Monitoring and APM tools for startups on AWS in 2026: how each works, setup effort and cost

Datadog: is it worth it for a small team, and does it support OpenTelemetry?

Datadog is the most complete product here and the easiest for a new engineer to find their way around, but the bill grows on logs and custom metrics. It does support OpenTelemetry, through its own Agent, a standard collector or direct OTLP.

How it works. The Datadog Agent collects host and container metrics, receives traces from Datadog's tracing libraries or OpenTelemetry, and ships logs. On ECS Fargate the docs say the Agent "needs to be run as an additional container within each task definition to be monitored", and logs go through a FireLens log router container. Datadog now recommends its Agent with the embedded DDOT collector for OpenTelemetry users, and also takes OTLP from a standard collector or directly.

Effort. Fast to a first dashboard on EKS or EC2, slower on Fargate because every task definition changes. The integrations catalogue is the strength: AWS, databases, queues and third-party services show up with ready-made dashboards.

Cost. On 5 November 2026: Infrastructure Pro $15 per host a month billed annually ($18 on-demand), APM $31 per host ($36), log ingestion $0.10 per GB and indexing $1.70 per million events for 15 days. The free plan covers up to five hosts with one-day metric retention. Our illustrative 10-host startup costs $1,100 a month at list price and $468 trimmed; the line-by-line maths, host counting rules, Fargate per-task billing and the startup credit program are in our Datadog pricing guide, which we won't repeat here.

Pick it when you have credits, several products to correlate (APM, RUM, synthetics, security), or engineers who already know it, and someone will watch log indexing and tag cardinality. Datadog vs New Relic, Dynatrace and Grafana is covered in each of those sections below.

Sentry: is error tracking enough before you buy an APM?

For most teams before product-market fit, yes: Sentry for errors plus CloudWatch for logs and alarms costs $0 to $26 a month in Sentry fees, and it is one of the cheapest useful tools on this page.

How it works. An SDK in your app catches exceptions with the stack trace, release and user context, and groups them into issues. It also does tracing, session replay, logs, cron and uptime checks. On Lambda you can add Sentry's layer, which Sentry calls "a quick way" with limited configuration, or bundle the SDK for full control.

Effort. An hour or two per app: install the SDK, set the DSN and release, upload source maps for a JavaScript frontend, route alerts to Slack.

Cost. Developer is free for one user with 5,000 errors. Team is $26 a month billed annually ($29 monthly) with unlimited users, 50,000 errors, 5 million spans, 50 replays and 5 GB of logs; Business is $80 ($89). Reserved errors above the Team quota cost $0.00029 each up to 100,000, falling with volume to $0.00012 above 20 million. Plans, quotas and self-hosted Sentry's real cost are in our Sentry pricing guide.

Sentry vs Datadog, in short: not either-or for most startups. Sentry is better at "which exception, since which deploy"; Datadog is better at "which host, service or query". Start with Sentry plus CloudWatch; add an APM later. The Sentry pricing guide compares Sentry vs Datadog on price and overlap. Pick Sentry's EU region at sign-up if you need it: Sentry says the location can't be changed afterwards.

Honeycomb: how does it work, and what does it cost?

Honeycomb is a tracing backend built on OpenTelemetry for teams who debug by asking new questions. It is free up to 20 million events a month, and Pro starts at $150 a month for 50 million.

How it works. You send OpenTelemetry traces with as many attributes as you like (customer, plan, feature flag), and query any combination of them fast. That is its strength over dashboards built in advance.

Effort. Standard OpenTelemetry: SDK in each service, a collector sidecar or daemon, point it at Honeycomb's endpoint (US or api.eu1.honeycomb.io for the EU instance).

Cost. Free covers 20 million events a month. Pro starts at $150 a month for 50 million events. Honeycomb's docs say "each span" counts as an event, so a request that produces 3 spans is 3 events, and that events are kept for 60 days. From 1 July 2026 the Pro rate is $3.00 per million events, against $1.30 on legacy plans; legacy customers move by 31 December 2026 unless they commit to a year at 20% over their old rate. For our example, 20 million requests at 3 spans each is 60 million events: sample to stay under 50 million and it is $150 a month. Honeycomb lists Pro volumes up to 750 million, so at the $3.00 per million rate the unsampled 60 million would be about $180, if billed linearly; confirm with Honeycomb.

