Skip to main content

Command Reference

Running juicefs by itself and it will print all available commands. In addition, you can add -h/--help flag after each command to get more information, e.g., juicefs mount -h.

juicefs COMMAND [options]

COMMAND could be:
auth authorize a filesystem
mount mount a filesystem
umount umount a filesystem
info show internal information of a path or inode
status show status of a volume
stats show real time performance statistics of JuiceFS
summary Show data and files distribution of a directory
sync sync between two storages
lsof list recent opened files
load load metadata from a previously dumped JSON file
webdav Start a WebDAV server
gateway start an S3-compatible gateway
import import files from existing object storage
rmr remove all files/directories recursively
snapshot create or remove snapshots
grep parallelized file pattern searcher
profile analyze log (Experimental)
bench run benchmark (via go binary)
benchmark run benchmark
objbench run benchmarks on an object storage
doctor collect diagnose information
fsck check consistency of a volume
gc garbage collector of objects in data storage
warmup build cache for target directories/files
version show the version


juicefs auth

Authenticate the client against JuiceFS Console, and fetch or update configuration file for the JuiceFS Volume, and stored in $HOME/.juicefs/$VOL_NAME.conf. This command is useful when you need to deploy JuiceFS via script, or use self-hosted object storage for onpremise deployment.


juicefs auth NAME [options]

juicefs auth myjfs --token xxx --access-key xxx --secret-key xxx


NAMEName of your file system.
--token=TOKENToken of your file system, obtain in JuiceFS Web Console.
--bucket=BUCKETWhen a file system is first created, object storage bucket is already specified in the web console. When in need, client can override this setting by specifying this option, in order to for example, connect to object storage in special network environments.

Same as the file system bucket setting in the web console, you can either specify the bare bucket name string, or the full endpoint starting with https:// or http://. These two method have different implications:

  • If a bare bucket name is specified, client will automatically conduct a network test to choose between internal/external endpoints (if service provider supports), for example, when using OSS beijing region, passing --bucket=test causes client to automatically choose between and, according to the network test result.
  • Explicitly specifying --bucket=, then client will respect this setting, without trying to deduce the full endpoint by itself.
--access-key=ACCESSKEYObject storage access key, see How to get AK/SK. An interactive prompt will occur to collect this info if missing in the command line options. Leave empty if you have configured host to access object storage without credentials.
--secret-key=SECRETKEYObject storage secret key, see How to get AK/SK. An interactive prompt will occur to collect this info if missing in the command line options. Leave empty if host have been configured to access object storage without credentials.
--session-token=SESSIONTOKENOptionally provide object storage session token.
--storage-class=valueStorage class for the destination storage, if supported. Read relevant docs for more, e.g. S3.
--bucket2=BUCKET2Optionally provide the name or endpoint for the secondary bucket For Data replication, same format and caveats as --bucket.
--access-key2=ACCESSKEY2When data replication is enabled, use this parameter to specify the access key for the secondary bucket. Learn more at Cross-region data replication.
--secret-key2=SECRETKEY2When data replication is enabled, use this parameter to specify the secret key for the secondary bucket. Learn more at Cross-region data replication.
--session-token2=SESSION_TOKEN2the session token to access the bucket (optional)
--passphrase=PASSPHRASEPassphrase for encrypted key, see Data Encryption for more.
--subdir=SUBDIRSpecify a sub-directory for mount, you can also create ACL to restrict client access to specific sub-directories, learn more at Client Access Control
--conf-dir=CONFDIRDirectory to store configuration files, default to ~/.juicefs)
--no-updateClient will by default obtain latest config from JuiceFS Web Console, set to true to disable this behavior, and then client will no longer talk to JuiceFS Console when mounting.

juicefs gc

Deal with leaked objects, and garbage fragments produced by file overwrites.


juicefs gc [command options] VOLUME-NAME

# Check and print results, but don't actually delete anything
juicefs gc $VOL_NAME

# Delete leaked objects
juicefs gc $VOL_NAME --delete


--deletedelete leaked objects (default: false)
--threads=10, -p 10Number of concurrent threads, default to 10.

juicefs fsck

Check consistency of file system.


