On Thu, 12 Jan 2017, John Spray wrote: > On Thu, Jan 12, 2017 at 8:12 AM, liuchang0812 <liuchang0812@xxxxxxxxx> wrote: > > Hi, all > > > > In our documentation, it says that: > > > > Ceph’s logging levels operate on a scale of 1 to 20, where 1 > > is terse and 20 is verbose. > > > > http://docs.ceph.com/docs/master/rados/troubleshooting/log-and-debug/#subsystem-log-and-debug-settings > > > > > > But, it seems that there are some logging levels 25 : > > > > > > mds/MDBalancer.cc:276: dout(25) << "=== got heartbeat " << > > m->get_beat() << " from " << m->get_source().num() << " " << > > m->get_load() << dendl; > > mds/CInode.cc:4189: dout(25) << __func__ << dendl; > > mds/MDCache.cc:11656: //dout(25) << "saw depth " << d << " " << > > *dir << dendl; > > mds/MDCache.cc:11666: //dout(25) << " saw sub " << **p << dendl; > > mds/journal.cc:373: dout(25) << "EMetaBlob::add_dir_context(" << > > dir << ") maybe " << > > > > > > So, shoud we update our documentation, or update logging level to 20? > > Probably neither -- the >20 log messages are intentionally obscure > things (although I do accept that there is a reasonable argument that > if we would never turn something on in a real system then we perhaps > should not have it in the code...) FWIW, the guideline I tend to use: 10 - O(requests) 20 - O(requests) or O(internal structures). e.g., we dump pg log entries and dump every element other ptoential large data structures at 20 30 - O(bytes). several bits of code hexdump actual read/write data at 30) You could update the doc to mention that there are levels >20 in some rare cases and that they are extremely verbose. sage
