From transactions to analytics: where do we go from here?

A presentation at DataEngBytes 2026 in March 2026 in Auckland, New Zealand by Marta Paes

Slide 1

Slide 1

From transactions to analytics Where do we go from here? DataEngBytes 2026 Marta Paes Sr. Product Manager

Slide 2

Slide 2

2 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 3

Slide 3

6 years later… Why am I still talking about CDC? 3 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 4

Slide 4

Itʼs personal 🔁 Periodic polling Circa 2014 4 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY OLTP OLAP

Slide 5

Slide 5

Itʼs personal �� 🔁 Periodic polling Circa 2014 5 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY OLTP OLAP

Slide 6

Slide 6

Itʼs personal �� �� 🔁 Periodic polling Circa 2014 6 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY �� OLTP OLAP

Slide 7

Slide 7

Itʼs personal ⚡ Log replication inserts, updates, deletes Circa 2018 OLTP WAL Write-Ahead Log 7 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY OLAP

Slide 8

Slide 8

Itʼs personal ⚡ Log replication inserts, updates, deletes Circa 2021 OLTP WAL Write-Ahead Log 8 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY OLAP

Slide 9

Slide 9

Itʼs personal ⚡ Log replication inserts, updates, deletes Today OLTP WAL Write-Ahead Log 9 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY OLAP

Slide 10

Slide 10

What this talk is about 10 01 How it all started 02 Where we hit a wall 03 What works for the 99% 04 Where do we go from here? ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 11

Slide 11

01 How it all started

Slide 12

Slide 12

How it all started Like most things, at ✨ internet-scale companies ✨ 2013 Databus LinkedIn), Wormhole Facebook), MoSQL Stripe 12 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 13

Slide 13

How it all started Like most things, at ✨ internet-scale companies ✨ Maxwell Zendesk), Bottled Water Confluent) 2013 2015 Databus LinkedIn), Wormhole Facebook), MoSQL Stripe 13 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 14

Slide 14

How it all started Like most things, at ✨ internet-scale companies ✨ Maxwell Zendesk), Bottled Water Confluent) 2016 2013 2015 Databus LinkedIn), Debezium Red Hat), Wormhole Facebook), MySQL Streamer Yelp MoSQL Stripe 14 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 15

Slide 15

How it all started Like most things, at ✨ internet-scale companies ✨ Maxwell Zendesk), Bottled Water Confluent) 2016 2013 2015 15 Debezium Red Hat), Wormhole Facebook), MySQL Streamer Yelp ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY 2019 2018 Databus LinkedIn), MoSQL Stripe Spinal Tap Airbnb DBLog Netflix)

Slide 16

Slide 16

02 Where we hit a wall

Slide 17

Slide 17

Debezium as the standard Canonical change events, battle-tested at scale, still going strong after 10+ years 2016 17 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY Databases MySQL, Postgres Kafka required? Yes Snapshots Blocking, full table Delivery guarantees At-least-once only Direction Source only Governance Red Hat Deployment Kafka Connect

Slide 18

Slide 18

Debezium as the standard Canonical change events, battle-tested at scale, still going strong after 10+ years 2016 18 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY 2026 Databases MySQL, Postgres 10+ connectors Kafka required? Yes Optional (Engine, Server) Snapshots Blocking, full table Incremental, parallel Delivery guarantees At-least-once only Exactly-once possible Direction Source only Source, Sink Governance Red Hat Commonhaus Foundation Deployment Kafka Connect Kafka Connect, Engine, Server, Operator K8s

Slide 19

Slide 19

“Hey, what if… 19 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 20

Slide 20

“Hey, what if… 20 ● Abstract some complexity of deploying and operating Debezium from the end user. ● Enable advanced features like schema evolution using existing primitives. ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 21

Slide 21

“Hey, what if… 2020 Flink CDC 2021 Confluent CDC 2022 2022 Streamkap 21 RisingWave CDC ● Abstract some complexity of deploying and operating Debezium from the end user. ● Enable advanced features like schema evolution using existing primitives. ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 22

Slide 22

03 What works for the 99%

Slide 23

Slide 23

“We first tried to implement Postgres CDC in-house using Debezium but it was way too complex. I knew a managed product built by engineers, whose goal in life is to transform bits from Postgres into ClickHouse, would be better than anything we could do ourselves. 60TB Aurora database Hundreds of customers moving data with CDC 23 10B updates/month ms query latency

Slide 24

Slide 24

What the 99% needs The average user will trade off speed for convenience 🛠 Easy to deploy & manage 24 🚀 �� Built to scale Full-featured ● ClickOps for fast onboarding ● Parallel snapshotting ● Graceful degradation ● Managed service or binary ● OOTB schema evolution ● Resilient to Day 2 operations ● Infrastructure-as-Code IaC ● Transactional consistency ● Monitoring & notifications ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 25

Slide 25

CDC as a feature, not a product Can you spot the trend? ● 2024 Create a user experience that is tailored to long-time + SQL users. ● Have more control over performance and semantics. 2021

  • Fivetran 25 Snowflake 2024 2023
  • HVR ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY Datavolo Databricks
  • Arcion ClickHouse PeerDB

Slide 26

Slide 26

26

Slide 27

Slide 27

04 Where do we go from here?

Slide 28

Slide 28

Where do we go from here? Things Iʼm excited about for the upcoming 6 years 🦀 Next-gen CDC tools 28 🚀 �� Data lake CDC Unified data stacks ● PeerDB ● Best-of-breed OLTP ● Delta Lake CDC (via CDF) ● Supermetal ● Best-of-breed OLAP ● Iceberg CDC ● Artie ● CDC as a background task ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 29

Slide 29

Next-gen CDC tools Supermetal as a challenger to first-gen tools “I also have to mention the difference in developer experience: having a single binary and not relying on operators, clusters, and much configuration/glue code felt amazing!ˮ Yaroslav Tkachenko (source) 29 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 30

Slide 30

Unified data stacks With some additional magic tricks ✨ 30 ● Postgres for transactions ● ClickHouse for analytics ● CDC as a background task ● pg_clickhouse for pushdown ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 31

Slide 31

Unified data stacks With some additional magic tricks ✨ ● Postgres for transactions ● ClickHouse for analytics ● CDC as a background task ● pg_clickhouse for pushdown Join the Private Preview waitlist! 31 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 32

Slide 32

Data Lake CDC Transactional workloads, just…different? 32 ©2026 CLICKHOUSE INC., CONFIDENTIAL & PROPRIETARY

Slide 33

Slide 33

See you in 6 years!

Slide 34

Slide 34

If any of this sounds interesting… ✓ Sign up for ClickHouse Cloud ✓ $300 credit ✓ Check out our careers page!