Ceph » rgw

using radosgw 15.1.0 from prebuilt el8 (centos) octpus rpms at http://download.ceph.com/rpm-octopus/el8/x86_64/ceph-radosgw-15.1.0-0.el8.x86_64.rpm

Normal s3 access works fine for a given user. But when using sts to assume a role, I get 'InvalidArgument" in the aws response. I am using aws-cli/1.18.16 Python/3.6.8 Botocore/1.15.16, on cent8.1

I have a role created with an assume-role-policy-doc that includes my own user as someone who can assume the role. There is also an access policy set for the role. My standard creds work fine for s3 operations, and are set in my awscli config as well as in my boto3 code in my test below

For the cli:

aws --debug sts --asuume-role --role-arn 'arn:aws:iam:::role/rolename' --role-session='sessname' --region=''

gives <Error><Code>InvalidArgument</Code>.....</Error> in the aws response.

The exact same error occurs with boto3 and:

assume_role_obj = sts_client.assume_role(RoleArn='arn:aws:iam:::role/rolename', RoleSessionName='SessName')

I have in ceph.conf

rgw sts key = abcdefghijklmnop
rgw s3 auth use sts = true

If I replace the RoleArn with a bogus one, I get 'NoSuchEntity' in the error response instead of 'InvalidArgument'

Setting rgw debug = 20, the sts errors in the rgw log file have:

sts:assume_role executing
ERROR: one of role arn or role session name is empty
sts:assume_role completing
sts:assume_role op status=-22

The -22 is consistent with the InvalidArgument, but I did pass in both RoleArn and RoleSessionName!!!! (The debug output shows them in the POST going out....)

Any ideas? Thanks!