Lemonade plans to put tools from eBPF startup Groundcover in production this year in a bid to reduce instrumentation work and other overhead for its small DevOps team.
The DevOps team at a New York-based insurer will soon place a big bet on a startup's Kubernetes monitoring tools, which could partially replace Datadog because of an approach to data gathering that requires no instrumentation.
Lemonade Inc., founded in 2015, sells renter, homeowner, car, pet and term life insurance through a digital platform that uses automation and analytics to keep its labor costs low, even as it has grown to more than 1 million paying customers. This approach extends to the company's DevOps team, primarily located in Israel, which supports some 200 developers with five DevOps engineers.
Observability data is key to driving the automation required to maintain this ratio while ensuring resilient systems for about 100 application services, said Avital Tamir, DevOps lead at Lemonade. Most of that data has been gathered through Datadog's software, which covers the IT infrastructure as well as application performance management (APM).
The company grew over the last five years at a compound annual growth rate of 150% and from 23 employees at its founding to more than 1,000. So did the cost of observability in terms of software licensing and developer time for manual instrumentation, Tamir said.
"The billing for Datadog just kept growing and growing. And looking a few years ahead, we kind of figured we need to start doing something that's a little more cost effective," Tamir said. "Not only that but also to open up the possibility of storing metrics for long periods of time … with high cardinality and without going bankrupt."
Tamir didn't specify how much spending on Datadog had been projected to grow long term. But he did say its licensing costs were expected to triple this year compared with last year. That put the goal of storing more data for longer to feed historical analytics engines largely out of reach.
Groundcover's dashboard puts eBPF data from Kubernetes into context and prioritizes alerts.
From drawings on napkins to APM tools in production
Tamir joined Lemonade two years ago. A few months into his tenure there, he heard about an early-stage company called Groundcover, based in Tel Aviv. Groundcover's founders were working on using a Linux kernel-based utility, Extended Berkeley Packet Filter (eBPF), to gather application performance data.
This is not unique. Demand for eBPF support to monitor and manage Kubernetes applications, from the network infrastructure where open source Cilium has made headway to service meshes and observability tools, has been generating growing buzz in the industry since 2020. Existing APM vendors such as Datadog and New Relic can also gather data via eBPF and automate the instrumentation of apps for eBPF observability through an open source utility named Pixie.
Just the fact that there's no query language you need to learn -- just these nice screens that take you through the work of showing you where things went wrong -- is very impressive to me.
Avital TamirDevOps lead, Lemonade Inc.
However, while Pixie makes instrumenting applications easier, what intrigued Tamir about Groundcover's planned APM tools was their promise not to require any instrumentation of Kubernetes applications, automated or otherwise, to monitor them.
Instead, Groundcover installs an agent via a Daemonset on Kubernetes clusters and uses information flowing out of the Kubernetes API to put that data in context for each application. Groundcover's agent also performs some analytics at the cluster level before sending data to its central repository. This reduces traffic on the network, trims the amount of data that must be stored for long-term analysis and prioritizes alerts according to criticality.
Lemonade now uses Groundcover's dashboards displayed on TV screens to show developers data about their apps in test and development environments. Not having to instrument applications separately for each programming language has allowed Lemonade's DevOps team to roll it out, update it and feed data to its packaged dashboards for these environments quickly and easily so far, Tamir said.
"Just the fact that there's no query language you need to learn -- just these nice screens that take you through the work of showing you where things went wrong -- is very impressive to me," he said.
Groundcover, founded in 2021, raised $24.5 million in seed and Series A funding in September. Its primary focus to date is on Kubernetes applications. Datadog covers the full IT stack from application to cloud infrastructure as well as IT security analytics.
Lemonade has yet to use the tool in production but plans to do so this year, according to Tamir.
"We've been watching [Groundcover] grow from drawings on napkins … and it feels like it's just getting started," he said. "I really would like to see it mature over the next year into a product that can actually rival some of the big observability players. But they still have some stuff they don't do yet, like custom metrics or observing cloud resources like AWS."
Still, if all goes well with a production rollout this year, Groundcover will begin to replace Datadog, starting with APM, Tamir said.
"We're getting a head start," he said. "Redesigning the stack is mostly us looking five years ahead. ... And the key idea there is not [for Groundcover] to be the best data analytics platform but a really great troubleshooting platform that helps you surface problems and show you what went wrong before you even know what to look for."
Beth Pariseau, senior news writer at TechTarget, is an award-winning veteran of IT journalism. She can be reached at [email protected] or on Twitter @PariseauTT.
