Todd Lipcon has posted comments on this change.

Change subject: Enable GSSAPI for servers and ExternalMiniCluster
......................................................................


Patch Set 11:

(4 comments)

http://gerrit.cloudera.org:8080/#/c/4765/11/src/kudu/integration-tests/external_mini_cluster.cc
File src/kudu/integration-tests/external_mini_cluster.cc:

Line 145:                           "could not create client principal");
> Are the status' returned from these methods not detailed enough?  Maybe we 
hrm.. they're not super specific - they'll say the subprocess exited but not 
why. The advantage of doing it here vs in the underlying method is that we get 
more context info here (i.e that it's the client principal that failed, and not 
some server principal), plus we only need one RETURN_NOT_OK_PREPEND whereas the 
underlying methods have several RETURN_NOT_OK()s


http://gerrit.cloudera.org:8080/#/c/4765/11/src/kudu/integration-tests/external_mini_cluster.h
File src/kudu/integration-tests/external_mini_cluster.h:

Line 123:   bool enable_kerberos;
> I think we need to have a discussion about what different configuration opt
Yea, I agree that this and the gflag that I added elsewhere in this patch 
probably will end up changing a bit as we add other components of security and 
want to either have more fine-grained control or an overall "enable all the 
security stuff" flag.


http://gerrit.cloudera.org:8080/#/c/4765/11/src/kudu/rpc/connection.cc
File src/kudu/rpc/connection.cc:

Line 55: DEFINE_bool(server_require_kerberos, false,
> Why is 'server' in the flag name? Flags are only used server side.
But servers also act as clients of other servers, and this only affects the 
server side of negotiation.

I agree that we need to re-evaluate the flags before shipping, but not sure we 
could default to requiring authentication. Many people don't have Kerberos 
infrastructure, and we don't have any other user authentication mechanism in 
scope (nor is there really any secondary standard that's in use in the Hadoop 
ecosystem).

Are you OK adding this as is, given I marked it experimental with the TODO 
below?


http://gerrit.cloudera.org:8080/#/c/4765/11/src/kudu/util/subprocess.h
File src/kudu/util/subprocess.h:

Line 71:   void SetEnvVars(std::map<std::string, std::string> env);
> This is good, but it looks like we aren't using it anywhere?
We are from external_mini_cluster.cc:638 where we pass the appropriate krb5 env 
vars into the daemons


-- 
To view, visit http://gerrit.cloudera.org:8080/4765
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I595469e9cc8b2b2f57e9726014eeeb8112595801
Gerrit-PatchSet: 11
Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-Owner: Todd Lipcon <[email protected]>
Gerrit-Reviewer: Alexey Serbin <[email protected]>
Gerrit-Reviewer: Dan Burkert <[email protected]>
Gerrit-Reviewer: Kudu Jenkins
Gerrit-Reviewer: Tidy Bot
Gerrit-Reviewer: Todd Lipcon <[email protected]>
Gerrit-Reviewer: Will Berkeley <[email protected]>
Gerrit-HasComments: Yes

Reply via email to