[Pvfs2-developers] server crash on startup with millions of files

[Sam Lang]

On Mar 15, 2007, at 2:12 PM, Pete Wyckoff wrote:

> slang at mcs.anl.gov wrote on Wed, 14 Mar 2007 17:23 -0500:
>>> slang at mcs.anl.gov wrote on Wed, 14 Mar 2007 16:11 -0500:
>>>> Let me know how these changes look, and if someone gets a chance to
>>>> look at performance differences, that would be great.
>>>
>>> Sounds good from the description.  I'll take a look at the encoding
>
>> @@ -642,10 +642,11 @@ struct PVFS_servreq_readdir
>>      PVFS_ds_position token; /* dir offset */
>>      uint32_t dirent_count;  /* desired # of entries */
>>  };
>> -endecode_fields_4_struct(
>> +endecode_fields_5_struct(
>>      PVFS_servreq_readdir,
>>      PVFS_handle, handle,
>>      PVFS_fs_id, fs_id,
>> +    skip4,,
>>      PVFS_ds_position, token,
>>      uint32_t, dirent_count)
>
> Each one of these structs must be fully padded in case something
> comes after it.  Instead of adding another pad after dirent_count to
> do this, you could nestle fs_id + dirent_count together, before or
> after the 64-bit token.  Actually maybe that's not true.  Nothing
> will come after this.  But you can get rid of the pad by rearranging
> a bit.

Right it looks like there are a lot of requests that just skip the  
pad on the last field if its 32 bits because its always going to come  
last.  I'll switch around the fields to get rid of the skip.
>
> Otherwise looks great.  I skipped big sectionss of whitespace only
> changes.

Most of that is probably just getting the indentation right.  I've  
configured vim to show me when there are tabs in a file instead of  
spaces.  If its something I'm editing, I've been re-indenting
the entire file.

> And you probably don't want the nightly-tests.cfg to go
> in.

Hmm...How did that get in there?  Doing too many things at once...

Thanks!

-sam

>
> 		-- Pete
>