Release Notes
This document provides the release history of JuiceFS Cloud Service. For JuiceFS Community Edition release notes, please refer to our GitHub page.
JuiceFS Client maintains good forward and backward compatibility. However, when upgrading across major versions (for example, 5.1 → 5.2), compatibility issues may occasionally arise. If you need to roll back after a major upgrade, see juicefs version.
JuiceFS 5.4.5 (2026/11/17)
- New
- mount: Added the
--idmappedoption to support Linux ID-mapped mounts (requires Linux 6.12+) - mount: Added the
--no-internaloption to disable access to internal virtual files on FUSE mounts - Added the
--cpu-affinityglobal option on Linux to bind client processes to specified CPU cores - Added
--log-rotate-sizeand--log-rotate-filesoptions to automatically rotatejuicefs.log - Python SDK: Added
importto import files from object storage, and addeduse_iorpcandiorpc_modeparameters
- mount: Added the
- Improved
- mount: Support concurrent decommissions of multiple cache nodes and reduce memory usage during cache migration
- mount: Sync and drop kernel page cache after cache writes
- mount: Improve distributed cache communication performance
- mount: Validate cache keys received from cache group peers
- mount: Support local and global upload/download bandwidth limits for secondary object storage (
bucket2) - mount: Support updating
shardsand storage class configurations from the console - mount: Support upgrading both daemon and worker processes
- mount: Increase the default open file descriptor limit (
nofile) to 1,000,000 - clone: Automatically create missing parent directories for the destination path
- dump: Added the
--parallel(-p) option to speed up metadata dumping with parallel requests - RDMA: Limit concurrent outgoing requests to reduce memory usage
- Upgraded
go-cephto v0.38.0
- Bugfix
- mount: Enforce subdirectory and whitelist boundaries for internal commands (
info,summary,rmr,compact,clone,warmup, etc.), ensure whitelists are always case-sensitive, and fix root directory visibility when a whitelist is enabled - mount: Fixed audit log directory filtering when mounting a subdirectory
- mount: Fixed an issue where non-root users could not update permissions during on-demand import
- mount: Do not count clean EOF of compressed block reads as object storage errors
- mount: Prevent
linkandrenamerequests from being interrupted - mount: Fixed a data race in
utils.WithTimeoutwhen reading partial blocks - clone: Fixed compatibility with metadata services earlier than 5.4
- import: Fixed an issue where importing into the volume root directory failed
- rmr: Return a non-zero exit code when removal fails
- debug: Fixed command-line flag parsing and accurately detect internal configuration files
- sync: Fixed inaccurate rate limiting when
--limitis enabled - Python SDK: Fixed read errors after
merge, incorrect privilege checks for users in therootgroup, hostargvbeing overwritten when loading shared libraries, cross-platformerrnomapping issues, and issues with thebufferingparameter and NUL-byte path validation inopen - Hadoop SDK: Fixed a data race when updating superuser and supergroup configurations
- mount: Enforce subdirectory and whitelist boundaries for internal commands (
JuiceFS 5.4.4 (2026/10/19)
- New
- Support configuring audit logs to record operations only for specified directories
- Support coexisting with the JuiceFS Community Edition client when mounting via
/sbin/mount.juicefs - Python SDK: Support loading user and group mappings from custom
passwdandgroupfiles, and support additional cache and readahead options
- Improved
- mount: Improved cache node commission to reduce cache misses during scaling
- mount: Improved data caching and request forwarding during cache node decommission
- mount: Faster handling of cache node topology change events and asynchronous hash ring balancing
- mount: Prevent compaction from exhausting the object storage upload concurrency limit
- mount: Avoid invalidating the local kernel page cache for the file descriptor currently being written
- mount: Clear client metadata cache when the mirror node becomes stale
- mount: Check the modification time of
mount.gzfirst during client upgrades and improve upgrade logging - heal: Sync data only one-way from the source volume to the mirror volume for mirrored file systems, and unify replication metrics
