ClickHouse Makes a MergeTree Part Visible with One Directory Rename
A MergeTree part is written to a temp directory through a stack of write buffers and published by one atomic directory rename.
Internals
ClickHouse Parses URLs with a Hand-Written State Machine
ClickHouse’s URL functions share a hand-rolled byte scanner instead of using a regex engine.
ClickHouse Prunes Granules with a Bloom Filter Skip Index
ClickHouse’s bloom_filter skip index keeps one bitset per granule and derives k bit positions from one 64-bit hash.
ClickHouse Runs Regular Expressions Four Different Ways
ClickHouse analyses a regexp before running it, and often never runs RE2 at all.
ClickHouse Shares a CPU Budget Between Queries with Fair CPU Slots
ClickHouse shares a global CPU-slot budget between queries with a max-min-fair scheduler, where each query’s slot ceiling is mutable and raised only when the pipeline finds real …