[Pvfs2-developers] extra trove error messages

Phil Carns pcarns at wastedcycles.org
Wed May 31 02:50:46 EDT 2006


>> Hi Phil,
>>
>> I think that's from the crdirent, which does a keyval_read to  verify 
>> that the directory doesn't exist, and only creates it if  ENOENT is 
>> returned.  At the level of keyval_read_op_svc, its  impossible to know 
>> that though.  I guess I can just remove the  error reporting call, 
>> although it would be nice to have it there  for cases where we 
>> actually want to read an entry we expect to be  there and print an 
>> error if we don't find one (corrupted db).
>>
> 
> Actually I take that back.  The only place in the keyval.c code we  
> print an error is in read_list.  So it probably isn't a problem there  
> since read_list is only used by get_eattr.  So the keyval_read_op_svc  
> only prints a warning if DBPF_KEYVAL_DEBUG is enabled.
> 
> It looks like the error you're seeing might be in dspace_getattr,  which 
> gets called from lookup and prelude.  If we do a lookup to  verify that 
> a file doesn't exist before creating it , I could imagine  seeing that.  
> In any case the problem is the same as before.  I could  print a warning 
> if debugging is enabled.
> 
> Alternatively, I could create a trove flag (TROVE_NOENT_NON_FATAL)  for 
> those operations where we actually want to return ENOENT as a non- fatal 
> response.

I'm not sure what the right answer is- it just caught me off guard to 
see those messages.  Any idea why they only show up on stderr (meaning 
that the server has to be run with -d)?  With that being the case it 
seems like the messages probably aren't very likely to be seen regardless.

-Phil


More information about the Pvfs2-developers mailing list