- debug: Added the
--file-limitoption to limit the size of the diagnostic archive, and include thread IDs (tid) in collected kernel stacks
- Bugfix
- mount: Fixed a potential
ENOENTerror whenopenorgetattrraces with an overwritingrename - mount: Fixed a byte counter overflow when rebalancing large amounts of cached data
- dump: Fixed an issue where the progress bar was not updated when dumping to a compressed file
- find: Fixed a data race when traversing split directories with
--follow-links(-L) that could cause missing entries in the output - RDMA: Fixed a potential nil pointer dereference when creating a new connection fails
- RDMA: Continue releasing connection resources even if
rdma_disconnectfails
- mount: Fixed a potential
JuiceFS 5.4.3 (2026/9/22)
- New
- mount: Added the
--commissionoption to gracefully add a cache-group node - heal: Support checking and repairing replicated data in a specified directory
- mount: Added the
- Improved
- Added metrics for local object storage access during data replication
- find: Improved output performance
- mount: Improved bulk cache node registration to prevent widespread cache invalidation during registration
- Perform one-way data synchronization only for mirrored file systems
- Improved the accuracy of data volume metrics during data synchronization
- Allow client mount options to be adjusted online
- Improved metadata proxy stability
- warmup: Comply with bandwidth limits for both background tasks and object storage
- Bugfix
- Fixed an issue where RDMA connections could fail to be established
- Fixed a potential deadlock when closing RDMA connections
- Fixed an issue where smoothly upgrading older read-only clients could result in nested mount points (introduced in 5.4.1)
- Fixed an issue where mounting with fstab could prevent diagnostic information from being collected
- Fixed an issue where directory whitelist restrictions could become ineffective when the root directory was split
- Fixed an issue where an unexpected interruption during data migration between cache disks could leave incomplete cached data
- Fixed a port listening failure when sharing block devices
JuiceFS 5.4.2 (2026/9/2)
- New
- Support setting a white list of directories for tokens
- Support configuring mirrored file systems to sync objects from the source on demand
- Improved
- mount: Added the
blockdev-portoption to specify the listening port for block devices - mount: Automatically optimize operating system network parameters to better support short-lived connections
- mount: Optimized connection management for cache nodes to better handle high-load scenarios
- mount: Reduced long-tail latency for cache nodes under high memory pressure
- mount: Hide token values in system mount options
- mount: Support reading encrypted tokens from configuration files
- mount: Ignore object-not-found errors for
Getrequests triggered by compaction - warmup: Support warming up only a specified range within a file
- sync: Automatically update expired local configuration files
- RDMA: Improved connection stability in multi-NIC environments
- mount: Added the
- Bugfix
- mount: Fixed an issue where abnormal cache files could cause the process to crash
- mount: Fixed an issue where some requests could be interrupted during smooth upgrades
- mount: Fixed an issue where file handle metrics could become negative
- restore: Fixed an issue where failed items were not counted in the progress bar
- find: Fixed an issue where some files could be missed when traversing split directories
- Fixed an issue where attempting to write to files in compatibility mode might not return an error as expected
- Fixed an issue where removing block devices could cause the client process to crash
JuiceFS 5.4.1 (2026/8/12)
- New
- mount: Added the
--decommissionoption to gracefully remove a cache-group node - mount: Added the
--cache-asideoption to enable cache-aside mode - mount: Added the
--writeback-threshold-sizeoption to set the maximum writeback block size - mount: Added the
--read-onlyoption to enable read-only mode - mount: Added the
--mem-limitoption to limit client process memory - mount: Added the
--merge-cacheoption to enable the new merged cache mode - mount: Support SM encryption for data
