Windl, Ulrich wrote:
> Hi again!
> 
> I've implemented a Nagios compatible monitor for the MDB freespace, but I'm 
> trying to understand:
> The "unused pages" are non-fragmented, obviously
> The "free pages" may be scattered all over, but are completely free
> May there exist partially free pages that are not accounted in the monitor 
> info?

No. There is no such thing as a partially free page.

> Also for a huge change request are contiguous pages needed, or is only the 
> number of available pages important?
> 
> Does MDB care about fragmentation at all?

LMDB is a key/value store, large values are stored on contiguous pages. 
Fragmentation will affect
the search for contiguous pages for large values.

back-mdb may change multiple underlying records for a given LDAP modification. 
Indices are small,
and unaffected by fragmentation. Other things may have varying sizes.
> 
> Kind regards,
> Ulrich Windl
> 
>> -----Original Message-----
>> From: Howard Chu <h...@symas.com>
>> Sent: Friday, April 25, 2025 6:16 PM
>> To: Windl, Ulrich <u.wi...@ukr.de>; openldap-technical@openldap.org
>> Subject: [EXT] Re: MDB monitoring: How many free pages actually?
>>
>> Windl, Ulrich wrote:
>>> Hi,
>>>
>>>
>>>
>>> Considering this monitoring information, I wonder:
>>>
>>> olmMDBPagesMax: 12800
>>> olmMDBPagesUsed: 9229
>>> olmMDBPagesFree: 3810
>>
>>> Is the number of actual free pages 12800 - 9229 + 3810, i.e. "unallocated
>> (12800 - 9229)" plus "usable" (3810)?
>>
>> Yes. "Pages Free" are used pages that are reusable.
>>
>>> How many free pages are needed before updates stop working?
>>
>> That depends on the size of an update.
>>
>>
>> --
>>   -- Howard Chu
>>   CTO, Symas Corp.           http://www.symas.com
>>   Director, Highland Sun     http://highlandsun.com/hyc/
>>   Chief Architect, OpenLDAP  http://www.openldap.org/project/
> 


-- 
  -- Howard Chu
  CTO, Symas Corp.           http://www.symas.com
  Director, Highland Sun     http://highlandsun.com/hyc/
  Chief Architect, OpenLDAP  http://www.openldap.org/project/

Reply via email to