SQL Server query plan using hash vs streamaggregate - sql-server

Question

I have three tables
CREATE TABLE [dbo].[caja]
(
[orden] [int] IDENTITY(1,1) NOT NULL,
[ejercicio] [int] NOT NULL,
[numero] [int] NOT NULL,
[tipo] [char](1) NOT NULL CONSTRAINT [DF_caja_tipo] DEFAULT ('N'),
[inicial] [int] NOT NULL,
[final] [int] NOT NULL,
[total] [int] NOT NULL,
CONSTRAINT [PK_caja]
PRIMARY KEY CLUSTERED ([orden] ASC),
CONSTRAINT [IX___caja__ejercicio_numero]
UNIQUE NONCLUSTERED ([ejercicio] ASC, [numero] ASC, [tipo] ASC),
CONSTRAINT [IX___caja__tipo_inicial]
UNIQUE NONCLUSTERED ([tipo] ASC, [inicial] ASC)
) ON [PRIMARY]
CREATE TABLE [dbo].[holograma]
(
[orden] [int] IDENTITY(1,1) NOT NULL,
[taller] [int] NOT NULL,
[tipo] [nchar](1) NOT NULL,
[inicial] [int] NOT NULL,
[final] [int] NOT NULL,
[total] [int] NOT NULL,
[fecha] [smalldatetime] NOT NULL,
CONSTRAINT [PK_holograma]
PRIMARY KEY CLUSTERED ([tipo] ASC, [inicial] ASC)
)
CREATE TABLE [dbo].[Tally]
(
[N] [int] IDENTITY(1,1) NOT NULL
CONSTRAINT [PK_Tally_N]
PRIMARY KEY CLUSTERED ([N] ASC)
)
Tally table contains one million records from N=1 to 1,000,000
Caja table contains a list of valid values to insert into holograma table,
Example:
orden | ejercicio | numero | tipo | inicial | final
888 | 2015 | 74 | R | 50144001 | 50144660
889 | 2015 | 75 | R | 50144661 | 50146660
and holograma:
taller | tipo | inicial | final | total | fecha
160 | A | 50144651 | 50144750 | 100 | 15/04/2015 <--values of two caja's
missing data
49 | A | 50144826 | 50145025 | 200 | 15/04/2015
I'm trying to get the missing data. Using the example must show me from 50144751 to 50144825, counting 75 numbers.
The problem is with the count aggregate, it's taking to much time when I delimit the values. This is my query
declare #tipo nchar(1)
,#numero int
,#ejercicio int
,#largo int
;
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED;
select #tipo='A' , #numero=176;
select #ejercicio=2015, #largo=10000000;
with
c as ( /* La caja a buscar */
select tipo, numero, inicial / #largo as serie, inicial as cInicial, final as cFinal, total, entregados, orden as cOrden
from caja
where 1=1
and (tipo=#tipo)
and (numero=#numero)
and ejercicio=#ejercicio
)
, h as ( /* Los hologramas que corresponden a esa caja */
select serie, inicial - (serie*#largo) as hInicial, final - (serie*#largo) as hFinal
, h1.orden as hOrden
, cOrden, cInicial, cFinal
from holograma as h1
inner join c
on h1.tipo=c.tipo
and h1.inicial>=40000000
and (h1.inicial between c.cInicial and c.cFinal or h1.final between c.cInicial and c.cFinal)
)
, t2 as ( /* se usa para corregir */
select n
from tally
inner join c on (n between cInicial- (serie*#largo) and cFinal - (serie*#largo))
)
, t as ( /* Generar los números individuales según la ENTREGA de hologramas */
select serie,n as nHolograma
, hOrden, h.cOrden
from t2
inner join h on (n between hInicial and hFinal)
)
, e as ( /* cuantos hologramas por caja se han entregado. este se usa para corregir tabla caja */
select cOrden, COUNT(nHolograma) as totalG
from t
group by cOrden
)
select * from e
And the query plan is
https://pastee.org/kpt3t
but if I change in the "t" subquery from "t2" table to "tally" table
, t as ( /* Generar los números individuales según la ENTREGA de hologramas */
select serie,n as nHolograma
, hOrden, cOrden
from tally, h
where 1=1
and (n between hInicial and hFinal)
)
the query is almost instant. this is the query plan
https://pastee.org/hfpnz
The difference is the StreamAggregate using tally table to Hash using t2 subquery (line 196) .
I use the "t2"-subquery to delimit to the values of the current caja.
Why does the aggregate change? It's one minute of difference. 1 second with delimiting the numbers to 1:02 minutes delimiting.