What would you use for ECS logs and APM on a $4k monthly budget?

Before comparing vendors, get a realistic baseline for data volume. Measure logs per day, trace volume, retention, and how many services actually need full tracing. The task count alone does not tell you what the bill will be.

Then price the same workload across a few managed services, including ingestion, retention, APM, and any per-container or per-host charges. Also test whether sampling traces and filtering noisy logs can keep the flash-sale month from turning into a nasty surprise.

If low maintenance is the main constraint, I would not treat a self-hosted stack as automatically cheaper. ClickHouse-based observability or a Grafana and OpenTelemetry setup can be cost-effective, but someone still owns upgrades, storage, alerting, and keeping the dashboards usable. With two people, that operational time is part of the price.

A managed ClickHouse-based service could be worth a trial, but check its default log verbosity and retention behavior. A noisy default configuration can erase the savings quickly. I would pilot it with a representative service before committing the whole fleet.

Look at Elastic-based managed offerings as well as the big APM vendors. Some price mainly by data volume rather than charging separately for every task, which may suit this setup. Compare the complete bill, though, including the data tier, retention, and any limits on metrics or tracing.

For the developer experience, try the actual workflows your team needs: finding a request across logs and traces, filtering by service, and sharing a useful view. A flexible interface is only helpful if developers can use it without asking your two DevOps engineers to build every query for them.

Consider splitting the job instead of forcing one product to do everything. A focused error-monitoring and APM tool may work for the core services, while a lower-cost log platform handles the other tasks. That can fit a tight budget, but it adds another interface and makes correlation between logs and traces something you need to test carefully.

There are also workable open-source routes, including Grafana-based stacks, but the trade is engineering time for license savings. Given your team size, I would only go that way if someone can clearly own maintenance and the developers are comfortable learning the workflow.