- mount: Support QPS limiter for distributed cache groups
- mount: Added monitoring metrics for FUSE read/write latency
- import: Added the
--trackoption to manage the full lifecycle of imported objects - import: Added the
--delete-dstoption to remove files that no longer exist in object storage - restore: Added
--start-timeand--end-timeoptions to filter the time range for trash recovery - sync: Added the
--delete-src-afteroption to delete source objects after syncing - clone: Added the
--threadsoption to adjust concurrency - gc: Added the
--sort-bufferoption to support external sorting in dry mode - fsck: Added the
--sort-bufferoption to support external sorting - merge: Added the
--files-fromoption to read the list of paths to merge from a file - Windows: Added the
AddWriteEaAccessWinFsp mount option
- mount: Added the
- Improved
- mount: Optimized thread scheduling and lock management for better performance
- mount: Hide token values in
psoutput - mount: When
--writebackis enabled, also build a read cache for writeback blocks - import: No longer updates atime when reading on-demand imported directories
- warmup: No longer updates directory atime during warmup
- clone: Deprecated the
--deleteand--forceoptions - sync: Deprecated the
--time-selectoroption - gc/destroy: Allow unordered object listings
- RDMA: Optimized performance and stability
- RDMA: Fall back to TCP when the last RDMA device becomes unavailable
- Bugfix
- mount: Fixed an issue that could cause mounting to hang on Linux kernel 7.1.3
- mount: Fixed an issue where the client might fail to start when security mechanisms such as SELinux are enabled
- Fixed an issue where blocks were not cached as expected in some cases when
cache-large-writewas enabled - Fixed an issue where object checksums might not be verified as expected when local cache was disabled
- Fixed an issue where modifying files in compatibility mode could incorrectly report an error
- Fixed an issue where write metrics were lost when using memory as data cache
- Fixed an issue where data writes spanning chunks might not be committed in the expected order
- Fixed an issue where the available space ratio could be incorrectly retrieved when the cache disk was formatted with Btrfs
- Fixed the incorrect FIFO file type character returned by the API
JuiceFS 5.3.14 (2026/7/30)
- Improved
- clone: Added the
threadsoption to adjust the number of threads used for concurrent cloning - Require writes to the primary storage to succeed when data replication is enabled
- clone: Added the
JuiceFS 5.3.13 (2026/7/22)
- Improved
- sync: Added the
remote-copyoption to support server-side copy - sync: Optimized path resolution performance
- sync: Added the
- Bugfix
- Fixed a distributed cache false error that could occur when imported object names contained
# - WebDAV: Fixed a potential null pointer crash when calling
staton a file - object/oss: Fixed an issue where CRC verification might not be executed correctly
- Fixed a distributed cache false error that could occur when imported object names contained
JuiceFS 5.3.12 (2026/7/3)
- Improved
- RDMA: Fixed a potential connection issue when a node runs as both an RDMA server and a client
- Bugfix
- Fixed incorrect inode numbers displayed by the Linux
findcommand (introduced in 5.3.10)
- Fixed incorrect inode numbers displayed by the Linux
JuiceFS 5.3.11 (2026/6/24)
- Improved
- RDMA: Support configuring multiple IP addresses on a single NIC
- RDMA: Prefer the
rdma linkcommand to discover available RDMA devices - RDMA: Improved error handling for better stability during failures
- Bugfix
- mount: Fixed a potential file descriptor leak during a smooth upgrade
- mount: Fixed an issue where a smooth upgrade in foreground mount mode could cause the new process to hang
- mount: Fixed an issue where files closed early during readahead could lead to incorrect
openfilescounts - Fixed a potential memory leak during data rebalancing in distributed cache nodes
- Fixed an issue where, with many cache-group nodes, some secondary cache-group nodes could be incorrectly registered to the primary cache group
- Fixed an issue where only half of the rate limit took effect for background tasks during cross-region replication
JuiceFS 5.3.10 (2026/6/9)
