Ceph

These are pretty much always errno numbers, which should be translated by the cli via strerror_r().

I wasn't aware that `-2` or `-1` where common values (couldn't find any references to it in the docs I searched).

However, the problem is that, as a user, I don't understand what the problem is so I can correct whatever step is needed.

When I first opened this ticket, the problem was that there was no keyring in the host where I was running the command. And yet,
the I get the same output when I don't have sufficient permissions:

$ rados lspools
2015-02-16 06:23:32.768509 7fd06a1df7c0 -1 monclient(hunting): ERROR: missing keyring, cannot use cephx for authentication
2015-02-16 06:23:32.768514 7fd06a1df7c0  0 librados: client.admin initialization error (2) No such file or directory
couldn't connect to cluster! error -2
[ubuntu@magna123 ~]$ sudo !!
sudo rados lspools
rbd
data
metadata

The error should attempt to identify what is wrong so that the user can amend the mistake.

For the first case it could be:

Could not find any keyrings for current user in $PATHS
Failed to authenticate to the cluster.

For the second scenario where I don't have enough permissions:

Cannot read keyring in $PATH: Permission denied.

Which makes it crystal clear to what the problem is and with a format that mimics what happens when
running a command (like `ls`) on a directory that the current user doesn't have permissions.

You want to get different errors when you're not allowed to read a keyring versus when it doesn't exist?

I guess in some cases that can be identified, but quite often you're just denied access to the directory...

I would say anything that is possible to identify (and get back to the user) would be better than the current output.

See #5634, too. Sigh.

Negative errnos are the convention back from kernel calls, so Ceph uses them internally too. See cpp_strerror() for the handling convention.

can we enclose the errno along with the path name into an exception and catch it in the caller? that would be very handy, IMO. and the error message from bufferlist is swallowed by MonMap::read(). so an Exception with a str and an errno is what we want. right?

but at this moment, i will translate the errno using cpp_strerror() in the up-most caller in CLI. so we will have something like:

$ rados lspools
couldn't connect to cluster: (2) No such file or directory

adding the Exception will change the signature of the interface of Rados. so will print the error message to log.

it would be fantastic to understand what is it that it expected and couldn't find (besides the actual error)

yeah, i will print the path of file when ceph server/client is trying to open it but fails to do so.=)