Configuration

Radish uses a single YAML file (radish.yml) to centralize all tunable parameters. This makes it easy to adapt the server to different environments — development, testing, production — without touching any source code.

---

The Config File

The default configuration lives at radish.yml in the project root:

# radish.yml

network:
  host: "127.0.0.1"
  port: 9000

persistence:
  dir: "persistence"
  snapshots_subdir: "snapshots"
  aof_subdir: "aof"
  aof_filename: "radish.aof"
  num_snapshot_shards: 256

background_tasks:
  sync_interval_sec: 5
  cleaner_interval_sec: 0.1

concurrency:
  num_lock_shards: 256

ttl_cleanup:
  sampling_threshold: 100000
  sample_percentage: 0.10

data_limits:
  list_display_limit: 50

---

Parameter Reference

Network

Parameter	Default	Description
host	127.0.0.1	Address the server binds to. Use 0.0.0.0 to accept connections from any interface (required for Docker)
port	9000	TCP port the server listens on

Persistence

Parameter	Default	Description
dir	persistence	Root directory for all persistence data
snapshots_subdir	snapshots	Subdirectory (relative to dir) for RDB shard files
aof_subdir	aof	Subdirectory (relative to dir) for the append-only file
aof_filename	radish.aof	Name of the AOF file
num_snapshot_shards	256	Number of RDB shard files. Must match concurrency.num_lock_shards

Background Tasks

Parameter	Default	Description
sync_interval_sec	5	Seconds between RDB snapshot syncs. Lower values mean less data loss on crash but more disk I/O
cleaner_interval_sec	0.1	Seconds between TTL expiration cleanup runs. Controls how quickly expired keys are reclaimed

Concurrency

Parameter	Default	Description
num_lock_shards	256	Number of ReadWriteLock partitions in the ShardedLock. Must match persistence.num_snapshot_shards

TTL Cleanup

Parameter	Default	Description
sampling_threshold	100000	If the total number of keys is below this value, the cleaner checks all keys for expiration. Above this threshold, it samples a random subset
sample_percentage	0.10	Fraction of keys to sample when above the threshold (e.g., 0.10 = 10%)

Data Limits

Parameter	Default	Description
list_display_limit	50	Maximum number of elements returned by L_GET. Prevents accidentally dumping huge lists over the wire

---

How It Works

Configuration is loaded once at startup via init_config!(). The resulting RadishConfig struct is stored in a global CONFIG ref that every component reads from:

struct RadishConfig
    host::String
    port::Int
    persistence_dir::String
    snapshots_subdir::String
    aof_subdir::String
    aof_filename::String
    num_snapshot_shards::Int
    sync_interval_sec::Float64
    cleaner_interval_sec::Float64
    num_lock_shards::Int
    sampling_threshold::Int
    sample_percentage::Float64
    list_display_limit::Int
end

If the YAML file is missing, all parameters fall back to their default values — so Radish works out of the box without any configuration.

---

Using a Custom Config Path

Both the server and client runners accept an optional third argument to specify a custom config file:

# Default config (radish.yml in project root)
julia server_runner.jl

# Custom config path
julia server_runner.jl 0.0.0.0 9000 /etc/radish/production.yml

Command-line arguments for host and port override the values from the config file, giving you layered configuration: YAML defaults < CLI overrides.

---

Important Constraints

num_lock_shards and num_snapshot_shards must be equal. The snapshot system uses the same hash function as the ShardedLock to partition keys into shards. If these values don’t match, snapshot files and lock partitions will be misaligned, leading to incorrect incremental saves.

---

Tuning Guide

Scenario	What to Change
Development	Defaults are fine. Low traffic, small datasets
High write throughput	Lower sync_interval_sec (e.g., 1) to reduce data loss window. Increase shard count if contention is high
Large datasets (millions of keys)	Increase sampling_threshold and/or lower sample_percentage to reduce cleaner overhead
Docker / remote access	Set host to 0.0.0.0
Memory-constrained	Lower list_display_limit to reduce response sizes