Forum Discussion

Anonymous's avatar
Anonymous
Not applicable
3 years ago

Loading data from ADLS Delta Table into Spark Dataframe is slow

Hi there, 

 

I'm experiencing slow read times when loading data from delta tables into data frames using PySpark in Synapse notebooks. 

 

This does not include the time taken for the Spark cluster to spin up. 

 

The delta table I am loading data from is relatively small, approximately 1 million rows and it takes about 30 seconds to load these rows into a dataframe.

 

Compared to SQL server this is very slow.

 

The simple syntax I'm using is:

 

df = spark.read.format("delta").load(deltasource).select("field1","field2","field3").
 
Data is text codes and dates - but is not especially wide
 
I am not doing any processing on this dataframe yet - just loading it. 
 
Are there any likely candidates for why the data frame loading speed is so slow? Synapse Serverless is much faster at loading this dataset as well. 
 
Thank you.
 
 
 
 
 
 
data engineering
data warehouse

1 Reply

  • ananthkrishna99's avatar
    ananthkrishna99
    Advocate I
    3 years ago

    Hi Anonymous 

    Where are you executing this query, is it in Fabric/Synapse. If it's in synapse what is the spark pool size used to run the notebook?

    If you are using Fabric, what type of environment is it, Trail/Dedicated Capacity, if it's dedicated capacity what is the size of sku and node size, if it's trail what is the node size used?

    In Synapse serverless, how did you test it, is it simply by select * from table or any other way?