Product @ ClickHouse Thinking about agentic experience & identity
Joined April 2015
- Tweets68
- Following64
- Followers377
- Likes3.4K
1/5 ClickHouse 26.10 (@ClickhouseDB) can read less data for GROUP BY queries ordered by key with LIMIT. The aggregation's Top-K boundary lets the scan skip rows and granules whose keys cannot enter the result.
4/5 All rows of the retained groups still contribute to their aggregates. ORDER BY count() DESC is a different problem; this is a boundary on group keys, not on aggregate scores.
This is a first for me; I sent an email to an "enterprise" vendor we use, asking a simple question. It was clearly forwarded directly to LLM that proceeded to randomly delete data across our account. Then sent me a slop response that didn't even answer my question.
1/4 ClickHouse 26.10 can pack an array of yes/no values into a bitmask, with a lambda deciding each bit. arrayPackBitsToUInt64(c -> c = 'Y', ['Y', 'N', 'N', 'Y']) returns 144, whose packed byte is 10010000.
3/4 String keeps the whole bitstream, FixedString(n) pads or truncates it to n bytes, and UInt64 keeps at most 64 bits. The group variants pack several low bits per element, so 4-bit codes [1, 2, 3, 15] become 0x123F.
Agree with every word from Proton's CTO
1/5 Monitoring dashboards with moving time windows can reuse query-condition-cache entries in ClickHouse 26.10, even when the filter uses now(). The cache key is rounded while the actual query filter stays exact.
4/5 An aligned bound such as today() - 10 can reuse entries in the same grid cell, while an unaligned sliding lower bound primes the next cell. An unaligned upper bound such as time < now() doesn't get that later-query reuse.
Al Brown retweeted
My 7+ years of #podcast charts and rankings from Apple and Spotify is finally migrated completely off MongoDB and onto ClickHouse and Postgres!
I've cut the umbilical cord and I look forward to seeing how much lower my next @Hetzner_Online bill will be now that I'm not paying for 9 TB of volume storage!
(I'll soon be able to shut down some extra servers, too.)
This should save me about $700/month. And this also means I can finally implement a backup strategy because not only is it much easier to back up 400 GB than 4.5 TB, but ClickHouse and Postgres are also a lot smarter for backing up, supporting incremental backups!
Now to move the rest of my apps off MongoDB and be done with it forever.
I don't think we should fight. The two companies are complementary. Since you're not doing this AI stuff, I really think Snowflake should consider getting its AI tokens through Databricks. You can get both cheap OSS models as well as frontier (OAI/Ant/Gemini/Grok) models from our platform. It all goes through Unity Gateway which has Smart Routing, Budgets, and Access Control. You can even use Omnigent with any harness, including CoCo. Most of our customers used this to reduce token spend ~30% last few months while token count went up. Genie lets you then track, control, manage that spend, all your financials, marketing, etc. Quick POC?
3/4 SELECT a - b and SELECT b - a get the same unordered hash, as can SELECT 1 and invalid text such as 1 SELECT. This is deliberately lossy workload grouping.