juicefs fsck [command options] VOLUME-NAME

juicefs fsck $VOL_NAME


--threads=10, -p 10Number of concurrent threads, default to 10.
--repairFix the prefix of objects.

juicefs dump

Dump metadata into a JSON file. Different from JuiceFS CE, JuiceFS EE require no manual metadata dump in order to make a backup, it's already handled by our engineers in the server end. Thus, this command is used mainly for debugging, or migration between CE and EE.


juicefs dump [command options] PATH JSON-PATH

# Export metadata to meta-dump.json
juicefs dump /jfs meta-dump.json

# Export metadata for only one subdirectory of the file system
juicefs dump /jfs/subdir meta-dump.json


PATHMount point path.
JSON-PATHExport file path. If the filename ends with .gz, it will be automatically compressed.
--gzipCompress with gzip.

juicefs load

Load metadata into an empty file system, often used for migrations between JuiceFS CE and EE.


juicefs load [command options] PATH [FILE]

# Import the metadata backup file into the file system
juicefs load /mnt/jfs meta-dump.json.gz

juicefs version

Show the JuiceFS version information. The update details of each version can be viewed on the "Release Notes".


juicefs version [options]

# Upgrade client version and seamless restart
juicefs version -u -r


-u --upgradeUpgrade the JuiceFS client to latest version.
-r --restartExecute seamless remount, if client is upgraded.

juicefs quota

Manage directory quota. Different from Community Edition, JuiceFS Enterprise Edition manages quota directly from the web console, and users should not directly run this command from the terminal. The purpose of this command is to stay consistent with CE interface, so that CSI Driver may approach quota management using the same code.

juicefs heal

Repair data when using data replication. Execute a bidirectional synchronization on object storage for both file system.


juicefs heal VOL-NAME [options]


--start=valueSpecify object storage key range.
-v --verboseDisplay more log.
--upload-limit=0Maximum bandwidth for upload in Mbps, default to 0 (unlimited).
--conf-dir=/root/.juicefsa directory for configuration (default: /root/.juicefs)


juicefs status

Show status of JuiceFS.


juicefs status [command options] VOLUME-NAME

juicefs status $VOL_NAME


--session=0 -s 0Show detailed information of one session, specified using a session ID (get from the SID field from the status output).

juicefs stats

The stats command reads the metrics of the JuiceFS client and prints the changes of each metric per second.

For more introduction and usage of this command, see "Real-time Performance Monitoring".


juicefs stats [options] MOUNTPOINT


--schema=ufmcoSchema string that controls the output, default to “ufmco”, its meaning:
u: usage
m: meta
c: blockcache
o: object
g: go
--interval=1Update interval in seconds, default to 1.
--verbosity=0Verbosity level, increase to see more information.

juicefs profile

Performs basic profiling based on access log, and writes comprehensive statistics to command line output. Use this command to quickly identify the overall file system access pattern of your application, and assess performance.

For more introduction and usage of this command, see "Real-time Performance Monitoring".


juicefs profile [options]


-x PATH --path=PATHThe directory path of access log, default to /jfs.
-f FILE --file=FILEThe filename of the access log, default to .ophistory. Refer to Access Log for different types of access log.
-g GROUP_BY --group-by=GROUP_BYGroup the output by a specific attribute, default to cmd, choose from uid, pid, cmd.
-s SORT_BY --sort-by=SORT_BYSort the output by a specific column, default to total_time, choose from group, number, avg_time, total_time.
-u FILTER_BY_UID --filter-by-uid=FILTER_BY_UIDFilter the output by one or multiple keywords on the UID column, please separate multi keywords by comma (,).
-p FILTER_BY_PID --filter-by-pid=FILTER_BY_PIDFilter the output by one or multiple keywords on the PID column, please separate multi keywords by comma (,).
-w WINDOW_SIZE --window-size=WINDOW_SIZEAnalysis time window, accept float value, default to 60 seconds.
-i FLUSH_INTERVAL --flush-interval=FLUSH_INTERVALFlush output interval, accept float value, default to 2 seconds.

juicefs info

Show internal information of a path or inode.


juicefs info [options] [PATH|INODE]

# Lookup single file
juicefs info /jfs/foo.txt

# Lookup directory
juicefs info /jfs/bar/

# Reverse lookup on inode
juicefs info 754051


PATH, INODETarget file path or inode.
-n, --plainShow numbers in plain format.

