ClickHouse Prices Distributed Plans in a Cascades Memo: Work, Network and Sequential Cost
ClickHouse’s new Cascades optimizer treats a distributed query as a search problem.
Clickhouse
ClickHouse Starts the Pipeline with One Thread: Work Stealing Queues and Downscaling Under Load
ClickHouse doesn’t hand you max_threads threads. It starts with one, measures how much parallel work the pipeline actually creates, and grows the pool only when idle…
ClickHouse Packs Short Strings by Size Class: 24 Bytes Fit in Three UInt64s
ClickHouse never stores a std::string in a hash table cell. It splits keys into size classes and packs strings of up to 24 bytes into plain integers, so your GROUP BY…
ClickHouse Aggregates Sorted Keys Without a Hash Table: Counting Runs and Pushing LIMIT Down
If your table is sorted by the GROUP BY key, ClickHouse can skip the hash table entirely: it counts each contiguous run of equal keys and emits it.
ClickHouse Merges Columns One at a Time: A Byte per Row Caps Merges at 127 Parts
ClickHouse’s vertical merge splits a merge into two stages: merge the key and record where each output row came from (one byte per row), then gather each payload column…