- Improved
- mount: Added the
read-sessionsoption to control the maximum number of segments for sequential file reads - gateway: Support setting file ownership via custom
HEADER - warmup: Added the
order-byoption to control output sorting order - sync: Support recording checkpoints for multipart-uploaded objects
- RDMA: Improved the strategy for selecting local NIC
- RDMA: Support hot unplug of RDMA devices
- RDMA: RDMA is now enabled solely based on the
rdma-networkoption - Optimized lock implementation to improve random small-block read performance under high load
- mount: Added the
- Bugfix
- mount: Fixed a potential memory leak when applications do not close directories as expected
- mount: Fixed a potential deadlock in the FUSE layer during process startup or shutdown
- mount: Fixed an issue where stale file sizes might be returned when reading directories and writing files occur simultaneously (introduced since 5.2.20)
- sync: Fixed an issue that could occur when decrypting small files
- Fixed an issue in Debian environments where client restart could fail due to an improperly set
PATHenvironment variable - Fixed an issue where readahead-related metrics could display as zero
- Fixed an issue where checksums might be calculated incorrectly when reading from local data cache
- Fixed an issue where caller line numbers were not displayed correctly in some logs (introduced since 5.3.9)
JuiceFS 5.3.9 (2026/5/27)
- Improved
- mount: Improved mount startup speed
- mount: Support hot reloading access credentials for object storage
- mount: When data replication is enabled and the primary storage network times out, prefer the secondary storage for a period of time
- mount: Support prefetching data objects from the secondary cache group
- warmup: Added the
--samplesoption to estimate the ratio of cached data by sampling - warmup: No longer updates file atime
- sync: Support encryption and decryption for transferred objects
- sync: Support periodic checkpoint writes for fast task restart
- import: Allow using file names as
Listprefixes - Reduced condition variable wait time in client
- Automatically fall back to TCP when RDMA connection is abnormal
- Added count metrics for warning and error level logs
- Bugfix
- Fixed an issue where smooth upgrade might fail when old processes could not exit normally
- warmup: Fixed an issue where results might be incorrect when checking multiple paths with
--check - sync: Fixed a false error that might occur when synchronizing the same symlink file concurrently
- find: Fixed an issue where empty directories might not be correctly identified when using
--empty - Windows: Fixed a potential client crash when integrating with AD services
JuiceFS 5.3.8 (2026/5/11)
- Improved
- find: Support
-printfto customize output format - sync: Reduce concurrent
lookupandresolverequests - import: Added the
delete-dstoption to remove previously imported files that are no longer needed - Update atime when accessing files using the
opencacheoption - Improved access log file rotation policy with support for specifying file size
- Improved RDMA usability
- object/oss: Support private link format
- find: Support
- Bugfix
- gateway: Fixed an issue where
HEAD path-to-file/did not return an error as expected - Fixed an issue where the checksum of local data cache files might be calculated incorrectly
- gateway: Fixed an issue where
JuiceFS 5.3.7 (2026/4/21)
- New
- Added the
findsubcommand for quickly filtering files that meet certain criteria
- Added the
- Improved
- mount: Added the
--access-logoption to write access logs to a specified file - mount: Ignored symlink cache configuration to avoid retaining old processes after a smooth restart (introduced since 5.3.5)
- gc: Added the
false-positive-rateparameter to adjust the false positive rate of the Bloom Filter - gc: Faster listing from Ceph
- sync: Allowed using the
jfs://protocol header without relying on configuration files, directly using the meta URL instead - Added
maxvalue for some FUSE and distributed cache metrics
- mount: Added the
- Bugfix
- Fixed an issue where the
debugsubcommand might fail to collect client logs - object/S3: Use the default credential chain in
autoS3Regionwhen the access key is empty
- Fixed an issue where the
JuiceFS 5.3.6 (2026/4/1)
- New
