I work for two companies: one has a large budget for Datadog, while the other is a small organization with only three engineers. We currently run Prometheus, Grafana, Alertmanager, and OpenSearch ourselves, but the operational overhead is becoming difficult to justify. Maintaining the Kubernetes cluster, storage, upgrades, and observability components takes too much time and the costs are rising.
We would like to move to a managed, cloud-based platform that correlates metrics, logs, traces, and alerts without Datadog-level pricing. We tried SigNoz, but had a poor experience with its documentation, support, and UI. New Relic felt too fragmented, and Better Stack lacked several features we need. We are not completely opposed to self-hosting if the maintenance burden is genuinely low.
Our workloads are mostly Kubernetes, with some legacy services running on EC2 and virtual machines at another provider. A particularly important requirement is reliable host-down alerting. With push-based OpenTelemetry systems, we have struggled to generate an alert that identifies exactly which host stopped reporting, unlike the straightforward pull-based behavior of Prometheus, Zabbix, and Nagios. What managed platforms or low-maintenance deployments would you recommend?
5 Answers
For host-down detection, an OpenTelemetry collector receiving data cannot automatically know whether a missing host is dead, disconnected, or simply delayed unless you add an explicit heartbeat or inventory signal. Keep a small pull-based probe or synthetic check for infrastructure availability, and use the telemetry platform for application signals. Alternatively, emit a regular heartbeat per host and alert on absence grouped by the host identifier, making sure the hostname or instance ID is preserved as a stable attribute. This hybrid approach usually works better than trying to force a purely push-based pipeline to behave like Prometheus.
VictoriaMetrics Cloud is worth evaluating if cost and Prometheus compatibility matter more than having the most polished interface. It can preserve much of the existing metrics workflow and is generally cheaper than the larger full-suite vendors. It will not necessarily provide the same level of built-in correlation or automatic instrumentation as Datadog, so compare the tracing, logging, and alerting features carefully.
Grafana Cloud is probably the least disruptive move. You can keep Prometheus-compatible dashboards and queries, send metrics with remote_write, and use managed log and trace components alongside them. It removes most of the infrastructure maintenance without forcing the team to learn an entirely new system. The main thing to watch is pricing: Kubernetes cardinality can grow very quickly, especially with node, pod, container, and kube-state-metrics data. Audit labels, scrape intervals, and unused series before committing.
Also compare Dynatrace, Splunk Observability Cloud, Honeycomb, Coralogix, and Dash0. The right choice depends heavily on whether your biggest requirement is logs, traces, automatic instrumentation, or metrics cost. Coralogix can be attractive when log volume dominates, while Dynatrace and Datadog tend to be stronger for agent-based discovery and correlation. Get actual pricing from your expected ingestion, retention, host count, and active-series volume instead of relying on headline prices.
If the team already operates Kubernetes and is comfortable with Helm, the LGTM stack can be a reasonable self-managed option. It provides a broad set of metrics, logs, traces, and profiling tools, but it is not a zero-maintenance solution. Running storage, upgrades, retention, backups, and capacity planning still requires engineering time. For a three-person team that wants to stop babysitting observability, managed Grafana services may be a better tradeoff.
That is our situation exactly: we already run a dedicated Kubernetes cluster for observability, but compute, storage, and upgrade work are becoming too expensive and time-consuming. We are looking for something mature enough for production rather than another platform to experiment with.