Kiro doesn't send failure notifications. Start with Test the connection. The test names the cause of most configuration errors. Then use the checks on this page.
- The policy is published. In Controls, changes apply only after you choose Publish. Under Destinations overview, the export's status should start with On, such as On, over the public internet.
- The policy is assigned. Kiro applies the policy only to profiles that it is assigned to.
- Users are active. Kiro sends real-time events only when users work in Kiro. Daily exports cover only users who were active on that day.
- The first daily export has run. Kiro sends the CSV report at 02:00 UTC, and OpenTelemetry metrics at 00:30 UTC (02:00 UTC before November 26, 2026). Each covers the previous day.
| Cause | Fix |
|---|---|
| The bucket is not in the Kiro profile's Region or account | Create the bucket in the profile's Region and in the account where users are subscribed |
The bucket policy doesn't let q.amazonaws.com write | Add the bucket policy. Check that aws:SourceAccount and aws:SourceArn match your account and Region |
| The S3 URI has no prefix | Add a prefix. Kiro can't write to the root of a bucket |
Check these causes in order:
- Region or account. The secret and its AWS Key Management Service (AWS KMS) key must be in the same account and Region as your Kiro profile.
- Missing permission.
q.amazonaws.comneedssecretsmanager:GetSecretValueon the secret, andkms:Decryptandkms:DescribeKeyon the key. Granting only one of them fails the same way as granting neither. - Wrong protocol. gRPC to an HTTP-only endpoint, or HTTP/protobuf to a gRPC-only endpoint, fails with no error. Grafana Cloud, Datadog, and Amazon CloudWatch need HTTP/protobuf.
- Wrong temporality. Grafana Cloud rejects delta metrics. Choose Cumulative.
- Malformed header. Use one header in
Name=valueformat. For a base64 credential, the decoded value has exactly one:. - Amazon CloudWatch only. The metrics are in the account that issued the API key. They don't appear in the classic metrics console. Query them with PromQL.
- Datapoints too old. Before November 26, 2026, the daily metrics arrive about two hours old. Dynatrace and Grafana Cloud reject them as too old. Until then, use real-time events for these platforms, or fill the gap from the CSV report. See Check your platform's time window.
Check these causes in order:
- Region or account. The secret and its KMS key must be in the same account and Region as your Kiro profile.
- Missing permission.
q.amazonaws.comneeds a grant on both the secret and the key. - The metrics secret was used. The events secret must point to your collector's logs base URL, not to a metrics URL such as
https://otlp.datadoghq.com/v1/metrics. - Wrong path. Give the base URL. Kiro adds
/v1/logs. For Grafana Cloud, the base URL ends in/otlp. - Malformed header. Use comma-separated
Name=valuepairs, with no comma inside a value. CloudWatch Logs needs all three headers. - CloudWatch Logs only. Check that the log group and log stream exist. Bearer token authentication must be on for the log group. The API key must come from the account that owns the log group.
- Wrong Dynatrace prefix. Use
Api-Tokenwith an access token, andBearerwith a platform token.
kiro.chat.message events arrive, but kiro.tool.call events don't, or the _client CSV columns and metrics are empty.
| Cause | Fix |
|---|---|
| The client is too old | Update Kiro IDE to the latest release. Update Kiro CLI to 3.0.0 or later, or to 3.1.0 or later when Collect metrics per user is off. See Client versions |
| An old process is still running | Quit every IDE window and kiro-cli process, and start them again |
| Kiro CLI uses the V2 agent engine | Run kiro-cli settings chat.agentEngine v3 |
| The prompts didn't use tools | Send a prompt that reads a file or runs a command |
When the private connection is not ready, the test sends nothing, and the result shows Endpoint not ready yet or Connection not ready yet. The detail message tells you why.
