The Canyon ReviewSelf-hosted observability stacks, ranked by total weekend cost
Grafana LGTM, VictoriaMetrics, SigNoz, Zabbix and Netdata ranked for teams leaving the Datadog bill behind. Weekends counted honestly.
You want VictoriaMetrics plus Grafana if metrics are the job, and SigNoz if you need traces in the same box. The full Grafana LGTM stack is the right answer for a platform team and the wrong answer for whoever is reading this to escape a SaaS bill, and the difference is measured in weekends.
Context, since this ranking exists because of an invoice: our observability cost teardown found the SaaS tier pricing out mid-size teams with cardinality to burn. The self-hosted escape is real. It is also not free; it is prepaid, in operations. Criteria: ops burden 40%, resource appetite 25%, coverage (metrics/logs/traces) 20%, docs and community 15%. Scores illustrative, August 2026.
The table
| Rank | Stack | Covers | The weekend truth |
|---|---|---|---|
| 1 | VictoriaMetrics + Grafana | metrics | quietest ops on this list |
| 2 | SigNoz | metrics, traces, logs | one box, one upgrade path |
| 3 | Grafana LGTM (Loki/Tempo/Mimir) | everything | four systems wearing one logo |
| 4 | Zabbix | metrics, alerts | old, complete, aesthetically 2009 |
| 5 | Netdata | per-node metrics | superb node view, fleet story thinner |
Notes per rung
VictoriaMetrics wins on the metric that governs self-hosting: how often it demands attention. Single binary, Prometheus compatible, frugal with RAM at cardinalities that make Prometheus sweat, and upgrades that have never once eaten our weekend. Pair with Grafana and most teams' observability needs are done.
SigNoz is the right pick when traces are non-negotiable and you want one product instead of a federation. ClickHouse underneath means real resource appetite, but the appetite buys the all-in-one upgrade path, and the project's docs respect your time.
The LGTM stack is what we run when someone else is paid to run it. Each component is excellent; together they are four distributed systems with four scaling behaviors, and the team that adopts them casually meets each one's personality during a separate incident. Platform teams with headcount: yes. Everyone else: rungs one and two exist.
Zabbix will outlive us all. It monitors everything, alerts reliably, and looks like it. If your estate is servers rather than services, its age is mostly an asset: every problem you will have was solved in its forums a decade ago.
Netdata gives the best per-node view in open source, full stop. As the fleet story it asks for their cloud, which is a different review.
The skip
Running the full LGTM constellation for a five-person team because the blog posts made it look modular. It is modular the way an airport is modular. Rung one, a Sunday afternoon, and the Datadog invoice becomes a memory with a Grafana dashboard on top of it.