Delay rows coming into DLT pipeline

Mathias — Wed, 29 May 2024 12:01:30 GMT

Backgroundand requirements: We are reading data from our factory and storing it in a DLT table called telemetry with columns sensorid, timestamp and value. We need to get rows where sensorid is “qrreader-x” and join with some other data from that same table and store elsewhere. The qr code’s are coming in with very low latency, much less than some of the sensors they should be joined with. There is need for a delay to wait for the other data coming in before processing these rows.

Suggestion: Can we create a DLT pipeline that would be run as batch every 5 minutes and only read rows that have timestamp older than x minutes?

@Dlt.table( comment="A table summarizing counts of the top baby names for New York for 2021." ) def top_baby_names_2021(): return ( dlt.read("baby_names_prepared") .filter(expr("Year_Of_Birth == 2021")) )

When looking at the above example code, I assume the pipeline would consider all rows coming in as handled, even though they are filtered out. So is there a way to put the filtered rows back into the pipeline if they are too fresh?

Re: Delay rows coming into DLT pipeline

raphaelblg — Wed, 29 May 2024 15:51:31 GMT

Hi @Mathias,

I'd say that watermarking might be a good solution for your use case. Please check Control late data threshold with multiple watermark policy in Structured Streaming.

If you want to dig-in further there's also: Spark Structured Streaming Programming Guide - Handling Late Data and Watermarking.

There are other ways to achieve what you're aiming for, I think it's more of a design decision.

topic Re: Delay rows coming into DLT pipeline in Data Engineering

Delay rows coming into DLT pipeline

Re: Delay rows coming into DLT pipeline