# MinIO AIStor EDGE.2026-03-03T18-23-09Z Released: 2026-03-03 This EDGE release delivers significant performance improvements to distributed locking and object reads, resolves critical concurrency and data-integrity bugs across replication, erasure, IAM, and inventory subsystems, and closes two security vulnerabilities. It also introduces new S3-compatible APIs, multi-NIC internode networking, and MegaRAID SMART health monitoring for enterprise storage hardware. --- ## Downloads ### Binary Downloads | Platform | Architecture | Download | | -------- | ------------ | ------------------------------------------------------------------------ | | Linux | amd64 | [minio](https://dl.min.io/aistor/minio/edge/linux-amd64/minio) | | Linux | arm64 | [minio](https://dl.min.io/aistor/minio/edge/linux-arm64/minio) | | macOS | arm64 | [minio](https://dl.min.io/aistor/minio/edge/darwin-arm64/minio) | | macOS | amd64 | [minio](https://dl.min.io/aistor/minio/edge/darwin-amd64/minio) | | Windows | amd64 | [minio.exe](https://dl.min.io/aistor/minio/edge/windows-amd64/minio.exe) | ### FIPS Binaries | Platform | Architecture | Download | | -------- | ------------ | ------------------------------------------------------------------------ | | Linux | amd64 | [minio.fips](https://dl.min.io/aistor/minio/edge/linux-amd64/minio.fips) | | Linux | arm64 | [minio.fips](https://dl.min.io/aistor/minio/edge/linux-arm64/minio.fips) | ### Package Downloads | Format | Architecture | Download | | ------ | ------------ | -------------------------------------------------------------------------------------------------------------------------------- | | DEB | amd64 | [minio_20260303182309.0.0_amd64.deb](https://dl.min.io/aistor/minio/edge/linux-amd64/minio_20260303182309.0.0_amd64.deb) | | DEB | arm64 | [minio_20260303182309.0.0_arm64.deb](https://dl.min.io/aistor/minio/edge/linux-arm64/minio_20260303182309.0.0_arm64.deb) | | RPM | amd64 | [minio-20260303182309.0.0-1.x86_64.rpm](https://dl.min.io/aistor/minio/edge/linux-amd64/minio-20260303182309.0.0-1.x86_64.rpm) | | RPM | arm64 | [minio-20260303182309.0.0-1.aarch64.rpm](https://dl.min.io/aistor/minio/edge/linux-arm64/minio-20260303182309.0.0-1.aarch64.rpm) | ### Container Images ```bash # Standard docker pull quay.io/minio/aistor/minio:EDGE.2026-03-03T18-23-09Z podman pull quay.io/minio/aistor/minio:EDGE.2026-03-03T18-23-09Z # FIPS docker pull quay.io/minio/aistor/minio:EDGE.2026-03-03T18-23-09Z.fips podman pull quay.io/minio/aistor/minio:EDGE.2026-03-03T18-23-09Z.fips ``` --- ## Security & Compliance ### Security Updates - **Fixed unauthenticated log streaming** — `APILogsHandler`, `ErrorLogsHandler`, and `AuditLogsHandler` lacked authentication checks, allowing any client with network access to stream server logs containing bucket names, object keys, access keys, and full HTTP request/response details. All three endpoints now enforce authentication (#3065) - **Fixed OpenID audience check bypass** — when the audience check returned an error, execution continued and valid credentials were still issued. The blast radius was limited to callers with a properly signed token but credentials must not be issued after a failed audience check (#3059) - Updated `golang.org/x/net` to address GO-2026-4559 (20f14134) - Updated dependencies to resolve `govulncheck` findings (e3c83dd6) ### Software Bill of Materials (SBOM) This release includes comprehensive SBOM documentation in multiple formats: - [SPDX JSON](sbom-EDGE.2026-03-03T18-23-09Z.spdx.json) - Standard SBOM format - [CycloneDX JSON](sbom-EDGE.2026-03-03T18-23-09Z.cyclonedx.json) - Security scanner compatible - [Go Modules](go-modules-EDGE.2026-03-03T18-23-09Z.txt) - Human-readable dependency list --- ## New Features - **UpdateObjectEncryption API** — implements S3 in-place encryption type change, allowing objects to transition between SSE-S3, SSE-KMS, and no encryption without re-uploading data (#3021) - **Multi-NIC internode networking** — adds support for multiple network interfaces to achieve aggregate internode bandwidth. Introduces config v3 format where NIC addresses live directly in pool node definitions. Existing v1/v2 configs are auto-migrated on startup (#2968) - **MegaRAID SMART health monitoring** — adds SMART health data collection for physical drives behind Broadcom/LSI MegaRAID controllers (HPE ProLiant, Dell PowerEdge, Cisco UCS) via the `megaraid_sas` kernel driver ioctl. Previously these systems reported `smartMetricStatusCollectionError = 4` for all drives (#3190) - **Kubernetes