chDB (@chdb_io) profile picture

https://teapawt.org/t.co/vlD2sAhx3c

localhost
Joined September 2023
Photo by chDB (@chdb_io)Photo by chDB (@chdb_io)Photo by chDB (@chdb_io)Photo by chDB (@chdb_io)Photo by chDB (@chdb_io)Photo by chDB (@chdb_io)Photo by chDB (@chdb_io)Photo by chDB (@chdb_io)Photo by chDB (@chdb_io)Photo by chDB (@chdb_io)
With import chdb.datastore as pd, an agent's first look stays cheap: df.head() runs as LIMIT 5; len(df) and df.dtypes never build the DataFrame. pandas loads it all first. 5M tool-call rows, 960 MB Parquet: pandas: 0.7-1.3 s, 2-3 GB RSS chDB: 0.1 s, 0.5 GB Same 5 rows.
1
60
chDB chdb-core v26.11.1 is out, on ClickHouse 26.11. Polars in and out: query a DataFrame, LazyFrame or Series in place, get a Polars DataFrame back. pip install -U chdb-core chdb.query("SELECT sum(a) FROM Python(lf)", "polars") github.com/chdb-io/chdb-core…
1
4
464
Agent needs memory for 100 questions at once (eval run, multi-step plan)? Skip the 100 vector queries. Put the query embeddings in a chDB table and JOIN: 100 x top-3, 1M memories, per user: loop: 430 ms one JOIN: 190 ms JOIN + user IN (...): 48 ms Same 300 rows each way.
1
4
177
Your next on-call incident is a dating sim. High Cardinality lets you date Logs, Metrics and Traces, with chDB running ClickHouse in your browser. One of them only shows up to 1% of your dates. Naturally. highcard.love/
74
Parquet integrity test: flip one byte inside a checksummed data page, then sum the column. In a June run, chDB 4.1.8 raised "Page CRC checksum verification failed" by default. pyarrow ships the check but leaves it off, so it returned a wrong sum. github.com/flying-coyote/sdw…
2
130
CI lesson from Cerberus by @tsouza: chDB already spreads one query across cores, so more test processes per runner backfired (8 on an 8-core box ran slower than 4). Their 900+ PromQL roundtrip fixtures now run ~3 processes per runner over 6 runners. github.com/tsouza/cerberus/b…
1
104
chDB retweeted
Someone built a dating sim where you date Logs, Metrics and Traces. Traces only shows up to 1% of your dates 😂 It runs real ClickHouse in your browser via chDB WASM. You can turn on SQL to see the checks behind the story. I had to check. It works. highcard.love/
1
1
7
450
Tip for agents that only get a shell tool: don't let them cat a big CSV into context. Give them one line: python -m chdb "SELECT status, count(), quantile(0.95)(ms) FROM file('access.csv') GROUP BY status" TSV 2M rows, 112 MB: 0.26s end to end, 4 rows back. No server.
90
Everr's local mode runs a fork of the upstream OTel ClickHouse exporter on embedded chDB. @everrlabs swapped the network connection for an in-process chDB handle and turned batch inserts into JSONEachRow, so laptop and cloud share one schema. github.com/everr-labs/everr/…
3
1
4
210
If a distributed #ClickHouse® query is slow, total runtime only tells you so much. What was each node actually doing? 🤔 We traced a GROUP BY across 2 shards/2 replicas to see which nodes handle the query & how partial aggregates get merged in this demo: hubs.la/Q04DS01W0
1
1
134
Nix users can now pin chDB like any other dependency: engram packages libchdb, the Python module and a chdb-rust example, hash-pinned and checked on nixpkgs unstable, 26.05 and 25.11 (x86_64-linux for now). github.com/levigross/engram
71
Give your data agent a cheap "did anything change?" tool before it diffs rows: SELECT _file, count() n, sum(cityHash64(*)) h FROM file('exports/*.parquet') GROUP BY _file Order-independent, no ETL. On 10M rows in chDB: 0.22s, 0.4 GB. A full two-way EXCEPT took 3.2s and ~4 GB.
1
111
Every coding-agent session shouldn’t have to start from zero. Loom is an open-source agent memory system built on ClickHouse. It keeps what agents learn connected to a live code index, so the next session starts with context that still matches the code. Meet Loom: clickhouse.com/blog/introduc…
18
13
4
84
7,106
chDB retweeted
Welcome aboard!
🚀 I’m incredibly excited to share that Xata is joining ClickHouse! Building Xata over the past five years has been an incredible journey. I’m especially proud of the technology we’ve built, the hard engineering challenges we’ve tackled, and the amazing team behind it all.
2
1
9
661
Replying to @auxten
Xata's copy-on-write Postgres branches next to ClickHouse feels like a natural pairing: spin up a branch of prod for a preview env and point analytics at it without touching the primary. Curious whether that branching ends up in ClickHouse's managed Postgres.
1
1
1
117
chDB retweeted
Replying to @monicasarbu
It is exciting to imagine: Xata, our managed PostgreSQL team, and pg_chdb coming together to create a product with such incredible technical vision!
1
2
8
200