Home › AWS › AWS data services › Managed Grafana
AWS data services · BI and dashboards · about 9 min read
Amazon Managed Grafana
Open-source Grafana run by AWS: live time-series boards for engineers over CloudWatch, Prometheus and fast stores.
In plain words
What it is
Grafana is a popular open-source tool for drawing live charts of numbers that change over time. Engineers use it to watch servers, queues, sensors and pipelines. A chart in Grafana is called a panel, a page of panels is a dashboard, and each panel asks a data source a question every few seconds. Grafana does not store data itself. It reads from other systems, such as Amazon CloudWatch, Prometheus, a time-series database or a SQL database.
Amazon Managed Grafana is Grafana that AWS runs for you. You create a workspace, which is your own Grafana server, and AWS handles the servers, scaling, patching and upgrades. People sign in with the company identity provider through IAM Identity Center or SAML. AWS can set up permissions so the workspace reads CloudWatch, X-Ray, IoT SiteWise, Timestream, OpenSearch and Amazon Managed Service for Prometheus without stored passwords.
A concrete example: a tyre plant runs a fast lane that turns curing-press sensor readings into alerts within seconds. The on-call data engineer wants one screen that shows, for every plant, how far behind the Kinesis readers are, whether the Flink job is checkpointing, how many DynamoDB writes are throttled and the latest press temperatures. A Managed Grafana dashboard reads CloudWatch for the first three and the live store for the last. It refreshes every ten seconds on a screen in the operations room.
Grafana is not a business BI tool. It is good at many lines over time and at alert rules on those lines. It is weak at the things a sales or finance user wants, such as pivot tables, drill-down through a product hierarchy, row-level rules per customer and emailed reports. Keep that line clear and the tool stays useful.
The gauges on a machine's control panel: quick to read, always live, and useless for writing the monthly report.
Big picture
Where it fits
The lit lane is where Managed Grafana works. Boxes on the left feed it; boxes on the right use what it produces.
Managed Grafana sits in the Use lane, but its audience is engineers, not business users. On the reference platform it is optional. The streaming page lists it as 'Pursue for engineers' time-series boards' on the live store, and the real-time page names Grafana as a cheap way to refresh an operations board every one to five seconds.
It reads two kinds of source. The first is CloudWatch, which watches the AWS movers: DMS lag, AppFlow failures, SiteWise ingestion and Kinesis reader lag. The second is the fast lane's live store, such as DynamoDB through a plugin, Timestream for InfluxDB, or a fast analytic store if you run one. It can also query a Databricks SQL warehouse with the Databricks data source plugin, but that keeps a warehouse awake and is rarely worth it for second-level refresh.
The main operations dashboard on the reference platform is built in Databricks AI/BI over the prod_ops tables, because those tables join AWS mover health with Databricks job runs, quality results and cost. Grafana does not replace that view. It adds a live, seconds-level board for the people who run the streaming lane, where waiting five minutes for the metric stream is too slow.
How it works
The ideas you need
| Idea | What it means |
|---|---|
| Workspace | Your own logically isolated Grafana server inside Managed Grafana, with its own URL, users and data sources. |
| Data source | A connection Grafana queries, such as CloudWatch, Prometheus, OpenSearch or a SQL database. Grafana stores no data itself. |
| Panel and dashboard | A panel is one chart or table; a dashboard is a page of panels that share a time range and refresh rate. |
| Refresh interval | How often each panel re-runs its query. Short intervals look live but multiply load and cost on the source. |
| Grafana alerting | Rules that evaluate a query on a schedule and send a notification when a condition holds for a set time. |
| Editor and viewer | User roles. Managed Grafana bills per active user, and editors or admins cost more than viewers. |
| Service account | A non-human identity with a token, used by automation such as Terraform to create dashboards and data sources. |
| Plugin | An add-on that gives Grafana a new data source or panel type. Managed Grafana supports a curated set and lets admins install more from the catalog. |
Choosing
When to use it, and when not
Use it when
- Engineers need a live board of CloudWatch metrics across many AWS accounts and regions on one screen.
- You run Prometheus, for example on EKS, and want dashboards and alerts without running Grafana yourself.
- The fast lane needs a seconds-level operations view of reader lag, checkpoint time and throttles, refreshed every few seconds.
