2.1.58

• Fix: [*]SCAN - fix possible NRE scenario if the iterator is disposed with an incomplete operation in flight (by @mgravell)
• Fix: [*]SCAN - treat the cursor as an opaque value whenever possible, for compatibility with redis-cluster-proxy (by @mgravell)
• Adds: [*]SCAN - include additional exception data in the case of faults (by @mgravell)

2.1.55

• Adds: Identification of assembly binding problem on .NET Framework. Drops System.IO.Pipelines to 4.7.1, and identifies new System.Buffers binding failure on 4.7.2 (by @mgravell)

2.1.50

• Adds: Bind directly to sentinel-managed instances from a configuration string/object (#1431 by @ejsmith)
• Adds: last-delivered-id to StreamGroupInfo (#1477 by @AndyPook)
• Change: Update naming of replication-related commands to reflect Redis 5 naming (#1488 by @mgravell, #945 by @mgravell)
• Fix #1460: IServer commands that are database-specific (DBSIZE, FLUSHDB, KEYS, SCAN) now respect the default database on the config (#1468 by @mgravell)
• Library updates (by @mgravell)

2.1.39

• Fix: Mutex around connection was not "fair"; in specific scenario could lead to out-of-order commands (#1440 by @kennygea)
• Fix #1432: Update dependencies (by @mgravell)
• Fix: Timing error on linux (#1433 by @pengweiqhca)
• Fix: Add auth to command-map for Sentinel (#1428 by @ejsmith)

2.1.30

• Build: Fix deterministic builds (#1420 by @mgravell)

2.1.0

• Fix: Ensure active-message is cleared (#1374 by @hamish-omny)
• Adds: Sentinel support (#1067 by @shadim, #692 by @lexxdark)
• Adds: IAsyncEnumerable<T> scanning APIs now supported (#1087 by @mgravell)
• Adds: New API for use with misbehaving sync-contexts (more info)
• Adds: TOUCH support (#1291 by @gkorland)
• Adds: Condition API (transactions) now supports SortedSetLengthEqual (#1332 by @phosphene47)
• Adds: SocketManager is now more configurable (#1115 by @naile)
• Adds: NRediSearch updated in line with JRediSearch (#1267 by @tombatron, #1199 by @oruchreis)
• Adds: Support for CheckCertificatRevocation configuration (#1234 by @BLun78 and @V912736)
• Adds: More details about exceptions (#1190 by @marafiq)
• Adds: Updated StreamCreateConsumerGroup methods to use the MKSTREAM option (#1141 by @ttingen)
• Adds: Support for NOACK in the StreamReadGroup methods (#1154 by @ttingen)
• Adds: Event-args now mockable (#1326 by @n1l)
• Fix: No-op when adding 0 values to a set (#1283 by @omeaart)
• Adds: Support for LATENCY and MEMORY (#1204 by @mgravell)
• Adds: Support for HSTRLEN (#1241 by @eitanhs)
• Adds: GeoRadiusResult is now mockable (#1175 by @firenero)
• Fix: Various documentation fixes (#1162 by @SnakyBeaky, #1135 by @ttingen, #1203 by @caveman-dick, #1240 by @Excelan, #1245 by @francoance, #1159 by @odyth, #1311 by @DillonAd, #1339 by @vp89, #1336 by @ERGeorgiev)
• Fix: Rare race-condition around exception data (#1342 by @AdamOutcalt)
• Fix: ScriptEvaluateAsync keyspace isolation (#1377 by @gliljas)
• Fix: F# compatibility enhancements (#1386 by @NickCraver)
• Fix: Improved ScriptResult null support (#1392 by @mgravell)
• Fix: Error with DNS resolution breaking endpoint iterator (#1393 by @mgravell)
• Tests: Better docker support for tests (#1389 by @ejsmith, #1391 by @NickCraver)
• Tests: General test improvements (#1183 by @mtreske, #1385 by @NickCraver, #1384 by @NickCraver)

2.0.601

• Adds: Tracking for current and next messages to help with debugging timeout issues - helpful in cases of large pipeline blockers

2.0.593

• Performance: Unify spin-wait usage on sync/async paths to one competitor
• Fix #1101: When a ChannelMessageQueue is involved, unsubscribing via any route should still unsubscribe and mark the queue-writer as complete

2.0.588

• Stability/Performance: Resolve intermittent stall in the write-lock that could lead to unexpected timeouts even when at low/reasonable (but concurrent) load

2.0.571

• Performance: Use new arena allocation API to avoid RawResult[] overhead
• Performance: Massively simplified how ResultBox<T> is implemented, in particular to reduce TaskCompletionSource<T> allocations
• Performance: Fix sync-over-async issue with async call paths, and fix the SemaphoreSlim problems that this uncovered
• Performance: Reintroduce the unsent backlog queue, in particular to improve async performance
• Performance: Simplify how completions are reactivated, so that external callers use their originating pool, not the dedicated IO pools (prevent thread stealing)
• Fix: Update Pipelines.Sockets.Unofficial to prevent issue with incorrect buffer re-use in corner-case
• Fix: KeyDeleteAsync could, in some cases, always use DEL (instead of UNLINK)
• Fix: Last unanswered write time was incorrect
• Change: Use higher Pipe thresholds when sending