Skip to main content
World Today News
  • Home
  • News
  • World
  • Sport
  • Entertainment
  • Business
  • Health
  • Technology
Menu
  • Home
  • News
  • World
  • Sport
  • Entertainment
  • Business
  • Health
  • Technology
Atrium Tests Snowflake MySQL CDC: 30-Second Latency, Preview Status, Cascade Delete Gap

Atrium Tests Snowflake MySQL CDC: 30-Second Latency, Preview Status, Cascade Delete Gap

November 18, 2026 Technology Desk Technology

The Snowflake Connector for MySQL pushed an update and an insert into Snowflake in about 20 to 30 seconds in a hands-on test published this month by Atrium AI — a latency window tight enough to keep an operational dashboard current, and one a batch pipeline would routinely miss by hours.

The Tech TL;DR:

  • Atrium AI loaded four Sakila tables (599 customers, 1,000 films, 16,044 rentals) into Snowflake with matching row counts, then saw an update and an insert replicate in roughly 20 to 30 seconds with no manual step.
  • The connector remains a Preview feature; per Snowflake’s documentation, general availability is “currently not on our product roadmap,” and Snowflake directs new MySQL replication work to the generally available Openflow Connector for MySQL.
  • Rows deleted via MySQL’s ON DELETE CASCADE never reach the binary log, so they never reach Snowflake — a data-consistency gap teams must plan around before production.

Test uses MySQL binary log replication

That framing opens Atrium AI’s report, and it captures why the test matters. A one-time export shows what was true when someone ran it; change data capture (CDC) keeps the destination current by replaying individual changes instead of re-copying whole tables. MySQL already records every insert, update, and delete in order in its binary log. A CDC tool reads that log and replays what happened.

Atrium AI built the test on a MySQL 8 database running in Docker with binlog logging configured as it would be in production, loaded with Sakila, MySQL’s reference dataset for a fictional DVD rental business.

The connector has two parts: a native application inside Snowflake installed from Snowflake Marketplace, and an Agent — a small program running wherever the database lives, in this case a second Docker container on the same laptop. According to Atrium AI, one Agent can serve multiple MySQL data sources it can reach directly, so a central Agent can cover several databases without new infrastructure on each server.

The arrangement relied on a straightforward handoff: Snowflake produces a configuration file containing a temporary credential, which the Agent swaps for a permanent key pair the first time it starts up. Atrium AI reported that the Agent then contacted Snowflake at brief, recurring intervals and discovered newly registered tables without any need for a restart.

Four tables, three stages, 30 seconds

The team pointed the Agent at four tables: customers, films, rentals, and payments. Within seconds it read each table’s structure — column names, data types, primary keys — with no manual definitions. It also mapped two MySQL types with no native Snowflake equivalent: ENUM and SET both landed as VARCHAR.

Every table moved through three stages: reading the table structure, copying existing data, then watching for new changes with no set end. Within a couple of minutes, all four tables had finished the initial copy and moved into ongoing CDC, with row counts matching on both sides.

Status indicators don’t prove changes are flowing, so Atrium AI made three changes in MySQL and watched for them in Snowflake. An updated customer name appeared with no manual trigger. A new customer insert was read from the binlog and replicated on its own.

InnoDB’s foreign-key constraint blocked a delete of a customer holding payment records at the source — the failed delete produced no change event, and the connector rightly replicated nothing, mirroring the true state of the source.

Test detail Result
Tables replicated 4 (customers, films, rentals, payments)
Source rows 599 customers, 1,000 films, 16,044 rentals
Live change latency About 20–30 seconds
ENUM / SET handling Mapped to VARCHAR

Cascading deletes never reach the binlog

In the Atrium AI setup the schema stopped the delete from happening at all, though MySQL additionally offers ON DELETE CASCADE, in which removing a parent row takes its child rows with it. Per Snowflake’s documentation, such cascade-driven deletions leave no trace in the binary log, meaning Snowflake receives only the parent row’s deletion while the dependent rows remain.

Atrium AI points out that Snowflake makes this statement with reference to the Openflow Connector for MySQL, which reads the identical binlog, so it should be treated as a limitation common to both connectors.

Openflow also ignores TRUNCATE statements, per Snowflake’s documentation, so rows removed that way would likewise stay behind in Snowflake.

The docs add other hard edges: tables without primary keys are not supported, a single transaction must fit into a binary log message of no more than 4GB or the source table is marked permanently failed and requires a full snapshot reload, and the ” character cannot appear in replicated schema, table, or column names.