Pick it when your hard problems are "why is it slow for these customers only", and you keep logs and AWS metrics in CloudWatch next to it.

Dynatrace pricing: what does it cost, and is it better than Datadog for a startup?

Dynatrace charges by host memory, $58 a month per 8 GiB host on Full-Stack Monitoring, drawn down from an annual commitment. Our illustrative startup comes to about $648 a month.

How it works. OneAgent instruments hosts and processes automatically. On Fargate an initoneagent container copies the agent into a shared volume and your app loads it with LD_PRELOAD. It also takes OpenTelemetry over OTLP.

Cost. The rate card on 5 November 2026: Full-Stack Monitoring $58 a month per 8 GiB host (billed at $0.01 per memory-GiB-hour), Infrastructure Monitoring $29 per host, Foundation & Discovery $7. Logs $0.20 per GiB to ingest, $0.0007 per GiB-day to retain and $0.0035 per GiB scanned. All of it draws down a Dynatrace Platform Subscription: a minimum annual commitment whose size the page doesn't state. The trial is 15 days.

Our example on 8 GiB hosts: 10 × $58 = $580, logs 300 × $0.20 = $60, retention about 300 GiB × 30 days × $0.0007 = $6.30, queries 500 × $0.0035 = $1.75: about $648 a month before any trace volume beyond what Full-Stack includes. Because pricing follows memory, the same ten hosts at 16 GiB cost $1,168 for the host line alone.

Datadog vs Dynatrace. Both are full platforms. Dynatrace charges by host memory and does more automatically once OneAgent is in; Datadog charges by host and product and gives you more manual control. For a startup the annual commitment is the deciding factor: Dynatrace is built for companies that sign one. We rarely recommend it below 15 engineers.

Elastic APM: when is it the right choice, and what does it cost?

Elastic is the strongest choice when searching logs is most of your debugging. Elastic Observability Serverless ingests from $0.07 per GB, so our example's logs and traces start at under $40 a month.

How it works. Elasticsearch underneath, with APM, metrics and synthetics on top. Data arrives through EDOT, Elastic's OpenTelemetry distribution, with generally available SDKs for .NET, Java, Node.js, PHP, Python, Android and iOS, and a collector built into Elastic Agent from 9.5.

Cost. Elastic Observability Serverless lists two tiers. Logs Essentials ingests from $0.07 per GB; Complete (which adds SLOs, machine learning and synthetics) from $0.09 per GB for logs and traces and $0.023 for metrics, with retention from $0.019 per GB-month. These are "from" prices. On them, our example's 300 GB of logs and 50 GB of traces on Complete come to about $31.50 of ingest plus about $6.65 a month to keep 350 GB: under $40, which is a floor, not a quote.

Pros. Log search is what Elastic does best, and Ireland and Frankfurt are among its Serverless regions. Cons. More product to learn than a startup needs if you only want alerts and traces, and a self-managed cluster is real work. Pick it when you already know Elasticsearch, or log search drives most of your debugging.

What is SigNoz, and is it a good open source APM?

SigNoz is an open source APM built on OpenTelemetry and ClickHouse, with logs, traces and metrics in one UI. You can self-host it for free or use SigNoz Cloud from $49 a month, about $131 for our example.

How it works. Logs, traces and metrics in one UI, stored in ClickHouse, with OpenTelemetry as the only way in. The self-hosted Docker setup runs ClickHouse, ClickHouse Keeper, PostgreSQL, an OTel collector listening on 4317 and 4318, and the SigNoz server, and needs "at least 4GB of memory allocated to Docker". There are install guides for Kubernetes on AWS and for ECS.

Cost. SigNoz Cloud Teams starts at $49 a month, which includes $49 of usage: logs and traces at $0.30 per GB with 15 days' retention, metrics at $0.10 per million samples. Our example: 300 GB of logs ($90), 50 GB of traces ($15) and 262.8 million metric samples ($26.28) is $131.28 a month. Startups under three years old, under 30 people and under $6 million raised can get it for $19 a month for the first 12 months. Data regions are US, EU and India. The Community Edition is free to self-host, with community support only.

