Spot instances usage in Azure Databricks

noorbasha534 — Tue, 04 Feb 2025 11:21:13 GMT

Hi all,

as per the below article -

https://community.databricks.com/t5/technical-blog/optimize-costs-for-your-data-and-ai-workloads-with-azure-and-aws/ba-p/66241

1. it is possible to choose the number of spot instances using 'availability' parameter. But, I see from the documentation it is a string with pre-defined values

azure_attributes.availability

string

Controls the compute uses on-demand or spot instances (ON_DEMAND_AZURE or SPOT_WITH_FALLBACK_AZURE).

2. Now, regarding, 'spot_bid_max_price' parameter - documentation says, ''You can also set this to -1 (the default), which specifies that the instance cannot be evicted on the basis of price. The price for the instance will be the current price for spot instances or the price for a standard instance.''

So, this means as the bid price changes, Azure Databricks will try to match with the price and therefore the word 'current price' is used above? Also, 'standard instance' refers to on-demand instance?

Appreciate if someone can shed some light on this.

Re: Spot instances usage in Azure Databricks

Alberto_Umana — Tue, 04 Feb 2025 12:30:51 GMT

Hello @noorbasha534,

Thanks for your question!

1. 'availability' Parameter:

The 'availability' parameter in Azure Databricks controls whether the compute uses on-demand or spot instances. The values for this parameter are:

ON_DEMAND_AZURE: This value indicates that the compute will use on-demand instances.
SPOT_WITH_FALLBACK_AZURE: This value indicates that the compute will use spot instances with a fallback to on-demand instances if spot instances are not available.

2. 'spot_bid_max_price' Parameter:

The 'spot_bid_max_price' parameter allows you to set the maximum price for spot bids. The documentation states that you can set this to -1 (the default), which means that the instance cannot be evicted based on price. The price for the instance will be the current price for spot instances or the price for a standard instance. This implies that Azure Databricks will try to match the bid price with the current spot price, and if the spot price exceeds the bid price, the instance will fall back to the price of a standard (on-demand) instance.

In summary, the 'current price' refers to the fluctuating price of spot instances, and 'standard instance' refers to on-demand instances.

topic Spot instances usage in Azure Databricks in Data Engineering

Spot instances usage in Azure Databricks

Re: Spot instances usage in Azure Databricks

1. 'availability' Parameter:

2. 'spot_bid_max_price' Parameter: