Summary
- TF AWS provider bumped to
6.71.0(from6.65.0) - New API groups:
s3tables,devopsagent,shieldandoam, plus new
resources inbackup,cognitoidp,fsx,networkfirewall,opensearchandsesv2 - Fixes for resources that got stuck or never stopped updating:
ProxyEndpoint.rds,
CatalogTableOptimizer.glue,User.mq, and RDS Performance Insights KMS key references
Field paths below apply to both the cluster-scoped (*.aws.upbound.io) and namespaced
(*.aws.m.upbound.io) variants of each resource unless a version is called out.
⚠️ Bugfix API change
User.mq — pending removed from spec (#2299)
pending was listed under spec.forProvider / spec.initProvider, but it is only ever
reported by AWS. It's gone from the spec and stays available under status.atProvider.pending.
If your manifests set it, the API server simply drops it on apply.
This comes with a fix: User.mq no longer calls UpdateUser on every reconcile when
replicationUser or consoleAccess is left unset, or while a change is pending on the
broker.
⚠️ Notable resource changes
Cluster.rds, Cluster.docdb — at most 3 availabilityZones
AWS never accepted more than 3, and the provider now rejects a longer list up front. A
resource listing 4 or more zones fails to reconcile until you trim the list to 3 or fewer.
Instance.ec2 — placement changes no longer rejected
Changing placementGroup, placementGroupId or placementPartitionNumber now stops the
instance, moves it and starts it again, keeping the same instance ID. Expect a short
downtime instead of a brand-new instance. Placement changes for spot instances are still rejected.
GlobalReplicationGroup.elasticache — changing the primary now fails over
Changing primaryReplicationGroupId used to be rejected. It now
fails over to the member you named, which may be in another region. Double-check
before changing this field or retargeting its reference.
Instance.rds — multiTenant is one-way
The new multiTenant field converts an RDS for Oracle instance to the multi-tenant
configuration. This can't be undone: switching it back to false will be rejected.
Existing instances will show multiTenant: false in their spec after the first reconcile.
Harness.bedrockagentcore — check delete permissions
When AWS answers a delete with "access denied", the provider now treats the harness as
already gone. If your ProviderConfig's IAM role lacks bedrock-agentcore:DeleteHarness,
the resource disappears from Kubernetes but stays in AWS. Make sure the permission is
granted.
ProxyEndpoint.rds — fix your external name if the endpoint was stuck (#2308)
ProxyEndpoints whose name differed from their proxy's name got stuck after creation: the
provider wrote the proxy name into the crossplane.io/external-name annotation and
could never find the endpoint again. That's fixed for new resources, but existing
annotations aren't repaired automatically.
If an affected endpoint's annotation holds the proxy name, set it to the endpoint name.
Do this before allowing management policies beyond Observe, otherwise the provider
will try to create a new endpoint named after the proxy.
RDS Performance Insights KMS key references (#2269)
performanceInsightsKmsKeyIdRef / …Selector on Instance.rds, Cluster.rds and
ClusterInstance.rds resolved to the bare key ID instead of the key ARN, so the resource
was modified on every reconcile. New resolutions now use the ARN.
🆕 New resources
A new s3tables API group:
| Resource | Group |
|---|---|
Table
| s3tables
|
TableBucket
| s3tables
|
TableBucketPolicy
| s3tables
|
TableBucketReplication
| s3tables
|
TablePolicy
| s3tables
|
TableReplication
| s3tables
|
S3TablesNamespace
| s3tables
|
New devopsagent, shield and oam API groups:
| Resource | Group | PR |
|---|---|---|
AgentSpace
| devopsagent
| #2317 |
Asset
| devopsagent
| #2317 |
Protection
| shield
| #2312 |
ProtectionHealthCheckAssociation
| shield
| #2312 |
Link
| oam
| #2310 |
SinkPolicy
| oam
| #2310 |
Additions to existing groups:
| Resource | Group | PR |
|---|---|---|
LogicallyAirGappedVault
| backup
| #2314 |
LogDeliveryConfiguration
| cognitoidp
| #2317 |
OntapVolume
| fsx
| #2278 |
ContainerAssociation
| networkfirewall
| #2288 |
VPCEndpoint
| opensearch
| #2298 |
AuthorizeVPCEndpointAccess
| opensearch
| #2298 |
Tenant
| sesv2
| #2323 |
TenantResourceAssociation
| sesv2
| #2323 |
shieldresources require an active AWS Shield Advanced subscription on the account.
Link.oamandSinkPolicy.oamcomplete CloudWatch cross-account observability alongside
the existingSink.oam.
🚀 Improvements
Harness.bedrockagentcorepublishes its ARN in connection details (#2301). Set
writeConnectionSecretToRefand your workloads can read the ARN they need for
InvokeHarnessfrom the Secret, instead of looking it up by name.- Each provider package only watches its own CRDs (#2274), reducing unnecessary
watch traffic in clusters with many provider packages installed.
