The cheapest log service is not necessarily the one with the lowest price per gigabyte. One vendor meters the data you send, another charges for what you keep searchable, and a third also bills for the data each query scans. Start with a recent incident: what did the on-call engineer search for, how far back did they need to look, and who needed access? Those answers tell you what to price.
| Tool | Start here when | Check before you buy |
|---|---|---|
| Grafana Cloud Logs | You want managed Loki and already work in Grafana or can learn LogQL. | The Free plan includes 50 GB ingested per month and 14 days of retention. Pro starts at $19 per month, includes 50 GB of log ingest and retains logs for 30 days. Check the separate rates for processing, writing and retaining logs above the included usage. |
| Amazon CloudWatch Logs | Your application already writes to AWS log groups and responders work in the AWS console. | AWS lists a 5 GB Free Tier for logs, covering ingestion, archive storage and Logs Insights data scanned. Above the included usage, estimate each meter in your AWS Region, not just ingestion. |
| Datadog Log Management | Your team already uses Datadog to investigate application and infrastructure incidents. | Count indexed events by retention policy, not just ingest bytes. Check your monthly index commitment, what happens above it, and which excluded events stay available outside indexed search. |
| Better Stack | A small team wants log search and incident response in the same product. | The free tier includes 3 GB of logs per month for three days. Price pay-as-you-go ingestion and retention in your region, or check the bundle limits. Do not count every telemetry reader as a paid responder. |
Compare each bill on the same workload
Grafana Cloud: The logs pricing section separates matching the retention window from staying under the ingest allowance. A team sending 40 GB per month can fit inside the Free ingest cap, but cannot use that plan to search a log from 20 days ago. Compare the $19 Pro starting point if that older search is required. Before upgrading, run a real filter in Loki against data labelled by service and environment and check whether the team can find the affected deployment without scanning unrelated streams.
CloudWatch Logs: AWS CloudWatch pricing distinguishes ingestion, stored logs and Logs Insights scans. Start from the actual log groups for one service. A query across a short incident window and a query across a month of multiple groups can scan very different amounts. Check the billed region and which log class your workload uses before using the pricing calculator. Set a retention policy on the trial log groups rather than assuming that deleting a dashboard deletes its underlying logs.
Datadog: Log Management Billing describes a monthly commitment on indexed log events and a 50% premium on on-demand usage above that commitment. Index documentation describes retention, daily quotas and filters. A filtered-out event does not become searchable in that index just because it reached Datadog; excluded logs can still flow through Live Tail or to an archive. Test whether the logs you would exclude include the one event needed to explain a failed request. Check ingest and archive costs separately in the terms for your account.
Better Stack: Better Stack pricing lists U.S. pay-as-you-go logs at $0.15 per GB ingested and $0.08 per GB per month retained. Its U.S. monthly Nano bundle is $45 with a 40 GB log allowance and 30-day log retention, alongside separate trace and metric allowances. Compare the bundle and usage-based options before assuming that you need the bundle for logs alone. Telemetry team members are listed at $0; a responder with incident-management access is a different role.
Separate ingestion, search and archive
Write down three quantities from the same sample: bytes sent, events kept in the searchable store and days those events remain searchable. Then estimate query or restore costs at the rate the engineer will use during an incident. Stored bytes may differ from ingested bytes, and daily traffic can vary. Do not calculate thirty days of storage by multiplying a single day's raw input by thirty unless the vendor meters that number.
For example, suppose an app sends 40 GB over a month and the team needs 30 searchable days to investigate late customer reports. Grafana Free misses the retention requirement even though the volume fits. Better Stack's free tier misses both. The U.S. Nano allowance fits the stated log volume, but compare it with the region-specific pay-as-you-go rate and confirm how it handles overflow. With CloudWatch, record the storage estimate and sample query scans for your Region. With Datadog, count the subset of those events that must remain indexed. These are ways to estimate the same requirement, not a claim that one vendor is cheapest for every app.
Fill out a cost estimate you can revisit
Make a worksheet with two columns, one for the current setup and one for the candidate. Keep the same service, environment, region and retention requirements in both. Enter the ingest volume and indexed event count separately, note whether either quantity is quoted in decimal GB or another unit, and enter the longest retention you actually need in interactive search. If some logs go to object storage after that window, use a separate archive row. Record the date and billing period next to every rate so you can tell a price change from a workload change later.
For the 40 GB workload, do not add Better Stack's U.S. Nano price to the U.S. usage-based rate as if the same logs incur both charges. Estimate the bundle, then separately estimate pay-as-you-go with the measured billable ingest and retained volume. On AWS, run the candidate Logs Insights query over a narrow time range and record bytes scanned; multiply by the expected executions only for that query pattern. A dashboard that reruns the query during an incident needs its own estimate. If you do not have a measurement yet, mark the amount unknown instead of setting the estimate to zero.
Give the worksheet a normal month and a busy month. For the busy case, use a past deployment that raised log volume and model its observed duration. If Datadog's indexed events would cross the commitment or a daily quota, price the extra indexed events under the account's terms and check which events responders would lose if a quota cut in. The engineering lead can then approve a real spend range and choose when to revisit it, rather than relying on a one-day sample.
Decide which logs to keep before adding retention
Split the sample by job: operational errors for on-call, routine success events for trends, and any records your organization must retain. Keep the evidence needed to reconstruct a failure in the searchable path. Before adding a filter, pick one error and confirm it is still possible to join it to the deployment or request that produced it. For routine high-volume events, compare using a metric or a representative sample; do not treat that sample as complete evidence for an individual customer incident.
Make ownership explicit: the service owner decides which fields enter the collector, the on-call lead verifies the alert and search path, and the budget owner reviews the usage report after the first billing cycle. If an archive is required, have a permitted teammate retrieve a test record and note the steps, time and expected charges. Only treat the archive as a substitute for interactive retention if that recovery path meets the team's use case.
Run one incident trial before migrating
- Select a sanitized sample from one deployment. Include a failed request, its preceding warning, a status event and enough routine traffic to test the filters. Do not copy tokens, cookies or raw customer identifiers to a trial account.
- Have a second engineer trace the failed request using what support would actually provide: an approximate time, service and request ID. Check whether they can find the deployment and the corresponding error without broader data access.
- Repeat the search against an older sample and a noisy sample. Check which events are searchable, which are archive-only, what the queries scan and which limits might be exceeded in a busy week.
- Record the ingest, retention, query and responder costs separately. Add engineering work for collectors and alert updates. If the current setup answers the trial without a cost or retention problem, keep it and fix the slow or expensive part first.
If the incident repeatedly returns as an application defect, use the bug tracking comparison to decide how to hand it off. If the gap is what happened in the browser before the request, compare session replay tools separately rather than keeping more backend logs by default.