Ceph » CephFS

Found this while working on the downstream BZ https://bugzilla.redhat.com/show_bug.cgi?id=2331118
Please take a look at https://bugzilla.redhat.com/show_bug.cgi?id=2331118#c25

Problem Description:

I looked into the code and found that the cephfs-journal-tool updates the expire_pos,write_pos,read_pos to the new_start (old_write_pos + 1) and writes
this out to the journal header. So when the mds restarts after recovery dentry, it would fetch new udpated exire_pos, write_pos, read_pos, so the trimming
happens from new expire_pos. The old journal log objects never get trimmed!!!

Please find the relevant code below of cephfs-journal-tool.

---------------
int Resetter::reset()
{
...
...

lock.lock();
  uint64_t old_start = journaler.get_read_pos();                                               
  uint64_t old_end = journaler.get_write_pos();                                                
  uint64_t old_len = old_end - old_start; 
  cout << "old journal was " << old_start << "~" << old_len << std::endl;                      

  uint64_t new_start = round_up_to(old_end+1, journaler.get_layout_period());                  
  cout << "new journal start will be " << new_start
       << " (" << (new_start - old_end) << " bytes past old end)" << std::endl;                

  journaler.set_read_pos(new_start);
  journaler.set_write_pos(new_start);
  journaler.set_expire_pos(new_start);
  journaler.set_trimmed_pos(new_start);                                                        
  journaler.set_writeable();   

  cout << "writing journal head" << std::endl;
  journaler.write_head(new C_SafeCond(mylock, cond, &done, &r));

....
}
--------------------

We have the following output of the journal reset command executed at the customer site.
------------
bash-5.1$ cephfs-journal-tool --rank=ocs-storagecluster-cephfilesystem:0 journal reset              
old journal was 1305835407849959~27424540185
new journal start will be 1305862836584448 (4194304 bytes past old end)
writing journal head
writing EResetJournal entry
done

-------------

Need to figure out if can the above and delete the unused stale journal log objects.