Some tests have m=2,k=2 and this will break them. Sometimes even if we
have 5 up osds, we end up with 4 and CRUSH gets picky, so build in a
buffer and only do this if we have 6 up.
We don't have an easy way from here to see what the min up osds for healthy
is... basically this map discontinuity test just sucks.
Signed-off-by: Sage Weil <sage@redhat.com>
rgw: fix wrong debug related to user ACLs in rgw_build_bucket_policies()
Reviewed-by: Adam Emerson <aemerson@redhat.com>
Reviewed-by: Casey Bodley <cbodley@redhat.com>
--single-version-externally-managed is used for packaging python eggs
to deb. so the metadata can be stored in the egg, and can be accessed
by user if necessary, see [1]. but this does not apply to users who
just want to "make install" without packaging Ceph.
the problem is, if we pass --single-version-externally-managed to
setup.py, we need to pass --record or --root also. otherwise setuptools
complains:
error: You must specify --record or --root when building system packages
so in this change, we choose not to pass
"--single-version-externally-managed", unless $DESTDIR is set.
both "ceph.spec.in" and "debian/rules" set $DESTDIR when running "make
install". so we can always check $DESTDIR to tell if current "make
install" is launched by the packaging script or not. this might not be
accurate, but good enough. assuming enduser does not use DESTDIR when
running plain "make install".
--
[1] https://wiki.debian.org/Python/FAQ#How_should_we_package_Python_eggs.3F
Signed-off-by: Kefu Chai <kchai@redhat.com>
If there is no session key, then we should not instantiate a session
handler, or else we'll later crash with something like
0> 2018-11-19 08:52:27.787 7f6b305e4700 -1 /home/sage/src/ceph.mimic/src/auth/Crypto.h: In function 'std::size_t CryptoKey::encrypt(CephContext*, const in_slice_t&, const out_slice_t&)' thread 7f6b305e4700 time 2018-11-19 08:52:27.784394
/home/sage/src/ceph.mimic/src/auth/Crypto.h: 164: FAILED assert(ckh)
ceph version 13.2.2-411-g4bfd25addd (4bfd25addd) mimic (stable)
1: (ceph::__ceph_assert_fail(char const*, char const*, int, char const*)+0x102) [0x7f6b36ad7e72]
2: (()+0x2ab007) [0x7f6b36ad8007]
3: (CephxSessionHandler::_calc_signature(Message*, unsigned long*)+0x297) [0x7f6b36e13007]
4: (CephxSessionHandler::check_message_signature(Message*)+0x6b) [0x7f6b36e131ab]
5: (AsyncConnection::process()+0x1dc5) [0x7f6b36bfbc55]
6: (EventCenter::process_events(unsigned int, std::chrono::duration<unsigned long, std::ratio<1l, 1000000000l> >*)+0x6a5) [0x7f6b36c0abb5]
The specific scenario is:
- new (nautilus) OSD connects to heartbeat port on an older OSD (<= mimic)
and sends an authorizer
- mimic osd ms_verify_authorizer succeeds, but does not set up the
session key because mimic doesn't care.. it blindly sets isvalid = true.
- mimic osd later creates a session handler
- mimic osd later tries to validate a message and crashes
The fix is to not instantiate the session handler if there is no key. Note
that this happens *after* we have authenticated the request (called
ms_verify_authorizer), which is responsible for populating the session
key while verifying the authorizer (if it is authenticating at all), so
this session_key is always set on an authenticated connection.
Fixes: http://tracker.ceph.com/issues/36443
Signed-off-by: Sage Weil <sage@redhat.com>
The %bcond_with and %bcond_without macros are confusing to folks who
don't do a lot of RPM packaging work. Let's try to help these folks out!
Signed-off-by: Nathan Cutler <ncutler@suse.com>
Fedora 29 still ships a Python 2 binary, but some of Ceph's build
dependencies are only available in py3 versions there. In other
words, from F29 on, it is no longer possible to do a py2 Ceph build
on Fedora, even if a python2 binary exists on the system.
If that were not enough, the Python 2 that ships with Fedora 29 is
linked against a non-compatible version of OpenSSL.
Before this commit, install-deps.sh was overriding the spec file's
Python build setting based on the presence or absence of a python2
binary. As the bug cited below indicates, this was not a good idea.
It's better for the spec file to be explicit about which OS versions
are py2 and which are py3, and just stick to that.
Fixes: http://tracker.ceph.com/issues/37301
Signed-off-by: Brad Hubbard <bhubbard@redhat.com>
Signed-off-by: Nathan Cutler <ncutler@suse.com>
and extract state transition out of repeat_handle_connect()
in this change, the connect/handle-connect loop is restructured, to
avoid ad-hoc state changes in helper functions. this pave the road to
explicit state transtion using named states.
also, exception is thrown instead in handle_connect_reply(), we should
not proceed in case of failures. and we need do error handling in the
named state in future.
currentl, `state` is set to `state_t::open` in `start_connect()` and
`start_accept()`, the next step is to set it in a named state.
Signed-off-by: Kefu Chai <kchai@redhat.com>
* refs/pull/24844/head:
osd: drop PGBackend::Listener::get_epoch().
osd: massively switch to get_osdmap_epoch().
osd: switch the return type of PG::get_osdmap().
Reviewed-by: Sage Weil <sage@redhat.com>
* refs/pull/24867/head:
mon: compensate for not having STDLIB_MAP_SPLICING
Reviewed-by: Kefu Chai <kchai@redhat.com>
Reviewed-by: João Eduardo Luís <joao@suse.de>
