How to tune long running Queries on Azure Synapse DW (Dedicated Pool) - Part 1 ~ datablogs

Sunday, August 28, 2022

How to tune long running Queries on Azure Synapse DW (Dedicated Pool) - Part 1

 |   |  No comments  |  , , , ,

Three hours of long running pipeline is reduced to run in 35 Minutes when we given right kind of top bottom optimization like our Body !!! 

Yes Sometimes if we give proper core and cross workouts you can run 10km in 55 Minutes !!! 

Its happened to me But Not Now 😆😆😆 

Oh Oh we are away from Data ground !!! Lets Move into Azure Synapse tunning,

In the starting phase every piece of code and pipeline process was really trouble us to optimize it , whatever we do on tunning it will run as same time 

Waited long time until each process to complete as its running 3 hours !!!

But Reecha blog helped something to start with basic check and given some hope to fine tune it 

Remember below piece of monitoring code always help us to too dig more 

--Check Long Running Query in Azure Synapse

SELECT *
FROM sys.dm_pdw_exec_requests
WHERE status not in ('Completed','Failed','Cancelled')
  AND session_id <> session_id()
ORDER BY submit_time DESC;
--Find the distributed query plan steps for long running query

SELECT * FROM sys.dm_pdw_request_steps WHERE request_id = 'xxxxxx'  ORDER BY step_index; 

***Replace your request_id from first query result
Sample Running Query :


Initially we thought data volume is huge , so we are expecting this much delay on complete . But distributed query Plans are given more finishing time for each queries in Azure Synapse

We will go step by step process of tunning , 

Multiple ways we can see the Execution plan of your Azure Synapse Queries

  1. Using Azure Console --> Click you Dedicated Pool --> Click Query Activity --> Each Queries you have Query Plans 

        2.Using SSMS , you can run below query and get your explain plan in XML Format 

                EXPLAIN select * from dbo_datablogs.vw_fact_transferdata (nolock)



So ,We have figured out the execution plan and further need to fix below things to make it faster 

We need to reduce data movement operations ( Shuffle Move , Trim Move , Partition Move ) and needs to have proper indexing on your business tables 

Based on your data distribution on tables and indexing , queries will be executed on SQL Pool . We need to distribute the tables accordingly . So Major part of the tunning on table level only 

Once we done that in proper way we have achieved your milestone ,

How to we distribute table in Azure Synapse ? 

Two types of distribution is available in Azure Synapse . Round Robin and Hash Distribution 

Round Robin Distribution
  • By Default , If you create table it will create in Round Robin Distribution
  • Table rows are distributed in all distribution 
  • Main purpose of this table to improve loading speed we can use round robin distribution
  • If your table is just junk or doesn't have proper any constraints and keys use round robin distribution
CREATE TABLE [dbo_datablogs].[ControlTable_blogs](
                [Id] [int] IDENTITY(1,1) NOT NULL PRIMARY KEY NONCLUSTERED NOT ENFORCED,
                [SourceObjectSettings] [nvarchar](max) NULL) 
WITH ( CLUSTERED COLUMNSTORE INDEX,  DISTRIBUTION = ROUND_ROBIN )

Hash Distribution
  • If your table is having frequent DML operations ( Insert , Update , Delete ) use hash distribution
  • It will increase performance 200% of your queries 
  • It will distribute the data to nodes based on your distributed column 
CREATE TABLE [dbo_datablogs].[ControlTable_blogs](
                [Id] [int] IDENTITY(1,1) NOT NULL PRIMARY KEY NONCLUSTERED NOT ENFORCED,
                [SourceObjectSettings] [nvarchar](max) NULL) 
WITH ( HEAP,  DISTRIBUTION = HASH(Id))

Replicate 

This is another way to storing the table in SQL Pool 
  • Full copy of the table will be available in all distribution to avoid data movement 
  • Table is having less size use replicate option
CREATE TABLE [dbo_datablogs].[ControlTable_blogs](
                [Id] [int] IDENTITY(1,1) NOT NULL PRIMARY KEY NONCLUSTERED NOT ENFORCED,
                [SourceObjectSettings] [nvarchar](max) NULL) 
WITH ( HEAP,  DISTRIBUTION = REPLICATE )

Thumb of Rules of Creating tables 

First Level : Find the absolute unique column or else use partial unique column for creating table as Hash Distribution 

Second Level : If your table doesn't have proper unique columns , try to create the table with used columns in Join , Group By , Distinct , Over and Having Clauses on Procedures or Select Clauses 

Lets examine First Level , 

When creating the table with Round Robin and created table with below script ,

CREATE TABLE dbt_datablogs.fact_transferdata
WITH
  (
    CLUSTERED COLUMNSTORE INDEX,  
    DISTRIBUTION = ROUND_ROBIN   
  )  
AS SELECT * FROM dbt_datablogs.vw_fact_transferdata

Its taken 28 Minutes to complete


So we figured out good candidate key and created table with below script 

CREATE TABLE dbt_datablogs.fact_transferdata
WITH
  (
    CLUSTERED COLUMNSTORE INDEX,  
    DISTRIBUTION = HASH(tranferaddressid)
  )  
AS SELECT * FROM dbt_datablogs.vw_fact_transferdata

Its taken 4 Minutes to complete 


We will see the huge difference when creating proper keys 

Lets examine Second Level ,

When we don't have proper keys , based on your business logics or select clause join conditions we can try to evaluate with multiple columns and create proper hash function to improve performance 

Lets do the smaller table tunning on this blog , you can try the same in your own environments , 

Below table is small but its having shuffle move and its delaying the process 11 Minutes ,







Then based on shuffle columns , add the proper hash function and recreate the table 

CREATE TABLE dbt_datablogs.fact_memberinfo
WITH
  (
    Heap,  
    DISTRIBUTION = HASH(infoid)
  )  
AS SELECT * FROM dbt_datablogs.vw_memberinfo

Once created the table again examined the Execution plan for the same statement and process is finished with in a seconds 



This is not an end , Also we have tunned few things on Data Factory and Azure Synapse Workload Management to get expected performance , we will see further on next blog 

To learn more in-depth , Please use below references from Microsoft Site as well 

References : 


Share:

0 comments:

Post a Comment