| Detail starts with | Meaning | What to do |
|---|---|---|
| "Kiro has not set up a private connection to endpoint service …" | Kiro has no connection for this endpoint service. Usually Kiro's account is not an allowed principal. Waiting doesn't help | Ask your vendor, or the owner of your own endpoint service, to allow Kiro's account. Then publish again |
| "This profile is not using Kiro's private connection to endpoint service …" | The published policy doesn't select this endpoint service | Publish private connectivity with this endpoint service, and turn on the export |
| "Kiro's private connection to this endpoint service was set up in …" | The Region in the test differs from the published Region | Correct Endpoint service Region |
| "Kiro is still creating …", "… is still being set up", "… does not have a private DNS name yet", or "… is in its warm-up period" | The connection is being built | Test again. The connection can take up to about 10 minutes after you publish with AWS PrivateLink, or 15 minutes with Amazon VPC Lattice. If it doesn't clear, check that the endpoint service owner accepted the connection and that the service has a private DNS name |
| "Kiro's private connection to this endpoint service is in state …" | The endpoint service owner rejected the connection, or it failed. Waiting doesn't help | Ask the endpoint service owner to check Kiro's connection request |
| "Kiro has not set up a private connection to resource configuration …" | Kiro has no connection for this resource configuration | Share the resource configuration with Kiro, then publish again |
| "Kiro has not connected to resource configuration …" | Kiro didn't find your resource share | Check the share's Region, account ID, and resource configuration |
| "Resource configuration … is connected under a different collector hostname …" | The test hostname differs from the published hostname | Use the published hostname. To change it, see Behavior and limits |
| "Kiro's connection to resource configuration … failed and is being replaced" | The association failed. Waiting doesn't help | Check the resource configuration and its resource gateway, then publish again |
| "Kiro's connection to resource configuration … is active, but … is not resolvable …" | DNS for your hostname is not ready in Kiro's network yet | Test again in a minute |
| "Kiro has not built the private connection yet, so nothing was sent …" | You tested a private mode before you published | Publish, and test again when the connection is ready, up to about 10 minutes after you publish with AWS PrivateLink, or 15 minutes with Amazon VPC Lattice |
| "Kiro checked your setup and found nothing missing …" | You tested before you published. Your setup looks complete, and Kiro hasn't built the connection yet | Publish, and test again when the connection is ready, up to about 10 minutes after you publish with AWS PrivateLink, or 15 minutes with Amazon VPC Lattice |
| "Kiro's AWS account cannot see endpoint service …" | Kiro's account is not an allowed principal of the endpoint service, or the name is wrong. Waiting doesn't help | Check the name. Ask the endpoint service owner to allow Kiro's account, then publish again |
| "Kiro could not confirm that the endpoint service has a verified private DNS name …" | The endpoint service may have no private DNS name, or the name is not verified yet | Ask the endpoint service owner to set and verify the private DNS name, then publish |
| "Resource configuration … is not shared with Kiro's AWS account, or the share invitation expired …" | Kiro didn't find your AWS RAM share | Share the resource configuration with Kiro's account, then publish within 12 hours |
| "Kiro cannot tell how long your private connection has been available" | Kiro can't confirm that the endpoint's warm-up period is over | Test again. The connection can take up to about 10 minutes after you publish to become ready |
| "Kiro cannot create the private connection: … does not offer any availability zone …" | The endpoint service has no Availability Zone that Kiro can use in your Region. Waiting doesn't help | Ask the endpoint service owner to add Availability Zones to the service, then publish again |
| "… endpoint service is not available to Kiro's AWS account" | Kiro's account can't use the endpoint service. Waiting doesn't help | Ask the endpoint service owner to allow Kiro's account, then publish again |
| "Resource configuration … is not owned by this profile's AWS account" | The resource configuration belongs to another account | Create and share the resource configuration from the account that holds your Kiro profile, then publish again |
| "Collector hostname … cannot be used for private connectivity" | The hostname breaks a hostname rule, for example it has a vpce label | Choose a hostname that meets the rules in Choose the collector hostname, then publish again |
| A detail that says the Amazon VPC Lattice connection is in its warm-up period | The association is active, and Kiro is waiting before it sends | Test again. The connection can take up to about 15 minutes after you publish to become ready |
| "Testing a private connection is not available in this Region yet." | Kiro can't run a private test in this Region | Check that data arrives at your collector |
| Symptom | Cause | Fix |
|---|---|---|
| AWS PrivateLink: TLS errors, or no data | The secret uses the collector's public hostname or a vpce.amazonaws.com name | Use the endpoint service's private DNS name |
| AWS PrivateLink: never ready | Kiro's account is not allowed, the endpoint service owner didn't accept the connection, or the service has no private DNS name | Ask the endpoint service owner to check all three, then publish again |
| AWS PrivateLink: the vendor has no private DNS name | The vendor expects a private hosted zone in your VPC, which Kiro can't use | Run your own collector, share it through Amazon VPC Lattice, and forward the data to your vendor |
| AWS PrivateLink, cross-Region: never ready | The endpoint service doesn't support your profile's Region, or it is in another AWS partition | Ask the endpoint service owner to add your Region to the service |
| Amazon VPC Lattice: never ready | The share names the wrong account, is in the wrong Region, came after you published, or comes from an account that doesn't own the resource configuration | Share from the owning account to Kiro's account for your Region, then publish again |
| Amazon VPC Lattice: never ready, and you published more than 12 hours after you created the share | The share invitation expired | Add Kiro's account to the share again, and publish within 12 hours |
| Amazon VPC Lattice: never ready, and you published a correct share within 12 hours | Another Kiro customer already uses the collector hostname | Choose a hostname in your own domain. Turn off the export, publish, wait about 5 minutes, and then publish the export with the new hostname. Switch to Public internet instead only if your network rules allow it. See Behavior and limits |
| Amazon VPC Lattice: deliveries time out | The port is not 443, 4317, or 4318, or a security group blocks it | See Choose a collector port and Security groups |
| Amazon VPC Lattice: TLS errors | The certificate is self-signed, from a private certificate authority, or doesn't cover the hostname | See Use a private certificate authority |
| Amazon VPC Lattice: data stops after you change the secret | The host in the secret no longer matches Collector hostname. Kiro compares them before each delivery, and sends nothing when they differ | Use the same hostname in both. To change the hostname, see Behavior and limits |
| A load balancer in front of the collector returns errors | The protocol doesn't match the target group | For gRPC, use protocol version gRPC. For HTTP, use HTTP1 or HTTP2 |
| The test says "Kiro did not send the test. OTEL_EXPORTER_OTLP_ENDPOINT must be an https:// URL …" | The secret's endpoint starts with http://, or its host is an internal IP address or a one-word name | Use an https:// URL that your certificate covers |
Page updated: November 16, 2026
Troubleshoot user activity exports