Fragmented free space in empty LMT ASSM tablespace?

Question

select count(*) from dba_extents where file_id=127;COUNT(*)
0
select count(*) from dba_free_space where file_id=127;COUNT(*)
2
select block_id, blocks from dba_free_space where file_id=127;
  BLOCK_ID     BLOCKS
       128     507904
    508032     139616
select extent_management, segment_space_management, allocation_type
from dba_tablespaces
where tablespace_name=(select tablespace_name from dba_data_files where file_id=127);EXTENT_MANAGEMENT     SEGMENT_SPACE_MANAGEMENT     ALLOCATION_TYPE
LOCAL     AUTO     SYSTEM
alter tablespace ... coalesce;
select count(*) from dba_free_space where file_id=127;COUNT(*)
2
select version from product_component_version where product like 'Oracle%';VERSION
11.2.0.2.0
How is that possible?

Answer 1

user5066799 wrote:
Thanks Jonathan, but, sorry - have not got - what do you mean? Do you mean that space management blocks occupy first 128 blocks of datafile or do they occupy blocks after last free extent block? The only management blocks I expect to find are these first 128 blocks (1M) which as I understand are datafile header (was 64K in earlier versions, but in 11.2 is 1M, with 8K results in 128 blocks).
Or do you mean the fact of using autoallocate extent size influences this matter? Should we consider "autoallocate start size" of 64K in relation to something?Your interpretation of what I was saying is correct - but I got the scale wrong, there's a different explanation of what you're seeing.
If you dump the first few blocks of the file, in your case using something like:
alter system dump datafile 129 block min 2 block max 6;you will see that block 2 will be the "file space header block" and the next blocks will be file space bitmap blocks.
Each file space bitmap block (in an 8KB block) loses a couple of hundred bytes to block headers and other control details, then has one bit for each 64KB of the data file, where the 64KB unit is dictated by the autoallocate option
Sample dump:
buffer tsn: 14 rdba: 0x01c00002 (7/2)
scn: 0x0b86.40279c3f seq: 0x01 flg: 0x04 tail: 0x9c3f1d01
frmt: 0x02 chkval: 0x3100 type: 0x1d=KTFB Bitmapped File Space Header
File Space Header Block:
Header Control:
RelFno: 7, Unit: 8, Size: 655360, Flag: 1
AutoExtend: NO, Increment: 0, MaxSize: 0
Initial Area: 126, Tail: 655359, First: 0, Free: 81904
buffer tsn: 14 rdba: 0x01c00003 (7/3)
scn: 0x0b86.40279b45 seq: 0x01 flg: 0x04 tail: 0x9b451e01
frmt: 0x02 chkval: 0x0afb type: 0x1e=KTFB Bitmapped File Space Bitmap
File Space Bitmap Block:
BitMap Control:
RelFno: 7, BeginBlock: 128, Flag: 0, First: 0, Free: 63488
buffer tsn: 14 rdba: 0x01c00004 (7/4)
scn: 0x0b86.40279b47 seq: 0x01 flg: 0x04 tail: 0x9b471e01
frmt: 0x02 chkval: 0xcafb type: 0x1e=KTFB Bitmapped File Space Bitmap
File Space Bitmap Block:
BitMap Control:
RelFno: 7, BeginBlock: 508032, Flag: 0, First: 0, Free: 63488 Note particularly the "Free: 63488" - which is bits, each bit represents 8 blocks for a total of 507904. Allowing for the first 128 blocks in the file this is why the map in block 7/4 has a BeginBlock of 508032.
A single extent in dba_free_extents is, apparently, not allowed to cross a space management block.
Regards
Jonathan Lewis
P.S. Since there are 126 available file space bitmap blocks, and 507904 blocks allowed per bitmap block, then when the file hits 63,995,904 blocks (488.25 GB) Oracle will have to add a secondary space management area. (At least, that's if you grow the file to that size; if you pre-create the tablespace at that something above that size then perhaps Oracle will allocate 2M of file header.)
P.P.S Here's a quote from my errata pages for Practical Oracle 8i: "Allowing for a little overhead, a single 8K block can hold information about 63,488 extents,..." so not a lot has changed in the interim, apart from the fact that the header is now 1MB rather than 64KB - possibly to cater for bigfile tablespaces, possibly to help ASM and Exadata align their extents and AUs.
Edited by: Jonathan Lewis on Dec 7, 2012 6:30 PM