Home › AWS › AWS data services › IoT Events and IoT Analytics
AWS data services · Ingestion and CDC · about 11 min read
AWS IoT Events and AWS IoT Analytics (end of support; what to use instead)
Two retired AWS IoT services: a device state-machine detector and a device-data analytics store. Both reached end of support; this page shows replacements.
In plain words
What it is
This page covers two AWS services that you may still meet in older designs, blog posts and exam material, but that you can no longer use. AWS IoT Events let you draw a state machine, called a detector model, for each device. A press could be in Normal, Warning or Overheated, and rules moved it between states when readings crossed limits for long enough, then sent an alert. AWS IoT Analytics was a managed pipeline for device data: a channel took raw messages, a pipeline cleaned them, a data store kept them, and a data set ran SQL or a notebook over them on a schedule.
Both reached their end. IoT Analytics stopped taking new customers on 24 July 2024 and reached end of support on 15 December 2025. IoT Events stopped taking new customers on 20 May 2025 and reached end of support on 20 May 2026. After those dates the consoles and resources are no longer accessible, so any data left inside them is gone unless it was exported first. Several of AWS's own documentation pages for the services have since been taken down as well.
Take an example. A tyre plant used IoT Events to raise an alert when a curing press stayed above its pressure limit for more than 30 seconds, and IoT Analytics to keep a year of press readings for a weekly SQL report. Today the same plant sends readings through SiteWise Edge to Kinesis Data Streams. A small Flink job keeps a 30-second window per press and writes the alert to DynamoDB and an on-call topic. A Databricks streaming table lands the same readings in bronze, and the weekly report runs on a gold table. That is the reference platform's design, and nothing in it needs either retired service.
Two shops on the high street that have closed their shutters for good: one was the night watchman who rang a bell when a machine misbehaved, the other was the archive clerk who kept and summarised the machine logbooks; the watchman's job now goes to a live monitoring desk and the clerk's to the main library.
Big picture
Where it fits
The lit lane is where IoT Events and IoT Analytics works. Boxes on the left feed it; boxes on the right use what it produces.
On the reference platform, IoT data moves in one direction. Plant gateways running SiteWise Edge on Greengrass, and later IoT Core for connected products, put readings on Kinesis Data Streams, the stream belt. Two readers share it. The fast lane is Managed Service for Apache Flink with DynamoDB for state and latest values, pushed to screens through API Gateway WebSocket or AppSync. The platform lane is a Databricks streaming table that writes bronze, silver and gold every 15 minutes. Firehose can also write buffered files to S3 for sources that only need minutes.
IoT Events used to sit in the fast lane and IoT Analytics in the platform lane. The site's streaming guide already lists IoT Events as Drop (retired) among the live processor options, and the streaming options table lists it as end of support. Their jobs map onto services that are already on the platform. That is why this page is mostly about what to use instead and how to move.
If you inherit a design that still draws either box, treat it as a gap to close, not a component to operate. Check where the alerts went, which reports read the data set, and whether any history was exported to S3 before the cut-off.
How it works
The ideas you need
| Idea | What it means |
|---|---|
| Detector model | The IoT Events state machine: states, the conditions that move a device between them, timers and actions such as sending to SNS or Lambda. |
| Detector instance | One live copy of a detector model per key, usually per device, holding that device's current state and variables. |
| Channel, pipeline, data store, data set | The four IoT Analytics parts: receive raw messages, clean and enrich them, keep them, and query them on a schedule. |
| End of support | The date after which AWS no longer runs the service for anyone; the console and resources become inaccessible, not just unsupported. |
| Stateful stream processing | Code that remembers something between events, such as how long a press has been over its limit. Flink keeps that state safely with checkpoints. |
| Live lane and platform lane | The site's two readers of one stream: seconds-level alerts and screens in the live lane, and tables in the lakehouse every 15 minutes in the platform lane. |
| SiteWise alarms | Alarms on SiteWise asset properties that were built on IoT Events; AWS's migration guide moves them to CloudWatch alarms. |
Choosing
When to use it, and when not
Use it when
- Never for new work: both services are past end of support and cannot be used.
- Read this page when you inherit an architecture diagram or runbook that still shows IoT Events or IoT Analytics, so you can name the replacement for each box.
- Read it when a stateful device rule, such as over the limit for 30 seconds, must be rebuilt in Flink, Lambda or Databricks.
- Read it when old exam material or a vendor proposal still recommends either service, so you can push back with the dates.
Avoid it when
- Do not plan a migration that keeps any part running; there is nothing left to keep.
