prepare for the change to split the options by services
in future, osd will have its own osd.yaml.in, while crimson will consume
both osd.yaml.in and crimson.yaml.in.
Signed-off-by: Kefu Chai <kchai@redhat.com>
* refs/pull/39742/head:
client: make Inode to inherit from RefCountedObject
client: minor cleanup to Inode class to simplify the code
client: minor cleanup to Inode related code
Reviewed-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
we do not refresh the daemons on maintenance hosts so our info
on them is always outdated. Therefore, the best option is to
assume maintenance mode is working correctly and the daemons
are stopped
Signed-off-by: Adam King <adking@redhat.com>
While this argument is accepted it has always been ignored. To avoid
confusion it should at least be called out in the documentation.
Signed-off-by: Lee Yarwood <lyarwood@redhat.com>
This PR improves the readbility and elegance of
the "Advanced OSD Service Specifications" section
of the OSD chapter of the cephadm guide.
Signed-off-by: Zac Dover <zac.dover@gmail.com>
The `handle_signal()` way requires going through the reactor, and
thus is unsuitable for situations when it's no longer operational.
SIGSEGV generated as a result of actual fault (not `kill()`) is
the intended audience of this change.
Signed-off-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
Removing the call to engine's `exit()` should allow the singnal
handler to exit. In the case of `SIGSEGV` this would return back
to the problematic instruction, and thus retrigger the fault.
As our handler already restores `SIG_DFL`, it's expected to get
a core dump.
Signed-off-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
seastar sets the signal handler for SIGSEGV and SIGABRT, and its signal
handler prints the stacktrace, but the the stack frames are adresses,
which are not human readable without the help of addr2line.
since crimson is linked with -rdynamic option, we have the symbols added
to the dynamic symbol table already. let print out the symbolized
addresses instead using our own stacktrace utility.
Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit 6e26243b6db49bb9813f8b8aeade68da07dc6065)
Conflicts:
src/crimson/osd/main.cc
We were looking at anon_conns and accepting_conns without holding
the lock (deleted_lock is not sufficient).
Drop this test, and move the decrements:
- inc when we add to conns or anon_conns (no changes there)
- dec when we remove from deleted_conns (several different paths!)
Fixes: https://tracker.ceph.com/issues/49237
Signed-off-by: Sage Weil <sage@newdream.net>
Either ignore or terminate, otherwise it may confuse the
"create"/"remove" caller.
Fixes: https://tracker.ceph.com/issues/50395
Signed-off-by: Mykola Golub <mgolub@suse.com>
as per Sage, the caps are out of date (should be based on profiles) and
the keyring locations aren't useful information
Signed-off-by: Kefu Chai <kchai@redhat.com>
rgw_pool and sysobj are part of the RADOS backend, and do not belong in
the API. Instead, pull the objects using them into the API. These are:
RGWOIDCProvider, RGWRole, and a new object LuaScript.
Signed-off-by: Daniel Gryniewicz <dang@redhat.com>
after the yaml-to-cxx migration, osd_deep_scrub_stride is an OPT_SIZE,
and is hence represented with an uint64. so we need to cast `r` to
uint64_t before comparing it with this setting for silencing GCC
warning.
Signed-off-by: Kefu Chai <kchai@redhat.com>