juicefs summary

Show tree summary of target directory.


juicefs summary [command options] PATH

# Show with path
juicefs summary /jfs/foo

# Show max depth of 5
juicefs summary --depth 5 /jfs/foo

# Show top 20 entries
juicefs summary --entries 20 /jfs/foo


--depth=2 -d 2Depth of tree to show, 0 means only show root, defaults to 2.
--entries=10 -e 10Show top N entries, sort by size, defaults to 10.
--csvPrint summary in csv format.

juicefs doctor

Collect abundant debug information, including access log, client log, CPU stats, Goroutine stacktraces. When run into unexpected situations, JuiceFS team may asks you to collect information with this command, and send back the result to continue support.

We encourage you to also take a look at Troubleshooting Methods and Troubleshooting Cases, and try to solve some of the more frequently encountered problems by yourself.


juicefs doctor MOUNTPOINT

juicefs lsof

List recently (within 10 minutes) opened files in JuiceFS.


juicefs lsof PATH


juicefs mount

Mount the file system. You must provide the token of the file system (check the "File System Settings" in the website console), and the key pair to access the object storage (refer to How to Get Object Storage Access Key and Secret Key for details).

If the configuration file is not generated by the auth subcommand before mounting the file system, the client will ask you to provide the file system token and the key pair to access the object storage API separately through command line interaction.

You can mount JuiceFS as root or normal user, but due to permission issues, cache and client log locations will be different, read option descriptions below for more.


juicefs mount NAME MOUNTPOINT [options]


NAMEThe name of your file system.
MOUNTPOINTThe mount point for your file system.
Run in the foreground mode, client logs will print directly to console, often used in debugging.
Run in the background mode, which is the default behavior.
--no-syslogDisable syslog.
--log=PATHLog file location, defaults to /var/log/juicefs.log (when running as root) or ~/.juicefs/log/juicefs.log (when running as normal user).
--update-fstabadd or update entry in /etc/fstab, when applicable, see Auto-mount on boot.
--token=TOKENSpecify client access token.
--no-updateClient will by default obtain latest config from JuiceFS Web Console, set to true to disable this behavior, and then client will no longer talk to JuiceFS Console when mounting.
Display verbose log.
--subdir=SUBDIRSpecify a sub-directory for mount, you can also create ACL to restrict client access to specific sub-directories, learn more at Client Access Control

FUSE Options

--enable-xattrEnable extended attributes (xattr) support, default to false, which is recommended for most scenarios because xattr comes with its own overhead.
--enable-aclEnable POSIX ACL support, see Extended ACL.
--no-bsd-lockDisable BSD lock support.
--no-posix-lockDisable POSIX lock support.
--block-interrupt=1(default: 1) number of seconds to block interruption
--allow-otherAllow accesses from other users. This option is enabled by default when you are running as root user. You need to enable this feature explicitly otherwise.
-o FUSE_OPTSOther FUSE options. For example, enabling FUSE write cache through -o writeback_cache can greatly improve the performance of random writes and fragmented writes (requires Linux Kernel 3.15 and above).

Metadata Cache Options

--metacacheto=METACACHETO(default: 300) Expiration of the meta cache (seconds).
--metacacheEnable metadata caching in local memory.
--max-cached-inodes=MAXCACHEDINODES(default: 5000000) the max number of cached inodes
Open files using cached metadata, default to false, which means every open operation access metadata service to check for file changes. If file is frequently opened and scarcely changed, consider enabling --opencache to further improve read performance. For the same mount point, open cache is automatically invalidated upon file modifications.
File attribute cache timeout in seconds, default to 1.
File entry cache timeout in seconds, default to 1.
Directory entry cache timeout in seconds (default: 1).
--readdir-cachecache readdir in kernel (require Linux kernel 4.20+)

