Ceph

Have a fix in place, would be sending out a pull request soon.

This is basically intended behavior; under some circumstances the message to the monitors can get "replayed" and there's no way for the monitors to distinguish between that and an accidental create call.
What we could do is look at the osdmap on the client side and refuse to even try the create, maybe?

Greg, the check is actually present in Objecter::create_pool() which is not hit(may be due to a stale osdmap?). I have now moved the check to be in librados::RadosClient::pool_create() with a pre condition of 'wait_for_osdmap()', and that seem to solve the issue. This also makes it consistent with other pool operations like 'pool listing/pool delete etc..

Please let me know if that sounds right.

Hmm. That's probably fine, assuming it's wired up correctly (e.g., doesn't get blocked if the osdmap is up to date). Speed of pool creates isn't really an issue we worry about, so delaying it a little is fine.

This tracker has been fixed via the pull request https://github.com/ceph/ceph/pull/2209.

I should have rather mentioned the commit https://github.com/ceph/ceph/commit/448edb191ca38c22c00923d63a5a83c884c2f223