Cloud Storage FUSE configuration file

This document describes how to use a Cloud Storage FUSE configuration file to configure the behavior of Cloud Storage FUSE in a persistent manner. To use the configuration file, specify the path to the configuration file in the --config-file option when you run the gcsfuse command.

The Cloud Storage FUSE configuration file is a YAML file that uses the following format and fields. Behavior controlled by the configuration file fields can also be set using the Cloud Storage FUSE CLI.

Cloud Storage FUSE configuration file format and fields

The following YAML configuration example shows the format and fields available in Cloud Storage FUSE.

app-name:"APP_NAME"
logging:
file-path:"FILE_PATH"
format:FORMAT
severity:SEVERITY
log-rotate:
max-file-size-mb:MAX_FILE_SIZE
backup-file-count:BACKUP_FILE_COUNT
compress:COMPRESS
cache-dir:"CACHE_DIR"
file-cache:
max-size-mb:MAX_SIZE
cache-file-for-range-read:CACHE_FILE_FOR_RANGE_READ
enable-parallel-downloads:ENABLE_PARALLEL_DOWNLOADS
parallel-downloads-per-file:PARALLEL_DOWNLOADS_PER_FILE
max-parallel-downloads:MAX_PARALLEL_DOWNLOADS
download-chunk-size-mb:DOWNLOAD_CHUNK_SIZE
metadata-cache:
enable-nonexistent-type-cache:ENABLE_NONEXISTENT_TYPE_CACHE
negative-ttl-secs:ENABLE_NEGATIVE_TTL_SECS
stat-cache-max-size-mb:STAT_CACHE_MAX_SIZE
ttl-secs:TTL_SECS
type-cache-max-size-mb:TYPE_CACHE_MAX_SIZE
only-dir:"ONLY_DIR"
gcs-auth:
anonymous-access:ANONYMOUS_ACCESS
key-file:"KEY_FILE"
reuse-token-from-url:REUSE_TOKEN_FROM_URL
token-url:"TOKEN_URL"
gcs-connection:
billing-project:"BILLING_PROJECT"
client-protocol:CLIENT_PROTOCOL
custom-endpoint:"CUSTOM_ENDPOINT"
http-client-timeout:HTTP_CLIENT_TIMEOUT
limit-bytes-per-sec:"LIMIT_BYTES_PER_SEC"
limit-ops-per-sec:"LIMIT_OPS_PER_SEC"
max-conns-per-host:MAX_CONNS_PER_HOST
max-idle-conns-per-host:MAX_IDLE_CONNS_PER_HOST
sequential-read-size-mb:SEQUENTIAL_READ_SIZE
implicit-dirs:IMPLICIT_DIRS
file-system:
kernel-list-cache-ttl-secs:KERNEL_LIST_CACHE_TTL_SECS
ignore-interrupts:IGNORE_INTERRUPTS
dir-mode:"DIR_MODE"
file-mode:"FILE_MODE"
fuse-options:FUSE_OPTIONS
gid:GID
rename-dir-limit:RENAME_DIR_LIMIT
temp-dir:"TEMP_DIR"
uid:UID
foreground:FOREGROUND
gcs-retries:
max-retry-attempts:MAX_RETRY_ATTEMPTS
max-retry-sleep:MAX_RETRY_SLEEP
multiplier:"MULTIPLIER"
metrics:
cloud-metrics-export-interval-secs:CLOUD_METRICS_EXPORT_INTERVAL
prometheus-port:PROMETHEUS_PORT
debug:
log-mutex:LOG_MUTEX
exit-on-invariant-violation:EXIT_ON_INVARIANT_VIOLATION
write:
enable-streaming-writes:STREAMING_WRITES
global-max-blocks:MAXIMUM_GLOBAL_BLOCKS
read:
enable-buffered-read:ENABLE_BUFFERED_READ
global-max-blocks:MAXIMUM_GLOBAL_READ_BLOCKS
profile:PROFILE

Cloud Storage FUSE configuration file fields

The following table describes the fields you can specify in your Cloud Storage FUSE configuration file. Unless stated otherwise, all fields are optional.

