Summary
N-Drip was burning through $1,100 per month on AWS CloudWatch due to a high-cardinality architectural flaw. Automat-it’s FinOps team restructured their serverless monitoring pipeline, cutting CloudWatch costs by over 89% and eliminating approximately 9,500 unnecessary custom metrics, without sacrificing observability.
About N-Drip
N-Drip is an Israeli agrotech company founded in 2015 by Prof. Uri Shani, Dr. Ariel Halperin, and Ran Ben-Or.
N-Drip solutions help growers significantly reduce water and resource use while maintaining or improving productivity. N-Drip Irrigation is a gravity-powered micro-irrigation system that requires no external energy or pressure-based filtration. N-Drip Connect turns in-field data into clear irrigation insights and reliable recommendations, presented on an easy-to-use dashboard so growers can make data-driven decisions throughout the season.
The Challenge: Runaway CloudWatch Costs from a Hidden Architectural Flaw
N-Drip’s cloud infrastructure runs on AWS, with serverless Lambda functions processing data from thousands of agricultural sensors in the field. As the sensor fleet grew, so did an invisible cost problem:
- Spiraling CloudWatch bills: Monthly CloudWatch costs hit approximately $1,100, with 82% attributed to custom metric storage.
- High-cardinality metric explosion: Lambda functions were generating nearly 10,000 unique custom metrics. Every single one was billable.
- Synchronous API overhead: Every Lambda invocation made synchronous PutMetricData network calls, adding latency and cost to each execution.
- Balancing cost and visibility: N-Drip needed granular, sensor-level observability for troubleshooting and debugging in the field. Simply deleting metrics was not an option.
The Solution: FinOps-Driven Migration to AWS Embedded Metric Format
Automat-it’s FinOps team, led by Mika Shpigel (FinOps Engineer) diagnosed the root cause and provided the architectural blueprint for a migration path that preserved full observability while eliminating the cost overhead. This included:
- Root cause analysis: Automat-it pinpointed Lambda functions passing sensor_id as a CloudWatch Dimension via PutMetricData as the cardinality driver, turning every sensor into its own billable metric stream across CloudWatch.
- Migration to AWS Embedded Metric Format (EMF): Guided by Automat-it’s FinOps best practices, N-Drip’s developers refactored their Lambdas to output structured JSON logs asynchronously, with sensor_id as a plain text property rather than a CloudWatch Dimension. AWS extracts a single aggregated metric in the background. This eliminated per-sensor billing.
- CloudWatch Logs retention policy: Automat-it established a strict 7-to-14-day CloudWatch Logs retention policy, which N-Drip applied to prevent new storage costs from accumulating.
- Preserved sensor-level queries: The development team transitioned to CloudWatch Logs Insights for querying individual sensors on demand, maintaining 100% observability through a modern, cost-efficient architecture.
The Results: 89% CloudWatch Cost Reduction, Zero Custom Metric Waste
The migration delivered immediate, measurable impact across cost, performance, and architecture:
- 89% reduction in total CloudWatch costs: Monthly spend dropped from $1,103 to just $123 (a savings of approximately $980/month).
- 99.8% drop in Metric Storage costs: Plummeted from $903.56/month to a mere $1.62/month.
- ~9,500 unnecessary custom metrics eliminated: The high-cardinality dimension issue was completely resolved.
- PutMetricData API costs reduced to $0: An additional $125.03/month in API call costs was wiped out entirely.
- Improved Lambda execution performance: Removing synchronous PutMetricData network calls reduced invocation latency.
- 100% observability maintained: N-Drip’s development team retained full sensor-level debugging capability through CloudWatch Logs Insights.
Optimize your AWS cloud spend with Automat-it
Whether you’re scaling on AWS or managing production workloads, Automat-it’s combined FinOps and DevOps expertise ensures all money spent is accounted for.