NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI, #3748 most downloaded on PyPI
Valkey GLIDE Async client. Supports Valkey and Redis OSS.
Latest stable
2.6.0
Last release 16 days ago
06 Nov 2026
Ships fairly regularly
a new release about every 3 weeks
Nearly every release is documented
notes for 32 of 34 stable releases
1 version withdrawn
withdrawn after publishing
2 years old
116 releases · first in 2024
Drag across the ruler to compare, or tap a release to open it.
fix(java): clear jedis-compat-shared javadoc errors and deprecate substr by @xShinnRyuu in https://github.com/valkey-io/valkey-glide/pull/7191
Connection Pooling (Preview): Introduced client-instance pooling (ClientPool) and isolated execution scopes for Java, Python, Node.js, and Go, backed by a single shared Rust implementation in glide-core. Pools eliminate multiplexer contention under high concurrency, while scopes provide dedicated connections for WATCH/MULTI/EXEC and CLIENT TRACKING. Connection state is reset when a client is returned to the pool, scopes inherit the parent client's database, credentials, and compression settings, and abandon detection reclaims leaked borrows after a configurable timeout. This feature is released as a preview: APIs and behavior may change in future releases, and we welcome feedback.
Experimental RDMA Support for Python: Added opt-in RDMA (Remote Direct Memory Access) to the Python sync client through RdmaConfiguration, with register_rdma_region, rdma_get, and rdma_set APIs. RDMA is built on the new glide-rdma crate, a transfer layer over libfabric loaded at runtime, and is exposed through glide-core and the FFI layer.
Valkey 9.x Command Support: Added new commands and options for Java, Python, Node.js, and Go:
DELIFEQ for atomic compare-and-deleteHGETDEL, key-level NX/XX conditions for HSETEX, and the cluster_info INFO sectionXX option for SISMEMBER, and IFNE option for SETNode.js Native Binding Layer: The Node.js client now calls the Rust core through a direct NAPI layer instead of exchanging requests and responses with it over a Unix domain socket.
Client-Side Caching for MGET: Fully cached MGET requests are served locally, and partially cached requests fetch only the missing keys, preserving duplicate-key and nil semantics.
mTLS Client Certificates with Automatic Reloading: Added mutual TLS client certificate and private key support with automatic certificate reloading to Java, Node.js, and Go. Python, which already accepted in-memory PEM client certificates, adds file-based client certificates (client_cert_path / client_key_path) with the same automatic reloading. Java can also load the client identity from a keystore via useMutualTlsFromKeyStore.
AZ Affinity All Nodes Read Strategy: Added the AZ_AFFINITY_ALL_NODES read strategy to Go and Node.js, which is already available in Java and Python since 2.5.3. Reads are served by any node in the client's availability zone, both primaries and replicas.
Custom IAM Credential Providers: Added GlideCredentialProvider support to IamAuthConfig in Java, Python, Node.js, and Go, letting applications supply custom IAM credential sources instead of the default AWS credential chain.
Python OpenTelemetry Enhancements: Optional propagation of the active OpenTelemetry parent span to command, batch and script spans (enable_trace_context_propagation=True, off by default), and spans for script invocations (EVALSHA).
Python Unix Domain Socket Support: Python standalone clients (async and sync) can connect over a Unix domain socket with NodeAddress(unix_socket_path=...). TLS and cluster mode aren't supported over Unix domain sockets.
Zero-Copy Reads: Added a zero-copy receive path for GET/MGET in glide-core, zero-copy MGET buffers in the Python sync client, and GlideString.asReadOnlyByteBuffer() in Java for read-only access to binary payloads.
Jedis 4 Compatibility for Java: A new valkey-glide-jedis-4-compatibility artifact lets applications written against the Jedis 4.x API run on GLIDE, alongside the existing Jedis compatibility layer.
Core Performance: Lower per-command overhead on the core hot path (+12–16% cluster throughput in the change's benchmarks) and vectored, zero-copy sends for large command arguments.
Go Inflight Requests Limit: Exposed inflightRequestsLimit via WithInflightRequestsLimit, bringing the Go client to parity with Java, Python, and Node.js.
Improved Core Reliability: A circular MOVED (a node redirecting to its own endpoint, e.g. an old primary kept alive behind the same hostname after a failover) now replaces the stale connection instead of reusing it.
The Rust client (valkey-glide crate, preview) is versioned and released separately under rust/v* tags, so its changes aren't listed here.
SIGSEGV. The TLS cluster connect needs more stack than musl gives Go's threads by default. Workaround: build with CGO_LDFLAGS="-Wl,-z,stack-size=8388608" (or go build -ldflags="-extldflags=-Wl,-z,stack-size=8388608"). Standalone, non-TLS, and glibc/macOS builds are not affected. Tracked in #7637.The curated, user-facing list of every change in this release is in CHANGELOG.md. Pull requests are grouped below by binding; expand a section to see them.
<details> <summary><b>Core and cross-binding</b> (67)</summary>
create_client_from_uri by @jamesx-improving in https://github.com/valkey-io/valkey-glide/pull/6660create_client_from_uri by @Aryex in https://github.com/valkey-io/valkey-glide/pull/6714cluster_info INFO section and test Valkey 9.2 INFO fields by @currantw in https://github.com/valkey-io/valkey-glide/pull/7364</details>
<details> <summary><b>Java</b> (20)</summary>
AZ_AFFINITY_ALL_NODES read strategy by @jamesx-improving in https://github.com/valkey-io/valkey-glide/pull/7059</details>
<details> <summary><b>Python</b> (25)</summary>
BusyResourceError by @xShinnRyuu in https://github.com/valkey-io/valkey-glide/pull/6605_CompatFuture waiters from the owning trio thread by @xShinnRyuu in https://github.com/valkey-io/valkey-glide/pull/6685AZ_AFFINITY_ALL_NODES read policy by @sz-armin in https://github.com/valkey-io/valkey-glide/pull/6721SISMEMBER XX support by @currantw in https://github.com/valkey-io/valkey-glide/pull/7371libglide_ffi in dev.py build by @currantw in https://github.com/valkey-io/valkey-glide/pull/7431</details>
<details> <summary><b>Node.js</b> (12)</summary>
unix_socket_path in lazy_client test NodeAddress by @jamesx-improving in https://github.com/valkey-io/valkey-glide/pull/7421</details>
<details> <summary><b>Go</b> (11)</summary>
</details>
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.5.3...v2.6.0
Your coding agent can read these notes before it upgrades. Set up the MCP server →
FT.HYBRID (#7380)DELIFEQ (#6146)ToSingleValkeyArg and ToValkeyArgs for option types (#7381)Cmd::with_arg (#7313)glide-rdma crate, an RDMA transfer layer over libfabric (loaded at runtime), for later RDMA support (#7177)rdma feature: RdmaConfig in the connection request, Client::register_rdma_region / rdma_get / rdma_set / close_rdma, and the rdma_parked_bytes / rdma_leaked_bytes counters. Transfers route to the key's primary and have no request timeout. BLOB.RDMA_* commands are refused when sent any other way, in every build (#7412)rdma feature: rdma_available, rdma_usable, rdma_checksum, register_rdma_region over caller-owned memory, rdma_get / rdma_set (blocking, with region_closed saying whether the region can transfer again), and free_rdma_region with RdmaMemoryHold for memory the fabric still holds (rdma_memory_leaked says whether it is leaked rather than parked), plus the rdma_parked_bytes / rdma_leaked_bytes counters. Sync clients only; pools refuse an RDMA configuration, and rdma_close_regions cancels transfers in flight without freeing the client (#7458)RdmaConfiguration / RdmaProvider / RdmaSetChecksum in both client configurations, and register_rdma_region, rdma_get, rdma_set, rdma_available, rdma_usable and rdma_checksum on the sync client, transferring through windows of caller-owned memory. Published valkey-glide-sync wheels include RDMA; libfabric is loaded only when a client asks for it. The async client and ClientPool refuse an RDMA configuration (#7464)ClientPool call no longer gives up when another thread holds the pool lock; it waits for the lock instead. Go and Python pool metrics read from inside a client callback no longer come back as zeros, Java's acquire no longer backs off and retries while an idle client exists, Go, Python and Java close() have finished tearing the pool down when they return, and a command's activity refresh is no longer skipped, so the abandon monitor no longer sees a stale timer for a client in use (#7507)JedisPool (4.x and 5.x) now restores its configured database before lending a connection, so a borrower no longer inherits a previous borrower's select (#7604)valkey-glide-jedis-compat-shared JAR. The valkey-glide-jedis-compatibility and valkey-glide-jedis-4-compatibility POMs depend on it, and since 2.6.0-rc1 Maven failed to resolve them with Could not find artifact io.valkey:valkey-glide-jedis-compat-shared (#7603)lib-ver instead of unknown in CLIENT INFO responses (#7616)NAPI_RS_ENFORCE_VERSION_CHECK=1 instead of failing with "Cannot find native binding" (#7602)memoryview argument with its byte length (nbytes) instead of len(). A memoryview cast to a wider format (e.g. .cast("I")) or with more than one dimension was sent truncated, silently storing a partial value or key; this affected direct commands, batches, custom_command, scripts and cluster scan (#7560)TrimByMinId (xadd/xtrim), ExclusiveIdBound (xrange/xrevrange/xpending_range) and the FtHybridVsimArm field (ft.hybrid) send bytes, bytearray and memoryview values verbatim instead of their Python repr, which the server rejected. A bytes stream ID returned by xadd can now be used as a TrimByMinId threshold (#7558)zadd (async, sync and Batch/ClusterBatch) no longer stringifies bytes, bytearray and memoryview members into their Python repr; members are now passed through verbatim and only the scores are converted to strings (#7521)AZ_AFFINITY or AZ_AFFINITY_REPLICAS_AND_PRIMARY has no clientAZ (#7536)update_connection_password (or a user-issued AUTH/HELLO), instead of the password the client was created with: a scope connection built after the rotation sends the new password, and an idle one re-authenticates before its next command once the server has accepted the new password (immediate_auth, AUTH, HELLO); a store-only update leaves open scope connections as they are, like the client's own connection (#7334)GETSET compresses value when enabled (#7411)SET with IFEQ or IFNE when compression is enabled (#7365)ClientPool no longer settles permanently above its configured max_size after a burst of concurrent borrow/release. Warmup connects its min_idle clients asynchronously and the pool became acquirable before they landed, so an acquirer racing that window saw total_count still below max_size and spawned an on-demand connection redundant with a warmup connection in flight; the warmup slots are now reserved up front so a racing acquirer reads the pool as full and does not over-create (#7338)HSetExOptions rejects OnlyIfEquals and OnlyIfNotEquals instead of sending them to the server (#7237)MOVED (a node redirecting to its own endpoint, e.g. an old primary kept alive behind the same hostname after a failover), replace the connection instead of reusing it because it still answers PING, for single commands and pipelines. Commands routed to a node by address now wait for an in-progress reconnect instead of failing (#7333)glide_scope_try_acquire in its own retry loop. A cause that waiting cannot fix (parent closed mid-wait, lazily connected mTLS parent with no certificate material yet, a connection request that cannot produce a scoped connection) now raises the binding's disconnect or request-error class immediately instead of a "pool exhausted" timeout after the full deadline, and a deadline error names the cause of the final attempt. One acquire makes one dial: the acquire awaits the connection it reserved a slot for, so the attempt-token dedupe that let a binding's repeat poll recognise its own in-flight creation is removed along with the loops. glide_scope_try_acquire and glide_scope_next_attempt_token (and the JNI and napi counterparts) are replaced by glide_scope_acquire_async / glide_scope_acquire_blocking / glide_scope_acquire_pipe (the last delivers to the Python async client's shared pipe); Node's IsolatedScope.acquire takes timeoutMs where it took maxRetries (#7112)client_ids came from separate counters that both started at 1, so closing an ordinary client (including ClientPool.create's own connectivity probe) evicted a pooled client sharing its id — killing it outright, invalidating its scopes, or polluting its inflight accounting. Ordinary-client handles now draw from the same global allocator pooled ids use, so every id in the shared registry is unique (#6968)request_timeout, with no error) after an unrelated pool borrow was taken and released on the same event