What every ETL destination does with your semi-structured fields

PostgreSQL

Incremental: a watermark on each entity's updated-at, applied with INSERT … ON CONFLICT DO UPDATE

Worth knowing: PostgreSQL folds unquoted identifiers to lowercase, so Datrise normalizes mixed-case source fields to snake_case

Upsert generator: INSERT … ON CONFLICT DO UPDATE

MySQL

Incremental: a watermark on updated-at, applied with INSERT … ON DUPLICATE KEY UPDATE

Worth knowing: MySQL collation matters for CRM text, so Datrise lands utf8mb4 to preserve emoji and non-Latin characters

Upsert generator: INSERT … ON DUPLICATE KEY UPDATE

Microsoft SQL Server

Incremental: a watermark on updated-at, applied with a MERGE statement

Worth knowing: SQL Server defaults to a case-insensitive collation, so Datrise preserves original casing in a metadata column to avoid silent key collisions

Upsert generator: MERGE … WITH (HOLDLOCK)

Oracle Database

Incremental: a watermark on updated-at, applied with MERGE INTO

Worth knowing: Oracle treats an empty string as NULL, so Datrise distinguishes blank source values from missing ones during load

Upsert generator: MERGE INTO … USING

Supabase

Incremental: a watermark on updated-at, applied with INSERT … ON CONFLICT DO UPDATE

Worth knowing: Datrise lands into a dedicated schema and leaves row-level security to you, so synced tables don't inherit public access by accident

Upsert generator: INSERT … ON CONFLICT DO UPDATE

Neon

Incremental: a watermark on updated-at, applied with INSERT … ON CONFLICT DO UPDATE

Worth knowing: Neon separates compute from storage, so Datrise batches writes to keep autoscaling compute from cold-starting on every small change

Upsert generator: INSERT … ON CONFLICT DO UPDATE

PlanetScale

Incremental: a watermark on updated-at, applied with INSERT … ON DUPLICATE KEY UPDATE

Worth knowing: PlanetScale disallows foreign-key constraints by default, so Datrise models relationships by stable id columns rather than enforced FKs

Upsert generator: INSERT … ON DUPLICATE KEY UPDATE

Snowflake

Incremental: staged loads merged on stable id with MERGE, so credits scale with change volume, not table size

Worth knowing: Snowflake upper-cases unquoted identifiers, so Datrise standardizes on lower-case quoted names to keep column references stable

Upsert generator: MERGE INTO … QUALIFY-deduped USING

Google BigQuery

Incremental: appends to a staging table, then MERGE on stable id into the partitioned target

Worth knowing: BigQuery bills by bytes scanned, so Datrise partitions and clusters every table to keep query costs predictable

Upsert generator: MERGE with partition-pruned target

Amazon Redshift

Incremental: COPY from staged files, then a delete-and-insert merge on stable id

Worth knowing: Redshift performance hinges on dist/sort keys, so Datrise picks them from your entity ids and sync timestamps rather than defaulting to EVEN distribution

Upsert generator: DELETE … USING + INSERT (staging pattern)

Databricks SQL Warehouse

Incremental: a Delta MERGE on stable id, with change history available via time travel

Worth knowing: Datrise writes Unity Catalog–governed Delta tables, so lineage and permissions are managed centrally rather than per-notebook

Upsert generator: Delta Lake MERGE INTO

Azure Synapse

Incremental: COPY into staging, then a MERGE on stable id

Worth knowing: Synapse dedicated pools reward good hash-distribution choices, so Datrise distributes on entity ids to avoid data-movement-heavy joins

Upsert generator: MERGE (dedicated SQL pool)

ClickHouse

Incremental: inserts into a ReplacingMergeTree keyed on stable id, so the latest version wins on merge

Worth knowing: ClickHouse deduplicates asynchronously on merge, so Datrise uses ReplacingMergeTree and FINAL-safe queries rather than assuming immediate upserts

Upsert generator: INSERT into ReplacingMergeTree + FINAL

DuckDB

Incremental: rewrites changed entities into the local database (or Parquet) on each run

Worth knowing: DuckDB is single-writer and embedded, so Datrise produces a consistent file snapshot rather than concurrent streaming writes

Upsert generator: INSERT OR REPLACE / ON CONFLICT DO UPDATE

Amazon Athena

Incremental: writes new Parquet partitions and registers them in the Glue Data Catalog

Worth knowing: Athena bills per byte scanned and small files hurt, so Datrise compacts to right-sized Parquet rather than many tiny objects

Upsert generator: MERGE INTO (Iceberg table, engine v3)

Amazon S3 Data Lake

Incremental: writes new date partitions and compacts small files on a schedule

Worth knowing: A lake has no schema enforcement, so Datrise writes a schema manifest alongside the data to keep downstream engines consistent

Upsert generator: append partition + latest-version view

Azure Data Lake Storage

Incremental: writes new date partitions to the container and compacts on a schedule

Worth knowing: ADLS hierarchical namespace makes folder layout matter, so Datrise keeps a predictable entity/date path your Azure engines mount directly

Upsert generator: append partition + latest-version view

MongoDB

Incremental: upserts by stable id with updateOne(upsert) on the source primary key

Worth knowing: Mongo has no fixed schema, so Datrise keeps field types consistent across documents to avoid mixed-type query surprises

Upsert generator: bulkWrite updateOne { upsert: true }

Amazon DynamoDB

Incremental: PutItem/UpdateItem keyed on a partition key derived from the entity id

Worth knowing: DynamoDB rewards access-pattern-first key design, so Datrise sets partition/sort keys from your entity ids rather than scan-heavy defaults

Upsert generator: PutItem with ConditionExpression

Spreadsheets

Incremental: refreshes the tab's rows each run, preserving header order

Worth knowing: Sheets caps out around the low millions of cells, so Datrise lands a curated column set rather than every raw field

Airtable

Incremental: upserts records matched on a stable id field

Worth knowing: Airtable enforces per-base record and API rate limits, so Datrise batches writes and lands a focused field set

Upsert generator: PATCH /records with performUpsert

CSV Files

Incremental: writes a fresh, fully-typed CSV per entity each run

Worth knowing: CSV has no types, so Datrise emits a companion schema and quotes/escapes consistently so downstream loaders don't misparse commas and newlines