Modern Observability Stack

Grafana LGTM (Loki/Grafana/Tempo/Mimir), Datadog, Honeycomb, eBPF — the 2026 landscape.

Overview

The 2026 landscape

The observability market has consolidated around three commercial-grade options: the self-hosted Grafana LGTM stack, Datadog as the comprehensive SaaS, and Honeycomb as the traces-first SaaS.

Grafana LGTM stack (open-source, self-hosted)

Loki — Log aggregation, label-indexed (cheaper than Elastic)
Grafana — Visualization, dashboards, unified UI
Tempo — Distributed traces, scales horizontally
Mimir — Metrics, Prometheus-compatible, horizontally scalable
Plus — OpenTelemetry Collector for ingestion, Alertmanager for alert routing

The three options compared

Grafana LGTM — Open-source, self-hosted — Vendor-neutral, scales horizontally, cheap at high volume — Ops burden: you run it
Datadog — SaaS, comprehensive: APM, logs, metrics, traces, RUM, synthetics, database monitoring, security — Minimal setup, deep integrations, polished UI — Cost: famously expensive at scale; teams routinely face $200K+/year bills
Honeycomb — SaaS, traces-first; Charity Majors' company, built on wide structured events — Phenomenal for "why is this request slow?" via BubbleUp, which finds outlier dimensions automatically — Lighter on logs- and metrics-only use cases; event-based pricing

Other notable players

New Relic — Older, mature APM; moved to consumption-based pricing in 2020
Splunk — Enterprise log juggernaut; expensive; big in compliance and security
Dynatrace — APM-first, positioned on AI and automation
AWS CloudWatch, GCP Cloud Operations, Azure Monitor — Cloud-native and integrated, usually paired with one of the above

2024-2026 trends

eBPF-based observability — Pixie, Cilium Hubble, Parca, Pyroscope: kernel-level instrumentation without code changes; continuous profiling becoming standard
OpenTelemetry as default — Every new vendor accepts OTLP; instrumentation is portable
Cost pressure — High-cardinality metrics, logs at scale and APM traces add up; rebuilding for cost (sampling, aggregation, tiered storage) is a major theme
AI-assisted root cause — Datadog Watchdog, Honeycomb BubbleUp, New Relic AI Anomaly Detection; mostly "anomaly detection lite," but improving

How to pick

> [!TIP] Startups should pay for the time saved. At enterprise scale, cost optimization dominates.

Key concepts

LGTM stack: Loki / Grafana / Tempo / Mimir. Grafana Labs' open-source observability suite. Horizontally scalable, OTel-native.
Datadog APM: Dominant commercial vendor. All-in-one. Famously expensive at scale; $200K+/year is normal for mid-sized teams.
Honeycomb: Wide structured events instead of three pillars. Best-in-class for trace-driven debugging via BubbleUp automatic dimension analysis.
eBPF observability: Kernel-level instrumentation without code changes. Pixie, Cilium Hubble, Parca, Pyroscope. Becoming standard for continuous profiling.
OTel as default: OpenTelemetry is now the de-facto instrumentation standard. Every major vendor accepts OTLP; your code is portable.

When to use it

Choosing or migrating an observability stack — the 3-vendor decision
Cost-reduction projects — moving from Datadog to LGTM is a common path
New-service instrumentation — defaulting to OTel + LGTM = vendor-neutral starting point
Hiring — different stacks have different operational profiles

Interview questions

How do you choose between Datadog and the LGTM stack?
When does Honeycomb make sense over Datadog?
What is eBPF observability and why is it relevant?