loop. The async client drew its shared-pipe id from a binding-local counter starting at 1, colliding with a pooled client's id (drawn from core's global sequence, also starting at 1) in the process-wide pipe registry; the pooled client displaced the ordinary client there, so the ordinary client's responses were routed away and dropped. The ordinary async client now draws its id from the same global allocator, matching Java and Node (#7003)slotForKey returning the wrong cluster slot for a key with an empty hash tag. The first } after the first { now closes the tag, and an empty tag ({}) hashes the whole key, so a scope pinned with a key like {}user}:1 targets the correct primary (#7251)AUTH <username> <token> over plaintext TCP, where it could be captured and replayed. An IAM-configured client now fails to create when TLS is off — before any token manager exists, so initial AUTH, background refresh, and reconnect all fail closed — and a scope off an IAM parent is rejected before any connect when its request disables TLS. TLS-enabled IAM connections are unchanged. An empty IAM cluster_name or region is also now rejected at client creation, instead of failing later as an opaque connect timeout or server-side auth error (#7023)node/npm/win32-x64-msvc/ platform stub so the Windows native module is published to npm on each release; fix deprecate-on-failure which was targeting the non-existent @valkey/valkey-glide-windows-x64-msvc instead of the correct @valkey/valkey-glide-win32-x64-msvc; fix prepublishOnly script which used removed napi-rs CLI v3 flags (--config → --package-json-path, --skip-gh-release → --no-gh-release); move napi prepublish out of the prepublishOnly npm lifecycle hook into an explicit CI step (napi-rs v3.8.3 workspace boundary check prevents running from the npm lifecycle cwd node/npm/glide/) and add workspaces: [] to node/package.json — an empty array satisfies napi-rs v3.8.3's Array.isArray(manifest.workspaces) boundary check without enrolling platform stubs as npm workspaces (which would cause EBADPLATFORM errors on Linux runners) (#7179)max_total slot is now held by an RAII guard that returns it on every creation outcome — including a creation cancelled mid-handshake, which previously leaked the slot until scopes to healthy shards could no longer be opened — and decrements saturate at zero so a give-back cannot wrap into permanent exhaustion. Concurrent retries of a single acquire are deduped to one in-flight creation, so a retry storm against a slow shard can no longer consume the whole pool, while distinct concurrent acquires each still open their own connection (#6966, #7067)ClientPool no longer leaks a borrow's open transaction to the next borrower, nor discards connections from a pool on a non-zero database. The release-time reset sent DISCARD to a single node (leaving an open MULTI on the other primaries) and SELECT <configured_db> on every release (which a cluster server rejects for a non-zero db, failing the reset so the connection was thrown away); it now routes DISCARD to all cluster nodes and sends SELECT only when the database may have changed during the borrow (#7301)ClientPool release again discards a connection it can't reset instead of handing it to the next borrower. Since 2.6.0-rc2 a connection left with CLIENT REPLY OFF on any node, including a read replica or one cluster node, returned to the pool and every later command on it timed out; a connection that subscribed during the borrow is now discarded as well (#7544)close() now closes its connection. The client registry held a strong reference to every client, so clients created on short-lived event loops (asyncio.run() in a worker thread) kept their server connections open until the process exited. Such a client now emits a ResourceWarning, visible under PYTHONDEVMODE or -W error::ResourceWarning (#7122)ClientPool client dropped by idle-timeout eviction, by a failed state reset on release, or by a release cancelled mid-reset is now reported to the binding, which reclaims the adapter or handle it holds for that client. Previously only the abandon monitor reported discards, so every other drop left the binding's side (in FFI, a tokio runtime and its connection) alive until the pool was destroyed; under low traffic the pool accumulated one dead runtime per eviction for its lifetime (#7168)MOVED, while slots that share a primary now share its idle connections. A slot whose primary cannot currently be resolved (topology not yet fetched, or mid-resharding) fails the acquire with the reason logged instead of connecting to a configured seed node, and a full pool whose idle connections all target other primaries evicts its oldest idle connection instead of reporting exhaustion (#6975)MOVED (or silently succeeding when the two slots happen to share a primary). Previously only the second and later commands were constrained, letting a scope's physical connection and its pinned slot disagree; a scope opened without a routing key stays unconstrained (#7057)ClientPool invalidates the scopes its clients own — the scope pool and its outstanding scopes are removed from the registries, so a scope can no longer read or mutate keyspace after its parent is gone (#6889)max_total = N now permits N concurrent scopes instead of N-1. The pool reserved a slot against max_total and the caller then re-checked capacity after that reservation, so the last acquire was never given a connection and the borrower timed out (#6795)CLIENT SETNAME on scope release so a connection name set during a scope does not leak to the next borrower of the same connection (#6898)redis:// URL and re-parsed it, which rejected a bare IPv6 host (redis://::1:7801) and surfaced only an eventual pool-exhaustion timeout; it now builds a redis::ConnectionInfo directly like every other connection path, reusing the cluster path's host/port parse so the two cannot drift (#7124)databaseId instead of logging a warning and discarding it. A JedisPool configured for a non-zero database ran every command against database 0, silently writing to a database the caller did not ask for (#6994)PSUBSCRIBE and PUNSUBSCRIBE as readonly commands so cluster routing treats them consistently with SUBSCRIBE/UNSUBSCRIBE (#6756)BLPOP key 0 blocks instead of timing out at the request timeout), and a scoped connection whose blocking command timed out or was cancelled is discarded on release instead of being reused with a stale server-side waiter (#6780, #6794)ScopedConnection works on pooled clients (#6763)AllNodes in create_client_from_uri's read_from option (#6721)ClientPool.getClient now resolves the request timeout, inflight-requests limit, and credentials from the client configuration the same way directly-created clients do, so a pooled client's per-command timeout and IAM guards no longer silently differ from the same config used directly, and its host-side inflight limiter is configured from the client's limit rather than left off (#7153)ClientPool now builds its ConnectionRequest through the same serializer as directly-created clients, so pooled clients (and scopes taken on them) honor every field carried in the connection request instead of silently dropping the ones the pool's separate serializer never mirrored (circuit breaker, lazyConnect, reconnect jitterPercent, mTLS, compression, client-side cache and more). Pooled clients now also resolve connectionTimeout the same way directly-created clients do rather than reusing requestTimeout. The address resolver and a custom IAM credentials provider remain unsupported for pooled clients — they are per-client Java callbacks the pool cannot forward, and a pool configured with either is now rejected rather than silently ignoring the callback or falling back to the default credential chain (#6897)scopedConnection() and one-argument IsolatedScope.acquire(client), including on pool-borrowed