SigNoz vs Grafana. Pick SigNoz when you want an open source APM with one UI instead of four Grafana components, and a cheap hosted option to start on. Check the 15-day retention against what your incident reviews and auditor need.

Better Stack and Axiom: are they enough for logs, uptime and on-call?

For uptime checks, on-call and cheap log storage, yes. Neither replaces a full APM for a team with many services.

Better Stack started as uptime monitoring and status pages, and now adds on-call, logs, traces, metrics and error tracking. Data comes through its collector, which uses "eBPF-based auto-instrumentation" with OpenTelemetry underneath, or straight over OTLP. The free plan has 10 monitors, a status page, 3 GB of logs kept for 3 days and 100,000 exceptions. Responder seats are $29 a month on yearly billing. Logs and traces cost $0.10 per GB to ingest and $0.05 per GB-month to keep in Europe, or $0.15 and $0.08 in the US. Our example's 350 GB of logs and traces comes to about $80 a month in the US region or about $53 in Europe, plus seats for whoever is on call.

Axiom is a log and event store priced for volume: Axiom Cloud is a $25 a month platform fee with 1 TB of data loading, 100 GB of storage and 100 GB-hours of query compute included; the free Personal plan loads 500 GB with 30 days' retention. Our 300 GB of logs fits inside the $25.

Pick them when you want uptime checks and on-call in one cheap tool (Better Stack), or a low-cost home for a lot of logs (Axiom), next to CloudWatch metrics and Sentry.

Is SolarWinds AppOptics end of life? Yes, since 31 January 2026.

SolarWinds' migration guide says AppOptics "reached its End of Service Life on January 31, 2026": customers can no longer monitor applications with it or get support for it, and the AppOptics documentation now says it "is no longer accessible or under active development." The replacement is SolarWinds Observability SaaS, which the same guide says is "available at the same price as AppOptics" and includes a native OpenTelemetry framework.

SolarWinds Observability SaaS starting list prices on 5 November 2026, in USD per month: $27.50 per service for Application Observability, $15.75 per node for infrastructure, $5.00 per GB a month for logs and $70 per database instance. At $5 per GB, logs make it expensive for a log-heavy startup: our example's 300 GB would be $1,500 a month on its own. If you are moving off AppOptics anyway, instrument with OpenTelemetry and price SolarWinds against the others on this page.

How do you avoid monitoring vendor lock-in, and what do SOC 2 and EU data change?

Instrument with OpenTelemetry so that changing backends is collector configuration, not code. For SOC 2, any of these tools works if alerts, retention and access are set up as evidence, and for EU or UK data, pick the region before you send the first byte.

Lock-in. What ties you to a vendor is the instrumentation in your code and the alerts and dashboards built on it, not the contract. If services emit OpenTelemetry and a collector does the routing, changing backends is collector configuration plus rebuilding alerts. If they import Datadog's or New Relic's own libraries, it is a code change in every service. Instrument with OpenTelemetry from the start unless a vendor feature you need only works with its own agent. Our OpenTelemetry on AWS guide covers ADOT, collectors and Application Signals setup.

SOC 2. Auditors don't ask which tool; they ask for evidence. Typically that means alerts that page a named person and a record of what happened next, log retention that matches your written policy, audit logs for who changed what (CloudTrail on AWS), and access to the monitoring tool limited and reviewed. Any of these tools can produce that. What matters is short retention on cheap plans (SigNoz Teams keeps logs 15 days, Better Stack's free plan 3) and the vendor's own SOC 2 report, which your auditor will ask for as a subservice organisation; we didn't check each vendor's report for this page. Our SOC 2 on AWS page and the monitoring section of the production readiness checklist cover the rest.

EU and UK data. CloudWatch keeps data in whatever region you run, including London (eu-west-2). Grafana Cloud offers Germany, Ireland, Sweden and the UK. Sentry offers Frankfurt but the choice is fixed when you create the organisation. New Relic charges $0.05 per GB more in its EU data center. Datadog has an EU1 site in Germany and a UK1 site. Elastic Serverless runs in Ireland, Frankfurt and London; Dynatrace offers Ireland, with Frankfurt and London on request. If a customer contract names a region, decide before you send the first byte: moving telemetry between regions later usually means starting a new account.