brain

Open source APM

Properties6
statusseedling
created2026-09-30
updated2026-10-07
tagsobservability, apm, monitoring, self-hosted, oss
aliasesOSS APM
public

3 min read

Open source APM

Projects on my radar for self-hosted application performance monitoring — the ” SigNoz / OpenObserve / Coroot” shortlist. Common shape: all three combine metrics + logs + traces (and more) in a single tool, all self-hostable, all positioned as open-source alternatives to Datadog/New Relic.

Curation source: Openobserve — Open Source APM Tools (vendor blog — comparative claims are theirs, not mine; captured as clippings/openobserve-ai-blog-opensource-apm-tools.md).

The radar

  • signoz (OSS) — OpenTelemetry-native observability platform: logs, metrics, traces in one tool (APM, distributed tracing, log management, infra monitoring). ClickHouse-backed. ~32k stars, TypeScript.
  • openobserve (OSS) — Rust observability platform: logs, metrics, traces, RUM; claims ~140× lower storage than Elasticsearch, single binary, petabyte scale. ~22k stars, AGPL-3.0.
  • coroot (OSS) — Metrics + logs + traces + continuous profiling with AI-powered root-cause analysis and automatic SLO monitoring; eBPF-based zero-code instrumentation. ~8k stars, Apache-2.0, Go.

Growing

  • (open: how do the three compare on resource footprint for a homelab-sized deploy — the deciding constraint at home?)
  • (open: which instruments Python/uv apps with least friction — OTel SDK vs Coroot’s eBPF auto-instrumentation?)
  • (open: does Coroot’s eBPF approach need kernel access that rootless Podman can’t give?)
  • Ties into the homelab monitoring stack (VictoriaMetrics today — see homelab-monitoring) — is APM a complement or an overlap?
  • Clipping: clippings/openobserve-ai-blog-opensource-apm-tools.md (private)
  • Related: Ansible — fleet config side of observability

The enterprise reference point: Albert Heijn’s LGTM platform (2026-10-07)

KubeBuilders → Bart Rijnders, “Observability at Albert Heijn” (36 KB, captured via r.jina.ai — Medium-WAF’d): the Dutch supermarket’s central platform for 1,800 engineers, replacing ELK, Nagios, Dynatrace and Azure Monitor with one self-hosted LGTM stack (Grafana · Loki logs · Tempo traces · Mimir metrics; Alertmanager + Opsgenie alerts).

Why it matters to this note’s shortlist question:

  • The 2023 problems were exactly fragmentation: operational overhead of disjointed tooling + high MTTR because “teams could not easily pinpoint where downstream errors originated” — the cross-team tracing case, at scale.
  • Platform properties chosen: multi-tenant “SaaS-like” self-hosted · clear ownership boundaries (platform team owns the platform; engineering teams own their dashboards/alerts/oncall — “you build, you run it” as a grassroots DevOps transformation) · single source of truth (with the honest trade-off: “a large system that, if it would fail, would fail for everyone”) · zero-ticket onboarding.
  • Why LGTM won: teams already knew Grafana; OSS + Prometheus/OTel leadership; the “big tent” — connect existing datasources so adoption can be partial; APIs/Terraform/operators for automatable onboarding and ephemeral test environments. Cost: telemetry in cheap object storage, compute on spot VMs.
  • Homelab mirror: doombox already runs the same genre (VictoriaMetrics + VictoriaLogs + Grafana, per homelab-monitoring) — Albert Heijn is that stack’s Mimir-shaped sibling at 1,800-engineer scale, with the ownership-boundary doctrine this vault’s Homelabs lane rehearsees.

Graph View

Backlinks