- You want IoT SiteWise or Timestream data on a time-series chart for plant engineers.
- You already know Grafana and want sign-in through IAM Identity Center and no server to patch.
Avoid it when
- The audience is business users who want sales, margin or scrap reports with drill-down and emailed snapshots; use the main BI tool.
- The main question is history over months in gold tables; Databricks AI/BI or Power BI over the SQL warehouse fits better.
- Only a handful of CloudWatch charts are needed, which CloudWatch dashboards already give you without another tool.
- You would point second-level refresh panels at a Databricks SQL warehouse, which keeps it running all day for a wall screen.
Production grade
What goes wrong in production
| Problem | What breaks | What to do |
|---|---|---|
| Refresh storms | Twenty people open a dashboard with thirty panels on a five-second refresh, and CloudWatch GetMetricData calls and costs jump. | Set a minimum refresh interval per workspace, use longer intervals for history panels, and share one wall screen instead of many open tabs. |
| Second alert system | Grafana alerts and CloudWatch alarms both fire for the same Kinesis lag, with different thresholds and channels, and on-call ignores both. | Pick one owner per signal. Keep paging alarms in CloudWatch with SNS, and use Grafana alerts only for signals that live only in Grafana's sources. |
| Version upgrades | Moving from Grafana 10.4 to 12.4 removes legacy alerting, AngularJS plugins and API keys, and some panels break. | Test the upgrade on a copy of the workspace, migrate API keys to service accounts, and replace AngularJS panels before you upgrade production. |
| Dashboards edited by hand | Someone fixes a panel in the browser, and the next deploy overwrites it, or the change is never reviewed. | Keep dashboards as JSON in Git, deploy them with Terraform or the API through a service account, and give most people viewer rights. |
| Cross-account data access | The workspace in the ops account cannot see CloudWatch metrics in the plant or data accounts. | Use CloudWatch cross-account observability or assume-role data sources, created by the permission provisioning feature or by infrastructure code. |
| Quota surprise | A workspace allows 100 alert rules and 2,000 dashboards, and neither can be raised. | Group alerts by labels instead of one rule per plant, and split workspaces by audience when you approach the limits. |
Alternatives
What to compare it with
| Alternative | Pick the alternative when | Pick Managed Grafana when |
|---|---|---|
| CloudWatch dashboards | You need a few AWS metric charts and alarms and want no extra tool or licence. | You need many sources, rich templating, or one board across Prometheus, CloudWatch and a database. |
| Self-hosted Grafana | You need plugins Managed Grafana does not allow, or full control of the version. | You do not want to run, patch and secure a Grafana server and its database. |
| Azure Managed Grafana | You are on Azure and watch Azure Monitor and Azure Data Explorer. | You are on AWS; the AWS service has native IAM and CloudWatch wiring. |
| Grafana Cloud | You want Grafana's own hosted stack with logs, metrics and traces storage included, across clouds. | You want everything inside your AWS account boundary and on the AWS bill. |
| Databricks AI/BI dashboards | The question joins ops data with job runs, quality and cost in prod_ops tables, refreshed every few minutes. | You need second-level refresh on a wall screen over streaming metrics. |
| Quick Sight | Business users need filters, drill-down, embedding and email reports. | Engineers need live time-series and alert rules. |
Cost
How you pay
Managed Grafana is priced per active user per month in each workspace. Editors and admins are priced higher than viewers, and a user counts only in a month they sign in. Some enterprise plugins need an extra Grafana Enterprise licence. The bigger cost is often on the data sources: every panel refresh is a query to CloudWatch, Prometheus or a database, and those calls are billed by those services. Check the Managed Grafana pricing page for current prices in your region.
What drives the bill
- Number of active editors and admins each month.
- Number of active viewers each month.
- CloudWatch GetMetricData and Logs Insights queries from panels, which grow with panels, refresh rate and open screens.
- Optional Grafana Enterprise plugin licences.
Cost traps
- Giving everyone editor rights by default, which bills them at the higher rate.
- A wall screen with a five-second refresh over many CloudWatch metrics runs all month and drives CloudWatch API cost.
- Logs Insights panels scan log data on every refresh and are billed by data scanned.
- Panels over a Databricks SQL warehouse stop it from auto-stopping.
Prices change; check the AWS pricing page for your region before deciding.