Field Description Valid value Default value

app-name

 The application name of the mount. String value, for example: "my-bucket-mount". ""

file-path

 The path to the log file where logs will be written. If this field is unspecified, the logs are routed to stdout when Cloud Storage FUSE runs in foreground mode and to syslogs when Cloud Storage FUSE runs in background mode. String value, for example: "/var/log". ""

format

 Specifies the format of the log file.
  • text
  • json
 json

severity

The severity level you want Cloud Storage FUSE to generate logs for. The severity levels are ordered from lowest severity to highest severity. For example, when you specify warning, Cloud Storage FUSE generates logs for warnings and errors. Generally, we recommend using the info severity level.
  • off: disables all logging.
  • error: contains messages about serious issues that prevent Cloud Storage FUSE from completing an operation, or indicate a failure such as mount failures, permission errors, and critical input/output (I/O) errors.
  • warning: contains messages indicating potential issues that aren't critical but can lead to issues if not addressed. Potential issues include non-fatal connection issues, deprecation warnings, resource contention that might cause slight performance degradation, and retries for transient failures. This severity level also includes information provided in the error severity level.
  • info: contains logging information such as startup and shutdown messages, or information about successful mount operations, progress updates, and configuration settings. This severity level also includes information about the warning and error severity levels.
  • debug: contains logging information such as startup and shutdown messages, or information about successful mount operations, progress updates, and configuration settings. This severity level also includes the information provided in the info, warning, and error severity levels.
  • trace: contains granular details about each Cloud Storage FUSE operation and function call, outlining the gcsfuse interaction with the kernel FUSE driver and Cloud Storage. This severity level also includes the details provided in debug, info, warning, and error severity levels.
 info

max-file-size-mb

 The maximum size in megabytes (MB) that log files can reach before being rotated. Integer. The minimum value is 1. 512

backup-file-count

 The maximum number of rotated log files to retain, excluding the active file that logs are written to.
  • Integer
  • 0: Retains all rotated log files
 10

compress

 Specifies whether rotated log files are compressed using gzip. Boolean value: true, false true

cache-dir

Enables the file cache and specifies the directory for storing file cache data.

 A path, for example: "/tmp/gcsfuse-cache-path". An empty value indicates this field is disabled. This field is disabled by default.

max-size-mb

Specifies the maximum size in MiB that the file cache can use and lets you limit the total capacity the file cache can use within its mounted directory.
  • Integer
  • -1: Specifies the use of the cache's entire available capacity in the directory you specify for cache-dir. This is the default only if cache-dir is passed.
  • 0: Disables the file cache.
 -1

cache-file-for-range-read

 Determines whether the full object should be downloaded asynchronously and stored in the Cloud Storage FUSE cache directory when the first read is completed from a non-zero offset. This field should be set to true if you plan to perform several random reads or partial reads. Boolean value: true, false false

enable-parallel-downloads

Accelerates reads of large files by using the file cache directory as a prefetch buffer using multiple workers to download multiple parts of a file in parallel.

Parallel downloads are automatically enabled when you enable file caching. To learn more about file caching, see Use Cloud Storage FUSE file caching.

To learn more about parallel downloads and how to configure supporting properties, Parallel downloads.

 Boolean value: true, false. true

parallel-downloads-per-file

 Specifies the number of maximum goroutines to spawn per file to download the object from Cloud Storage into the file cache. Integer 16

max-parallel-downloads

 The maximum number of goroutines that can be spawned at any given time across all the download jobs of files.
  • Integer
  • -1: Specifies unlimited parallel downloads.
  • 0: Disables parallel downloads. Can only be used if --enable-parallel-downloads is not passed or is passed as false.
 Twice the number of CPU cores on your machine or 16, whichever is higher.

download-chunk-size-mb

 Specifies the size of each read request in MiB that each goroutine makes to Cloud Storage when downloading the object into the file cache. Integer 200

enable-nonexistent-type-cache

 Creates a type cache entry with the type NonexistentType if a file isn't found in Cloud Storage. If the file gets created in Cloud Storage but the NonexistentType entry for the file is cached, then Cloud Storage FUSE cannot request that file until the NonexistentType entry is removed from the type cache. Boolean value: true, false false

stat-cache-max-size-mb

 The maximum size memory that the stat cache can use, in MiB. The stat cache is always entirely kept in memory.
  • Integer. We recommend the following:
    • 32 if your workload involves up to 20,000 files.
    • If your workload is larger than 20,000 files, increase the size by values of 10 for every additional 6,000 files, where the stat cache uses an average of 1,500 MiB per file.
  • -1: Sets no limit, where the stat cache use as much memory as needed.
  • 0: Disables the stat cache.
 32

