SQL Server index and poor execution plan - sql

Question

I have an existing SQL Server database where I cannot modify the structure or the queries ran and am facing an issue with poor execution impacting performance and ultimately cloud database cost.
Kindly note my experience with SQL is quite limited and after multiple googling and trial and errors, still did not achieve acceptable result. Any tips or help is much appreciated, thank you all in advance. If you would like me to provide more information, feel free to comment and I will update the post accordingly.
The issue
I have two tables: Table1 and Table2. Table2 references Table1 via TABLE1_ID field and we run a SQL query extracting info from Table2 while filtering on Table1 ( INNER JOIN I believe).
Using the following query:
DECLARE #P1 datetime
DECLARE #P2 datetime
SELECT
dbo.Table2.VALUE
FROM
dbo.Table2,
dbo.Table1
WHERE
-- joins Table1/Table2
dbo.Table1.ID = dbo.Table2.TABLE1_ID
-- filters on Table1
AND dbo.Table1.TIMESTAMP between #P1 and #P2
My understanding would be that the database engine would first filter on Table1 then do the join with Table2, however, the execution plan I am seeing is using a Merge Join implying Table2 is fully scanned then joined with filtered results from Table1.
What I have tried
I have tried the following, attempting to identify the problem or optimize performance:
Optimization attempt Creating an FK constraint
Optimization attempt Creating other indexes with/without include columns
Issue identification Changing the query to select value from both Table1 and Table2 and see the difference
Re-creating the issue
The following script could allow you to re-create the database structure (please note it will insert 1M records into both tables):
CREATE TABLE [dbo].[Table1] (
[ID] [decimal](10, 0) IDENTITY(1,1) NOT NULL,
[VALUE] [nchar](10) NULL,
[TIMESTAMP] [datetime] NOT NULL
) ON [PRIMARY]
GO
ALTER TABLE [dbo].[Table1] ADD CONSTRAINT [DF_Table1_TIMESTAMP] DEFAULT (sysdatetime()) FOR [TIMESTAMP]
GO
CREATE UNIQUE CLUSTERED INDEX [IX_Table1_ID] ON [dbo].[Table1]
(
[ID] ASC
)
GO
CREATE NONCLUSTERED INDEX [IX_Table1_TIMESTAMP] ON [dbo].[Table1]
(
[TIMESTAMP] ASC
)
INCLUDE ([ID])
GO
CREATE TABLE [dbo].[Table2] (
[ID] [int] IDENTITY(1,1) NOT NULL,
[TABLE1_ID] [decimal](10, 0) NOT NULL,
[VALUE] [nchar](10) NULL
) ON [PRIMARY]
GO
CREATE NONCLUSTERED INDEX [IX_Table2_TABLE1_ID] ON [dbo].[Table2]
(
[TABLE1_ID] ASC
) INCLUDE ([VALUE])
GO
Declare #Id decimal(10,0) = 1
DECLARE #Now datetime = SYSDATETIME()
While #Id <= 1000000
Begin
Insert Into dbo.Table1 values ('T1_' + CAST(#Id as nvarchar(10)), DATEADD (ss, #Id, #Now))
Insert Into dbo.Table2 values (#Id, 'T2_' + CAST(#Id as nvarchar(10)))
Print #Id
Set #Id = #Id + 1
End
GO
Then you can try to run the following query:
DECLARE #P1 datetime
DECLARE #P2 datetime
SELECT
dbo.Table2.VALUE
FROM
dbo.Table2,
dbo.Table1
WHERE
dbo.Table1.ID = dbo.Table2.TABLE1_ID
AND dbo.Table1.TIMESTAMP between #P1 and #P2

Answer 1

My understanding would be that the database engine would first filter on Table1 then do the join with Table2,
Wrong. SQL is a descriptive language, not a procedural language. A SQL query describes the result set, not the methods used for creating it.
The SQL parser and optimizer are responsible for generating the execution plan. The only requirement is that the results from the execution plan match the results described by the query.
If you want to control the execution plan, then SQL Server offers hints, so you can require a nested loop join. In general, such hints are used to avoid nested loop joins.
Actually, your query is reading the index. This is a more efficient way of "filtering" the data than actually reading the data and filtering. This looks like an optimal execution plan.
Further, don't use commas in the FROM clause. Use proper, explicit, standard, readable JOIN syntax.