- Added the
attrsubcommand to manage extended attributes of files
- Added the
- Improved
- Enhanced client session reconnection checks to prevent connecting to the wrong cluster
- Automatically upgrade all extended-version binaries when upgrading the client
- Reduced client process memory usage
- Removed forced timeout interruption for
flockto avoid returningERESTARTSYSerrors
- Bugfix
- Fixed an issue where object storage GET request latency was not being counted
- Fixed an issue where the
stageBlockBytesmetric was not correctly counted when writeback was enabled - Fixed an issue where slow requests might not be correctly counted in container environments
JuiceFS 5.3.5 (2026/3/19)
- Improved
- mount: Support adjusting mount timeout via the
JFS_MOUNT_TIMEOUTenvironment variable - mount: Disable kernel symlink cache to avoid incorrect symlink targets
- Reduced client CPU usage
- Improved peak performance for small I/O random reads
- Allow iorpc and tcp to share the same port
- When background tasks are disabled, also disable the scheduled sync task for cross-region data replication
- mount: Support adjusting mount timeout via the
- Bugfix
- import: Fixed an issue that the configuration file might not be read correctly when using the
--conf-dirparameter - Fixed an issue that Ceph could not be accessed normally when the
JFS_NO_CHECK_OBJECT_STORAGEenvironment variable was set - SDK: Fixed a potential infinite loop when recording operation logs
- import: Fixed an issue that the configuration file might not be read correctly when using the
JuiceFS 5.3.4 (2026/3/10)
- New
- Support limiting the number and total size of trash files
- Improved
- Support metrics for FUSE read/write latency
- Optimized memory usage for GUID mapping and added local cache
- Release op lock when client disconnects
- merge: Automatically retry after defragmentation for large files
- gc: Reduced memory usage during runtime
- Bugfix
- Fixed an issue that when open cache is enabled, the first chunk of a file might read stale data
- Fixed an issue that some synchronization mechanisms did not work correctly when cross-region data replication was enabled
- clone: Fixed an issue that batch cloning might result in incomplete clones
- clone: Fixed a potential panic when encountering a directory without access permissions during cloning
JuiceFS 5.3.3 (2026/3/2)
- Improved
- Increased the maximum number of files per batch during clone
- sync: Optimized temporary file naming and limited directory-level cache entries
- RDMA: Automatically fall back to TCP when no device is available
- RDMA: Support setting handshake timeout and improved memory management and stability
- RDMA/iorpc: Multiple memory and connection stability optimizations (page management, synchronous write responses, cancel request handling)
- Bugfix
- Fixed an issue that incorrect file length might be used when calculating cache file checksums
- sync: Implemented idempotency when large files uploaded in chunks finish syncing, and fixed a potential panic on errors
- Windows: Fixed WinFsp version detection issue
JuiceFS 5.3.2 (2026/2/2)
- New
- sync: Support external rate limiting
- Improved
- sync: Reduced memory usage during directory traversal
- Improved RDMA compatibility and stability
- Bugfix
- Fixed an issue that cache could become ineffective on older kernels because tmpfs does not support directio
JuiceFS 5.3.1 (2026/1/14)
- New
- mount: Added the
--cache-scan-intervaloption to adjust cache disk scan interval - umount: Support
--update-fstabto remove corresponding configuration - fsck: Allow checking data integrity for a specific path
- debug: Support using
--pidto specify process ID - Support ioctl to set a file or directory to read-only or append-only
- Added
--trace-fileto enable and collect tracing - Allow read-only access to on-demand imported file systems through the S3 API
- Support importing arbitrary buckets from MinIO
- Build cache for data imported from other buckets
- Support accessing cache data through RDMA
- Support limiting QPS when accessing object storage
- Python SDK: Added APIs for
status,warmup,merge, andinfo - Python SDK: Added the
follow-linkparameter to clone API
- mount: Added the
- Improved
- mount: Add
nofailin fstab by default to ignore mount failures - status: Filter out tool sessions by default