- Do not rebuild the IoT Analytics pipeline shape one to one in new services; a lakehouse with bronze, silver and gold already does the job with fewer parts.
- Do not move every detector model into Lambda code by default; long windows and many devices need real state, which is Flink's job.
Production grade
What goes wrong in production
| Problem | What breaks | What to do |
|---|---|---|
| Data lost at the cut-off | History left inside an IoT Analytics data store or detector state left in IoT Events became inaccessible after end of support. | Check whether exports to S3 were taken before the dates. If not, rebuild history from the plant historian or the raw stream archive, and write the gap into the data contract. |
| Hidden alert paths | An IoT Events action sent alerts to SNS, Lambda or email, and nobody noticed when they stopped after 20 May 2026. | List every on-call alert by name and test each one end to end in the replacement. Add a heartbeat alert that fires if no alerts or readings arrive for too long. |
| State machines rewritten as stateless code | Detector logic such as for 30 seconds is moved to a Lambda that sees one batch at a time, so alerts are missed or doubled. | Use Flink keyed state and event-time windows for duration rules. Lambda reading Kinesis is fine for simple threshold alerts; AWS's own migration example keeps per-device state in DynamoDB. |
| Timer behaviour differs | IoT Events timers fired on wall-clock time; a Flink job uses event time and watermarks, so late or missing readings change when alerts fire. | Decide per rule whether silence itself is an alert, add processing-time timers for no-data rules, and replay a recorded day through both designs to compare alerts. |
| SiteWise alarms orphaned | SiteWise asset alarms ran on IoT Events, so they stopped being evaluated when it was retired. | Follow the AWS guide to move SiteWise property alarms to CloudWatch alarms, or move the rule into the Flink live lane. |
| Reports that read IoT Analytics data sets | Dashboards and notebooks pointed at data set contents in S3 or QuickSight, and broke silently. | Repoint them to gold tables in Unity Catalog, or to Athena over the same S3 data, and add a freshness check. |
| Copying the old pipeline shape | Teams rebuild channel, pipeline and store as three new services and end up with more moving parts than before. | Land raw readings once in bronze, clean in silver, aggregate in gold. One lander, one run log. |
Alternatives
What to compare it with
| Alternative | Pick the alternative when | Pick IoT Events and IoT Analytics when |
|---|---|---|
| Managed Service for Apache Flink plus DynamoDB (replaces IoT Events) | You need duration rules, windows, joins between machines, or exactly-once alerts within seconds. | The rule is a simple threshold on one reading and nobody on the team will own a Flink job. |
| Lambda on Kinesis with DynamoDB state and EventBridge schedules (AWS's example migration) | A handful of simple per-device rules; it is the pattern AWS's detector migration guide describes. | Many devices with long windows; batch-at-a-time code makes state and timers hard to get right. |
| Kinesis or Firehose to S3, then Databricks (replaces IoT Analytics) | Device history and reports belong next to the rest of the plant data, governed in Unity Catalog. | There is no lakehouse at all and one small team only needs ad hoc queries. |
| Firehose to S3, then Athena (lighter IoT Analytics replacement) | A small AWS-only setup needs SQL over device files with no cluster. | Data must join with SAP, MES and quality data in governed tables. |
| Azure Stream Analytics or Fabric Activator, and Azure Data Explorer | The platform runs on Azure; they cover detectors and device analytics there. | You are on AWS; keep one cloud. |
| Databricks Structured Streaming with state | The team wants one engine and seconds are not needed, or real-time mode is adopted later. | Seconds-level alerts are needed today; the site keeps Flink in the live lane for now. |
Cost
How you pay
Neither service bills any more, because neither runs. The cost question is what the replacements cost. Flink is billed per Kinesis Processing Unit hour while the application runs, DynamoDB by reads, writes and storage or by provisioned capacity, Lambda by requests and duration, Kinesis by shard hours or by data volume in on-demand mode, Firehose by data ingested and any format conversion, and Databricks by DBUs plus the cloud compute. Athena is billed by data scanned. List prices vary by Region, so check the pricing pages for yours.
What drives the bill
- How many Flink processing units the live lane needs, which depends on devices, rules and state size.
- DynamoDB write volume, which is far lower if the job writes only changes of state rather than every reading.
- Kinesis shards or on-demand volume, shared with every other reader of the stream.
- How often the Databricks platform-lane job runs and how large its compute is.
Cost traps
- Rebuilding each detector as an always-on Flink application instead of one application that runs many rules keyed by device.
- Writing every reading to DynamoDB when only state changes and latest values are needed.
