cubridcluster-discuss

[Cubridcluster-discuss] [Open Discussion] about count star without 'where' conditions

[SourceForge.net <no...@so...>]

The following forum message was posted by piaosongmei at http://sourceforge.net/projects/cubridcluster/forums/forum/1129809/topic/4049458:

as i've described at #305, for 
[code]select count(*) from table[/code] 
there're as bellow process:

1. non-partition no-index table:
    [code]xasl->proc.buildvalue->agg_list->flag_agg_optimize[/code]  is set
as 1. So it not scan out tuple but just read stats info of heap file (get heap
header to read heap_hdr->estimates.num_pages and heap_hdr->estimates.num_recs).
So out solution is if the heap file is belong to remote host, then send command
NET_SERVER_HEAP_GET_CLASS_NOBJS_AND_NPAGES to remote server to get objects
count.

2. non-partition index table:
    the same as case 1.

3. partition no-index table:
    [code]xasl->proc.buildvalue->agg_list->flag_agg_optimize[/code]  is set
as 0. So, it will scan out each record to increase
[code]xasl->proc.buildvalue.agg_list.curr_cnt[/code]. So out solution is: if
this table is belong to remote server, send XASL about [code]select * from
table[/code] to remote server to fetch records(it's just temporary solution.
it need to improve performance at M3).

4. partition index table:
    same as case 3.


my doubt is: why partition table do not read stats info same as non-partition
table? can we just read each partition stats info and summarize them?

[Cubridcluster-discuss] [Open Discussion] RE: about count star without 'where' conditions

[SourceForge.net <no...@so...>]

The following forum message was posted by dong1 at http://sourceforge.net/projects/cubridcluster/forums/forum/1129809/topic/4049458:

Let me explain more about our question.
for sql query as:
[code]select count(*) from t1[/code]

we observed that, 
if t1 is non-partition table,  current CUBRID code just get stats info about
the count number for optimization. like case1 and case 2.
if t1 is partition table ,current CUBRID code do scan , do not get stats info.
like case 3 and case 4.

When we implement for global table,  I think we should follow the same behavior
just like current CUBRID code.
But We think non-partition table and partition table should have the same
behavior.
That means, if non-partition table can get stats info, why partition table can
not?

so, our questions are:
1 why partition table have to do scan, do not like non-partition table , which
get stats info directly?
2 if we implement global partition table, can we get stats info from servers
, just like local non-partitiont table?


Hi, Park, This it the question That I have metioned to you in monday's VTC meeting,
How do you think?

Thanks a lot!

[Cubridcluster-discuss] [Open Discussion] RE: about count star without 'where' conditions

[SourceForge.net <no...@so...>]

The following forum message was posted by iamyaw at http://sourceforge.net/projects/cubridcluster/forums/forum/1129809/topic/4049458:

Ok. I have understood what you are saying.
I'll look into source code to find out what's the problem and the reason.
BTW, if I do instantly replying, frankly speaking, the existing implementation
of partition feature is a little bit dumb.