On the Inner Ring side, the container processor now validates and applies an optional eACL table when processing CreateContainerV2 notary events. The feature depends on changes introduced in NeoFS SDK RC20.
Performance optimizations
Four performance improvements target read-path operations across different storage scenarios.
The release optimizes GET requests for erasure-coded objects that do not require full EC recovery (PR #3996). When the requested data is already present on the local node or the first data part holder, NeoFS now streams directly from that source instead of assembling all EC parts.
Request forwarding has also been reworked (PR #4008, #4021). When a storage node forwards GET, HEAD, SEARCH, or RANGE requests to another node, it previously re-signed the forwarded request on every attempt. Signing now happens at most once per request, with additional improvements to node selection logic. Primary holders are tried first for REP containers, with shuffling for load balancing.
Split object assembly now prefetches child objects concurrently while streaming (PR #4014, #4042). Large multi-part objects previously fetched each child sequentially. A new pipelined approach uses a configurable look-ahead window to pre-fetch subsequent children in the background while one child streams to the client, reducing end-to-end GET latency for large files.
FSTree range reads have been optimized to use proper seeking instead of copying skipped bytes to a no-op sink (PR #4016), reducing CPU overhead on RANGE operations for non-compressed objects.
Bug fixes
The release addresses seven bugs, with three of particular note.
Ranged object reads from the FSTree backend could return more payload bytes than requested (PR #4046, #4051). The range-limiting logic did not correctly account for bytes already present in its read buffer prefix. The same PR also fixed incorrect remote EC part ranges used during policer checks, which shared a root cause in range boundary calculations.
GET requests marked as payload-only could be incorrectly denied during eACL rechecks (PR #4024, #4029). When a request omitted the object header, the node failed to match against attribute-based eACL rules, blocking access for users relying on those rules.
The neofs-ir epoch metric was not populated at startup and failed to update after FS chain connection loss (issue #4035), causing monitoring dashboards to display incorrect epoch data.
Additional fixes addressed a storage node panic on startup without the Control service configured, and protobuf scanning and data race issues in EC GET response handling.
Dependencies and upgrade
NeoFS Node v0.54.0 updates its dependency on NeoGo to v0.120.0, which patches multiple security vulnerabilities, and on NeoFS SDK to RC20, which provides the underlying support for atomic eACL creation.
No configuration changes or metabase migrations are required to upgrade from v0.53.0. Binaries are available for Linux, macOS, and Windows on both amd64 and arm64 architectures.
The full release notes can be found at the link below:
https://github.com/nspcc-dev/neofs-node/releases/tag/v0.54.0
neonewstoday.com