- Keeping a separate raw copy in Firehose files and in bronze for the same stream with no reader of the files.
- Forgetting that a Lambda polling Kinesis uses read throughput that other readers also need.
Prices change; check the AWS pricing page for your region before deciding.
Security and governance
Keeping it safe
- Give the Flink application and any Lambda functions their own IAM roles, each limited to the one stream they read and the one DynamoDB table and SNS topic they write.
- Encrypt Kinesis, DynamoDB and S3 with the platform's KMS key; Flink application state and snapshots are encrypted by the service.
- Run Flink and Lambda inside the VPC with VPC endpoints for Kinesis, DynamoDB, S3 and STS, so alert traffic never needs a public route.
- Record every change to alert rules in Git and every deployment in CloudTrail, so an auditor can see who changed a threshold and when.
- If old IoT Analytics exports in S3 hold operator IDs or other personal data, scan them with Macie and tokenise before loading into bronze.
Operating it
What to watch, and the limits
| Watch | Why |
|---|---|
| Flink uptime and lastCheckpointDuration | Whether the detector job is running and checkpointing; a failing checkpoint means state could be lost on restart. |
| Kinesis GetRecords.IteratorAgeMilliseconds | How far behind the live lane is; alerts are only as fresh as this number. |
| Alerts sent per hour per rule | A rule that suddenly sends zero or ten times its normal count needs a look; this is the check IoT Events users often lacked. |
| DynamoDB ThrottledRequests | Throttled writes mean state or latest values are falling behind. |
| Bronze freshness per plant | Time since the last reading landed in the lakehouse; it replaces the old IoT Analytics data set run status. |
| Limit | Value and notes |
|---|---|
| IoT Analytics new customers | Closed from 24 Jul 2024 |
| IoT Analytics end of support | 15 Dec 2025; console and resources no longer accessible |
| IoT Events new customers | Closed from 20 May 2025 |
| IoT Events end of support | 20 May 2026; console and resources no longer accessible |
Limits and quotas checked 2026-11-11; many can be raised through Service Quotas.
Decisions
Situations and verdicts
| Situation | Choice | Verdict |
|---|---|---|
| Inherited design with IoT Events detectors for press alerts | Rebuild the rules in the Flink live lane with DynamoDB state | Pursue: Flink is already the site's live processor; one application holds all rules keyed by press. |
| Three simple threshold alerts on one small line | Lambda reading Kinesis | Pursue with care: cheap and simple, but move to Flink as soon as a rule needs a duration or window. |
| Weekly device reports that ran on IoT Analytics data sets | Gold tables in Databricks fed by the stream | Pursue: the reports join with production and quality data that already live in the lakehouse. |
| Keep any part of IoT Events or IoT Analytics running | Either retired service | Drop: both are past end of support and cannot run. |
FAQ
Questions people ask
Is AWS IoT Events discontinued?
Yes. AWS IoT Events stopped taking new customers on 20 May 2025 and reached end of support on 20 May 2026. The console and resources are no longer accessible.
When did AWS IoT Analytics end?
AWS IoT Analytics closed to new customers on 24 July 2024 and reached end of support on 15 December 2025. AWS's migration guide pointed to Kinesis Data Streams, S3, Glue and Athena as replacements.
What replaces IoT Events detector models?
AWS's migration example uses IoT Core rules into Kinesis Data Streams, a Lambda function that evaluates the state logic, a DynamoDB table for device state and EventBridge schedules for timers. On this site's platform the same job goes to Managed Service for Apache Flink, which handles windows and state more safely at scale.
What happened to SiteWise alarms?
SiteWise alarms were built on IoT Events. AWS's migration guide moves alarms on SiteWise properties to CloudWatch alarms, with SiteWise MQTT notifications as another route.
Does Databricks replace IoT Analytics?
For analytics, yes. Land readings from Kinesis or S3 in a streaming table, then clean and aggregate them in silver and gold. You get history, SQL, notebooks and governance in one place.
Sources
Checked against
- AWS docs: IoT Analytics end of support
- AWS docs: IoT Analytics migration guide
- AWS docs: IoT Events end of support
- AWS docs: Migration procedure for IoT Events detector models
- AWS docs: Migration procedure for SiteWise alarms in IoT Events
Checked 2026-11-11.
Related
Where it connects
Streaming on AWS and Databricks · Streaming options at a glance · Real-time analytics · Amazon Managed Service for Apache Flink · AWS IoT SiteWise and SiteWise Edge · Amazon Timestream · All AWS data services
← IoT Core · IoT Greengrass →