What floci-io/floci shipped
Written by FoxPlug from the repository's own releases; not affiliated with Floci. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/floci-io/floci; Floci did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of November 2, 2026
What shipped
- Added EC2 Instance Connect with SendSSHPublicKey, verified by the unmodified AWS agent reading keys from IMDS. Pull request #7473
- Bedrock now serves ListFoundationModels, GetFoundationModel, ListInferenceProfiles and GetInferenceProfile instead of UnknownOperationException. Pull request #7675
- CreateDBSnapshot and RestoreDBInstanceFromDBSnapshot now work for MySQL and MariaDB instances instead of being refused with InvalidDBInstanceState. Pull request #7747
- Security Hub BatchImportFindings and GetFindings now import and query ASFF findings per account and Region. Pull request #7692
- Step Functions now evaluates the execution role on DynamoDB, S3 putObject, RDS Data and Redshift Data tasks under IAM enforcement. Pull request #7722
- REST API Gateway now counts usage plan calls per key, reports them in GetUsage and refuses calls over quota or throttle with 429. Pull request #7665
- DocumentDB now models manual cluster snapshots and both restores with mongodump archives. Pull request #7585
- Redshift clusters now support audit logging of connection, user and user activity logs to S3. Pull request #7699
- API Gateway REST now supports create, read, list, update and delete for documentation parts. Pull request #7712
- ECR registry requests no longer stall during background garbage collection runs. Pull request #7437
Why it matters
This week adds several services developers were missing outright: EC2 Instance Connect, Bedrock foundation model and inference profile listing, MySQL and MariaDB RDS snapshots, Security Hub finding import and DocumentDB cluster snapshots. Usage plan accounting in API Gateway and execution role checks on Step Functions tasks make local behaviour closer to what teams see in real accounts. Several fixes also unblock stalled workflows, including ECR requests stalling during garbage collection and Lambda containers orphaned by earlier runs.
Changelog entry
- Feat(ec2-instance-connect): add SendSSHPublicKey, verified by the unmodified AWS agent Pull request #7473
- Feat(bedrock): list and get foundation models and system-defined inference profiles Pull request #7675
- Feat(rds): CreateDBSnapshot and RestoreDBInstanceFromDBSnapshot for MySQL and MariaDB Pull request #7747
- Feat(securityhub): import and query ASFF findings Pull request #7692
- Feat(stepfunctions): evaluate the execution role on DynamoDB, S3, RDS Data and Redshift Data tasks Pull request #7722
- Fix(apigateway): count usage plan calls per key and enforce quota and throttle Pull request #7665
- Feat(docdb): support cluster snapshots and restores Pull request #7585
- Feat(redshift): support audit logging to S3 Pull request #7699
- Feat(apigateway): add documentation part operations Pull request #7712
- Fix(ecr): prevent registry request stalls during background garbage collection Pull request #7437
- Fix(lambda): remove containers orphaned by previous runs Pull request #7734
- Fix(apigateway): filter GetUsagePlans by keyId Pull request #7655
This week in Floci: EC2 Instance Connect works with the unmodified AWS agent, Bedrock lists foundation models and inference profiles, RDS snapshots cover MySQL and MariaDB, Security Hub imports ASFF findings, and API Gateway usage plans now count per key and enforce quota.
Floci shipped a busy week. EC2 Instance Connect now answers SendSSHPublicKey and works with the unmodified AWS agent. Bedrock serves ListFoundationModels, GetFoundationModel, ListInferenceProfiles and GetInferenceProfile. RDS CreateDBSnapshot and RestoreDBInstanceFromDBSnapshot now work for MySQL and MariaDB. Security Hub imports and queries ASFF findings. Step Functions checks the execution role on DynamoDB, S3, RDS Data and Redshift Data tasks. API Gateway counts usage plan calls per key and enforces quota and throttle, and now supports documentation parts. DocumentDB does cluster snapshots and restores, Redshift does audit logging to S3, and ECR registry requests no longer stall during garbage collection.
