Patches #3
Patches #3
Conversation
|
I cannot think of a better way to handle this. |
libhdfs3/src/rpc/RpcChannel.cpp
Outdated
| @@ -761,8 +761,8 @@ void RpcChannelImpl::buildConnectionContext( | |||
| std::string euser = key.getAuth().getUser().getEffectiveUser(); | |||
| std::string ruser = key.getAuth().getUser().getRealUser(); | |||
|
|
|||
| if (!key.getAuth().getUser().hasEffectiveUser()) | |||
| euser = principal; | |||
| if (!euser.empty()) | |||
There was a problem hiding this comment.
Not sure this change is correct. There is also still a call down below.
I played around with this a little today. I think this works if real and effective are both "user" but fails if effective is "user@DOMAIN" , even though that will get mapped to user also.
There was a problem hiding this comment.
I think behavior we want is:
1 Kerberos or simple, set effective to the effective user name if set, otherwise real. (for kerberos we get the real user name from the ticket).
2 For simple, if there is also a real user, set that.
This is a bit complicated, because I think in the UGI, real user name can be the effective user name and getUser is the original principal, so it won't match up to Java code exactly.
@bdrosen96 , do you think there is a better way to do this than to walk through the patches one-by-one like this? Having based the repo on HAWQ, I don't know how a I could cherry-pick or otherwise apply your changes.