Databricks broadcast joins: when the memo beats the meeting (and when it kills a task)
The Databricks Data EngineerShow Notes
Two engineers on the same team join the same big orders table to the same small lookup table. Same cluster, same data, one line of code different. Sarah's finishes in the time it takes to get a coffee. Mike's dies, and the error isn't about the data at all. A task ran out of memory building a hash map.
Broadcast joins get passed around as a tip instead of a model: small table equals fast, flip the switch when a join runs long. But the mechanism in almost every write-up is years out of date, and the size Spark checks is not the size that has to fit in memory.
In this episode: - How to explain broadcast versus shuffle joins at standup, in one sentence, without a whiteboard - What actually happens on a current Databricks runtime when a join broadcasts, and why the popular explanation stopped being true - Why the size estimate Spark trusts is not the size that lands, and where to read the real payload - What adaptive query execution rescues you from, and the two places its hands are tied - The question to ask before you add a broadcast hint, and the failure signature that tells you a broadcast is what broke
This episode is for Databricks data engineers who write joins every week and treat the broadcast hint as a speed switch. Whether you're mid-level and tired of guessing why one join flies and an identical one falls over, or senior and about to be asked in an interview why Spark chose a sort merge join, you'll walk away able to predict the call before you run the query.
--- Helping 18,000+ Databricks data engineers become seniors: interview like seniors, execute like seniors, think like seniors.
Follow The Databricks Data Engineer for new episodes every week.
LinkedIn: linkedin.com/in/jrlasak Newsletter: dataengineer.wiki
#DataEngineering #Databricks #DataEngineer #CareerGrowth #ApacheSpark #DeltaLake1
More from The Databricks Data Engineer
The rollout is done. Every production table migrated to Unity Catalog, every one owned by a team group, and the revenue table carries a certified check mark. The platform lead posts in Slack: fully governed. Monday morning, finance replies in the thread. The revenue on the finance dashboard and the revenue in the sales deck are about three percent apart. Which one's real?
Two proposals sit on a head of data's desk, both about the nightly pipeline that feeds the finance dashboard. One swaps the full reload for incremental loads with Auto Loader: it finishes in a fraction of the time and it's cheaper every month after the first. The other keeps the full reload, puts it on a bigger job cluster and starts it an hour earlier. It costs more. Forever. She asks one question, "What happens if this breaks on quarter close?", and approves the second one.
Your nightly Databricks job ran long, so you open the Spark UI. Two hundred tasks. One hundred ninety-nine finished in about a minute. One is still running, and it's been running for forty. You double the cluster and rerun it. Same forty minutes. Nearly twice the bill.
Sam is four years into a Databricks role, up for senior this cycle. Merged pull requests on the ingestion layer. The nightly job he rescued at three in the morning when the orders table stopped landing. Then his manager asks for one call where the answer wasn't obvious, he framed it, and he lived with the result. Sam starts three different sentences. None of them finish.
Open the workspace. Start a cluster. Get coffee. Come back, and sometimes it's ready. If you've been doing this a few years you can feel that wait in your shoulders, and it's the least interesting part of the story.