Week of October 26, 2026
What shipped
- Added the EventBridge V2 service with routing to SQS consumer subscribers, reworked filtering to support the $or clause, and fixed IAM to share the events space for v2. Pull request #6962
- Pipes can now deliver to a Redshift cluster or Serverless workgroup through the Data API. Pull request #7165
- In docker runner mode, s3-dist-cp steps now run instead of failing and taking their cluster down. Pull request #7192
- In docker runner mode, a step's Properties, core-site and spark-log4j2 classifications are now applied instead of only being stored and shown. Pull request #7226
- The npm native registry route now retains a missing package from a repository's public:npmjs external connection and serves it instead of returning 404. Pull request #7198
- Inspector2 gained the Disable operation, with DISABLING to DISABLED transitions per resource type and enabled scan types preserved during selective disablement. Pull request #7181
- CreateFlowLogs with a CloudWatch Logs destination now keeps the log group, and a create without LogDestinationType no longer reports s3. Pull request #7197
- CreateDBCluster now stores Tags and KmsKeyId, so ListTagsForResource and DescribeDBClusters return them. Pull request #7225
- An NLB listener created with Protocol TCP now gets a dedicated raw TCP relay instead of an HTTP server that answered 400. Pull request #7044
- ListMetrics now returns at most 500 metrics with a NextToken when more remain, on the Query and both JSON targets. Pull request #7236
Why it matters
This week filled in notable service behaviour: an EventBridge V2 service and Redshift delivery for Pipes widen what can be wired together, while EMR steps in docker mode now actually apply their properties and run s3-dist-cp. Several read-back bugs were closed, so tags, policies, engine versions and flow log settings that callers wrote now come back from describe and list calls. Docker and SDK test fixtures were also tightened so platform and network assumptions stop leaking between tests.
Changelog entry
- Added EventBridge V2 with routing to SQS consumer subscribers and $or filter support; IAM now shares the events space for v2. Pull request #6962
- Pipes can deliver to a Redshift cluster or Serverless workgroup through the Data API. Pull request #7165
- S3-dist-cp steps now run in docker runner mode instead of failing. Pull request #7192
- Docker runner mode now applies a step's Properties, core-site and spark-log4j2 classifications. Pull request #7226
- Npm packages can now be retained from a repository's public:npmjs external connection. Pull request #7198
- Inspector2 added the Disable operation with per-resource-type transitions and selective disablement. Pull request #7181
- CreateFlowLogs now keeps the CloudWatch log group and defaults the destination type instead of reporting s3. Pull request #7197
- CreateDBCluster now stores Tags and KmsKeyId for read-back. Pull request #7225
- NLB TCP listeners now relay raw TCP instead of running an HTTP server. Pull request #7044
- ListMetrics now paginates at 500 results with a NextToken. Pull request #7236
- CreateCacheCluster now keeps EngineVersion and AutoMinorVersionUpgrade for memcached and redis/valkey. Pull request #7177
- CloudFront cache and origin request policies now read Name and Comment from the config root only. Pull request #7252
This week: EventBridge V2 with SQS routing, Pipes to Redshift via the Data API, s3-dist-cp steps running in docker mode, npm packages retained from public external connections, and read-back fixes for flow logs, DB cluster tags and ListMetrics pagination.
This week's changes to Floci: EventBridge V2 is in, with routing to SQS consumer subscribers, $or filtering and IAM fixed to share the events space. Pipes can now deliver to a Redshift cluster or Serverless workgroup through the Data API. In docker runner mode, s3-dist-cp steps run instead of failing and taking the cluster down, and a step's Properties, core-site and spark-log4j2 are now applied. On the registry side, npm packages can be retained from a repository's public:npmjs external connection. Inspector2 gained Disable with per-resource-type transitions, and TCP listeners on a network load balancer now relay raw bytes instead of answering HTTP 400. Read-backs are more faithful: flow logs keep their CloudWatch log group, DB clusters store their tags and KMS key, cache clusters keep En
Week of October 19, 2026
What shipped
- IAM enforcement now checks a request against the resources it names instead of a wildcard for namespaces without hand-built ARNs. Pull request #6511
- GetParametersByPath now applies label filters to parameter history so a release label returns the versions carrying it. Pull request #6560
- AWS::EC2::PrefixList is now provisioned instead of stubbed, with create producing a customer-managed list that Ref and route targets resolve to. Pull request #6481
- The Kinesis event source mapping poller now applies BisectBatchOnFunctionError to split a failed batch. Pull request #6523
- Lambda gains an init-timeout-seconds setting so a newly started runtime gets its own budget for its first invocation. Pull request #6442
- GetSubscriptionAttributes now omits FilterPolicyScope for subscriptions without a filter policy. Pull request #6532
- Lambda Web Functions now delete the oldest unused revisions once a function passes 50, as AWS does. Pull request #6518
- ECS adds opt-in dynamic awsvpc host ports so tasks sharing a container port can coexist while stable host publishing stays the default. Pull request #6509
- UpdateUsagePlan now patches throttle, quota and stage throttle paths with API Gateway's errors and a lock against concurrent patches. Pull request #6418
- The ELBv2 readiness helpers retry transient TLS handshake exceptions within the existing ten-second deadline. Pull request #6538
Why it matters
This week fixes several cases where a request was answered differently from AWS: SSM label filters were ignored, IAM checks used a wildcard for many namespaces, and SNS returned a filter policy scope it should not. It also closes gaps in asynchronous paths, from Kinesis batch bisection to Lambda init budgets and ECS host ports. Pagination, queue redrive storage and usage plan locking also got attention, which matters to anyone running Terraform or SDK tests against the project.
