Words

Your word, the word you will see in Databricks documentation, and what it means for your work

Not yet written. The structure below is settled; the prose is not.

One discipline makes this page worth writing: source every entry from an error message or a documentation page you would actually meet, not from a glossary. The test for an entry is whether it would have saved somebody a search.

Bidirectional, because the translation is needed in both directions. Being overwhelmed by documentation is very often a vocabulary problem.

Your words

sf, terra, purrr, furrr, polygon, raster, CRS, EPSG, replicate, resample, fit, tune.

These need no explanation, only a mapping. The entry for each says what the platform calls the thing, or says plainly that the platform has no word for it, which is itself useful: a reader who stops searching for the Databricks term for a resample has got what she came for.

The platform’s words

Driver, executor, worker, node, partition, serialisation, pushdown, UDF, artifact, volume, DBFS, catalog, schema, access mode, cluster lifecycle, warehouse.

This is where the explaining budget goes. Each entry has to do two things: say what the word means, and say what it changes about her work. “Driver” is not a definition exercise. It is the machine her code runs on unless she does something specific, and that is the fact that makes the rest of the site readable.

Three of these carry a consequence sharp enough that the entry should state it outright rather than leave it to a job page. A partition is the unit of parallel work, so the count of them, and not the count of rows, decides how much runs at once. Serialisation is the cost paid to move work to another machine, which is why distributing can be slower than not distributing when there is no other machine to reach. And an access mode decides which languages a cluster will accept, in a way that is narrower than it first appears.

No domain nouns belong in this table. A catchment is not a Databricks word and needs no translation.

Where a word means two things

The page should carry a short list of terms the reader will meet with two incompatible meanings, because these cost more search time than unfamiliar words do. “Worker” is the clearest case: a worker machine in a cluster, and a forked R worker process inside one machine, are different things, and most parallel-R writing means the second while most Databricks writing means the first. A reader who does not have that distinction cannot read Run a Monte Carlo simulation correctly, because the whole page turns on it.

Next

If a word sent you here from the middle of something, go back to it. If the word was about where your code runs, Sending things to where your code runs is the page that uses this vocabulary most heavily; if it was about the machine, see Work out what the cluster you have can do.