negative-ttl-secs

Defines the time to live (TTL) in seconds of negative stat cache entries, which store results for non-existent files in the cache.
  • Integer representing seconds, for example: 10 (10 seconds).
  • 0: Disables negative stat caching.
  • -1: Allows unlimited negative stat caching and disables a TTL expiration.
 5

ttl-secs

 Defines the time to live (TTL) in seconds of cached metadata entries.
  • Integer representing seconds, for example: 30 (30 seconds).
  • -1: Bypass TTL expiration and serve files from the cache whenever they're available.
  • 0: Use the most up-to-date file. Using this value issues a Get metadata call to make sure that the object generation for the file in the cache matches what's stored in Cloud Storage.
 60

type-cache-max-size-mb

 The maximum size in MiB per directory that the type cache can use. The type cache is always entirely kept in memory.
  • Integer. We recommend the following:
    • 4 if the maximum number of files within a single directory from the bucket you're mounting contains 20,000 files or less.
    • If the maximum number of files within a single directory that you're mounting contains more than 20,000 files, increase the value by 1 for every 5,000 files, which is an average of around 200 bytes per file.
  • -1: Specifies no limit and lets the type cache use as much memory as needed.
  • 0: Disables the type cache.
 4

only-dir

 Mounts only a specific directory within a bucket. A path, for example: "/etc/gcsfuse.yaml".

anonymous-access

 Disables authentication for requests. This field should be set if you're using a custom endpoint that doesn't support authentication. This field should also be set if you're using Cloud Storage FUSE with public buckets. Boolean value: true, false false

key-file

 Specifies an absolute path to the credential JSON key file for authenticating requests to Cloud Storage. By default, Cloud Storage FUSE uses Application Default Credentials to authenticate requests. A URL. When this field is not set, Application Default Credentials are used.

reuse-token-from-url

 Specifies whether to reuse the token acquired from --token-url. Boolean value: true, false true

token-url

 Specifies a URL for getting an access token when the --key-file is absent. A URL.

billing-project

 Specifies a project to use for billing when the mounted bucket is accessed. This field is often required when mounting a bucket enabled with Requester Pays. String value representing a "project ID". ""

client-protocol

 Specifies the protocol used for communicating with the Cloud Storage backend. http1

custom-endpoint

 Specifies an alternative custom endpoint for fetching data. The custom endpoint must support the equivalent resources and operations as the Cloud Storage JSON endpoint, storage.UNIVERSE_DOMAIN_NAME:443. If a custom endpoint isn't specified, Cloud Storage FUSE uses the global Cloud Storage JSON API endpoint, storage.googleapis.com:443. If authentication isn't supported on the custom endpoint you specify, set the anonymous-access field to true to bypass authentication. An endpoint, for example: storage.googleapis.com:443.

http-client-timeout

 Specifies how long the Cloud Storage FUSE HTTP client can wait to get a response from the server before timing out. Duration, for example: 1h10m10s for 1 hour, 10 minutes, and 10 seconds. 0s specifies no timeout. 0s, which specifies no timeout

limit-bytes-per-sec

 Specifies the bandwidth limit at which Cloud Storage FUSE can read data from Cloud Storage, measured over a 30-second window. "-1", which specifies no limit.

limit-ops-per-sec

 Specifies a limit for operations performed per second, measured over a 30-second window. Floating point number. -1 specifies no limit. "-1"

max-conns-per-host

 Specifies the maximum number of TCP connections allowed per server. This becomes effective when --client-protocol is set to http1. 0

max-idle-conns-per-host

 Specifies the maximum number of idle connections allowed per server. This becomes effective when --client-protocol is set to http1. Integer between 0 and 2147483647. 0 specifies no limit on TCP connections. 0

sequential-read-size-mb

 Specifies the chunk size of the data to be downloaded from Cloud Storage, in megabytes (MB). Integer between 1 and 1024. 200

implicit-dirs

 Implicitly includes folders and managed folders. See Files and directories in the Cloud Storage FUSE GitHub documentation for more information. Boolean value: true, false false

kernel-list-cache-ttl-secs

 Enables the list cache and defines the time to live (TTL) in seconds of cached list entries. The list cache is kept in memory in the page cache, which is controlled by the kernel based on available memory.
  • Integer representing seconds, for example: 10 (10 seconds).
  • 0: Disables list caching.
  • -1: Bypasses entry expiration and always returns the list response from the cache when it's available.
0

ignore-interrupts

 Instructs Cloud Storage FUSE to ignore system interrupt signals, like SIGINT triggered by Control+C. This prevents signals from terminating in-flight operations. Boolean value: true, false. true

dir-mode

 Permissions bits for directories, in octal. Integer between 000 and 777 (inclusive). "755"

file-mode

 Specifies permissions bits for files, in octal. Integer between 000 and 777 (inclusive). "644"

fuse-options

 Specifies additional system-specific mount options.

