External tools like logrotate which use killall rely on the "ceph-mds" name to
find it in /proc/*/comm. This is normally the case but when ceph-mds respawns
using /proc/self/exe (on Linux), its name will change to "exe". This makes
logrotate fail to signal ceph-mds the log needs to be reopened and will lead to
the log growing until the disk is full.
This patch unconditionally sets the name so on respawn the name is correct (and
in any other possible scenario, such as a misnamed executable). Note, there is
still a very small race between execve and ceph_pthread_setname where the
process name is wrong.
Problem was introduced by
66a122025f6cf023cf7b2f3d8fbe4964fb7568a7.
Fixes: http://tracker.ceph.com/issues/19291
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit
4f177bb6b72cf9c8eb363051b27496c026b345f0)
Conflicts:
src/ceph_mds.cc - use pthread_setname_np() instead of
ceph_pthread_setname(), drop compat.h include
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
+#include <pthread.h>
#include <iostream>
#include <string>
int main(int argc, const char **argv)
{
+ pthread_setname_np(pthread_self(), "ceph-mds");
+
vector<const char*> args;
argv_to_vec(argc, argv, args);
env_to_vec(args);