An active Agent continuously reads all binlog events even for tables not added for replication, and per Snowflake’s documentation the Agent might crash due to lack of available memory if the log contains very large events such as updates of BLOB-like columns. Authorization to the source database works only via user and password; private-key authorization is not supported.

Connector journals enable Type 2 modeling

Underneath the tables analysts query, the connector keeps an independent record of each insert, update, and delete performed since replication began, captured in sequence within its dedicated JOURNALS schema. The day-to-day tables show only the latest version of each row — a Type 1 slowly changing dimension, which Atrium AI calls the right default for operational reporting.

When the question becomes “what was true on this date,” Type 1 is not enough. Because the journal already holds every change captured, building a Type 2 dimension with start and end dates means modeling on data already in hand.

Atrium AI notes the connector does not build the Type 2 table itself, and history from before replication began is not captured — which is why most teams should start CDC early, since history accumulates from day one.

Where Openflow fits for production

One clean test is a proof of concept; for production planning, connector status matters most. Snowflake’s documentation states the Preview connector’s move to general availability is “currently not on our product roadmap,” that “support for future bug fixes and improvements are not guaranteed,” and that the “new solution is available as Openflow Connector for MySQL and includes better performance, customizability, and enhanced deployment options.”

Openflow supports replication from MySQL 8.0 and 8.4 sources as well as MariaDB 11.4 and newer, either in near real time or on a scheduled basis, and it carries every table through the identical set of three stages Atrium AI watched during the test.

Snowflake’s engineering team explains how Openflow reads the MySQL binlog and streams changes into Snowflake, and Atrium AI advises running the test again on Openflow before anything goes to production, since it is configured differently and has its own limits.

Atrium AI flags four checks for an Openflow implementation: timing (which tables need near real time versus scheduled refresh), capacity (sizing the runtime for sustained change volume), row identity (each table needs a primary key or a NOT NULL unique index, or a logical key configured), and deletes (cascading deletes and TRUNCATE do not replicate).

Working through those checks is what separates a working demo from an architecture a business can depend on — the kind of evaluation work a Snowflake data engineering practice runs for every connector and migration, and the kind where vetted managed database consultants and data migration specialists earn their keep when a multi-source migration spans systems with different lag tolerances.

For teams replicating now, the Preview connector’s configuration is documented on Snowflake’s official connector documentation, and the Sakila test schema remains the reference dataset for MySQL CDC benchmarking, with the Agent distributed as a Docker image for containerized deployment in your own network.

-- Verify replication state and journal capture in Snowsight
SELECT TABLE_NAME, REPLICATION_STATE
  FROM CONNECTOR_UI.MYSQL.TABLES
 WHERE TABLE_NAME IN ('customers','films','rentals','payments');

-- Type 2 dimension skeleton built from the connector's journal
CREATE OR REPLACE TABLE customer_scd2 AS
SELECT
  j.customer_id,
  j.name,
  j.__snowflake_inserted_at AS valid_from,
  LEAD(j.__snowflake_inserted_at)
    OVER (PARTITION BY j.customer_id
          ORDER BY j.__snowflake_inserted_at) AS valid_to
FROM customer_journal j;

Preview today, Openflow tomorrow

The contrast at the center of this story is stark: the connector Atrium AI tested performed well end to end, and the vendor itself says that connector has no path to general availability. Teams building on the Preview app are building on a feature whose own documentation warns bug fixes are not guaranteed — while the binlog constraint on cascading deletes applies to both paths, Preview and GA alike.

The FAQ below answers the questions a reader would still ask after the test results.

Always consult with certified IT and cybersecurity professionals before altering enterprise networks or handling sensitive data.

Share this:

  • Share on Facebook (Opens in new window) Facebook
  • Share on X (Opens in new window) X

Keep reading

  • Memory Spot Prices End 14-Month Climb, Weighing on Samsung and SK Hynix Shares
  • Democrats Boycott Senate Vote to Advance Republican FCC Nominee Danielle Thumann Severs
  • Cleveland Clinic Tests Virtual Reality for Cancer Distress (newsdirectory3.com)

Related

Search:

World Today News

World Today News is your trusted source for global journalism — breaking headlines, in-depth analysis, and reporting from around the world.

Quick Links

  • Privacy Policy
  • About Us
  • Accessibility statement
  • California Privacy Notice (CCPA/CPRA)
  • Contact
  • Cookie Policy
  • Disclaimer
  • DMCA Policy
  • Do not sell my info
  • EDITORIAL TEAM
  • Terms & Conditions

Browse by Location

  • GB
  • NZ
  • US

Connect With Us

© 2026 World Today News. All rights reserved. Your trusted global news source directory.
For contact, advertising, copyright, issues email: [email protected]

Privacy Policy Terms of Service