Using Begin & Commit inside a Stored Procedure

  • Hi,

    I have a stored procedure where i update table "A" with matching records from table "B". Before Update there is a "begin tran" and after several updates and inserts in table "A" there is a "Commit" Statement. Now my question is the table "B" is a live table where there will be lot of Inserts and updates continuously happening. In this case will it block the other transactions which are using table "B" ? As i am updating "A" by referencing "B" will it block "B" also????

     

    Thanks in Advance!

     

  • it depends on lock granularity

    in worst case scenario whole both tables will be locked for other sessions  till the end of tran

    https://www.mssqltips.com/sqlservertutorial/3305/what-does-begin-tran-rollback-tran-and-commit-tran-mean/

     

     

     

  • Thank you!

  • Also worth noting, modifying two tables within a transaction can be a vector towards deadlocks if another procedure or query modifies the same two tables, but in a different order.

    "The credit belongs to the man who is actually in the arena, whose face is marred by dust and sweat and blood"
    - Theodore Roosevelt

    Author of:
    SQL Server Execution Plans
    SQL Server Query Performance Tuning

  • "after several updates and inserts..."

    Maybe it's worth asking... does your procedure specify 'set xact_abort on'?  When a proc has multiple DML statement in a transaction and without xact_abort setting can result in tricky situations.  Without that setting it's possible a rollback will not be complete if the procedure throws an exception.  If the proc only contains a single DML statement (unless it's 'merge') it's possibly ok to omit (although I never do).  "raises a run-time error" = the old way of saying "throws an exception".

    The docs say:

    "When SET XACT_ABORT is ON, if a Transact-SQL statement raises a run-time error, the entire transaction is terminated and rolled back."  "OFF is the default setting."

    https://docs.microsoft.com/en-us/sql/relational-databases/sql-server-transaction-locking-and-row-versioning-guide?view=sql-server-2017

    Aus dem Paradies, das Cantor uns geschaffen, soll uns niemand vertreiben können

  • And use try..catch

  • Steve Jones - SSC Editor wrote:

    And use try..catch

    I hear a whole lot of people making that recommendation.  Try/Catch is usually not necessary, especially in the presence of SET XACT_ABORT ON, which will auto-magically produce the correct ROLLBACK for the entire explicit transaction regardless of the number of statements in the explicit transaction block .

    Further, most people do a really terrible job in their Catch blocks and don't even produce what just letting SQL properly handle the error does.

    --Jeff Moden


    RBAR is pronounced "ree-bar" and is a "Modenism" for Row-By-Agonizing-Row.
    First step towards the paradigm shift of writing Set Based code:
    ________Stop thinking about what you want to do to a ROW... think, instead, of what you want to do to a COLUMN.

    Change is inevitable... Change for the better is not.


    Helpful Links:
    How to post code problems
    How to Post Performance Problems
    Create a Tally Function (fnTally)

  • Not necessary, but I'd argue it's a good practice and habit to get into. It clearly sets out the transaction, which can be lost in code, and it allows you to include some logging if necessary in a catch block.

     

  • Agree try/catch is primarily for error capture.  Partial rollbacks are not common in my experience.  Sql clients could cache partial form input however they do it.

    In theory something could be done after the rollback inside the catch block.  Here's a  NoSql-ish example:

    1. Create a heap table with 2 columns, col1_name and col2_name
    2. Create a unique index on col1_name and specify ignore dupe keys
    3. Insert two records (it will work)
    4. Insert the same two records (they are ignored)
    5. Create a procedure to do 2 more inserts.  One insert appends '_next' and then the second insert appends '_last' to existing values.  The second insert throws a 'divide by 0 exception' on purpose and then the proc selects all records.  The catch block of the procedure does 3 things:

      1. Returns all error metadata generated
      2. Rollback the transaction
      3. Selects all records

    6. execute the procedure

    With the exception thrown the proc returns 2 rows and all the error fields.  If commented out the line 'and col1_name=1/0;' the proc returns 6 rows.  If the rollback is commented out in the proc, the error is: "mismatching number of BEGIN and COMMIT statements".  If the 'select * from dbo.tests' is placed before the rollback it will return 4 rows.  If placed after the rollback it will return 2 rows.

    In words the proc says: "here's some data, put it with all the other data (if you don't already have it) and then return everything (no matter what and if something goes wrong tell me exactly what happened)."

    drop table if exists dbo.test;
    go
    create table dbo.test(
    col1_namenvarchar(1024) not null ,
    col2_namenvarchar(1024) not null);
    create unique nonclustered index ndx_test_unq_col1 on dbo.test(col1_name)
    with
    (ignore_dup_key = on);
    go

    insert test(col1_name, col2_name) values('this', 'that'), ('these', 'those');
    insert test(col1_name, col2_name) values('this', 'that'), ('these', 'those');

    drop proc if exists dbo.sptest
    go
    create proc dbo.sptest
    as
    --set nocount on;
    set xact_abort on;

    begin transaction
    begin try
    insert test(col1_name, col2_name)
    select
    col1_name + '_next',
    col2_name + '_next'
    from
    test;

    insert test(col1_name, col2_name)
    select
    col1_name + '_last',
    col2_name + '_last'
    from
    test
    where
    col1_name not like '%_next'
    --and col1_name=1/0;

    select * from test;
    commit transaction;
    end try
    begin catch
    select
    error_number() error_number,
    error_severity() error_severity,
    error_state() error_state,
    error_line () error_line,
    error_procedure() error_procedure,
    error_message() error_message,
    xact_state() xact_state;

    --select * from test;
    if xact_state()=-1
    rollback transaction;

    select * from test;
    end catch

    set xact_abort off;
    --set nocount off;
    go

    exec sptest;

    Aus dem Paradies, das Cantor uns geschaffen, soll uns niemand vertreiben können

Viewing 9 posts - 1 through 8 (of 8 total)

You must be logged in to reply to this topic. Login to reply