Configuration

aki reads its configuration the way Redis does. You set directives at startup, and you change most of them at runtime over the wire.

At startup you pass the common ones as flags to aki server (see the CLI reference). At runtime you use the CONFIG family.

redis-cli CONFIG GET maxmemory
redis-cli CONFIG SET maxmemory 256mb
redis-cli CONFIG REWRITE
redis-cli CONFIG RESETSTAT

CONFIG GET reads a directive (globs work, so CONFIG GET max* returns all matches). CONFIG SET changes one. CONFIG REWRITE persists the live values. CONFIG RESETSTAT clears the runtime statistics.

aki registers the full Redis directive surface, around 207 directives, so tooling that enumerates CONFIG GET * sees everything it expects. Not all of those change behavior. The next two sections split them: the ones that do, and the ones that are accepted for compatibility but inert.

Directives that change behavior

These actually change how aki runs.

Directive	Notes
maxmemory	Memory ceiling before eviction kicks in.
maxmemory-policy	Eviction policy: noeviction, allkeys-lru, volatile-ttl, and so on.
appendonly	Turn the append-only log on or off.
appendfsync	Durability policy: always, everysec (default), or no.
no-appendfsync-on-rewrite	Pause fsyncs during a log rewrite to keep latency steady.
aof-load-truncated	Tolerate a half-written record at the tail on load. Default yes.
aof-timestamp-enabled	Write #TS:<ms> annotations before records.
save	Snapshot points, as seconds changes pairs.
databases	Number of logical databases. Default 16, set at file creation.
maxclients	Maximum connected clients. Default 10000.
requirepass	Password for the default user.
timeout	Close idle client connections after this many seconds.
tcp-keepalive	TCP keepalive interval.
hash-max-listpack-entries, hash-max-listpack-value	Hash small-encoding limits.
list-max-listpack-size	List small-encoding limit.
set-max-intset-entries	Integer-set encoding limit.
set-max-listpack-entries, set-max-listpack-value	Set small-encoding limits.
zset-max-listpack-entries, zset-max-listpack-value	Sorted-set small-encoding limits.
notify-keyspace-events	Keyspace notification flags.
acl-pubsub-default	Default channel access for new users. Default resetchannels.
client-query-buffer-limit	Maximum size of a single client query buffer.
proto-max-bulk-len	Maximum bulk string size accepted on the wire.
lua-time-limit, busy-reply-threshold	How long a script may run before it is reported as busy.
slowlog-log-slower-than	Microsecond threshold for the slow log.
slowlog-max-len	Maximum slow-log length.
latency-monitor-threshold	Millisecond threshold for the latency monitor.

A few defaults worth stating, since pages elsewhere refer to them: appendfsync is everysec, aof-load-truncated is yes, acl-pubsub-default is resetchannels, databases is 16, and maxclients is 10000.

Accepted for compatibility, not acted on

A large set of directives, around 103, stay registered so CONFIG GET and CONFIG SET report the full Redis surface, but they do not change behavior. The subsystem they configure does not exist in aki. Setting one is honest about this: the value is stored and returned, and nothing reads it. This matches how Redis itself treats configs its current build does not act on.

The inert groups, and why:

• Manual memory management (active defragmentation, lazyfree frees, maxmemory-clients). The Go runtime allocates, frees, and compacts memory on its own, so there is nothing for these to drive.
• Clustering (the cluster-* family). aki is a single node and runs no cluster bus.
• TLS transport (the tls-* family, metrics-tls). aki ships no TLS listener. Terminate TLS with a proxy in front.
• IO threading (io-threads, io-threads-do-reads, io-uring). aki serves each connection on its own goroutine and lets the Go scheduler place the work. There is no shared IO pool to size.
• Daemonization (daemonize, supervised, pidfile, proc-title-template, and similar). aki is one process started by its command, not a forking daemon.
• Listener tuning (bind, tcp-backlog, unixsocketperm, protected-mode). The address and socket are chosen when the server binds, not re-read from the live store.
• Storage engine internals (page-size, buffer-pool-size, wal-autocheckpoint, the aki-* checkpoint knobs, aki-filename). The pager and WAL run on built-in defaults, and the page geometry is fixed in the file header when the file is created.
• Compression and encryption (compression, rdbcompression, encryption, and their keys). aki stores values as is and ships no codec, which would pull in non-stdlib code.
• RDB and AOF writer knobs that do not map (rdb-compat, rdb-save-incremental-fsync, aof-rewrite-incremental-fsync, aof-use-rdb-preamble). aki's snapshot is its own file codec, and its AOF rewrite is inline, so these writer steps do not exist.
• Replication transport tuning (repl-diskless-sync, repl-timeout, repl-backlog-ttl, the replica-announce-* pair, and similar). aki's replica link is its own command stream, not the Redis fork-and-socket path.
• Output buffer limits (client-output-buffer-limit). aki has no growing per-client output buffer to bound. A slow consumer slows the writer rather than piling up in a buffer.
• Listpack node sizing (list-compress-depth, stream-node-max-bytes). The flat store keeps each value in one structure, so there are no listpack node chains to split. The element-count encoding limits above are wired.
• Collation (locale-collate). The collation tables live in a non-stdlib package, and aki is zero-dependency, so sort stays on byte and numeric order.

There are a handful of always-on or aki-specific entries too: sanitize-dump-payload (aki always fully parses a payload), shutdown-timeout (SHUTDOWN exits with no fork or replica handoff), and the aki-only active-expire-* and acllog-max-entry-bytes entries.

The full authoritative breakdown lives in the spec. The short version: if a directive names a subsystem aki does not have, you can set it, tooling will see it, and it will not do anything.

For how the durability directives behave in practice, see the persistence guide.