with recommendations. 

"We use an AWS tool that recommends savings plans and how to revise instances for cost savings," Green said. "We'd like to see that as part of the Harness tool as well."

Other Harness users that previewed CE, such as Choice Hotels, have said they'd also like to see the tool add proactive cloud cost analysis, but Green said his team uses CE in staging environments to generate such estimates ahead of production deployments.

Harness plans to add predictive cost estimates based on what resources are provisioned for deployments, a company spokesperson said. The Continuous Efficiency platform already forecasts cloud costs for apps and clusters, and later releases will predict usage based on seasonality and trends.