Changelog entry
- IAM enforcement checks requests against the resources they name for more namespaces. Pull request #6511
- GetParametersByPath applies label filters to parameter history. Pull request #6560
- CloudFormation provisions AWS::EC2::PrefixList instead of stubbing it. Pull request #6481
- Lambda Kinesis event source mapping splits failed batches with BisectBatchOnFunctionError. Pull request #6523
- Lambda adds init-timeout-seconds so a cold runtime gets its own init budget. Pull request #6442
- SNS reports FilterPolicyScope only for subscriptions with a filter policy. Pull request #6532
- Web Functions delete the oldest unused revisions past 50. Pull request #6518
- ECS adds opt-in dynamic awsvpc host ports for tasks sharing a container port. Pull request #6509
- UpdateUsagePlan patches throttle, quota and stage throttle settings. Pull request #6418
- RDS PostgreSQL restores terminate target sessions before replaying the snapshot. Pull request #6550
- DynamoDB returns set members in DynamoDB's order. Pull request #6520
- ListBackupVaults and other listing calls return pages with continuation tokens. Pull request #6541
Weekly roundup: IAM checks now use the resources a request names, SSM label filters work on parameter history, PrefixList is provisioned, Kinesis batch bisection is honored, and Lambda gets an init timeout setting.
This week in Floci: IAM enforcement now checks a request against the resources it acts on rather than a wildcard for many namespaces. GetParametersByPath applies label filters to parameter history, and AWS::EC2::PrefixList is provisioned instead of stubbed, so Ref and routes resolve to a real list. The Kinesis poller honors BisectBatchOnFunctionError, Lambda adds an init-timeout-seconds budget for cold runtimes, and Web Functions delete the oldest unused revisions past 50. Smaller fixes cover SNS filter policy scope, ECS awsvpc host ports and usage plan patching.
Week of October 12, 2026
What shipped
- Cognito now supports Google identity provider federation, including callback, token exchange and userinfo attribute mapping. Pull request #5868
- S3's own access check, which evaluates bucket policies and ACLs, now runs under the s3 enforce-auth setting alongside the IAM checks. Pull request #5913
- EMR now answers ListReleaseLabels, DescribeReleaseLabel and ListSupportedInstanceTypes instead of reporting them as unsupported. Pull request #5748
- CloudFormation now removes the displaced bucket when an AWS::S3::Bucket rename is a replacement, unless UpdateReplacePolicy is Retain. Pull request #5762
- The AWS::Lambda::Function provisioner now reads LoggingConfig and passes it to Lambda; before, every stack-created function logged in Text. Pull request #5902
- An HTTP API's CORS configuration now applies to integration responses, not just the preflight. Pull request #5935
- CopyPackageVersions is now bridged for Maven, npm, pypi and nuget as well as the generic format. Pull request #5888
- CloudWatch now publishes SNS alarm actions on alarm state transitions, for automatic evaluation and SetAlarmState. Pull request #5918
- An Express execution now ends at its maximum execution time instead of continuing to run on its worker thread. Pull request #5864
- Complete container stop and remove operations are now limited to eight concurrent owners across service pools. Pull request #5860
Why it matters
This week closes gaps that broke common infrastructure workflows: an EMR module could not pick a release, a bucket rename left two buckets behind, and an HTTP API passed its preflight but failed browsers on real responses. It also extends identity and access: Cognito accepts Google providers, and S3 bucket policy checks now sit under the same enforcement switch as IAM. Several fixes address hangs and wrong answers that surfaced only under load or across regions.