Object Storage Options

--get-timeout=60Max number of seconds to download an object, default to 60, may need to increase this in low bandwidth conditions, see Read / Write error.
--put-timeout=60Max number of seconds to upload an object, default to 60, may need to increase this in low bandwidth conditions, see Read / Write error.
(default: 30) number of retries for network failure. Wait between retries increase linearly, but when set to over 30, wait becomes a fixed 10 seconds delay. The actual total timeout depends on specific timeout settings for each type of requests.
--max-uploads=20Upload concurrency, defaults to 20. This is already a reasonably high value for 4M writes, with such write pattern, increasing upload concurrency usually demands higher --buffer-size, learn more at Read/Write Buffer. But for random writes around 100K, 20 might not be enough and can cause congestion at high load, consider using a larger upload concurrency, or try to consolidate small writes in the application end.
--max-downloads=200Max number of concurrent downloads, default to 200, which is usually large enough for clients and does not require further adjustments. This option controls client global download concurrency, but for single file read, a hard limit of 128 blocks is imposed to avoid uncontrolled resource usage in large file sequential scenarios.
Maximum QPS for deletion, default to -1 (unlimited), set to 0 to disable deletion. Use this if for example, some object storage limit API calls.
--flush-wait=5The time delay between automatic flush to object storage, default to 5 seconds. When slowly writing a big file, frequent flushes will cause fragmentation, consider increasing this delay to avoid write amplification.
--upload-limit=0Maximum bandwidth for upload in Mbps, default to 0 (unlimited). Apart from bandwidth control, this parameter can also be used for system load control with high throughput scenarios.
--download-limit=0Maximum bandwidth for download in Mbps, default to 0 (unlimited). Apart from bandwidth control, this parameter can also be used for system load control with high throughput scenarios.
--externalPrefer to use the external entrypoint of the object storage service. This means the traffic of JuiceFS will go through the public network instead of the internal network.
--internalPrefer to use the internal entrypoint of the object storage service. This means the traffic of JuiceFS will go through the internal network instead of the public network. (Applicable to some object storages that separate internal and external network domain names, e.g. Aliyun OSS, UCloud UFile, etc.)
--rsa-key=PATHPath to RSA private key (PEM). Passphrase will be asked or provided with JFS_RSA_PASSPHRASE environment variable.
--flipWhen data replication is enabled, use secondary bucket as primary bucket. Learn more at Cross-region data replication.

Cache Options

Read Cache to understand JuiceFS's cache design.

--buffer-size=300Total size of memory used for read/write buffering in MiB default to 300, learn more at Read/Write Buffer
--prefetch=1When a block is randomly read by a small offset range, the whole block is scheduled for download asynchronously, this is called prefetch in JuiceFS. This options controls prefetch concurrency, which defaults to 1, and can be disabled by setting to 0. Note that:
  • Prefetch can cause read amplification in some scenarios
  • When used with dedicated cache cluster, clients do not read from object storage directly (but through cache group members instead), so this option is meaningless for --no-sharing clients.
--writebackEnable client write cache, data will be written to local disk first, and then uploaded asynchronously in the background, default to false. Client write cache brings data security risks, use with caution.
--cache-dir=CACHEDIRCache directory, defaults to /var/jfsCache or $HOME/.juicefs/cache depending on the current user.

Separate multiple directories by :, add quotation mark if using wildcard match (*). e.g. --cache-dir '/data*/jfsCache:/mydata*/jfsCache'.

Moreover, you can specify /dev/shm to use the memory device (defaults to half of node memory), or use the string memory to use client process memory.
--cache-size=102400Local cache size limit in MiB, default to 102400 (100GiB). When multiple directories are used as cache directory, --cache-size limits the total size sum (split evenly). Set to 0 to disable disk cache altogether, but note that this causes an extra 100MB memory overhead, to grant client a minimum cache space. Read troubleshooting resource usage for more.
--free-space-ratio=0.2Minimum free space ratio on cache disk, default to 0.2. This option doesn't support being set to 0, to achieve maximum usage, you can set a value close to 0 (such as 0.01). Also if Client write cache is enabled, this option also controls write cache size
--cache-mode=0600File permission for cached blocks, default to 0600.
--cache-partial-onlyOnly cache files and reads smaller than block size, default to false. Enable this when you need to frequently access a same set of small files, or when disk throughput is smaller than object storage throughput.
--verify-cache-checksum=fullChecksum level for cache data. After enabled, checksum will be calculated on divided parts of the cache blocks and stored on disks, which are used for verification during reads. The following strategies are supported:
  • none: Disable checksum verification, if local cache data is tampered, bad data will be read;
  • full (default): Perform verification when reading the full block, use this for sequential read scenarios;
  • shrink: Perform verification on parts that's fully included within the read range, use this for random read scenarios;
  • extend: Perform verification on parts that fully include the read range, this causes read amplifications and is only used for random read scenarios demanding absolute data integrity.
--cache-eviction=2-random Added in v5.0.1Cache eviction algorithm, supports the following strategies:
  • 2-random (default): Evict cache using 2-random algorithm, i.e. randomly choose two blocks, and evict the one with a lower atime (last access time). For most real-world scenarios, this algorithm approximates LRU.
  • none: Disable cache eviction altogether, new cache will not write once cache capacity is drained. You'll have to manually free up cache space.

