
Tao Wan
Co-founder at Oodle. Previously at Snowflake, Roku and Amazon.

Part 7
JSON and 217 small files
A ClickHouse JSON column limited to 5 paths had 217 mark files per part. How cold mark caches slowed our log searches, and the setting that fixed it.

Part 6
Move to Altinity?
Our 2025 PoC of moving logs from ClickHouse Cloud to Altinity BYOK: query results, the original pros and cons one-pager, and why we did not move.

Part 5
Materialized views that count wrong
Our ClickHouse MVs got the totals right and the filters wrong. Merge keys, lagging minutes, agent queries, and how to check your own MVs.

Part 4
Queries, inserts and merges share the same CPU
Two 14 minute log searches starved ClickHouse inserts and merges, and agents' LIKE scans made 46.7% of queries fail.

Part 3
Merges That Run Out of Memory
After a downsize, two ClickHouse merges failed on the memory limit for 19 hours. What to check before shrinking a service.

Part 1
Too many parts
The three ways ClickHouse throws TOO_MANY_PARTS, what fixed each one, and the alerts to set up before inserts start failing.

Part 2
TTL
Raising retention rewrote a whole ClickHouse table, and expired data kept the CPU busy every 4 hours. Two TTL settings fixed both.