clients (#6962)create_client_from_uri so bracketed hosts (e.g. redis://[::1]:6379) resolve correctly. url::Url::host_str() returns the literal with its surrounding brackets ([::1]), which tokio::net::lookup_host cannot resolve, causing connections to IPv6 literals to hang until the connect timeout. The URI parser now uses the unbracketed canonical form via Ipv6Addr::to_string().fork(), the flush thread is gone but OnceLock prevented reinitialization, causing commands in forked child processes (e.g. PySpark workers) to hang indefinitely. Now detects fork via PID comparison and reinitializes the pipe. Using a parent's client object in a forked child now raises ClosingError on command paths and skips the FFI call on close(), instead of silently hanging. (#6673)create_client_from_uri now preserves a URI username without a password to enable IAM authentication (#7193)create_client_from_uri so credentials containing URI-reserved characters (@, :, /, ?, #, %, +, space, non-ASCII) authenticate correctly (#6659)recovery_requests_queue_size option (default: 1000) controls the queue depth. (#6640)_CompatFuture waiters from the owning trio thread (#6685)ClosingException: Client is shutting down. The JVM runs all shutdown hooks concurrently, and GLIDE's hook previously tore the client down and blocked new requests, racing with (and aborting) legitimate work such as persisting state before exit. Deterministic teardown remains available via the client's close(). (#4809)test-release in pypi-cd.yml when only one package is published manually, so a skipped sibling publish job no longer causes post-publish validation to be skipped entirely (#6542)BaseClient.__aenter__ return type to Self (from the widened "BaseClient" introduced in 2.5.0). Entering the async context manager (async with await GlideClusterClient.create(...) as client) now preserves the concrete subclass for static type checkers, matching create(). (#6531)anyhow to 1.0.103 to fix RUSTSEC-2026-0190, an unsoundness advisory in anyhow::Error::downcast_mut() that can trigger undefined behavior (#6364).gitignore from the released module so consumers who commit vendor/ keep the generated artifacts (internal/protobuf/*.pb.go, rustbin/**, lib.h) (#6441)BLPOP key 0 is never. A reused scope connection also gets a fresh scope id, and a blocking scope command keeps the abandon monitor from reclaiming its pool-borrowed parent mid-command (#7000)0 is no longer dropped from the command sent to the server. zpopmin/zpopmax with count=0, OffsetOptions with end=0, StreamClaimOptions with idle/idle_unix_time/retry_count of 0, ExpirySet/ExpiryGetEx with a value of 0 (set, getex, hsetex, hgetex), FtSearchOptions/FtAggregateOptions with timeout=0 and FtAggregateSortBy with max=0 previously omitted the argument, so the server ran a different command (e.g. zpopmin(k, 0) popped and removed a member). Empty-string JsonGetOptions values are also forwarded (#7525)&b"key"[..]) as one argument instead of one argument per byte (#7313)ConfigurationError when the core's configuration validation rejects client creation: the Unix domain socket checks, the reconnect strategy, and the library name and version, all new in 2.6. Errors that 2.5.x already reported, such as an invalid compression level, keep ClosingError (RuntimeError for MonitorClient). The FFI ConnectionResponse gains a trailing connection_error_type field (#7535)FtAggregateReducer.to_list, .first_value, and .random_sample factory methods for the TOLIST, FIRST_VALUE, and RANDOM_SAMPLE FT.AGGREGATE reducers; an in_keys option on FtSearchOptions for FT.SEARCH's INKEYS; and test coverage for Vector Range queries (VECTOR_RANGE/YIELD_DISTANCE_AS) in FT.SEARCH and FT.AGGREGATE and for the bare * wildcard query. (valkey-search#932, #946, #949, #967, #985, #960)FLOAT16/BFLOAT16 vectors, FILTER, SORTABLE UNF, SCORE_FIELD, NOHL), FT.SEARCH/FT.AGGREGATE scoring (WITHSCORES/SCORER/ADDSCORES), FT.AGGREGATE's LOAD ... AS renaming and WITHCURSOR paging (ft.cursor_read/ft.cursor_del), and a new ft.hybrid API fusing a text/tag/numeric SEARCH arm with a VSIM vector-similarity arm via FtHybridCombine.RRF/.Linear/.Function. See #7335 for the full feature list, including the search.emulate-release >= 1.3.0 server-side requirement for a few of these. (valkey-search#1.3.0-rc1)NodeAddress(unix_socket_path=...), and accept valkey+unix:// URIs in create_client_from_uri (#4878)BaseClient and the command interfaces (#7401)cluster_info INFO section (Valkey 9.1) (#7199)enable_trace_context_propagation=True (disabled by default) (#7069)glide-core (#6864)logger_core to glide_logger (#7178)GlideCredentialProvider support to IamAuthConfig for custom IAM credential sources on direct clients. Pool configurations with custom providers are rejected; IAM without a custom provider continues to use the default AWS credential chain. Go/Python use the bounded C FFI adapter, while Java/Node use checked object bridges. (#7055, #6825)GlideString.asReadOnlyByteBuffer() for zero-copy, read-only access to binary payloads (#6600)GoStringN (one copy, interior NULs preserved) instead of GoBytes+string in GET/MGET response parsing, pubsub callbacks, MONITOR client/command strings, and script hashes (#6751)AZ_AFFINITY_ALL_NODES read policy (#6721)AZ_AFFINITY_ALL_NODES read strategy (#7059)AZAffinityAllNodes read strategy (#7104)inflightRequestsLimit configuration via WithInflightRequestsLimit, bringing the Go client to parity with Java, Python, and Node (#6385)send_scope_command() and release_client_async(). Pool release resets state (DISCARD + SELECT). Scopes inherit parent's current database, credentials, and compression. Circuit breaker and inflight limits enforced. Abandon detection reclaims leaked borrows after configurable timeout (default 5 min, skips blocking commands, 0 to disable). (#6338)command_with_route_info FFI entrypoint, accepting routing as a RouteInfo C-struct pointer instead of protobuf-encoded bytes — the same mechanism batch() already uses. Existing command, command_with_buffer, command_with_buffers, and invoke_script are unchanged. (#6494)valkey-glide and valkey-glide-sync packages to PyPI via Trusted Publishing (OIDC) with PEP 740 attestations, replacing API-token uploads (#6478)EVALSHA) so invoke_script calls appear in traces with DB semantic convention attributes (#6350)InfoSection enum for info_sections command (#7366)clientAz is now trimmed and validated for the AZAffinity, AZAffinityReplicasAndPrimary, and AZAffinityAllNodes read strategies. A whitespace-only clientAz now raises a ConfigurationError, and a padded value (e.g. " us-east-1a ") is trimmed before reaching the core so reads pin to that zone instead of falling back to all nodes (#7104)ci(node): stop npm-cd post-publish test failing on npm propagation (r…
ci(node): stop npm-cd post-publish test failing on npm propagation (r…
fix(core): decrement standalone total_clients once per client, not pe…
fix(core): decrement standalone total_clients once per client, not pe…
ci(node): drop win32 from npm CD matrix to unblock rc1 release
ci(node): drop win32 from npm CD matrix to unblock rc1 release (#7417)
Signed-off-by: Jeremy Parr-Pearson jeremy.parr-pearson@improving.com
fix(core): Move scope acquire wait into glide-core, delete binding po…
fix(core): Move scope acquire wait into glide-core, delete binding po…
Add AZ_AFFINITY_ALL_NODES read strategy for Java and Python (backport #7059 + #6721) by @jamesx-improving in https://github.com/valkey-io/valkey-glide
AZ_AFFINITY_ALL_NODES read strategy for Java and Python (backport #7059 + #6721) by @jamesx-improving in https://github.com/valkey-io/valkey-glide/pull/7106Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.5.2...v2.5.3
test(python): backport subscription timestamp flake fix
test(python): backport subscription timestamp flake fix (#6415)
Backport of 98f0d6c to release-2.5.
