Production serving profiles

Environment presets for bulk ingest vs query serving

ToraDB uses the same binaries for ingest and serving; tune behavior with environment variables. The demo applies presets via TORADB_PROFILE in demo/run.sh.

Profiles

Profile	TORADB_CACHE_*	TORADB_WARMUP_ON_START	Typical workload
ingest	TORADB_CACHE_INDEX_BYTES=256MB (low)	0	toradb-ingest, resume, bulk load
serving	TORADB_CACHE_AUTO=1 or explicit 4–8GB	1	API, repeated queries

Ingest

• Keep index cache small so ingest does not compete with segment writes.
• Disable warmup; no query path yet.
• After bulk load, run toradb-ingest resume --db PATH --table TABLE to build TBM3 sidecars, lexicons, and optional route index.

export TORADB_PROFILE=ingest
./target/release/toradb-ingest resume --db ./data/my_table --table passages

Serving

• Use TORADB_CACHE_AUTO=1 to size the mmap BM25 cache from available RAM (TORADB_CACHE_INDEX_FRACTION, default 0.20, capped by TORADB_CACHE_AUTO_CAP_BYTES, default 8GB).
• Override with TORADB_CACHE_INDEX_BYTES when you need a fixed budget.
• Enable TORADB_WARMUP_ON_START=1 for APIs so the first user query is not cold.

export TORADB_PROFILE=serving
export TORADB_CACHE_AUTO=1
TORADB_DEBUG=1 ./demo/run.sh

Cache variables

Variable	Default	Purpose
TORADB_CACHE_INDEX_BYTES	256MB	Fixed LRU budget for segment BM25 mmap blobs
TORADB_CACHE_AUTO	off	Set 1 to derive index_bytes from system RAM
TORADB_CACHE_INDEX_FRACTION	0.20	Fraction of RAM for auto budget
TORADB_CACHE_AUTO_CAP_BYTES	8GB	Upper cap for auto budget
TORADB_CACHE_DEBUG	off	Log chosen cache budget at startup

See also Search strategies and On-disk layout.