TokenReview** — adds a configurable `oidc` token validation mode via `MINIO_KUBERNETES_TOKEN_VALIDATION`. Unlike `tokenreview`, the `oidc` mode uses Kubernetes OIDC endpoints and requires no additional RBAC cluster roles (#3100) - **Self-service password change** — new `POST /minio/admin/v3/change-my-password` endpoint lets users change their own password without requiring `admin:CreateUser` policy permissions. Administrators can block this with the new `ChangeMyPasswordAdminAction` (#2835) - **`Prefer: return=full` header** — when a request includes `Prefer: return=full`, the server returns the complete object and ignores the `Range` header, responding with `Preference-Applied: return=full`. Enables cache/proxy layers to fetch whole objects in a single request (#2898) --- ## Performance Improvements - **Write lock coalescing** — write locks are now coalesced per node, mirroring existing read-lock coalescing. The first writer acquires the distributed lock via RPC; subsequent writers on the same node queue locally with zero additional RPCs until the current holder completes, significantly reducing RPC traffic under heavy write workloads (#3197) - **Batch lock refresh RPCs** — reduces distributed lock refresh traffic from N×M to M RPCs per cycle by coalescing all refresh calls to the same lock server into a single `BatchRefresh` RPC, eliminating per-lock refresh goroutines under high concurrency (#3060) - **Direct-write GetObject for plain objects** — for unencrypted, uncompressed objects, erasure decoded data is streamed directly to the HTTP `ResponseWriter` without an intermediate `io.Pipe` or goroutine. Encrypted and compressed objects retain the existing code path (#3072) - **Singleflight for hot metadata paths** — concurrent calls to data-usage reads, `HeadBucket` peer fan-outs, and multipart upload ID verification are now deduplicated via singleflight, reducing redundant disk and RPC operations under concurrent load (#3135) - **Singleflight for getObjectFileInfo** — concurrent `GetObject`/`HeadObject` requests for the same object share a single disk fan-out across all erasure set disks instead of each independently querying all drives. Results are deep-cloned to prevent mutation aliasing across handlers (#3118) - **NSScanner cache offloading** — per-bucket cache load/save and scan mode computation are now handled locally by each peer rather than marshaled and transferred over the wire through the coordinator. Reduces CPU and network overhead in large clusters (#3030) - **Prefer-local erasure read** — fixed dead code path where the prefer-local optimization was bypassed in distributed mode because `Hostname()==""` is never true for URL endpoints. Now uses `IsLocal()` so local drives are correctly preferred for reads (#3139) --- ## Bug Fixes ### Data Integrity - Fixed silent deletion of valid objects: three bugs composed to cause false `DeleteDanglingObject` on versioned erasure sets when a `CopyObject` (versionOnly=true) partially wrote a new versionID to only a subset of disks and a concurrent `HeadObject` polled the object. `pickLatestQuorumFilesInfo` now allocates a fresh error slice instead of mutating the caller's, and `writeAllMetadataWithRevert` correctly returns write-quorum errors regardless of the revert flag (#3230) - Fixed `CleanAbandonedData` writing stale metadata to drives that already held current data, potentially corrupting object metadata on healing (#3081) - Fixed erasure `ToFileInfo` panic on nil `PartActualSizes` — metadata written before September 2023 or without compression may have nil `PartActualSizes`; indexing it unconditionally caused panics during reads of those objects (#3200) - Fixed `GetObjectReader` data race between concurrent `Read` and `Close` calls (#3133) - Fixed `lazyPipeReader` goroutine outliving `GetObjectReader.Close` — the background erasure read goroutine could access object data after cleanup functions released the namespace read lock. `Close` now blocks until the goroutine exits before running any cleanup (4065f3e2) - Fixed `kvFields` substring collision on key matching — `strings.Index` caused shorter keys to falsely match inside longer keys (e.g. `speed=` matching within `idle_speed=`) (#3202) ### Replication - Fixed replication config cleared in the scanner when `ListBucketTargets` fails — this dropped the `Config` field causing the ILM evaluator to skip the `EdgeSyncBeforeExpiry` guard and expire objects with non-`COMPLETED` replication status (#3201) - Fixed replication tag filter