Signed-off-by: James Xin james.xin@improving.com
fix(ffi): backport to release-2.5 — release response payload buffers when parking arenas in ARENA_POOL by @superbaddude in https://github.com/valkey-i
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.5.1...v2.5.2
revert: fix(node): add pnpm-workspace.yaml
Reverts the pnpm-workspace.yaml workaround for napi v3 pre-publish
boundary error, which is no longer needed as we are reverting to napi v2.
Signed-off-by: affonsov 67347924+affonsov@users.noreply.github.com
Reverts the prepublishOnly --package-json-path fix, no longer needed
as we are reverting to napi v2.
Signed-off-by: affonsov 67347924+affonsov@users.noreply.github.com
Reverts the napi v3 prepublish CLI rename fixes, no longer needed
as we are reverting to napi v2.
Signed-off-by: affonsov 67347924+affonsov@users.noreply.github.com
Reverts the napi-rs v2 to v3 migration which introduced a glibc 2.28
regression for linux-gnu builds, breaking Amazon Linux 2 (glibc 2.26).
The root cause of the original musl build failure (Rust 1.98.0 adding
--fix-cortex-a53-843419) is addressed separately by pinning Rust to 1.97.0.
Signed-off-by: affonsov 67347924+affonsov@users.noreply.github.com
Rust 1.98.0 (released 2026-08-20) added --fix-cortex-a53-843419 to the
aarch64-unknown-linux-musl target spec. The napi-rs v2 zig linker wrapper
does not filter this flag, causing CI to fail with:
error: unsupported linker arg: --fix-cortex-a53-843419
Pin to Rust 1.97.0 (the last version before this flag was added) to
unblock the release-2.5 build pipeline while a proper long-term fix
(napi v3 migration without glibc regression) is tracked separately.
Signed-off-by: affonsov 67347924+affonsov@users.noreply.github.com
chacha20 0.10.0 was yanked from crates.io. Update to 0.10.2 to pass
cargo-deny advisories check. This fix was present in PR #6908 but was
lost when reverting the napi v3 migration.
Signed-off-by: affonsov 67347924+affonsov@users.noreply.github.com
Signed-off-by: affonsov 67347924+affonsov@users.noreply.github.com
test(python): clear stale WATCH in pooled-client teardown (backport #…
test(python): clear stale WATCH in pooled-client teardown (backport #…
feat(python): Add configurable lib_name and client_info_tag to client config (async + sync)
feat(python): Add configurable lib_name and client_info_tag to client config (async + sync) (#6389)
feat(python): Add configurable lib_name to client config
Expose a lib_name option on GlideClientConfiguration and
GlideClusterClientConfiguration (used by both the async and sync
clients). When set, it is sent via CLIENT SETINFO LIB-NAME at
connection establishment, enabling library/framework attribution in
CLIENT INFO / CLIENT LIST. When unset, the existing per-client defaults
(GlidePy / GlidePySync) are preserved.
The lib_name field already exists in the shared ConnectionRequest
protobuf (field 19) and is consumed by the Rust core (the Java client
sets it today), so this is a binding-only change that mirrors the
existing client_name handling and brings parity with Java's libName.
Adds async and sync tests asserting a custom lib_name appears in
CLIENT INFO.
Closes #6378
Signed-off-by: Omer Rubinstein omerrubi@amazon.com
Adds a client_info_tag configuration option (async and sync) that
appends a parenthesized tag to the library name while preserving the
underlying GLIDE identity, e.g. GlidePy(my-framework:1.2.3). This keeps
existing GLIDE-level adoption metrics intact while attributing a
wrapping library/framework, following the ioredis 'ioredis()'
convention.
The tag composes on top of the resolved lib_name (the client default or
a configured lib_name override) in each client wrapper. A no-whitespace
guard rejects invalid tags at config construction time, matching the
CLIENT SETINFO value constraints.
Adds async and sync tests asserting the default identity is preserved
with the tag appended.
Signed-off-by: Omer Rubinstein omerrubi@amazon.com
lib-name in CLIENT INFO depends on CLIENT SETINFO, added in server
7.2.0. The CI engine matrix runs redis 6.2 with run: always, so these
assertions would fail on the 6.2 leg. Add @skip_if_version_below(7.2.0)
to all four new tests, matching the existing guarded
test_register_client_name_and_version.
Addresses valkey-review-bot feedback (also guards the two
client_info_tag tests, which share the same dependency).
Signed-off-by: Omer Rubinstein omerrubi@amazon.com
Address review feedback on #6389:
Signed-off-by: Omer Rubinstein omerrubi@amazon.com
Collapse the wrapped test signatures to single lines (both fit within
black's 88-char limit), fixing the lint job failure under black 26.5.1.
Signed-off-by: Omer Rubinstein omerrubi@amazon.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
feat: Add clientInfoTag and libName to Java, Node, Go and Python clients (#6755)
feat(java): Add support for clientInfoTag
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Adds optional libName and clientInfoTag configuration options to the
Node.js GLIDE client (standalone + cluster). When clientInfoTag is set,
it appends a parenthesized tag to the effective library name sent via
CLIENT SETINFO LIB-NAME (e.g. GlideNode(my-framework:1.0)).
Whitespace validation throws ConfigurationError. Composition follows
the same pattern as Java and Python implementations.
Signed-off-by: Jonathan Neufeld jonathan.neufeld@improving.com
Adds optional WithLibName and WithClientInfoTag configuration methods to
the Go GLIDE client (standalone + cluster). When clientInfoTag is set,
it appends a parenthesized tag to the effective library name sent via
CLIENT SETINFO LIB-NAME (e.g. GlideGo(my-framework:1.0)).
Whitespace validation panics on invalid input. Composition follows the
same pattern as Java and Python implementations.