Security and governance
Keeping it safe
- Sign in through IAM Identity Center or SAML from the company identity provider; map identity groups to admin, editor and viewer roles.
- Use the workspace IAM role, with read-only policies scoped to the accounts and metrics needed, instead of stored keys for AWS data sources.
- Turn on network access control so only the company network or VPC endpoints reach the workspace, and use VPC connectivity for data sources in private subnets.
- Use service accounts with short-lived tokens for automation; API keys are not supported after the Grafana 12 upgrade.
- Workspace API calls are recorded in CloudTrail; include workspace admins in the quarterly access review.
- Keep personal data off operations boards. Plant metrics and lag do not need names; if a panel reads the live store, read aggregates, not records.
Operating it
What to watch, and the limits
| Watch | Why |
|---|---|
| Workspace status | Whether the workspace is ACTIVE, or in an UPDATING or FAILED state after an upgrade. |
| Data source errors | Panels showing query errors or timeouts, usually expired roles or blocked network paths. |
| Alert rule evaluation failures | Grafana alert rules that fail to evaluate and so never fire. |
| Active users by role | Editors and viewers per month, which is what you pay for. |
| CloudWatch API usage from Grafana | GetMetricData calls attributed to the workspace role in CloudTrail and the bill. |
| Limit | Value and notes |
|---|---|
| Workspaces | 5 per Region per account by default (adjustable) |
| Alert rules | 100 per workspace (not adjustable) |
| Dashboards and data sources | 2,000 each per workspace |
| Users | 10,000 provisioned, 500 concurrent per workspace |
| Service account tokens | 100 per workspace, including expired tokens |
Limits and quotas checked 2026-11-11; many can be raised through Service Quotas.
Decisions
Situations and verdicts
| Situation | Choice | Verdict |
|---|---|---|
| Operations-room screen for the streaming fast lane across six plants | Managed Grafana over CloudWatch metrics for Kinesis, Flink and DynamoDB, refreshed every 10 to 30 seconds | Pursue with care: right tool for live engineering boards, if refresh rates and editors are kept in check. |
| Daily operations dashboard of failed jobs, quality and cost | Databricks AI/BI over prod_ops tables | Drop Grafana here: the join of AWS and Databricks evidence lives in prod_ops. |
| Paging when DMS lag passes 15 minutes | CloudWatch alarm with SNS to the on-call pager | Pursue: keep paging in one place; Grafana can show the same metric without owning the alert. |
| Team already runs self-hosted Grafana on EC2 | Move to Managed Grafana unless a needed plugin is not allowed | Keep if you have it: move when the next major version upgrade is due. |
FAQ
Questions people ask
Is Amazon Managed Grafana a replacement for QuickSight or Power BI?
No. Grafana is for live time-series boards for engineers. Business dashboards with drill-down, row rules per customer and emailed reports belong in Power BI, Databricks AI/BI or Quick Sight.
Which Grafana version does Amazon Managed Grafana run?
New workspaces can use Grafana 12.4 since April 2026, and existing 10.4 workspaces can be upgraded in place since May 2026. Version 12 drops legacy alerting, API keys and AngularJS plugins, so test before upgrading.
Can Grafana query Databricks?
Yes, through a Databricks data source plugin that queries a SQL warehouse. Use it for slow-changing panels only, because a fast refresh keeps the warehouse running and adds cost.
How is Amazon Managed Grafana priced?
Per active user per month in each workspace, with editors and admins priced higher than viewers. The queries that panels send to CloudWatch and other sources are billed separately by those services.
Should alerts live in Grafana or CloudWatch?
On this platform, paging alarms for AWS movers live in CloudWatch with SNS. Use Grafana alerts only for signals that exist only in Grafana's sources, so each signal has one owner.
Sources
Checked against
- AWS docs: What is Amazon Managed Grafana
- AWS docs: Managed Grafana service quotas
- AWS What's New: Grafana 12.4 workspaces (April 2026)
- AWS What's New: in-place upgrade to Grafana 12.4 (May 2026)
- AWS: Managed Grafana pricing
Checked 2026-11-11.
Related
Where it connects
Streaming on AWS and Databricks · Real-time analytics · Amazon CloudWatch · Quick Sight · AWS and Databricks reference platform · All AWS data services
← Timestream · Quick Sight →