Leegin Bernads

DevOps Engineer

Groww

Bangalore

05.2021 - Current

Designed and architected a scalable, fault-tolerant observability stack using OpenTelemetry, Prometheus, and Grafana Mimir to handle over 120 Million active time series. Implemented robust guardrails on metrics and labels to prevent cardinality explosion, ensuring high performance, availability, and operational efficiency.
Led the design and management of a distributed tracing system, leveraging OpenTelemetry Agent and Collector, Refinery, and Grafana Tempo. Implemented head and tail-based sampling strategies to capture high-value traces, significantly improving issue detection and debugging efficiency. Enabled identification of 'unknown unknowns' and reduced Mean Time to Recovery (MTTR) by approximately 40%.
Architected and deployed a production-grade centralized logging system using Promtail and Grafana Loki, ingesting over 6 TB of logs daily. Implemented log-level-based segregation using regex matching and Promtail pipelines to enrich logs with Kubernetes metadata for efficient querying. Introduced log sampling based on log levels to optimize storage and enhance the signal-to-noise ratio, contributing to faster incident resolution and a measurable reduction in MTTR.
Developed a high-cardinality metric analysis service to track metric usage across multi-tenant environments. The service evaluates whether metrics are referenced in Grafana dashboards, alerting/recording rules, or queried within the last 30 days. This enabled the automatic classification of metrics as used or unused, improving observability hygiene, reducing noise, and optimizing resource utilization.
Architected and deployed a proprietary event-driven tool, named Heimdall, to enforce comprehensive labeling compliance policies for GCP resources across the organization. This solution effectively addressed the company's challenge of billing segregation. As a result of its success, plans are underway to release Heimdall as an open-source tool for wider adoption within the GCP cloud user community.
Implemented event-driven autoscaling using KEDA for applications, significantly enhancing scalability capabilities, while simultaneously reducing infrastructure costs.