Import data

Goal. Get existing data into Strata. About three minutes.

Import into one model

strata ./mydb arrow import --target kv ./out.parquet

imported 3 rows into kv from ./out.parquet (0 skipped)

--target is the model: kv, json, event, vector or graph. The receipt reports rows imported and rows skipped, and the skipped count is the one to check.

What the file needs

The simplest source of a valid file is an export, so a round trip needs no schema work at all:

strata ./source arrow export --primitive kv --format parquet ./data.parquet
strata ./target arrow import --target kv ./data.parquet

imported 3 rows into kv from ./data.parquet (0 skipped)

For a file you are producing yourself, key-value wants a key, a value, and how each is encoded:

key,key_encoding,value,value_encoding
user:1,utf8,Ada,utf8
user:2,utf8,Bob,utf8

JSON wants a key and the document as a string:

{"key": "user:1", "document": "{\"name\": \"Ada\", \"role\": \"engineer\"}"}

The version and timestamp columns an export includes are not required on import. Imported rows are new commits on the target, with the target’s own versions.

Import into a branch first

An import is a lot of writes at once, and doing it on a branch means you can look before it reaches anything that matters:

strata ./mydb branch fork default import-run
strata ./mydb --branch import-run arrow import --target kv ./data.parquet
strata ./mydb --branch import-run kv count
strata ./mydb branch merge import-run default

If the file was wrong, delete the branch instead of merging it. Nothing on default ever saw it.

Failure modes

Rows skipped. The receipt counts them. Usually a row is missing a key or has an encoding the target does not accept.

invalid_argument.executor.arrow_format. The extension and the content disagree, or the extension is unsupported. Use .parquet, .csv or .jsonl.

Everything imported but reads come back empty. Check the branch and space you imported into against the one you are reading.

Next

Migrate from SQLite puts this to work.