Distributed cache options

Read distributed cache for more.

--cache-group=CACHEGROUPCache group name for distributed cache. Nodes within the same group share cache, disabled by default.
--subgroups=0the number of sub-groups in a cache group (default: 0)
--group-ip=GROUPIPManually set IP address listened by distributed cache, in hosts with multiple network interfaces, use this option to specify the correct NIC. Can also use CIDR prefix to automatically select IP, e.g. use to match with
--group-portListen port for distributed cache, defaults to a large random port. If the specified port doesn't work, will automatically increase by 1 until a successful bind.
--group-networkNetwork interface for distributed cache. Default listening IP is the first private IP found in the network device list, if this isn't appropriate for you, use this option to specify a fixed network interface.
--group-weight=100Adjust group member weight.
--no-sharingWhen inside a cache group, only fetch cache data from others, but never share its own cache. Use this option on ephemeral mount points (like Kubernetes Pod). Default to false.
--fill-group-cacheNewly written blocks are sent to distributed cache, rather than local disk cache. This option is not affected by --no-sharing, i.e. clients using a dedicated cache cluster will still contribute their writes to the cache group. Since 4.9.22, if client write cache is enabled, the staging data will be sent asynchronously to the cache group as well, so that cache group members can also read data that's not yet persisted.
--cache-group-size=0Unit is MiB. JuiceFS Client performs readahead and prefetch, so for files smaller than this size, client will try to schedule all its data blocks into a single node, to maximize cache utilization.

Experimental Options


These options are currently experimental and subject to change in the future. Incorrect usage may pose data integrity risks, if you do need to run this command, seek help from a Juicedata engineer.

--min-inodes-diff=1048576When the inodes difference between the current zone and the zone with the least inodes exceeds this value, the metadata will be automatically balanced. Defaults to 1048576.
--min-dir-inodes=1048576When the number of directories in the current path exceeds this value, the metadata is automatically balanced. Defaults to 1048576.
--max-space=0Limit the displayed total space in GiB, defaults to 0, meaning that the true total capacity is displayed.

juicefs umount

Umount a JuiceFS file system. This command handles the platform differences, i.e. umount in Linux and diskutil unmount in macOS. Note that there's a similar hidden command called juicefs unmount, this is specially reserved for Kubernetes FlexVolume and not for normal use.

If you encounter any errors while unmounting JuiceFS, read the docs for troubleshooting steps.


juicefs umount PATH [options]


PATHThe mount point for your file system.
-f --forceUmount it forcibly.

juicefs gateway

Start an S3 gateway, read S3 gateway for more.


juicefs gateway [command options] VOLUME-NAME ADDRESS

export MINIO_ROOT_USER=admin
export MINIO_ROOT_PASSWORD=12345678
juicefs gateway testVolume localhost:9000


Apart from options listed below, this command shares options with juicefs mount, be sure to refer to mount as well.

--console-url=URLWeb URL address, only needed in on-premise environments.
--mountpoint=s3gatewaySpecify the JuiceFS mount point on host, needed only for absolute symlink resolution, defaults to s3gateway. See S3 Gateway and symbolic link.
--access-log=PATHPath for file system access log, default location is the .accesslog file under the root mount point.
--no-bannerOmit MinIO startup messages.
--multi-bucketsUse top level directories as buckets, default to false.
--keep-etagKeep object ETag when uploading.
--umask="022"Umask used for new files, default to 022.

juicefs webdav

Start a WebDAV server to access JuiceFS file system.

