Valdr

Valkey, in safe Rust. Alpha.

Single-node Valkey workalike. Safe-Rust data engine, RESP parser, and TLS stack. No fork(), no C module ABI. The Lua VM still wraps the upstream C interpreter — see MLUA_EXIT_PLAN.md for the path to replace it.

Coverage

  Counted assertions             ██████████  3,015 / 3,015
  Single-node-core blocks        █████████░  2,466 / 2,541   · 97%
  Full upstream Valkey suite     ██████░░░░  2,466 / 4,299   · 57%

The 1,758-block gap to 4,299 is out of scope — cluster, modules, sentinel, replication-integration, platform. Not failing. Verified 2026-05-28 against Valkey 9.1.0; bucketing in docs/TEST_AND_FEATURE_COVERAGE.md.

Scope

  Strings, lists, hashes, sets, sorted sets  ████████████  full
  Streams + consumer groups                  ████████████  full
  Pub/Sub (regular + sharded)                ████████████  full
  Transactions (MULTI/EXEC/WATCH)            ████████████  full
  Scripting (EVAL/EVALSHA/FCALL)             ████████████  via mlua
  ACL, AUTH, multi-DB                        ████████████  full
  Expiration, maxmemory eviction             ████████████  full
  RDB persistence                            ████████████  bidirectional
  TLS                                        ████████████  rustls 1.2/1.3, mTLS

  AOF persistence                            ███████░░░░░  alpha
  Replication                                ██████░░░░░░  alpha · partial-resync unproven

  Cluster mode                               ░░░░░░░░░░░░  not implemented
  Sentinel HA                                ░░░░░░░░░░░░  not implemented
  Loadable C module ABI                      ░░░░░░░░░░░░  not implemented

Performance

  GET    p=1     ░░░░░             0.95× / 0.92×
  GET    p=16    ▒▒▒▒▒▒▒           1.06× / 1.12×
  GET    p=100   █████████         1.32× / 1.53×

  Default-suite median across 23 commands             1.13× / 1.15×

Throughput ratios vs Valkey 8.1.7 / 9.1.0, same host (Apple M3 Max), warmed.

Full per-row data (9 pipeline depths + 23 commands)

vs Valkey 8.1.7
Workload	Valdr rps	Valkey rps	Ratio
Pipeline-depth · GET / PING_MBULK / SET · p = 1 / 16 / 100
GET p=1	149,925	157,233	0.954×
GET p=16	2,202,643	2,083,333	1.057×
GET p=100	5,555,556	4,201,680	1.322×
PING p=1	152,905	160,514	0.953×
PING p=16	2,267,574	2,232,143	1.016×
PING p=100	7,246,377	5,102,041	1.420×
SET p=1	150,376	159,236	0.944×
SET p=16	1,941,748	1,757,469	1.105×
SET p=100	3,610,108	2,695,418	1.339×
Default suite · pipeline = 100 · payload = 64 bytes
ping_inline	5,263,158	3,703,704	1.421×
ping_mbulk	7,142,857	5,555,556	1.286×
set	3,703,704	3,030,303	1.222×
get	4,761,905	3,846,154	1.238×
incr	4,166,667	4,000,000	1.042×
lpush	2,777,778	2,439,024	1.139×
rpush	2,702,703	2,702,703	1.000×
lpop	2,564,102	2,272,727	1.128×
rpop	2,380,952	2,500,000	0.952×
sadd	2,380,952	3,225,806	0.738×
hset	1,724,138	2,500,000	0.690×
spop	2,857,143	4,000,000	0.714×
zadd	1,923,077	2,439,024	0.788×
zpopmin	2,857,143	4,166,667	0.686×
lrange_100	176,367	129,032	1.367×
lrange_300	56,180	38,971	1.442×
lrange_500	34,153	23,175	1.474×
lrange_600	27,778	18,685	1.487×
mset	636,943	456,621	1.395×
mget	1,063,829	847,457	1.255×
xadd	1,123,596	1,388,889	0.809×
function_load	578,035	58,893	9.815×
fcall	847,458	1,428,571	0.593×

vs Valkey 9.1.0
Workload	Valdr rps	Valkey rps	Ratio
Pipeline-depth · GET / PING_MBULK / SET · p = 1 / 16 / 100
GET p=1	161,551	175,747	0.919×
GET p=16	2,590,674	2,304,148	1.124×
GET p=100	6,024,096	3,937,008	1.530×
PING p=1	179,533	177,305	1.013×
PING p=16	2,375,297	2,331,002	1.019×
PING p=100	7,407,407	4,950,495	1.496×
SET p=1	165,016	176,991	0.932×
SET p=16	2,028,397	1,607,717	1.262×
SET p=100	3,636,363	2,347,418	1.549×
Default suite · pipeline = 100 · payload = 64 bytes
ping_inline	5,263,158	4,000,000	1.316×
ping_mbulk	7,142,857	5,555,556	1.286×
set	3,846,154	2,564,102	1.500×
get	4,545,454	3,448,276	1.318×
incr	3,846,154	3,571,428	1.077×
lpush	2,564,102	2,380,952	1.077×
rpush	2,500,000	2,777,778	0.900×
lpop	2,500,000	2,173,913	1.150×
rpop	2,380,952	2,439,024	0.976×
sadd	2,325,581	3,125,000	0.744×
hset	1,754,386	2,500,000	0.702×
spop	2,857,143	3,703,704	0.771×
zadd	1,851,852	2,222,222	0.833×
zpopmin	2,702,703	3,703,704	0.730×
lrange_100	186,916	114,943	1.626×
lrange_300	62,461	37,722	1.656×
lrange_500	35,137	21,777	1.613×
lrange_600	29,308	18,123	1.617×
mset	628,931	442,478	1.421×
mget	1,030,928	740,741	1.392×
xadd	1,098,901	1,408,451	0.780×
function_load	588,235	56,593	10.394×
fcall	869,565	1,351,351	0.643×

Each adversary measured in its own run; Valdr-rps differs by ~3–8% between runs (normal variance on un-quantized rps numbers).

Try it

docker pull ghcr.io/ianm199/valdr:alpha
docker run -d --name valdr -p 6379:6379 ghcr.io/ianm199/valdr:alpha
printf 'PING\r\n' | nc 127.0.0.1 6379
# +PONG

docker run --rm --network host valkey/valkey:8-alpine \
  valkey-benchmark -h 127.0.0.1 -p 6379 -n 100000 -c 50 -P 100 -t get,set