[mgravell]

Adds client support for the Redis 8.12 BLESS command family (redis/redis#15649), gated behind SER011 (Server_8_12).

API

[Flags] public enum BlessFlags { None = 0, NoEvict = 1 }

// IDatabase / IDatabaseAsync
bool       KeyBless(RedisKey key, BlessFlags bless, CommandFlags flags = CommandFlags.None);   // BLESS SET
bool       KeyUnbless(RedisKey key, BlessFlags bless, CommandFlags flags = CommandFlags.None); // BLESS CLEAR
BlessFlags KeyBlessFlags(RedisKey key, CommandFlags flags = CommandFlags.None);                // BLESS GET

// IServer
IEnumerable<RedisKey> BlessedKeys(BlessFlags bless, int database = -1, int pageSize = 250, long cursor = 0, int pageOffset = 0, CommandFlags flags = CommandFlags.None); // BLESS SCAN

Also: RedisFeatures.Bless, and KeyNotificationType.Bless / Unbless for the bless / unbless keyspace events (class g). After merging main post-#3312 these sit at 66/67, after the SetBit..SDiffStore range that PR added, and both tokens are in the parser test table.

Notes

• BLESS is replica-eligible, so GET can be served by a replica; SET/CLEAR are raised to primary-only per message.
• Retry: SET/CLEAR are write-checked (the flag state converges), GET is read-only, SCAN uses the usual cursor rule.
• BlessFlags.None and undefined bits throw client-side. Unknown tokens in a GET reply are ignored.
• BLESS SCAN always sends COUNT, because the server default (1024) isn't our page size.
• BLESS is excluded from the Twemproxy and Envoy command maps.
• Design notes and deviations from the cross-client HLD: redis-developer/client-designs redis-oss/client-hld/bless/se-redis.md.

Testing

• Unit tests: result processor, exact wire bytes, routing/retry, key-prefix forwarding.
• Live tests (require: v8_12_0) pass on RESP2 and RESP3 against redislabs/client-libs-test:8.12-m02-int.
• 8.12 RC2 (2026-11-03): all Bless tests (live, round-trip, processor, key-prefix) pass against redislabs/client-libs-test:8.12-rc2. The NO-EVICT token, BLESS SCAN argument order, the missing-key no such key errors and the bless/unbless keyspace events were all re-verified by hand on RC2 and match the implementation.

Resolved (was: open before marking ready)

• RedisFeatures.v8_12_0 is now pinned to 8.11.241, which is what 8.12 RC2 reports — the 8.12.0 placeholder would indeed have made the live tests skip on RCs. The docker topology / CI default image is now redislabs/client-libs-test:8.12-rc2, with cluster-bus-port-protected-mode no added to the test cluster configs (RC2 introduces that check and otherwise refuses to start a plaintext cluster node).

RC2 ZSCAN duplicates (server behavior change, handled)

The only CI failures on the RC2 image are the 8 ScanTests.SortedSetScanLarge cases (all page sizes, RESP2+RESP3): a full ZSCAN iteration over a stable 2000-element sorted set returns 2001-2003 entries, with a few members duplicated. Reproduced with pure redis-cli against redislabs/client-libs-test:8.12-rc2 (deterministic per dataset; ZCARD is 2000, distinct members 2000), clean on redis:8.2, and HSCAN/SSCAN are unaffected — a ZSCAN behavior change in the 8.12 btree zset encoding, reported upstream as redis/redis#16158. SortedSetScanLarge now asserts the distinct member count, so CI is expected green; revisit if redis/redis#16158 is fixed before GA. Everything else was already green (6700+ tests).