The cloud-based CRM company Pipedrive has been relentlessly modernising its observability stack, first adopting Grafana visualisation and Grafana Mimir for Prometheus metrics, then recently completed a migration of its distributed tracing from a third-party SaaS provider to OpenTelemetry and Grafana Tempo, and its logging stack from Graylog to Grafana Loki.

Along the way, the team developed its own in-house library to include OpenTelemetry in its roughly 750 microservices.

Observability Platform Team Lead Karl-Martin Karlson shares Pipedrive’s journey migrating to the LGTM Stack in an infrastructure spanning 8 data centres, 5 physical locations, and over 20k Kubernetes pods.

Chapters

0:00 Introduction to Karl-Martin, Pipedrive, and Pipedrive's infrastructure

1:39 Pipedrive's observability stack before LGTM and OpenTelemetry

3:32 The challenges that Pipedrive faced with their previous stack

5:47 Pipedrive's distributed tracing use cases

6:28 What led Pipedrive to choosing OpenTelemetry and the LGTM Stack

7:14 The timeline to get it all done

8:50 Enabling OpenTelemetry

9:19 Pipedrive's architecture

10:34 Implementation challenges

11:46 Recorded metrics vs. raw metrics

13:26 Label cardinality: Understanding log and metric labels

14:00 How we made the data useful (visualizations!)

18:09 LGTM infrastructure by the numbers

19:19 Pipedrive's future plans with Grafana