- restore: Added progress bar
- warmup: Count data distribution by priority when using
--check - warmup: Added
--follow-linkoption - Improved response time for interactive operations such as info and warmup
- import: Fail immediately on List errors
- objbench: Fixed inaccurate logs
- gateway: Verify user-provided ETag when completing multipart uploads
- gateway: Compress client access logs
- sync: Allow configuring more client parameters (for example,
--mountpoint) - sync: Improved error handling when creating directories with the
jfs://protocol - sync: Improved performance of synchronizing large files with the
jfs://protocol - Improved cleanup speed of metadata cache
- Improved DirectIO with support for more types of read requests
- Automatically clean data cache when merging fragments
- Limit warmup bandwidth through background task throttling
- Collect read amplification metrics by default
- Verify file system name in config file during mount
- Use password saved in configuration file
- Allow customizing client hostname via the
HOSTNAMEenvironment variable - Reduced lock contention related to cache
- Support cache node IPs in CIDR format
- Respect
--io-retriesfor client reconnection attempts - Enable FUSE
EnableReadDirPlusAutoby default for kernel optimization - Ensure source cluster is connected before allowing access to writable mirror cluster to avoid errors
- Access source cluster when mirror is unavailable to avoid service interruptions
- Use mirrored file system name as mount point name
- Fixed an issue that metrics from multiple file systems could be mixed together
- Adjusted data replication monitoring to reduce noise from expected failed requests
- Automatically re-download client binary when a Ceph-capable client is corrupted
- Windows: Persist data on file close by default
- Windows: Improved background mount
- Windows: Added
--create-permmount option - Windows: Support customizing file system name
- Windows: Set the UID of administrator user to 0 (same as Linux root)
- Windows: Support DirectIO
- Windows: Optimized mount speed
- Optimized lock contention when evicting disk cache
- No longer retry timeout distributed cache requests
- mount: Add
- Bugfix
- Fixed an issue that statvfs might hang when
--block-interruptis configured - Fixed an issue that mount options in fstab could be incorrect
- sync: Fixed incorrect umask in cluster mode
- auth: Fixed potential failure when running auth concurrently
- Fixed inaccurate ping metrics in writable mirror mode
- Fixed an issue that scalar metrics with value 0 were not reported
- Fixed an issue that kernel cache could cause append writes to overwrite each other
- Fixed compatibility issues when source clusters of mirrored file systems use different versions
- Windows: Fixed GID mapping issue
- Fixed an issue that statvfs might hang when
JuiceFS 5.2.23 (2026/1/29)
- Improved
- Improved performance and stability of
statfswhen mounting subdirectories
- Improved performance and stability of
- Bugfix
- Fixed an issue that unexpected data could be uploaded to object storage when shared block device is enabled
JuiceFS 5.2.22 (2026/1/4)
- Improved
- Added
idle-writer-waitoption to reduce fragment generation during slow writes
- Added
- Bugfix
- Fixed an issue that checking Ceph support could exit unexpectedly
JuiceFS 5.2.21 (2025/12/22)
- Improved
- Python SDK:
clonesupports thefollow-linkparameter - Close idle distributed cache connections in time to release file descriptors
- Python SDK:
- Bugfix
- Fixed lock contention issue on cache nodes under high load
- sync: Fixed incorrect error count in logs
JuiceFS 5.2.20 (2025/12/8)
- Improved
- Interrupt in-flight requests when removing cache nodes to retry as soon as possible
- Increased significant digits in progress bar to 3
- Bugfix
- sync: Fixed compatibility issue with object storage that does not support URL encoding
- Fixed an issue that
readdircould cause data inconsistency - Fixed support issue of
WHITEOUTforrenameon macOS
JuiceFS 5.2.19 (2025/11/7)
- Improved
- Accelerated cache data evacuation when cache nodes go offline.
- Added support for adjusting the client's file descriptor limit via the
JFS_RLIMIT_NOFILEenvironment variable. - object/oss: added support for switching to v1 version signatures.
