Case studies

AWS platform and fleet monitoring for a commercial coffee roaster manufacturer

Cut monitoring costs by about 80 percent by moving from Logz.io to managed Grafana, then automated equipment onboarding and fleet updates.

Client and industry

Bellwether Coffee

Commercial coffee roaster manufacturing, United States

The problem

The client sells connected roasters that run in cafés and shops. The team needed to know when a machine in the field misbehaved, ship software to the fleet without bricking devices, and run the cloud side across several environments without hand-built infrastructure.

Roaster and cloud application logs went to Logz.io, which cost about $10,000 a month at the usage the client had reached. Much of that volume was debug logging nobody read.

What we did

  • One managed Grafana for roasters and cloud. We brought data from the roasters and the cloud application into a managed Grafana instance, with a custom Prometheus data source for the device metrics, and retired Logz.io.
  • Cleaner logs. We removed debug logs and other noise at the source, so only data the team uses is stored and paid for.
  • Dashboards and alerts. Grafana dashboards and alert rules cover the fleet and the cloud services side by side. Amazon CloudWatch metrics are exported in Prometheus format and shipped with the OpenTelemetry Collector.
  • AWS platform as code. Separate engineering, staging and production accounts, built from shared Terraform modules with Terragrunt: networking, Amazon ECS services, container registries, build pipelines and encrypted secrets.
  • Telemetry from the fleet. Roasters send logs and events through AWS IoT Core and Amazon Kinesis to AWS Lambda functions, which store them and forward them for search. A fleet data-health job flags machines that stop reporting.
  • Pipelines for new equipment. Adding a new type of equipment is automated through pipelines, so onboarding no longer relies on manual steps.
  • Safe device updates. A pipeline builds the roaster operating system image and versioned update bundles. Ansible playbooks install updates over the air and roll a machine back if a step fails.
  • Further projects. Once monitoring was in place, we took on more work for the client, including authentication.

Stack used

Grafana, Prometheus, Terraform, Terragrunt, AWS IoT Core, Amazon Kinesis, AWS Lambda, Amazon ECS, OpenTelemetry, Ansible and Python.

Duration

2 years

Outcome

Monthly monitoring cost fell from about $10,000 to about $2,000.

Moving from Logz.io to a managed Grafana instance with a custom Prometheus data source, and removing debug logs that were not needed, cut the monthly bill by roughly 80 percent.

Start with an Observability Audit

In one to two weeks we review your monitoring, logs and alerts and tell you what to fix first.

Request an Observability Audit