gid

 Specifies the Group Identifier (GID) owner of all inodes.
  • Integer representing a GID.
  • -1: The GID of the caller is used.
 -1

rename-dir-limit

 Allows the renaming of directories containing fewer descendants than the specified limit. Integer between 0 and 2147483647. 0

temp-dir

 Specifies a path to the temporary directory where writes are staged prior to being uploaded to Cloud Storage. A string path, for example: "/mnt/ssd/example-user-gcsfuse-temp-dir". "/tmp"

uid

 Specifies the User Identifier (UID) owner of all inodes.
  • Integer representing a UID.
  • -1: The UID of the caller is used.
 -1

foreground

 Runs the gcsfuse command in the foreground. Boolean value: true, false false

max-retry-attempts

 Specifies the maximum number of times an operation is retried if the operation fails, preventing endless retry loops.
  • Integer representing the maximum number of retry attempts, for example: 10.
  • 0: Allows unlimited retry attempts.
 0

max-retry-sleep

 Specifies the maximum duration that Cloud Storage FUSE is allowed to sleep in a retry loop with exponential backoff. Once the backoff duration exceeds the specified maximum duration, the retry continues with the specified maximum duration. Duration, for example: 1h5m50s (1 hour, 5 minutes, and 50 seconds) or 60s (60 seconds). 30s

multiplier

 Specifies the multiplier for exponential backoff between consecutive retries. Floating-point number "2"

cloud-metrics-export-interval-secs

Exports metrics to Cloud Monitoring with the specified interval.

 Integer representing a value in seconds, for example: 10 (10 seconds). 0 specifies no exporting. 0

prometheus-port

Exposes Prometheus metrics endpoint on the specified port and /metrics path.

 Integer representing the port you want to specify. 0

log-mutex

 Prints debug messages when a mutex is held too long. If this field is specified, the severity level of logs is automatically set to trace, which includes trace logs, debug logs, info logs, warning logs, and error logs. Boolean value: true, false. false

exit-on-invariant-violation

 Exits the program when internal variant violations are detected. Boolean value: true, false. false

enable-streaming-writes

 Controls the write path flow so that data is uploaded directly to Cloud Storage as its written instead of fully staging the write locally and uploading it upon close() or fsync(). For more information about streaming writes, see Reads/Writes in the Cloud Storage FUSE GitHub documentation. Boolean value: true, false. true

global-max-blocks

Specifies the maximum number of data chunks available for streaming writes across all files. Each file uses one 32 MiB data chunk by default.
  • Integer
  • 0: disables streaming writes.
  • -1: sets the value to unlimited chunks.
  • Default value for low-specification machines with limited resources: 4
  • Default value for high-specification machines with a large amount of resources: 1600

enable-buffered-read

 Specifies asynchronous prefetching of parts of a Cloud Storage object into an in-memory buffer, allowing subsequent reads to be served from the buffer instead of requiring network calls. Boolean value: true, false false

read:global-max-blocks

Specifies the maximum number of blocks available for buffered reads across all file handles.
  • Integer
  • 0: disables buffered reads.
  • -1: sets the value to unlimited blocks.
 40

profile

 Applies a predefined, optimized set of Cloud Storage FUSE configurations for caching, threading, and buffer sizes to help you achieve high performance for a specific workload type, such as training, serving, and checkpointing. To learn more about each predefined configuration based on your workload type, see Profile-based configurations for AI/ML workloads.
  • A string, for example, "aiml-checkpointing".
  • "aiml-training": optimizes performance for high throughput reads of large datasets and prevents Cloud GPUs and Cloud TPU hardware from waiting for data.
  • aiml-checkpointing: optimizes performance for high throughput writes for large files by drastically reducing the time it takes to save multi-gigabyte checkpoints, minimizing training pauses.
  • aiml-serving: optimizes performance for serving workloads by streamlining data access and applying caching mechanisms.
 ""

Back to top

Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates.

Last updated 2025年11月11日 UTC.