using OR instead of AND semantics — S3 spec requires all tags in an `` filter to match; previously an object matching any single tag was replicated (#3112) - Fixed last-hour replication metrics merge silently dropping all receiver-side data — the merge accumulator was initialized from a zero-value struct instead of the receiver, producing zeroed remote-side replication metrics (#3117) - Fixed `older-than` filter overflow in the replication reset handler — `days` is a `time.Duration` (nanoseconds) and dividing by 24 produced day counts in the trillions, making the filter non-functional and causing all objects to be resynced regardless of age (#3120) - Fixed site replication concurrency: indefinite blocking when peer clusters become unreachable, data races on shared removal state, and missing synchronization for concurrent removal attempts (#2708) - Fixed incorrect `Partial` status returned when `RemoveAll=true` and a peer is unreachable — local cleanup had already run so returning `Partial` caused unnecessary retry loops (#3198) ### IAM / Authentication - Fixed IAM broadcast leader election never selecting a leader and IAM state hash never reflecting mutations — `globalLocalNodeName` (bare `host:port`) was compared against grid targets (`http://host:port`), so no node was ever elected and periodic IAM state convergence never triggered (#3142) - Fixed data race on `rolesMap` reads in IAM — `HasRolePolicy`, `GetRolePolicy`, `IsAllowedServiceAccount`, and `IsAllowedSTS` read the map without holding `RLock()`, racing with the background OpenID initializer. Replaced with lock-free `xsync.Map` (#3091) - Fixed STS policy resolution order — JWT claims are now checked before PolicyDB in `IsAllowedSTS` and `TemporaryAccountInfo`, preventing stale PolicyDB entries from overriding claim-based policies (#3138) - Fixed race condition in `getPeerName` (#3178) - Fixed nil pointer dereference in `PeerPolicyMappingHandler` (#3040) - Fixed config changes being applied during rolling upgrades — newer config keys introduced by upgraded nodes can no longer be persisted until all nodes are upgraded and the cluster API version is bumped (#3169) - Fixed LDAP and OpenID purge routines starting before the respective services are reachable, potentially purging valid sessions on startup (#3002) ### Tables / Delta Sharing - Fixed Delta Sharing queries silently dropping files with deletion vectors — tables with row-level DELETEs returned only unaffected data files; DV presigned URLs also returned 404 because the binary file path was not resolved (#3105) - Fixed Delta Sharing deletion vector descriptors not being transformed for standard clients — raw `storageType = "u"` (Z85-encoded) descriptors are now rewritten to `storageType = "p"` with presigned URLs. Absolute-path DV descriptors are rejected (#3122) - Fixed `loadAllShards` returning 500 on concurrent namespace delete (#3150) - Fixed orphaned `metadata.json` files left when a commit loses the CAS race — asynchronous cleanup now removes the superseded file (#3182) - Fixed table transaction recovery direction — the system was rolling forward instead of back when recovering staged-to-live, potentially leaving tables in an inconsistent state (#2996) - Fixed warehouse creation returning 500 instead of 409 when the name conflicts with an existing bucket (#3156) - Improved error message when `mc mb` targets an existing Tables warehouse — now returns a descriptive conflict error instead of the generic `BucketAlreadyOwnedByYou` (#3157) - Fixed `Always set non-nil error to result` in Delta Sharing response paths (#2992) ### Inventory - Fixed `saveFinalJobStateWithRetry` overwriting concurrent control-plane changes — when a `CancelJob`, `SuspendJob`, or `PutConfig` changed job metadata between retry attempts, the executor's terminal state was written over the control-plane change (#3163) - Fixed stale `Lock` field left on jobs after a `Suspend → Resume` cycle, preventing schedulers from claiming the job on the next cycle (#3192) - Fixed goroutine leak and incorrect NDJSON manifest key names in inventory output (#2938) - Added panic recovery with automatic 1-minute backoff restart to the inventory scheduler and executor, preventing a single job panic from cascading across cluster nodes (#2522) ### Multipart / Object Operations - Fixed double XML response during `CompleteMultipartUpload` — when a precondition failure (`If-Match`/`If-None-Match`) occurs after the whitespace keep-alive goroutine has already written HTTP 200 and the `