Changelog entry
- Cognito: support Google identity provider federation in the authorization-code flow. Pull request #5868
- S3: bucket policy and ACL checks now run under the enforce-auth setting. Pull request #5913
- S3: a presigned copy now checks its source read against its credential. Pull request #5891
- S3: versionId=null now names only the null version on every object operation. Pull request #5887
- Emr: add ListReleaseLabels, DescribeReleaseLabel and ListSupportedInstanceTypes. Pull request #5748
- Cloudformation: replace an S3 bucket when BucketName changes and remove the old one. Pull request #5762
- Cloudformation: pass a function's LoggingConfig to Lambda. Pull request #5902
- Apigatewayv2: apply an HTTP API's CORS configuration to its actual responses. Pull request #5935
- Apigatewayv2: trust Floci's own CA when fetching a JWT issuer's keys. Pull request #5933
- Cloudwatch: publish SNS actions on alarm state transitions. Pull request #5918
- Stepfunctions: end an Express execution at its maximum execution time. Pull request #5864
- Codeartifact: bridge CopyPackageVersions to Maven, npm, pypi and nuget. Pull request #5888
This week: Cognito accepts Google identity providers, S3 bucket policy checks join the enforce-auth switch, EMR answers three release operations, and a CloudFormation bucket rename now deletes the old bucket. Plus CORS on HTTP API responses and SNS actions on alarm transitions.
This week in Floci: Cognito now supports Google identity provider federation, from callback through token exchange and attribute mapping, and S3's bucket policy and ACL checks now come under the same enforce-auth setting as the IAM checks. EMR answers ListReleaseLabels, DescribeReleaseLabel and ListSupportedInstanceTypes, so Terraform data sources that pick a release now work. CloudFormation removes the displaced bucket when an S3 bucket rename is a replacement, unless the update replace policy retains it, and a function's LoggingConfig now reaches Lambda. HTTP API CORS configuration now applies to integration responses, not only the preflight, and CloudWatch publishes SNS alarm actions on state transitions.
Week of October 5, 2026
What shipped
- Cognito now serves the OAuth token revocation endpoint on its own host and on custom domains, so signing a user out revokes the refresh token and every access and ID token. Pull request #4936
- CloudFormation now provisions Application Auto Scaling scalable targets and scaling policies, so stacks that attach a scaling policy to an ECS service no longer fail on those resources. Pull request #5407
- ElastiCache now provides serverless cache management APIs, with CreateServerlessCache, DescribeServerlessCaches, ModifyServerlessCache and DeleteServerlessCache over the AWS Query protocol, each backed by a hidden cluster. Pull request #5366
- CloudFormation now creates real CloudFront public keys, key groups and functions instead of reporting success for stub resources that created nothing. Pull request #5390
- Malformed CopyTags values are rejected before a DB cluster snapshot is copied, instead of being silently treated as false and creating a copy without source tags. Pull request #5472
- CreateSubnet now rejects IPv4 CIDRs that do not fit wholly inside the primary VPC block or an associated secondary block, returning InvalidSubnet.Range without creating a subnet. Pull request #5461
- S3 sends Access-Control-Allow-Credentials true when a request Origin matches a CORS rule through a specific origin, on both preflight and actual responses. Pull request #5375
- A resource that an UpdateStack adds now reports CREATE_IN_PROGRESS, then CREATE_COMPLETE or CREATE_FAILED, and DescribeStackResource shows CREATE_COMPLETE. Pull request #4932
- IAM now evaluates every action against the resource it acts on, so a policy statement that names one IAM resource binds to the action working on it. Pull request #5404
- The Docker call timeout is now configurable through floci.docker.api-timeout-seconds, defaulting to 300 seconds, so a shorter value shortens how long a call waits. Pull request #5431
Why it matters
This week broadens what stacks and clients can rely on: CloudFormation gained provisioners for Application Auto Scaling and real CloudFront public keys, key groups and functions, ElastiCache gained serverless cache APIs, and Cognito answers the OAuth token revocation endpoint. At the same time several validation gaps closed, so subnets outside a VPC block, IAM actions that name a resource, and malformed RDS flags are rejected instead of quietly accepted. For anyone running against Floci, that means fewer surprises from resources that reported CREATE_COMPLETE while nothing was created, and more requests behaving the way AWS does.