Signed-off-by: Jonathan Neufeld jonathan.neufeld@improving.com
Signed-off-by: Jonathan Neufeld jonathan.neufeld@improving.com
Signed-off-by: Jonathan Neufeld jonathan.neufeld@improving.com
Signed-off-by: Jonathan Neufeld jonathan.neufeld@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
fix: Address review comments on client_info_tag PR (#6755)
Signed-off-by: Jonathan Neufeld jonathan.neufeld@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
cargo clippy --all-features --all-targets \ --fix --allow-dirty --allow-stagedSigned-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Jonathan Neufeld jonathan.neufeld@improving.com
Co-authored-by: Neil Derraugh neil.derraugh@improving.com
Remove changes inherited from newer mainline features that are not available in the release 2.5 branch. This drops unsupported AZ-affinity and Python pool test coverage, along with orphaned TLS and Go configuration fields.
Add the missing Go test import and remove redundant Rust conversions so the backport builds cleanly with current lint tooling.
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Adapt the client identification test to the socket-based BaseClient
constructor used in release 2.5.
Remove a response-handling test inherited from the newer direct-NAPI
implementation.
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Neil Derraugh neil.derraugh@improving.com
Signed-off-by: Jonathan Neufeld jonathan.neufeld@improving.com
Co-authored-by: Omer Rubinstein omerrubi@amazon.com
Co-authored-by: Jonathan Neufeld jonathan.neufeld@improving.com
fix(ffi): backport to release-2.5 — release response payload buffers …
fix(ffi): backport to release-2.5 — release response payload buffers …
Remove the repository-level SECURITY.md in favor of the organization-level SECURITY.md (#6439) by @xShinnRyuu in https://github.com/valkey-io/valkey-g
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.5.0...v2.5.1
fix(core): make TimeoutWatchdog singleton fork-safe (cherry-pick #671 …
fix(core): make TimeoutWatchdog singleton fork-safe (cherry-pick #671…
fix(Core): buffer pending requests for all recovery paths
Signed-off-by: affonsov 67347924+affonsov@users.noreply.github.com
Nothing published for this version
deps(node): bump protobufjs to ^7.6.3 (CVE-2026-48712, CVE-2026-54269) by @xShinnRyuu in https://github.com/valkey-io/valkey-glide/pull/6274
Client-Wide Circuit Breaker: Introduced a comprehensive per-client circuit breaker system that detects when the GLIDE core is unhealthy and proactively rejects requests synchronously at the FFI boundary. Enhanced with structured timeout diagnostics and watchdog-informed recovery mechanisms to prevent thread parking and improve overall client resilience.
Python Async Performance Enhancement: Completely overhauled the Python async client architecture by replacing UDS (Unix Domain Socket) and protobuf transport with direct FFI requests and unnamed pipes for responses, delivering significant performance improvements and reduced latency.
Custom Socket Address Resolution: Added comprehensive support for custom socket address resolution when connecting to Valkey across all client languages, providing greater flexibility for complex network configurations and deployment scenarios.
Comprehensive New Command Support: Added new commands with full multi-language implementation:
MIGRATE single-key and multi-key variantsRESET, CLIENT PAUSE, CLIENT UNPAUSECLIENT CACHING, CLIENT TRACKING, CLIENT TRACKINGINFOMONITOR with dedicated MonitorClientSAVE, BGSAVE, BGREWRITEAOFREPLICAOF, FAILOVERLATENCY HISTORY, LATENCY LATEST, LATENCY RESETMEMORY DOCTOR, MEMORY MALLOC-STATS, MEMORY PURGE, MEMORY STATSEnhanced TLS Performance: Fixed critical TLS connection performance regression by optimizing initial connection handling and implementing per-process certificate caching, eliminating timeouts in multi-client and multi-process deployments.
Advanced Client-Side Caching: Implemented Phase 2 server-assisted invalidation via CLIENT TRACKING with comprehensive support across all client languages, enabling efficient cache invalidation strategies.
Improved Core Reliability: Enhanced connection management with circular MOVED handling, TCP deadlock fixes for large payloads, topology refresh improvements, and liveness-aware pipeline send-timeout mechanisms.
Enterprise-Grade Observability: Added structured timeout diagnostics that classify timeout causes and provide actionable corrective recommendations, along with dedicated timeout watchdog threads independent of Tokio runtime for improved debugging and monitoring capabilities.
Go OpenTelemetry Enhancement: Added support for connecting GLIDE command and batch spans to an application's existing OpenTelemetry trace context through OpenTelemetry.SetSpanContextExtractor, letting applications propagate active trace context from context.Context without adding a direct go.opentelemetry.io/otel dependency to the Go SDK.
Go Valkey Search Server Module Package: Refactored Valkey Search (FT) commands into a standalone glideft server module package, decoupling search functionality from client internals and establishing a scalable pattern for future server module support.
ft commands into a glideft package by @edlng in https://github.com/valkey-io/valkey-glide/pull/5855valkey-bundle image to a stable version by @edlng in https://github.com/valkey-io/valkey-glide/pull/6049CLIENT PAUSE and CLIENT UNPAUSE commands by @currantw in https://github.com/valkey-io/valkey-glide/pull/6068CHANGELOG formatting and consistency by @currantw in https://github.com/valkey-io/valkey-glide/pull/6142SAVE, BGSAVE, BGREWRITEAOF commands by @currantw in https://github.com/valkey-io/valkey-glide/pull/6165SAVE, BGSAVE, BGREWRITEAOF commands (Node, Python, Go) by @currantw in https://github.com/valkey-io/valkey-glide/pull/6188LATENCY HISTORY, LATEST, and RESET command support by @currantw in https://github.com/valkey-io/valkey-glide/pull/6257clean command to dev.py by @currantw in https://github.com/valkey-io/valkey-glide/pull/6320serverAssisted config and CLIENT TRACKINGINFO command (Node, Python, Go, Java) by @currantw in https://github.com/valkey-io/valkey-glide/pull/6344Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.4.2...v2.5.0
FatalSendError, while a genuinely dead connection still fails fast (#5446)MIGRATE support (#6293)CLIENT TRACKINGINFO command (#5961)MEMORY DOCTOR, MEMORY MALLOC-STATS, MEMORY PURGE, and MEMORY STATS commands (#6286)MIGRATE KEYS (multi-key) variant (#6063)MIGRATE KEYS (multi-key) variant (#6064)LATENCY HISTORY, LATENCY LATEST, and LATENCY RESET command support (#6206)FAILOVER and REPLICAOF command support (#6222)SAVE, BGSAVE and BGREWRITEAOF command support (#6095)FAILOVER and REPLICAOF command support (#6170)ClientCircuitBreakerConfiguration. Tracks error rate in a sliding window, trips when threshold is exceeded, and recovers automatically via optimistic HalfOpen with consecutive success validation. Java additionally performs a synchronous pre-check to prevent thread explosion under managedBlock(). (#5996)CLIENT PAUSE and CLIENT UNPAUSE command support (#6035)MonitorClient for the MONITOR command (#5977)MIGRATE KEYS (multi-key) variant (#6066)MIGRATE command support (#5933)MIGRATE command support (#5934)MIGRATE command support (#5935)SCAN when a read-from-replica AZ affinity strategy is configured. The slot map carries no AZ metadata, so these strategies now fall back to their documented round-robin behavior (replicas for AZAffinity, replicas plus primary for AZAffinityReplicasAndPrimary) instead of hitting todo!() (#5909)client_side_cache configuration to the URI-based client creation API (create_client_from_uri) — supports max_cache_kb, entry_ttl_ms, eviction_policy (LRU/LFU), and enable_metrics via JSON options; cache_id is auto-generated internally (#5860)jedis-4-compatibility) and shared jedis-compat-shared module for common Jedis 4/5 facade code (#5751)Nothing published for this version