Use the environment variable WEBDAV_USERNAME and WEBDAV_PASSWORD to set the username and password for WebDAV server.


juicefs webdav [command options] NAME ADDRESS

# Set username and password
export WEBDAV_USER=root

# Start WebDAV server on port 9007
juicefs webdav myjfs localhost:9007


Apart from options listed below, this command shares options with juicefs mount, be sure to refer to mount as well.

--cert-filecertificate file for TLS
--key-filekey file for TLS


juicefs benchmark

Run single benchmark, including read/write/stat big/small files.


juicefs benchmark [options] path-to-test


--dest=/jfs/benchmarkTemporary directory for benchmark test files, default to /jfs/benchmark.
--block-size=BLOCK_SIZE(default: 1MiB) block size in MiB
--bigfile-file-size=BIGFILE_FILE_SIZE(default: 1024MiB) size of big file in MiB
--smallfile-file-size=SMALLFILE_FILE_SIZE(default: 0.1MiB) size of small file in MiB
--smallfile-count=100(default: 100) number of small files

juicefs objbench

Run basic benchmarks on the target object storage to test if it works as expected.


juicefs objbench [command options] BUCKET


--storage=fileObject storage type (e.g. s3, gcs, oss, cos) (default: file)
--access-key=valueAccess Key for object storage (can also be set via the environment variable ACCESS_KEY)
--secret-key valueSecret Key for object storage (can also be set via the environment variable SECRET_KEY)
--session-token=valueSession token for object storage
--block-size=4096size of each I/O block in KiB (default: 4096)
--big-object-size=1024size of each big object in MiB (default: 1024)
--small-object-size=128size of each small object in KiB (default: 128)
--small-objects=100number of small objects (default: 100)
--skip-functional-testsskip functional tests (default: false)
--threads=4, -p 4number of concurrent threads (default: 4)

juicefs warmup

Download data to local cache in advance, to achieve better performance on application's first read. You can specify a mount point path to recursively warm-up all files under this path. You can also specify a file through the --file option to only warm-up the files contained in it.

If the files needing warming up resides in many different directories, you should specify their names in a text file, and pass to the warmup command using the --file option, allowing juicefs warmup to download concurrently, which is significantly faster than calling juicefs warmup multiple times, each with a single file.

The warmup command will not repeatedly download files that already exist locally (it will be skipped directly), so the displayed download speed may be higher than the actual download traffic. If the distributed cache is enabled, executing this command on any node in the cache group can achieve the same effect.


juicefs warmup [PATH] [options]

# Warm up all files in datadir
juicefs warmup /mnt/jfs/datadir

# Warm up selected files
echo '/jfs/f1
/jfs/f3' > /tmp/filelist.txt
juicefs warmup -f /tmp/filelist.txt


file containing a list of paths (one path per line)
-p 50
-c 50
Download concurrency, default to 50. Increase concurrency to speed up, but if the download bandwidth of the object storage is low, reduce concurrency to avoid download timeout errors.
--io-retries=1Max number of allowed retries, default to 1. Increase this value in poor network conditions to better handle failures.
--max-failure=0Controls the command exit code according to the number of download failures, the number of allowed failures is default to 0, meaning that upon any data block warmup failure, command exits with error (exit code 1). When faced with warmup errors, check client logs for troubleshooting.
Run in the background.
--evictEvict cached blocks for given path.
--checkcheck whether the data blocks for the given path are cached or not.

juicefs snapshot

Also called juicefs clone. This command takes a snapshot of your data by creating a mere metadata copy, without creating any new data in the object storage, thus snapshotting is very fast regardless of target file / directory size. Under JuiceFS, this command is a better alternative to cp, moreover, for Linux clients using kernels with copy_file_range support, then the cp command achieves the same result as juicefs snapshot.


Snapshot result is a metadata copy, all the files are still referencing the same object storage blocks, that's why a snapshot behaves the same in every way as its originals. When either of them go through actual file data modification, the affected data blocks will be copied on write, and become new blocks after write, while the unchanged part of the files remains the same, still referencing the original blocks.

