What is the difference between INNER, LEFT, and FULL OUTER?

INNER keeps only rows whose keys match in both files. LEFT keeps every left row, filling right columns with blanks where there is no match. FULL OUTER keeps all rows from both files, blanking the missing side.

Do the key columns need the same name?

No. You choose the left key column and the right key column independently, so you can join id against customer_id. Matching is on the values in those columns, not the header names.

What happens with duplicate keys?

If a key appears multiple times on the right, each left row that matches it produces one output row per matching right row — the same cartesian behaviour as SQL. Deduplicate first with the CSV Deduplicator if you want a single match.

Is the join fast on big files?

Yes. The right table is indexed into a hash map keyed on its join column, then the left table is scanned once, giving roughly linear O(n + m) performance instead of comparing every pair of rows.

CSV Joiner — Gera Tools

Merge two CSVs on a shared key

The CSV Joiner combines two CSV files the way a SQL join does. Match rows on a shared key column — for instance an order file against a customer file on a customer ID — and choose whether to keep only matches (INNER), every left row (LEFT), or every row from both sides (FULL OUTER). The right table’s key column is not repeated in the output.

How it works

Both files are parsed with a quote-aware reader. The right table is then indexed into a hash map whose keys are the values of its join column, with each key mapping to the list of right rows that have it. This is a classic hash join.

The left table is scanned once. For each left row the tool looks up its key in the hash map: every matching right row produces one combined output row (left columns followed by the right columns except the duplicate key). In LEFT and FULL OUTER mode, a left row with no match still emits one row with the right columns left blank. In FULL OUTER mode, any right rows whose keys were never matched are appended at the end with the left columns blank and the right key copied into the key position. Because the right side is indexed once and the left side scanned once, the work is O(n + m).

Example and notes

Joining:

id,name        id,city
1,Ann          1,London
2,Bob          2,Paris
3,Cara         4,Berlin

on id with an INNER join yields two rows (ids 1 and 2). A LEFT join adds Cara with a blank city; a FULL OUTER join additionally adds Berlin (id 4) with a blank name. If a key repeats on the right, expect one output row per match — deduplicate the right file first if you need exactly one.