What's Changed
New Contributors
- @emreocell made their first contribution in #2745
[1.5.43] — 2026-11-19
Added
- Cognito — app client secret rotation —
AddUserPoolClientSecret,ListUserPoolClientSecretsandDeleteUserPoolClientSecret. A client holds at most two secrets, each with an ID of<ClientId>--<epoch ms>. A generated secret is 51 lowercase letters and digits.SECRET_HASH, the token endpoint andGetTokensFromRefreshTokenaccept either active secret. A third givesLimitExceededException"Client secrets cannot exceed limit of 2 secrets.", deleting the last one givesInvalidParameterException"Cannot delete the only client secret.", and an unknown ID givesResourceNotFoundException"Client secret does not exist". Once the original secret is deleted,DescribeUserPoolClientno longer shows aClientSecret. Reported by @TheKerbycz. - Cognito —
POST /oauth2/revoke— revokes a refresh token and the tokens issued from it, so a later refresh fails withNotAuthorizedException"Refresh Token has been revoked". It takes a form body only (otherwise415with an empty body). A missing token answers400invalid_request"Invalid parameter in request", and a bad client secret answers401invalid_client"Invalid client credentials in request". An unknown token, or one already revoked, answers200with an empty body. Reported by @TheKerbycz.
Fixed
- Kinesis — CBOR timestamps past the year 9999 — a CBOR request carrying an epoch timestamp beyond 9999-12-31, such as the millisecond
StreamCreationTimestampthe Kinesis Client Library sends in multi-stream mode, failed whole withSerializationException. CBOR timestamps now reach the handlers as epoch numbers, as JSON ones do, which also fixesGetShardIteratorAT_TIMESTAMPover CBOR. Reported by @tabili4. - Kinesis — calls slowed as a stream grew — every put and read rebuilt each shard's whole record list to drop expired records, so a stream holding 200,000 records spent ~13 ms per call on it. Expiry now trims only the expired records at the front. Reported by @TLDart.
- Aurora DSQL — isolation levels — with
DSQL_STRICT=1, a transaction could begin atREAD COMMITTEDorSERIALIZABLE, and a plainBEGINran at Postgres'sread committed. Every transaction now runs atrepeatable read. Any other level inBEGIN,START TRANSACTION,SET TRANSACTIONorSET SESSION CHARACTERISTICSis refused0A000"Unsupported isolation level: …", anddefault_transaction_isolationandtransaction_isolationcan no longer be set, as on the live service. Reported by @toby6578. - Cognito — OIDC provider endpoints with a query — an OIDC provider whose
authorize_url,token_url,attributes_url,jwks_urioroidc_issuerhas a query string is refused withInvalidParameterException"Endpoint can not contain queries" on create and update. Reported by @TheKerbycz. - Cognito — federated sign-in —
/oauth2/authorizewithoutidentity_provider, for a client whose only provider is an external one, now redirects straight to that IdP. When the IdP's token endpoint refuses the code, the user returns to the app'sredirect_uriwitherror=invalid_requestanderror_description" Error - <error_description>". An unreachable token host gives the host name as the description. Before, the user got a Cognito error page. An OIDC user now stores only the claims the attribute mapping names, andemail_verifiedisfalseunless it is mapped. Reported by @TheKerbycz. - IAM —
aws:PrincipalArnandaws:PrincipalIsAWSService— for a role session,aws:PrincipalArnis now the role's ARN with its path, not the session ARN.aws:PrincipalIsAWSServiceisfalseon a direct call; before, it was missing, so aBoolcondition on it never matched. Reported by @TheKerbycz. - DynamoDB —
dynamodb:LeadingKeys— withAUTH=true, the partition key values ofGetItem,PutItem,UpdateItem,DeleteItem,Query,BatchGetItem,BatchWriteItemand the transaction items are now in the condition context. AForAllValues:StringEqualspolicy on them limits each caller to its own keys, and a refusal is 400AccessDeniedException. Reported by @TheKerbycz. - S3 — expected bucket owner —
x-amz-expected-bucket-ownerandx-amz-source-expected-bucket-ownerwere ignored. When the bucket belongs to another account, the request now answers403 AccessDenied"Access Denied". A value that is not a 12-digit account ID answers400 InvalidBucketOwnerAWSAccountID. Reported by @TheKerbycz. - S3 — SSE-KMS needs KMS permissions — with
AUTH=true, writing an object under a customer managed key (put, copy, multipart create) also needskms:GenerateDataKeyon the key. Reading it (get, copy source) needskms:Decrypt. A refusal is403 AccessDenied"User: is not authorized to perform: kms:GenerateDataKey on resource: because no identity-based policy allows the kms:GenerateDataKey action". Reported by @TheKerbycz. - Secrets Manager —
RotateSecretruns the rotation function —RotateSecretcopied the current value into a newAWSCURRENTversion itself, and the function was never invoked. It now invokes the function withcreateSecret,setSecret,testSecretandfinishSecret. Each event carriesClientRequestToken,RotationToken,SecretIdandStep. The function moves the stages, and the call answers with the newVersionId. WithRotateImmediatelyfalse, onlytestSecretruns. Reported by @TheKerbycz. - Cognito —
AdminLinkProviderForUsersource provider — aSourceUser.ProviderNamethat is not an identity provider of the pool is refused with 400InvalidParameterExceptionSourceProviderName must match a Provider that is configured for the User Pool, and the user keeps itsidentities. Contributed by @emreocell. Reported by @TheKerbycz. - Cognito —
SupportedIdentityProvidersof an app client —CreateUserPoolClientandUpdateUserPoolClientrefuse a name that is neitherCOGNITOnor an identity provider of the pool with 400InvalidParameterExceptionThe provider <name> does not exist for User Pool <pool id>., naming the first missing name in list order; the match is case-sensitive, also forCOGNITO. The client is not created or changed. Create the identity provider before the client that names it. Contributed by @emreocell. Reported by @TheKerbycz. - RDS —
CreateDBClusterstatus — theCreateDBClusterresponse reportedStatusavailable. It now reportscreating, as AWS does.DescribeDBClustersstill reports the status from the cluster's members, so a cluster with no members isavailableat once. Contributed by @Areson. - ECR — pull-through cache rule for an ECR upstream in the same account —
CreatePullThroughCacheRulefor an ECR upstream in the caller's account and another region asked for acustomRoleArn, and refused one with "Upstream registry URL and provided credential type do not match", so the rule could never be created. It is now created without a role, as on AWS, and the first pull copies the upstream image without one. Rules for an upstream in another account are unchanged. Contributed by @kharabasz.