ceph_fuse: fix daemonization when pid file is non-empty #13532
Conversation
|
@tchaikov would you mind reviewing this? I think you're more familiar with the preforker etc code |
|
will take a look shortly. |
src/ceph_fuse.cc
Outdated
| int r = forker.parent_wait(err); | ||
| if (r) { | ||
| cerr << "ceph-fuse" << err << std::endl; | ||
| if (global_init_prefork(g_ceph_context) >= 0) { |
There was a problem hiding this comment.
global_init_prefork() should always return 0 here. we can either add an assert, or ignore the the return value.
src/ceph_fuse.cc
Outdated
| r = forker.prefork(err); | ||
| if (r < 0) { | ||
| cerr << "ceph-fuse" << err << std::endl; | ||
| forker.exit(r); |
There was a problem hiding this comment.
the exit(3) call does not allow the destructors to be called. take the cct for example, its dtor won't be called this way. so i'd recommend use return instead.
src/ceph_fuse.cc
Outdated
| if (r < 0) { | ||
| cerr << "ceph-fuse" << err << std::endl; | ||
| } | ||
| forker.exit(r); |
|
@tchaikov updated |
|
I saw following crash after replacing forker.exit(r) with 'return r'. The monitor code uses forker.exit(r) too. Maybe it's intentional to avoid the destructors 2017-02-22 10:16:40.070124 7fae407ba240 -1 /home/zhyan/Ceph/ceph/src/log/Log.cc: In function 'void ceph::logging::Log::stop()' thread 7fae407ba240 time 2017-02-22 10:16:40.069483 /home/zhyan/Ceph/ceph/src/log/Log.cc: 431: FAILED assert(is_started()) ceph version 12.0.0-251-g2f6046c (2f6046cf2041468da66f4a4a03abd7d3bebb7461) 1: (ceph::__ceph_assert_fail(char const*, char const*, int, char const*)+0x102) [0x7fae37885fa2] 2: (()+0x2a1704) [0x7fae378aa704] 3: (CephContext::~CephContext()+0x4fb) [0x7fae37a3c74b] 4: (CephContext::put()+0x173) [0x7fae37a3c9f3] 5: (main()+0x96f) [0x5624194eb08f] 6: (__libc_start_main()+0xf1) [0x7fae34d12731] 7: (_start()+0x29) [0x5624194ec449] NOTE: a copy of the executable, or `objdump -rdS ` is needed to interpret this. |
|
oic, the reason we called |
|
there are only two users of global_init_postfork_start(), ceph-fuse and ceph-mon. For ceph-mon, the parent process does not call global_init_postfork_start(). |
tchaikov
dismissed
their stale review
as discussed over irc, i'd suggest to bring cbf18b1 back to start log in parent.
During daemonizing ceph-fuse, both the parent and child processes call global_init_postfork_start(). global_init_postfork_start() calls pidfile_write(). If pid file is non-empty, one pidfile_write() call fails because it can't lock the pid file. The fix is only child process calls global_init_postfork_start() Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
During daemonizing ceph-fuse, both the parent and child processes call
global_init_postfork_start(). global_init_postfork_start() calls
pidfile_write(). If pid file is non-empty, one pidfile_write() call
fails because it can't lock the pid file.
The fix is only child process calls global_init_postfork_start()
Signed-off-by: "Yan, Zheng" zyan@redhat.com