Although object storage data is not actually copied, snapshots takes up storage space, and will result in an increase in bill, pay special attention when taking snapshots on large size directories.


juicefs snapshot SRC DST [-f|--force] [-c|--copy]  # create a snapshot
juicefs snapshot -d DST [-f|--force] # delete a snapshot


SRCThe path to the file you are creating the snapshot from.
DSTThe locate where you want to store your snapshot.
-d --deleteRemove the snapshot. Only use this command on snapshots, deleting normal files using this command will result in permission errors. Deleted snapshots will bypass Trash completely. If you want to preserve the deleted snapshots in Trash like normal files, use juicefs rmr or simply rm -r.
-f --forceOverwrite existing snapshot, or remove the snapshot arbitrarily.
-c --copyUse current user's UID, GID and umask to create the snapshot. Otherwise it will use permission information of the file you are creating snapshot.

juicefs rmr

Remove all the files and subdirectories, similar to rm -rf, except this command deals with metadata directly (bypassing POSIX API), thus is much faster.

It will try to remove them using current UID/GID, ignore those that can't be removed, return the number of removed files/directories, the number of left ones.

If trash is enabled, deleted files are moved into trash. Files inside trash is still billable, read more at trash.


juicefs rmr DIR

juicefs sync

Sync data between two storage systems, read Data Migration for more.


juicefs sync [command options] SRC DST

# Sync object from OSS to S3
juicefs sync oss:// s3://

# Sync objects from S3 to JuiceFS
juicefs sync s3:// jfs://VOL_NAME/

# SRC: a1/b1,a2/b2,aaa/b1 DST: empty sync result: aaa/b1
juicefs sync --exclude='a?/b*' s3:// jfs://VOL_NAME/

# SRC: a1/b1,a2/b2,aaa/b1 DST: empty sync result: a1/b1,aaa/b1
juicefs sync --include='a1/b1' --exclude='a[1-9]/b*' s3:// jfs://VOL_NAME/

# SRC: a1/b1,a2/b2,aaa/b1,b1,b2 DST: empty sync result: a1/b1,b2
juicefs sync --include='a1/b1' --exclude='a*' --include='b2' --exclude='b?' s3:// jfs://VOL_NAME/

Both SRC and DST takes up the format [NAME://][ACCESS_KEY:SECRET_KEY[:SESSIONTOKEN]@]BUCKET[.ENDPOINT][/PREFIX], in which:

  • NAME: Can be all JuiceFS supported storage, e.g. s3, oss, even jfs. Using jfs:// allows you to directly transfer data from or into JuiceFS, bypassing the FUSE mount point, this saves resource under large scale data transfer.
  • ACCESS_KEY and SECRET_KEY: Object storage credentials, if special characters are included, use single quote to avoid shell interpretations.
  • SESSIONTOKEN: Optionally provide session token for the object storage.
  • BUCKET[.ENDPOINT]: Object storage endpoint, note that currently MinIO only supports path style (minio://[ACCESS_KEY:SECRET_KEY[:TOKEN]@]ENDPOINT/BUCKET[/PREFIX]).
  • [/PREFIX]: Optionally provide the prefix of the files to synchronize.

Apart from options listed below, this command shares options with juicefs mount, be sure to refer to mount as well.

Do notice that many of these shared options works exclusively with jfs scheme, for example, the sync command supports --writeback as well, but if the sync destination is a normal JuiceFS FUSE mount point, it won't automatically convert itself to a --writeback mount point simply because you are running a sync job with --writeback. To sum up, --writeback only works when sync destination is a jfs scheme address, and this is also true for many other options that's shared with mount.

