โ05-02-2024 02:43 AM
May be I am new to Databricks that's why I have confusion.
Suppose I have worker memory of 64gb in Databricks job max 12 nodes...
and my job is failing due to Executor Lost due to 137 (OOM if found on internet).
So, to fix this I need to increase executor memory.
But how to increase it, choosing more workers or choosing 128gb worker? which one will work.?
Help me on this..
โ05-02-2024 04:31 PM - edited โ05-02-2024 04:35 PM
Hello @amitkmaurya ,
Increasing compute resources may not always be the best strategy. To gain more insights into each executor's memory usage, check the cluster metrics tab and Spark UI for your cluster. If one executor has a much higher memory usage than the others, it could indicate a data skew issue.
Executor OOM issues can be caused by several factors, including poorly distributed partitions (skew), excessive GC and poorly optimized jobs overall. For a detailed root cause analysis and potential mitigation, please contact Databricks Support.
โ05-16-2024 11:17 PM
Hi @raphaelblg ,
I have solved this issue. Yes, in my case data skewness was the issue that was causing this executor OOM, so adding repartition just before writing resolved this skewness. I didn't change any workers or driver memory.
Thanks for your help..!!
โ05-02-2024 04:31 PM - edited โ05-02-2024 04:35 PM
Hello @amitkmaurya ,
Increasing compute resources may not always be the best strategy. To gain more insights into each executor's memory usage, check the cluster metrics tab and Spark UI for your cluster. If one executor has a much higher memory usage than the others, it could indicate a data skew issue.
Executor OOM issues can be caused by several factors, including poorly distributed partitions (skew), excessive GC and poorly optimized jobs overall. For a detailed root cause analysis and potential mitigation, please contact Databricks Support.
โ05-16-2024 11:17 PM
Hi @raphaelblg ,
I have solved this issue. Yes, in my case data skewness was the issue that was causing this executor OOM, so adding repartition just before writing resolved this skewness. I didn't change any workers or driver memory.
Thanks for your help..!!
Join a Regional User Group to connect with local Databricks users. Events will be happening in your city, and you wonโt want to miss the chance to attend and share knowledge.
If there isnโt a group near you, start one and help create a community that brings people together.
Request a New Group