Core: Improve performance of Client initial connection with TLS (#6219) by @jamesx-improving in https://github.com/valkey-io/valkey-glide/pull/6316
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.4.1...v2.4.2
Nothing published for this version
Backport #5916 to release 2.4: Core: Propagate per-command response timeout to multiplexed connection by @jeremyprime in https://github.com/valkey-io/
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.4.0...v2.4.1
Nothing published for this version
Nothing published for this version
Client-Side Caching Support: Full client-side caching support enabling local caching of frequently accessed data to reduce round-trips to the server
Client-Side Caching Support: Full client-side caching support enabling local caching of frequently accessed data to reduce round-trips to the server
Transparent Compression: Client-side transparent compression feature enabled. This feature is experimental
Valkey Search 1.2 Support: Comprehensive FT.* command support for the Valkey Search module
NodeDiscoveryMode for Standalone Clients: New configuration option to allow for better integration with proxies
JSON module support: Full support for Valkey JSON module
URI-Based Client Creation API: New FFI create_client_from_uri function accepting Redis/Valkey URI strings with optional JSON configuration
Performance Enhancements: Multiple performance enhancements in the Valkey Glide Core
FT.* commands by @edlng in https://github.com/valkey-io/valkey-glide/pull/5571glideft refactor into release branch by @edlng in https://github.com/valkey-io/valkey-glide/pull/5863Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.3.1...v2.4.0
tokio::time::sleep may not fire on time. The watchdog uses a separate OS thread to guarantee timeout delivery, preventing commands from hanging indefinitely when the async runtime is overloaded. (#5752)request_timeout (including blocking command timeouts) is now enforced at the connection level, ensuring commands time out reliably instead of waiting indefinitely with Duration::MAX. (#5581)refreshTopologyFromInitialNodes=true and the initial/seed node is unreachable (e.g., dead primary during failover), the topology refresh now queries healthy existing connections instead of failing silently. This complements #5812 to enable failover recovery when the seed node itself is the failed node. (#5814)value_from_split_pointer(high_u32, low_u32) with value_from_pointer(i64) in the Node.js Rust client to accept response pointers as a single integer, eliminating the high/low split that caused upper 32-bit truncation when protobufjs.util.Long was null (#5851)BaseClientCommands, GlideClientCommands, GlideClusterClientCommands, etc.) as a public package for testing and abstraction (#4900)NodeDiscoveryMode configuration for standalone clients — STATIC mode skips role detection for proxy compatibility (e.g., Envoy), DISCOVER_ALL mode discovers full topology (primary + all replicas) from any starting node (#5724)create_client_from_uri FFI function accepts Redis/Valkey URI strings and optional JSON configurationoptions parameter for ft.search() and ft.aggregate() is now optional. (#5571)numberOfEdges, vectorsExaminedOnConstruction, vectorsExaminedOnRuntime) matching the TypeScript type definitions (#5571)break for NUMERIC field case in FT.CREATE serialization, preventing fall-through to VECTOR case (#5571)PeriodicChecks configuration to AdvancedClusterClientConfiguration — supports PeriodicChecksEnabled, PeriodicChecksDisabled, and PeriodicChecksManualInterval modes for controlling cluster topology check intervals (#5842)lib-ver instead of unknown in CLIENT INFO responses (#5634)now_or_never() with proper poll() in recovery to fix busy-spin, remove loop{} from poll_flush, and fail pending requests immediately during recovery to prevent OOM under sustained partition (#5715, #5716)Logger.logger_level now returns a Level enum instead of an internal PyO3 type, consistent with the sync client (#5637)Nothing published for this version
Nothing published for this version
Nothing published for this version
Address issues with tests in CI (#5625) by @alexr-bq in https://github.com/valkey-io/valkey-glide/pull/5629
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.3.0...v2.3.1
IAMTokenProvider trait so the cluster reconnection loop fetches a fresh IAM token before each connection attempt, preventing AUTH failures when tokens expire during node downtime (#5663)cluster_params.password has a valid IAM token in refresh_slots_inner and reconnect_to_initial_nodes, not just the per-node reconnection loop (#5663)current_thread tokio runtime for sync FFI clients, eliminating cross-thread condvar wake overhead on every command (#5083, #5624)GLIDE_TOKIO_WORKER_THREADS env var to control tokio worker thread count for the async client runtime (default 1) (#5083)Dynamic PubSub Support: Added dynamic PubSub methods across all clients (Node, Java, Python Sync, Go) including subscribe, unsubscribe, psubscribe, pu
🚀 Key Features
Dynamic PubSub Support: Added dynamic PubSub methods across all clients (Node, Java, Python Sync, Go) including subscribe, unsubscribe, psubscribe, punsubscribe, ssubscribe, sunsubscribe with timeout support, getSubscriptions() for subscription state tracking, and pubsub_reconciliation_interval configuration.
mTLS Support: Added client certificate and private key support for mutual TLS (mTLS) authentication across Core and Python clients.
Java 8 Compatibility: Made the Java client compatible with Java 8, expanding support for legacy environments.
Java Uber JAR: Added uber JAR support for multi-platform builds, simplifying deployment across different architectures.
Read-Only Mode: Added read-only flag support in Core, enforcing no write commands and allowing connections without a primary node.
Go ACL Commands: Added comprehensive ACL server management commands (ACL CAT, ACL DELUSER, ACL DRYRUN, ACL GENPASS, ACL GETUSER, ACL LIST, ACL LOAD, ACL LOG, ACL LOGRESET, ACL SAVE, ACL SETUSER, ACL USERS, ACL WHOAMI).
Go Cluster Management: Added CLUSTER management commands (CLUSTER INFO, CLUSTER NODES, CLUSTER SHARDS, CLUSTER SLOTS, CLUSTER KEYSLOT, CLUSTER MYID, CLUSTER MYSHARDID, CLUSTER GETKEYSINSLOT, CLUSTER COUNTKEYSINSLOT, CLUSTER LINKS).