Changelog entry
- Feat(cognito): serve the OAuth token revocation endpoint Pull request #4936
- Feat(cloudformation): provision Application Auto Scaling targets and policies Pull request #5407
- Feat(elasticache): add serverless cache APIs Pull request #5366
- Feat(cloudformation): provision CloudFront public keys, key groups and functions Pull request #5390
- Feat(core): make the Docker call timeout configurable Pull request #5431
- Fix(rds): reject malformed cluster snapshot copy flags Pull request #5472
- Fix(ec2): reject subnet CIDRs outside the VPC address space Pull request #5461
- Fix(s3): send Access-Control-Allow-Credentials for matched CORS origins Pull request #5375
- Fix(cloudformation): report CREATE events for a resource an update adds Pull request #4932
- Fix(iam): evaluate every IAM action against the resource it acts on Pull request #5404
- Fix(dynamodb): mark time-to-live deletions on the stream as AWS does Pull request #4930
- Fix(cloudformation): honour PermissionsBoundary on AWS::IAM::Role and AWS::IAM::User Pull request #5266
This week: CloudFormation provisions Application Auto Scaling targets and policies plus real CloudFront public keys, key groups and functions. ElastiCache serverless cache APIs land, and Cognito serves OAuth token revocation.
This week in Floci: CloudFormation now provisions Application Auto Scaling scalable targets and scaling policies, and creates real CloudFront public keys, key groups and functions instead of reporting CREATE_COMPLETE for stubs. ElastiCache gained serverless cache management APIs, and Cognito now serves the OAuth token revocation endpoint on its own host and on custom domains. Validation also tightened: CreateSubnet rejects CIDRs outside the VPC address space, IAM evaluates every action against the resource it acts on, and malformed RDS snapshot copy, parameter reset and proxy boolean flags are rejected before anything is created or changed.
Week of September 28, 2026
What shipped
- Redshift Serverless now supports workgroups that actually run, letting Terraform, CloudFormation and SDK clients create a namespace and workgroup, connect with psql or JDBC, and run SQL through the Data API. Pull request #5012
- EKS now applies nodegroup labels, taints and capacity type to Kubernetes nodes, so pods with nodeSelector requirements schedule correctly and workloads tolerating nodegroup taints land on the right nodes. Pull request #5063
- CloudFormation now provisions AWS::Cognito::UserPoolUserToGroupAttachment instead of stubbing it, so users actually join groups instead of remaining unattached. Pull request #4806
- IoT now serves the custom authorizer management API with CreateAuthorizer, DescribeAuthorizer, UpdateAuthorizer, DeleteAuthorizer, ListAuthorizers, SetDefaultAuthorizer, DescribeDefaultAuthorizer and ClearDefaultAuthorizer. Pull request #4826
- EventBridge now supports API destination targets with CreateApiDestination, DescribeApiDestination, UpdateApiDestination and related operations. Pull request #5009
- CloudFormation now provisions AWS::Transfer::User, completing the Transfer service with User and Server support. Pull request #5076
- EMR can now resize running clusters with ModifyInstanceGroups and ModifyInstanceFleet, and list bootstrap actions with ListBootstrapActions. Pull request #4900
- EKS clusters no longer ship with a default local-path StorageClass, so PersistentVolumeClaims wait for a real CSI driver instead of silently provisioning local volumes. Pull request #5061
- IAM now models SSH public keys with UploadSSHPublicKey, GetSSHPublicKey, ListSSHPublicKeys, UpdateSSHPublicKey and DeleteSSHPublicKey. Pull request #5051
- REST Lambda authorizers now validate authorization policies, respect TTL caching, and reject invalid output instead of accepting the first Allow statement blindly. Pull request #5002
Why it matters
This week brings significant completeness improvements across managed services: Redshift Serverless gains actual compute workgroups, EKS nodes receive their metadata correctly, and several services like Cognito user groups and IoT custom authorizers move from stubs to fully functional implementations. These changes reduce silent failures where operations would succeed but do nothing.
