BigQuery destination for ELT & CDC

Overview

Integrate.io uses BigQuery’s streaming insert API to ingest data into BigQuery. This requires billing to be enabled on the Google Cloud project.

Requirements

Billing must be enabled on the Google Cloud project.
Ensure you have permission to create a service account with the BigQuery Data Editor role on the project.

Setting up BigQuery for ELT & CDC

Create service account

Creating a service account for FlyData to use requires the BigQuery Data Editor role.

Go to your project’s service accounts page and click + CREATE SERVICE ACCOUNT.

Google Cloud service accounts page with Create Service Account button

Enter the Service account name and add a description.

Entering service account name and description

Click CREATE AND CONTINUE.

Select the BigQuery Data Editor role.

Selecting the BigQuery Data Editor role for the service account

Select the BigQuery Job User role.
Click CONTINUE.

Click DONE.

The service account should have been created successfully. Find it in the service accounts list and click it.

Go to the KEYS tab.

Click ADD KEY and select CREATE NEW KEY.

Keys tab with Add Key dropdown showing Create New Key option

Select JSON and click CREATE.

Selecting JSON key type for the service account credentials

A JSON file should have been downloaded to your system. Upload it.We will fetch the client_email and private_key properties from the JSON credential file.

Limitations

BigQuery doesn’t support deduplication. Instead, the following columns are added to the destination table for future deduplication on the customer’s side. __offset - Offset column. Keeps track of the offset in incremental order to determine which existing row is the latest.
__op - Operation column. Tells what kind of operation is performed for a particular row (0 - Created, 1 - Inserted, 2 - Updated, 3 - Deleted) This query is to list the de-duplicated records:

SELECT customer.*
FROM (SELECT id,
             ARRAY_AGG(
                     t ORDER BY t.__offset DESC LIMIT 1
                 )[OFFSET(0)] customer
      FROM \`euphoric-effect-296205.northwind.customer\` t
      GROUP BY id);

This query is to upsert a staging table that Integrate.io is syncing to, into a production table used for querying:

MERGE INTO \`northwind\`.customers o
    USING (SELECT event.*
           FROM (SELECT customer_id,
                        ARRAY_AGG(
                                t ORDER BY t.__offset DESC LIMIT 1
                            )[OFFSET(0)] event
                 FROM customers_staging_1 t
                 GROUP BY customer_id)) as s
    ON o.customer_id = s.customer_id
    WHEN MATCHED AND s.__op = 3 THEN
        DELETE
    WHEN MATCHED THEN
        UPDATE SET o.company_name = s.company_name, o.__op = s.__op
    WHEN NOT MATCHED THEN
        INSERT ROW

After deleting tables on BigQuery, we have to wait (5+ minutes) to let the change propagate to all the machines in the BigQuery cluster. Otherwise, it might lead to some rows missing. This happens on resync requests. Please contact support@integrate.io to assist with the resync timing of user tables.

Table Naming

Source table names are converted to snake case when the destination tables are created in BigQuery. The conversion follows these rules:

All letters are lowercased.
An underscore is inserted at a lowercase-to-uppercase boundary (for example, MyTable becomes my_table).
An underscore is inserted at a digit-to-letter transition (for example, A100B becomes a100_b).
No underscore is inserted at a letter-to-digit transition (for example, ITEM01 stays item01).
Non-alphanumeric characters such as spaces and hyphens are replaced with underscores.

Note that the digit rule is asymmetric: a name is split only where a digit is followed by a letter, never where a letter is followed by a digit. This mainly affects short, code-style table names.

Source table	BigQuery table	Why
`INVOICES`	`invoices`	All letters, no transitions, only lowercased
`REV2`	`rev2`	Letter-to-digit only (`V` to `2`), which does not split
`ITEM01`	`item01`	Letter-to-digit only (`M` to `0`), which does not split
`Q1SALES`	`q1_sales`	Digit-to-letter transition (`1` to `S`) splits
`Order2023Items`	`order2023_items`	Splits at `3` to `I` (digit to letter), not at `r` to `2`
`MyTable`	`my_table`	Splits at the lowercase-to-uppercase boundary

Because of the lowercasing, source tables whose names differ only by case (for example, Orders and ORDERS) map to the same BigQuery table name.

Data Type Mapping - BigQuery

IP Allowlist

Initial Sync Process

Last modified on July 9, 2026

Amazon S3 destination bucket setup for ELT & CDC

MySQL destination for ELT & CDC

⌘I

Getting Started

Pipelines

Sources

Destinations

Monitoring

Security

Transforms

Data Observability

BigQuery destination for ELT & CDC

Overview

Requirements

Setting up BigQuery for ELT & CDC

Create service account

Limitations

Table Naming

Data Type Mapping - BigQuery

IP Allowlist

Initial Sync Process

​Overview

​Requirements

​Setting up BigQuery for ELT & CDC

​Create service account

​Limitations

​Table Naming

​Related

Data Type Mapping - BigQuery

IP Allowlist

Initial Sync Process

Overview

Requirements

Setting up BigQuery for ELT & CDC

Create service account

Limitations

Table Naming

Related