- object/oss: the
LISTinterface now includes the token parameter only when it is not empty.
JuiceFS 5.2.18 (2025/10/30)
- New
- Added the
case-insensitivemount option to support case-insensitive filenames. - Python SDK: added an API for listing files in the trash.
- Added the
JuiceFS 5.2.17 (2025/10/20)
- Improved
- The
synccommand now supports filtering by different types of timestamps. - The
synccommand will no longer be logged as a new session. - Python SDK: added quota management APIs.
- Python SDK: clone operations now support copying the attributes of source files.
- The
- Bugfix
- Fixed an issue that reference counts in the directory cache might not be properly reset during client reconnection.
- Fixed an issue that FUSE requests might be frequently interrupted.
- Fixed an issue that opening files on macOS might hang for a long time.
- Windows: fixed an issue that the
cache-diroption was not correctly applied when running in the background.
JuiceFS 5.2.16 (2025/10/11)
- Bugfix
- Fixed an issue that abnormal cache groups in certain scenarios could cause client data read failures.
JuiceFS 5.2.15 (2025/9/24)
- New
- Added support for two-level distributed caching.
- Improved
- Optimized the allocation of read-write cache in client memory to prevent performance degradation during severe contention.
- Added the ability to limit bandwidth when reading data from cache groups.
- Improved performance when deleting large directories using the
rmrcommand.
- Bugfix
- Fixed an issue that outdated attributes from the kernel cache might be read when accessing multiple mount points.
- Fixed an issue that failure to retrieve cache disk usage could cause the client process to crash.
- sync: fixed a potential deadlock issue when synchronizing large files.
JuiceFS 5.2.14 (2025/9/17)
- Improved
- Limit the memory usage of cache nodes to comply with
buffer-size.
- Limit the memory usage of cache nodes to comply with
- Bugfix
- Fixed the conflict between concurrent downloads.
- Fixed the issue that the client could crash when the available space for disk cache was too small.
- auth: fixed the issue that
authdid not prompt foraccesskeyandsecretkey.
JuiceFS 5.2.13 (2025/9/9)
- Improved
- When received multiple
GETrequests for the same object, the cache node now uses only one background concurrency. - Cache nodes no longer reuse data sent via zero-copy.
- Return an error directly when the checksum of cached data is insufficient, preventing client crashes.
- When received multiple
- Bugfix
- Fixed an issue that the
--block-interruptparameter was not correctly parsed during mount.
- Fixed an issue that the
JuiceFS 5.2.12 (2025/9/3)
- Improved
- Remove unnecessary checks on object storage encryption configuration for
fsckandgccommands. - Prefer using checksums stored on disk when distributed cache nodes send data.
- Optimize performance during concurrent access to the same data on cache nodes.
- Reduce lock contention when clients scan cache disks to build indexes.
- Improve handling of unknown error types for Alibaba Cloud OSS.
- gateway: Reduce memory usage when listing objects.
- Java SDK: Prefer Kerberos for authentication.
- sync: Automatically ignore non-existent files when using the
--files-fromparameter. - sync: Reduce the number of large files synchronized concurrently.
- sync: Allow using the
--cache-groupparameter when the source and destination are the same JFS volume.
- Remove unnecessary checks on object storage encryption configuration for
JuiceFS 5.2.11 (2025/8/21)
- Improved
- warmup: periodically print progress information when debug logging is enabled.
- Java SDK: added automatic renewal of Kerberos TGT.
JuiceFS 5.2.10 (2025/8/13)
- Improved
- Optimized the read-ahead strategy to better balance memory usage during concurrent read operations.
- Bugfix
- Fixed compatibility issues when older clients access the distributed cache.
- Windows: fixed an issue that integer parameters might not be applied correctly when running in the background.
JuiceFS 5.2.9 (2025/8/7)
- Improved
- Optimized memory usage balancing in read-write hybrid scenarios to mitigate read amplification.