Changelog entry
- Feat(redshift-serverless): add running workgroups and Data API WorkgroupName support Pull request #5012
- Feat(eks): apply nodegroup labels, taints and capacity type to the node Pull request #5063
- Feat(cognito): provision user group attachments Pull request #4806
- Feat(iot): manage custom authorizers Pull request #4826
- Feat(eventbridge): add ApiDestination target support Pull request #5009
- Feat(cloudformation): provision AWS::Transfer::User Pull request #5076
- Feat(emr): resize instance groups and fleets, list bootstrap actions Pull request #4900
- Feat(eks): stop shipping a default storage class in clusters Pull request #5061
- Feat(iam): model SSH public keys Pull request #5051
- Fix(apigateway): validate and cache REST Lambda authorizers Pull request #5002
- Fix(iam): resolve server certificate references in ELB and CloudFront Pull request #5075
- Fix(stepfunctions): resolve a Lambda task's function ARN in its own account and region Pull request #5079
This week: Redshift Serverless workgroups run, EKS labels and taints reach nodes, Cognito user groups attach, IoT custom authorizers work, EventBridge API destinations land, Transfer adds User, EMR resizes clusters, EKS drops the default storage class, IAM models SSH keys, REST a
Floci shipped significant progress across managed services this week. Redshift Serverless now supports workgroups that actually run, EKS applies nodegroup labels and taints to Kubernetes nodes, Cognito user groups attach instead of remaining stubbed, and IoT serves the custom authorizer management API. EMR can resize running clusters, EventBridge supports API destination targets, Transfer adds User support, and IAM models SSH public keys. REST Lambda authorizers now validate policies and respect caching instead of accepting invalid output blindly.
Week of September 21, 2026
What shipped
- ECS task containers now receive their IAM role credentials, enabling tasks to authenticate as their configured role. Pull request #4486
- Route 53 now resolves private hosted zone records from cluster containers, bridging the management plane and DNS serving. Pull request #4353
- S3 CreateBucket applies AWS's region rules consistently across all partitions. Pull request #4494
- SES list operations now support pagination with a shared mechanism covering ListTemplates and ListExportJobs. Pull request #4465
- KMS imported key material now supports elliptic curve key types: ECC_NIST_P256, ECC_NIST_P384, ECC_NIST_P521, and ECC_SECG_P256K1. Pull request #4428
- EC2 route tables now support InstanceId and NetworkInterfaceId route targets with automatic InstanceOwnerId population. Pull request #4416
- API Gateway import now honors the failonwarnings parameter and preserves parser warnings in responses. Pull request #4531
- Cloud Map DNS A record answers now include the configured TTL and distinguish NXDOMAIN from no-data responses. Pull request #4448
- IoT MQTT WebSocket bridge stabilizes concurrent client sessions and handles broker-initiated closes. Pull request #3816
- ElastiCache now accepts HELLO AUTH handshakes for protected endpoints. Pull request #4476
Why it matters
This week brings critical AWS feature coverage: ECS tasks can now authenticate as their roles, Route 53 serves private DNS to containers, and multi-partition support extends S3 and hostname handling across AWS regions. The additions span authentication, DNS resolution, and regional compatibility.
Changelog entry
- Feat(ecs): ECS task containers receive IAM role credentials from task-role sessions Pull request #4486
- Feat(route53): Route 53 resolves private hosted zone records from cluster containers Pull request #4353
- Fix(s3): S3 CreateBucket applies AWS region rules in every partition Pull request #4494
- Feat(ses): SES list operations support pagination for ListTemplates and ListExportJobs Pull request #4465
- Feat(kms): KMS imported key material supports ECC_NIST_P256, ECC_NIST_P384, ECC_NIST_P521, ECC_SECG_P256K1 Pull request #4428
- Feat(ec2): EC2 route tables support InstanceId and NetworkInterfaceId route targets Pull request #4416
- Fix(apigateway): API Gateway import honors failonwarnings and preserves parser warnings Pull request #4531
- Feat(cloudmap): Cloud Map DNS A answers include configured TTL Pull request #4448
- Fix(iot): IoT MQTT WebSocket bridge stabilizes concurrent sessions Pull request #3816
- Fix(elasticache): ElastiCache accepts HELLO AUTH handshakes Pull request #4476
- Fix(cognito): Cognito PasswordPolicy validates MinimumLength and removes default of 8 Pull request #4359
- Fix(sqs): SQS FIFO deduplication retained for full five-minute window after deletion Pull request #4460
- Fix(sns): SNS retries failed SQS subscription deliveries up to three times with DLQ support Pull request #4468
- Fix(core): AWS hostnames built and recognized in every partition Pull request #4451
- Fix(cloudformation): CloudFormation DeleteStack deletes UPDATE_FAILED resources Pull request #4470
- Fix(rds): RDS RestoreDBClusterFromSnapshot honors Port parameter Pull request #4326
- Fix(cognito): Cognito auth challenge sessions expire Pull request #4417
- Fix(cognito): Cognito USER_AUTH flow hides unknown users with PreventUserExistenceErrors Pull request #4478
- Feat(iam): IAM adds last-accessed reporting actions Pull request #4488
- Fix(elbv2): ELBv2 health checks use target group's HealthCheckPort Pull request #4432
Floci ships ECS task role credentials, Route 53 private DNS from containers, and multi-partition S3 region rules. EC2 route targets, KMS ECC keys, and ElastiCache HELLO AUTH also land.