--start=KEY, -s KEY, --end=KEY, -e KEYProvide object storage key range for syncing.
--exclude=PATTERNExclude keys matching PATTERN.
--include=PATTERNInclude keys matching PATTERN, need to be used with --exclude.
--limit=-1Limit the number of objects that will be processed, default to -1 which means unlimited.
--update, -uUpdate existing files if the source files' mtime is newer, default to false.
--force-update, -fAlways update existing file, default to false.
--existing, --ignore-non-existingSkip creating new files on destination, default to false.
--ignore-existingSkip updating files that already exist on destination, default to false.
--dirsSync empty directories as well.
--permsPreserve permissions, default to false.
--inplaceModify files in place, rather than delete and replace. If destination storage system is a JuiceFS volume with Trash enabled, overwritten files by default will be kept in Trash. Use --inplace to modify files in place and prevent Trash usage. Currently only jfs:// type destination storage is supported.
--links, -lCopy symlinks as symlinks default to false.
--delete-src, --deleteSrcDelete objects that already exist in destination. Different from rsync, files won't be deleted at the first run, instead they will be deleted at the next run, after files are successfully copied to the destination.
--delete-dst, --deleteDstDelete extraneous objects from destination.
--check-allVerify the integrity of all files in source and destination, default to false. Comparison is done on byte streams, which comes at a performance cost.
--check-newVerify the integrity of newly copied files, default to false. Comparison is done on byte streams, which comes at a performance cost.
--dryDon't actually copy any file.
--http-port=6070HTTP port for the pprof server to listen on, default to 6070.
--threads=10, -p 10Number of concurrent threads, default to 10. Higher concurrency requires more resource, particularly in large file scenarios, because more big files will be loaded into the process memory.
--list-threads=1Number of list threads, default to 1. Read concurrent list to learn its usage.
--list-depth=1Depth of concurrent list operation, default to 1. Read concurrent list to learn its usage.
--no-httpsDo not use HTTPS, default to false.
--storage-class=valueStorage class for the destination storage, if supported. Read relevant docs for more, e.g. S3.
--bwlimit=0Limit bandwidth in Mbps default to 0 which means unlimited.
--manager=ADDRManager node address used in distributed syncing, this is an internal option that's used in the executed command on the worker nodes.
--worker=ADDR,ADDRWorker node addresses used in distributed syncing, comma separated.

Metrics related options

--metricsAddress to export metrics, defaults to
--consulConsul address to register, defaults to

juicefs import

juicefs import scans the target object storage bucket, collects metadata of the existing files, and saves to the JuiceFS metadata engine. This gives JuiceFS access to the existing object storage files, but the file data will remain in the object storage as is. To learn more, see Import existing object storage files.


juicefs import URI DST [-v] [--name=<name>] [--mode=<mode>]

# Assuming the mount point being /jfs, and importing from the same bucket as the file system, in such case, the bucket name can be omitted
# But remember to use a prefix, to avoid importing data created by JuiceFS itself (the chunks directory), into JuiceFS, as it'll be meaningless inside JuiceFS
juicefs import /prefix /jfs/imported

# Assuming the bucket is in the same region as the file system
# Then the endpoint can be omitted to just the bucket name
juicefs import another-bucket/prefix /jfs/imported

# If the bucket is located in another region, the full endpoint must be specified
juicefs import /jfs

# Import files in mybucket by a prefix, into the specified directory within JuiceFS
juicefs import other-bucket/prefix /jfs/oss-imported


URIThe object storage URI of the target, format is the BUCKET[.ENDPOINT][/PREFIX]. Must use the same type of object storage as the file system itself, and should be accessible with the same set of credentials (access key, secret key), you cannot specify separate keys within this URI.
DSTImport destination, must be a JuiceFS mount point, and the file system must be using the same object storage as the URI.
--name=NAMEThe name of the file system. This is required in systems other than Linux.
--mode=MODEMode (Unix permissions) of imported files.
-vRun verbosely.
--conf-dir=/root/.juicefsa directory for configuration (default: /root/.juicefs)
--no-updateclient will by default obtain latest config from JuiceFS Web Console, set to true to disable this behavior (usually not needed)

juicefs grep

Parallelized file pattern searcher.


juicefs grep [PATTERN] PATH ... [options]


[PATTERN]Support simple pattern and basic regular expression.
PATHSearch path and files.
-j JOBS --jobs=JOBSNumber of parallel jobs (default is number of CPU cores).
-e PATTERNS --regexp PATTERNSThis option is most useful when multiple -e options are used to specify multiple patterns, or when a pattern begins with a dash (-).
-E extended-regexpInterpret pattern as an extended regular expression.
-HAlways print filename headers with output lines.
-n --line-numberEach output line is preceded by its relative line number in the file, starting at line 1.
-i --ignore-casePerform case insensitive matching. By default, it is case sensitive.
-v --invert-matchSelected lines are those not matching any of the specified patterns.