Read first: How data flows · Medallion pipeline.
What it is#
Bringing a file into the datalake through push mode ("You upload"): you send the file to the landing zone and Ingestia loads it into Bronze automatically. Supported formats: CSV, JSON (lines/JSONL or array), and Parquet.
CSV, JSON and Parquet only accept upload. There is no "we fetch" for a file format: for Ingestia to fetch files from a folder or a bucket of yours, use the object storage connectors — Amazon S3, Google Cloud Storage, Azure Blob Storage and compatible ones (MinIO, Wasabi, Cloudflare R2, Backblaze B2, DigitalOcean Spaces), plus SFTP, Google Drive, Dropbox, OneDrive/SharePoint and file over HTTP. All of them are in Limited access: each connector is released after it is validated with a real credential from your account. This tutorial covers upload, which is the simplest path.
What it is for#
Uploading one-off extracts, dumps, or exported spreadsheets without setting up a database connection — ideal for getting started quickly or for data that does not live in a connectable system.
When to use / when not to use#
- Use it when: you have a ready file (CSV/JSON/Parquet) and want to load it now.
- Do not use it when: the data updates on its own in a database/SaaS app (prefer a scheduled pull) — push is a load that you trigger.
Plan and permissions#
- Product core · access check · ingestion run kill switch.
Prerequisites#
- The file in CSV, JSON, or Parquet.
- For CSV: knowing the delimiter (comma, semicolon, tab, or pipe).
- For JSON: knowing whether it is lines (JSONL/ndjson) or a single array.
- Parquet is self-describing — no format options.
Step by step#
- In the console, go to ETL/ELT → Connections and choose the format: CSV, JSON, or Parquet (category Files), in You upload mode (push).
- Adjust the format options:
- CSV → Delimiter:
Comma (,),Semicolon (;),Tab, orPipe (|). - JSON → Format:
Lines (JSONL / ndjson)orSingle array. - Parquet → nothing to configure.
- CSV → Delimiter:
- Use the upload area to send the file to the landing zone.
- Trigger the load in Bronze tables → Ingest live (or let the schedule take care of it).
- Follow the steps live:
validar → landing → bronze → registrar.
Example (synthetic)#
A sales CSV with semicolon as the delimiter (vendas.csv):
pedido_id;data_pedido;uf;categoria;preco_unitario;quantidade;status
P-1001;2026-01-03;SP;Bebidas;12,90;3;pago
P-1002;2026-01-03;RJ;Limpeza;29,90;1;pago
P-1003;2026-01-04;MG;Bebidas;12,90;2;canceladoThe same data in JSONL (one line per object):
{"pedido_id":"P-1001","data_pedido":"2026-01-03","uf":"SP","preco_unitario":"12,90","quantidade":"3","status":"pago"}
{"pedido_id":"P-1002","data_pedido":"2026-01-03","uf":"RJ","preco_unitario":"29,90","quantidade":"1","status":"pago"}Expected result#
- The file reaches the landing zone and is loaded into Bronze with each column's type detected from its values (number, date, text) and the technical stamps. No row is discarded in Bronze.
- Watch out for digits-only identifiers (CPF without punctuation, CEP, product
code like
00123): detection reads them as a number and the leading zero is lost in Bronze. Send these fields with punctuation (071.518.203-05) or with a non-numeric character, and check the column after the first load. - The
registrarstep updates the catalog and the last sync. - Then run the pipeline to take it to Silver/Gold.
Automatic sanitization (why the load does not break over a column name)#
If a header has an accent, a space, or comes in empty ("preço", "nome do cliente"), a
direct load would fail (Invalid field name). When that happens, Ingestia sanitizes
the file and reloads it: lowercase, no accents, _ instead of spaces; an empty column
becomes coluna; a name that starts with a number becomes col_.... It also unwraps
envelopes ({meta..., data:[...]}) and converts arrays/pretty-JSON to NDJSON. Conservative
rule: an unparseable line is kept as is — Ingestia never invents data.
Limits and costs#
- Load into Bronze: default
batchSize5,000,maxRows200,000. - Loading into the datalake is billable.
Security#
- The file goes to your workspace's landing zone (
landing/{prefix}/...); it never crosses paths with another customer. - Do not put secrets or unnecessary PII in the file — PII columns are masked in profiling, but it is best not to bring in what you do not need.
Common errors#
| Symptom | Likely cause | What to do |
|---|---|---|
| All columns in a single one | wrong CSV delimiter | Select the correct delimiter (; is common in Brazil) |
| JSON does not load | wrong format selected | Switch between Lines (JSONL) and Single array |
Invalid field name | header with accents/spaces | Sanitization reloads it on its own; check the resulting names |
| CPF/CEP without the leading zero | the column has only digits and was read as a number | Send the field formatted (with dots/hyphen) and recreate the table; in Silver use LPAD(CAST(col AS STRING), 11, '0') for CPF |
Troubleshooting#
- On the run screen, check the steps
validar → landing → bronze → registrar. - In the Bronze layer, compare the number of ingested rows with the file's.
Related#
Last reviewed: 2026-09-08.