This week's Floci update strengthens container authentication and DNS: ECS tasks now receive IAM role credentials, Route 53 resolves private zones from cluster containers, and S3 applies region rules across partitions. Additional improvements include EC2 route target support, KMS elliptic curve imported keys, ElastiCache HELLO AUTH, and SES pagination.
Week of September 14, 2026
What shipped
- SageMaker training jobs now respect GPU instance types like ml.g5.xlarge instead of running on CPU. Pull request #4015
- EKS Pod Identity association management lets you map Kubernetes namespaces and service accounts to IAM roles. Pull request #4012
- Glue catalog resource policy and Data Catalog encryption settings are now supported. Pull request #4027
- Glue Data Catalog connections API is implemented with create, get, update, delete and test operations. Pull request #3939
- EventBridge can now route events to Step Functions state machines as direct targets. Pull request #4006
- Cognito USER_AUTH flow lets users choose between password, password SRP, email OTP or SMS OTP authentication. Pull request #3930
- Redshift supports S3 COPY with JSON format and manifest files, plus DynamoDB zero-ETL backfill. Pull request #3889
- AWS Batch CancelJob and TerminateJob operations stop containers and manage job lifecycle state correctly. Pull request #3922
- IAM and STS now use cryptographic random generation for access key secrets, session secrets, tokens and KMS grant tokens. Pull request #3960
Why it matters
This week brought significant AWS service coverage expansions—Glue catalogs and classifiers, EKS Pod Identity, SageMaker GPU support, EventBridge-to-Step-Functions integration, and Redshift JSON copying—alongside critical security fixes for credential generation and important lifecycle fixes for Cognito, Batch and other services. The work spans multiple AWS services with both new capabilities and correctness improvements.
Changelog entry
- Feat(sagemaker): support GPU-backed training jobs Pull request #4015
- Feat(eks): implement pod identity association management Pull request #4012
- Feat(glue): catalog resource policy and Data Catalog encryption settings Pull request #4027
- Fix(cognito): accept USER_SRP_AUTH in AdminInitiateAuth Pull request #3973
- Fix(cognito): omit AvailableChallenges from RespondToAuthChallenge responses Pull request #4025
- Fix(iam): use SecureRandom for secret credentials and tokens Pull request #3960
- Fix(eks): store and return encryptionConfig and logging on clusters Pull request #4007
- Feat(glue): Data Catalog connections Pull request #3939
- Feat(eventbridge): start Step Functions targets Pull request #4006
- Feat(cognito): support the USER_AUTH choice-based auth flow Pull request #3930
- Fix(dynamodb): enforce the 400KB item limit on every update surface Pull request #3991
- Feat(glue): add crawler classifier APIs Pull request #3985
- Feat(redshift): support S3 COPY JSON and manifest, and DynamoDB zero-ETL backfill Pull request #3889
- Feat(batch): support canceling and terminating jobs Pull request #3922
- Fix(kms): match the AWS error codes, messages and check order Pull request #3968
This week: SageMaker GPU support, EKS Pod Identity, Glue catalog/classifiers, EventBridge→StepFunctions, Redshift JSON COPY, Cognito USER_AUTH, Batch job control, plus cryptographic credential generation and dozens of correctness fixes.
This week's Floci updates span AWS service expansion and critical fixes. New: SageMaker GPU-backed training, EKS Pod Identity associations, Glue catalog policies and classifiers, EventBridge-to-StepFunctions routing, Redshift JSON/manifest COPY, Cognito USER_AUTH flow, Batch job cancellation. Security: switched credential generation to cryptographic randomness. Correctness: fixed Cognito auth flows, DynamoDB item limits, KMS error codes, CloudFormation lifecycle, ECS request handling, and dozens more.