Revert "[Authorization] Converge authz of ns policies from super-user to tenant-administrator (#13157)"#13463
Conversation
… to tenant-administrator (apache#13157)" This reverts commit 3cc8309.
michaeljmarshall
left a comment
There was a problem hiding this comment.
For example, I think that the publish rate limiting is to prevent a specific tenant from inconvenience to other tenants by publishing at a very high rate. It is a problem that tenant managers can raise this on their own.
This is a good point. My mistake was in assuming that a tenant admin should have permission to update all namespace policies.
I think we are likely missing policies at the tenant level. It seems like a tenant admin should be able to operate on all namespace policies. That being said, I agree that we should revert this commit now.
@yuruguo - the work doesn't need to be lost completely. We could update the default PulsarAuthorizationProvider implementation to align with correct permissions (i.e. what operations should be super user and what should be tenant admin). It might be worth discussing the "right" permission level on the mailing list.
|
@massakam Thanks for your clarification, I have a few questions.
@michaeljmarshall Thank you for your reply, I want to try to provide more information
Currently, we provide But we don’t use TenantOperation operation in default implementation-PulsarAuthorizationProviderMaybe we need to use it to optimize this piece of permission logic
There is a summary here but there is no update PIP 49: Permission levels and inheritance |
|
We have two things here:
Currently, we don't have any tenant-level limitations, so that the superuser is not able to limit the resources of the tenant. Another point is some policies only applied to a topic, not limit the whole namespace. The tenant admin still can create more partitions to request more resources of the cluster. |
As far as
That looks good to me.
I think so. At the very least, the superuser should be able to set an upper limit.
I think |
|
@massakam Thanks for your meticulous answer 👍
Maybe we can find the reason why the access permission was changed from super user to tenant administrator based on the submission history.
I agree with it. Finally, I am not opposed to revert the PR(#13157), but maybe it's not a good way to revert it directly or completely, the reasons are as follows:
Therefore, it may be a better solution to revert purposeful by a special PR, what do @eolivelli @315157973 @michaeljmarshall think? |
|
I lean towards reverting the whole commit and then creating new PRs to add the right functionality. However, I don't have a strong opinion here.
I think @massakam's point is that #13157 decreased the necessary permission level and that it has problematic consequences for things like tenant rate limits. I agree that we should leave the rate limit permissions at the superuser level for now until we make it possible to set an upper limit for a given tenant. We should figure this out before 2.10.0 gets released. |
|
FYI @cckellogg , @jerrypeng |
|
My suggestion is
Any thoughts? |
|
@codelipenghui @yuruguo |
|
I agree with the decision to revert it before 2.10.0 is released and to discuss this on the mailing list. Thanks for merging this, @nkurihar. |
Motivation
With #13157, some namespace policies can now be changed by tenant admins as well as superusers. I don't think this change is correct.
For example, I think that the publish rate limiting is to prevent a specific tenant from inconvenience to other tenants by publishing at a very high rate. It is a problem that tenant managers can raise this on their own.
Therefore, the privilege to change the publish rate limit should only be granted to the administrators of the entire Pulsar instance, i.e. the superusers.
Modifications
This reverts commit #13157.
Verifying this change
Documentation
no-need-doc