Node OpenTelemetry Enhancement: Added parent span context propagation support for OpenTelemetry integration.
Python Sync Performance: Added support for bytearray and memoryview as command argument types and response buffer support for get() to reduce memory copies and improve performance.
Python Sync OpenTelemetry: Added OpenTelemetry support with traces and metrics configuration for the sync client.
Java New Commands: Added EVAL_RO, EVALSHA_RO, SCRIPT DEBUG, KEYS, MIGRATE, WAITAOF, and comprehensive ACL commands.
🎉 New Contributors
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v1.2.0...v2.3.0
Thanks to everyone who contributed to this release!
@xShinnRyuu, @affonsov, @alex-rehnby-martin, @jduo, @avifenesh, @prashanna-freshworks, @jamesx-improving, @Sasidharan3094, @taylor-curran, @shohamazon, @hank95179, @Aryex, @liorsve, @pratheepkumar, @xdk-amz, @prateek-kumar-improving, @oxy-star, @Maayanshani25, @kaushikvijayakumar, @tdschwarz, @rzoutm, @naoki-tateyama, @yipin-chen, @invokermain, @Sylvain-Royer, @v3rm0n, @jh1231223, @jeremyprime, @danielr, @louwersj, @a-potapenko-g
bytearray and memoryview as command argument types to improve performance by reducing copies (#5492)convertMapToKeyValueStringArray and convertMapToKeyValueGlideStringArray to fix performance bottleneck and ArrayStoreException (#5602)hostname: "" in node metadata, which was used as the connection address instead of falling back to the IP.Failed to convert napi value Undefined into rust type u32 error (#5128)Nothing published for this version
Nothing published for this version
Nothing published for this version
Fix to allow auto-detect npm tag to handle RC and backport releases by @xShinnRyuu in https://github.com/valkey-io/valkey-glide/pull/5669
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.2.9...v2.2.10
[Backport-2.2] Fix/empty hostname cluster slots#5373 by @jamesx-improving in https://github.com/valkey-io/valkey-glide/pull/5397
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.2.7...v2.2.9
Core: improve topology refresh reliability and handle ReadOnly errors by @shohamazon in https://github.com/valkey-io/valkey-glide/pull/5225
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.2.6...v2.2.7
Cherry-picked into release-2.2 by @shohamazon in https://github.com/valkey-io/valkey-glide/pull/5185
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.2.5...v2.2.6
[Backport] [2.2] Rust: Rust 2024 compatibility fix (#5101) by @jamesx-improving in https://github.com/valkey-io/valkey-glide/pull/5139
Failed to convert napi value Undefined into rust type u32 error (#5128) by @jamesx-improving in https://github.com/valkey-io/valkey-glide/pull/5143Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.2.4...v2.2.5
Failed to convert napi value Undefined into rust type u32 error (#5128)Nothing published for this version
Due to recent NPM policy changes, there are issues publishing to NPM. There are no changes to the Node client since v2.2.1. Please continue to use v2.
Due to recent NPM policy changes, there are issues publishing to NPM. There are no changes to the Node client since v2.2.1. Please continue to use v2.2.1 for the Node client. Once we resolve the issue, we will publish the update to NPM.
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.2.2...v2.2.3
Nothing published for this version
Due to recent NPM policy changes, there are issues publishing to NPM. There are no changes to the Node client between v2.2.1 and v2.2.2. Please contin
Due to recent NPM policy changes, there are issues publishing to NPM. There are no changes to the Node client between v2.2.1 and v2.2.2. Please continue to use v2.2.1 for the Node client. Once we resolve the issue, we will publish the update to NPM.
rustls-pemfile usage to rustls-pki-types (#5037) by @jduo in https://github.com/valkey-io/valkey-glide/pull/5045sniffio (#5018) by @jamesx-improving in https://github.com/valkey-io/valkey-glide/pull/5044Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.2.1...v2.2.2
Nothing published for this version
Nothing published for this version
JAVA: Add Windows support to Java build matrix and workflows by @affonsov in https://github.com/valkey-io/valkey-glide/pull/5019
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.2.0...v2.2.1
Nothing published for this version
Nothing published for this version
Windows Support for Java Client: Migrated the Java client from Unix Domain Sockets (UDS) to JNI-based communication.
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.1.1...v2.2.0
Thanks to everyone who contributed to this release!
@avifenesh, @affonsov, @jamesx-improving, @liorsve, @jbrinkman, @Maayanshani25, @xShinnRyuu, @alex-bq, @omida-bitquill, @prateek-kumar-improving, @alexey-temnikov, @shohamazon, @manz, @jeremyprime, @currantw, @hank95179, @jh1231223, @jduo, @xdk-amz, @yipin-chen
go mod vendor support via CI-generated marker files (#4721)BaseClient.ts, GlideClient.ts, and GlideClusterClient.ts. (#4607)install-rust-and-protoc action to explicitly include the rustfmt and clippy components. (#4816)Nothing published for this version
Nothing published for this version
Python: improve UDS socket error handling by @jbrinkman in https://github.com/valkey-io/valkey-glide/pull/4755
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.1.0...v2.1.1
Nothing published for this version
Nothing published for this version
Jedis Compatibility Layer: Released a new compatibility layer for the Java client to greatly simplify migrating to Valkey GLIDE from Jedis.
Full Changelog: https://github.com/valkey-io/valkey-glide/compare/v2.0.1...v2.1.0
Thanks to everyone who contributed to this release!
@acarbonetto, @adarovadya, @affonsov, @alexr-bq, @amirreza8002, @Angraybill, @asafpamzn, @avifenesh, @barshaul, @BoazBD, @carthas, @cyip10, @EdricCua, @edlng, @eifrah-aws, @fenichelar, @Funk66, @GilboaAWS, @goller, @gulbaki, @ikolomi, @jamesx-improving, @janhavigupta007, @jeremyprime, @jhpung, @jbrinkman, @jonathanl-bq, @Juneezee, @karthiknathan06, @leszek-bq, @liorsve, @ljmcg, @Maayanshani25, @marcoreni, @meitalkra, @MikeMwita, @mo-amzn, @mohsen-alizadeh, @mortymacs, @mthemis-provenir, @niharikabhavaraju, @nihohit, @omangesg, @omida-bitquill, @prateek-kumar-improving, @rueian, @Sa1Gur, @SergioNR, @shohamazon, @thearchitector, @tjzhang-BQ, @umit, @X39, @yipin-chen, @Yury-Fridlyand
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
One column per month.