🐞 Bug fixes
CatalogTableOptimizer.gluecould never become ready (#2270). Every reconcile
after creation failed withid does not exist in tfstate. Fixed. The external name is
nowcatalogId,databaseName,tableName,type.ProxyEndpoint.rdsgot stuck after creation (#2308)- RDS Performance Insights KMS key references caused endless updates (#2269)
User.mqcalledUpdateUseron every reconcile (#2299)- Large numbers sent as
2.592e+06— numeric values stored as strings for Terraform
(≥ 1,000,000 or very small) were written in exponent form, which some AWS resources
can't parse. They're now written as plain numbers. - Resources configured only through
initProviderno longer show a permanent difference
on newer resources (Terraform Plugin Framework based) when AWS changes the value after
creation. --debuglogs now show controller-runtime panic stack traces (#2284).
Upstream fixes that come with the provider bump
TransitGatewayPeeringAttachment.ec2— no more forced replacement when AWS returns an
emptyoptionsblock.Instance.rds—dbNameandusernameno longer force a replacement after an Oracle
instance is converted to multi-tenant outside the provider.Scraper.amp— updates to scrapers with a CloudWatch destination no longer fail.SecretVersion.secretsmanager— no longer fails when an old version's KMS key is
disabled or deleted.Method.apigateway— fixed adeserialization failederror on read.AutoscalingGroup.autoscaling— fixed a crash whencapacityReservationTargetis empty.Harness.bedrockagentcore— no more "inconsistent result after apply" when
environmentVariablesis unset.GlobalCluster.rds— waits for the source cluster's promotion to finish, fixing a race
with dependent resources.ClusterCapacityProviders.ecs— deleting from an inactive cluster no longer fails.Pipe.pipes— updates no longer fail for self-managed Kafka sources without a root CA.Resource.cloudcontrol— retries when AWS throttles requests.
✨ New capabilities
New cross-resource references
| Resource | Field | Points at |
|---|---|---|
Function.lambda
| durableConfig.kmsKeyArn
| Key.kms
|
Backup.fsx
| volumeId
| OntapVolume.fsx
|
Newly accepted values
| Resource | Field | Value |
|---|---|---|
VPCEndpoint.ec2
| vpcEndpointType
| Tunnel
|
RuleGroup.networkfirewall
| ruleGroup.rulesSource.rulesSourceList.targetTypes
| AWS_ROUTE_53_RESOLVED_DOMAIN
|
Workspace.grafana
| grafanaVersion
| 13.2
|
WebACL.wafv2, RuleGroup.wafv2
| rule action in ruleJson
| Monetize
|
Fields that are no longer required
| Resource | Field |
|---|---|
Service.ecs
| deploymentConfiguration.lifecycleHook.hookTargetArn, …roleArn (still needed for AWS_LAMBDA hooks)
|
New fields
Paths are relative to spec.forProvider, and are mirrored on spec.initProvider and
status.atProvider unless noted. New fields are available on every served version.
| Kind.Group | New field(s) |
|---|---|
Service.ecs
| deploymentConfiguration.lifecycleHook.{targetType,timeoutConfiguration} — ECS PAUSE deployment hooks
|
WorkloadIdentity.bedrockagentcore
| tags
|
Topic.sns
| maximumMessageSize
|
AgentRuntime.bedrockagentcore
| platformVersion (V1 / V2)
|
NetworkInterface.ec2
| connectionTrackingSpecification — TCP/UDP connection tracking timeouts
|
DataSource.bedrockagent
| dataSourceConfiguration.managedKnowledgeBaseConnectorConfiguration.syncSchedule, vectorIngestionConfiguration.contextEnrichmentConfiguration
|
MetricAlarm.cloudwatch
| evaluationWindow.wallClockWindow — wall-clock evaluation windows
|
Cluster.rds
| publiclyAccessible (Multi-AZ DB clusters)
|
WebACL.wafv2, RuleGroup.wafv2
| monetizationConfig
|
Cluster.neptune
| networkType (IPV4 / DUAL)
|
Plan.backup
| status-only rule.ruleId
|
Instance.rds
| multiTenant
|
Function.lambda
| durableConfig.kmsKeyArn
|
Channel.medialive
| encoderSettings.thumbnailConfiguration
|
FirewallPolicy.networkfirewall
| firewallPolicy.route53ResolvedIpValidation
|
Model.sagemaker
| hubAccessConfig on model data S3 sources
|
ResourceConfiguration.vpclattice
| resourceConfigurationDefinition.cidrResource — for CIDR resource configurations
|
🔐 Dependency and security updates
| Dependency | v2.9.0 | v2.10.0 |
|---|---|---|
| Go toolchain | 1.27.1
| 1.27.4
|
golang.org/x/net
| v0.59.0
| v0.62.0
|
golang.org/x/crypto
| v0.57.0
| v0.59.0
|
github.com/getkin/kin-openapi
| v0.144.0
| v0.149.0
|
google.golang.org/grpc
| v1.83.2
| v1.85.0-dev
|
| Kubernetes libraries | v0.37.1
| v0.37.2
|
sigs.k8s.io/controller-runtime
| v0.25.2
| v0.25.3
|
| AWS SDK for Go v2 | v1.47.0
| v1.47.4
|
alpine runtime image
| 3.24.2
| 3.24.4
|