localPoolSize

Number of pooled connections held open to each node in the local datacenter. Wired to the driver's advanced.connection.pool.local.size (CONNECTION_POOL_LOCAL_SIZE) option (GH #80 / ISS-072); the driver's own default -- and this field's default -- is 1.

Each pooled connection multiplexes up to maxRequestsPerConnection concurrent requests, so the real per-node concurrency ceiling is localPoolSize * maxRequestsPerConnection. Raise this only after profiling actual per-node concurrency under real load -- more connections isn't free (each is a TCP socket plus driver-side heap bookkeeping), and going from 1 to a handful of connections per node is almost always enough headroom before maxRequestsPerConnection becomes the real bottleneck. Note Kandra uses the stock OSS DataStax driver, not a shard-aware ScyllaDB driver, so pool size alone can't route directly to the owning shard on wide nodes (see docs/issues/ISS-069-assorted-low-severity-findings.md item 4) -- widening the pool is a coarser lever than shard-aware routing would be, but it's the lever Kandra currently exposes.