Configuration File Format

Garage uses a TOML configuration file to define its behavior. This page documents all available configuration options.

Configuration File Location

The configuration file path is specified when starting Garage:

garage -c /etc/garage.toml server

Full Configuration Example

Here is a complete example showing all available options:

replication_factor = 3
consistency_mode = "consistent"

metadata_dir = "/var/lib/garage/meta"
data_dir = "/var/lib/garage/data"
metadata_snapshots_dir = "/var/lib/garage/snapshots"
metadata_fsync = true
data_fsync = false
disable_scrub = false
use_local_tz = false
metadata_auto_snapshot_interval = "6h"

db_engine = "lmdb"

block_size = "1M"
block_ram_buffer_max = "256MiB"
block_max_concurrent_reads = 16
block_max_concurrent_writes_per_request = 10
lmdb_map_size = "1T"

compression_level = 1

rpc_secret = "4425f5c26c5e11581d3223904324dcb5b5d5dfb14e5e7f35e38c595424f5f1e6"
rpc_bind_addr = "[::]:3901"
rpc_bind_outgoing = false
rpc_public_addr = "[fc00:1::1]:3901"
# or use rpc_public_addr_subnet to filter autodiscovery:
# rpc_public_addr_subnet = "2001:0db8:f00:b00:/64"

allow_world_readable_secrets = false

bootstrap_peers = [
    "563e1ac825ee3323aa441e72c26d1030d6d4414aeb3dd25287c531e7fc2bc95d@[fc00:1::1]:3901",
    "86f0f26ae4afbd59aaf9cfb059eefac844951efd5b8caeec0d53f4ed6c85f332@[fc00:1::2]:3901",
]

allow_punycode = false

[consul_discovery]
api = "catalog"
consul_http_addr = "https://127.0.0.1:8500"
tls_skip_verify = false
service_name = "garage-daemon"
ca_cert = "/etc/consul/consul-ca.crt"
client_cert = "/etc/consul/consul-client.crt"
client_key = "/etc/consul/consul-key.crt"
tags = [ "dns-enabled" ]
meta = { dns-acl = "allow trusted" }
datacenters = ["dc1", "dc2", "dc3"]

[kubernetes_discovery]
namespace = "garage"
service_name = "garage-daemon"
skip_crd = false

[s3_api]
api_bind_addr = "[::]:3900"
s3_region = "garage"
root_domain = ".s3.garage"

[s3_web]
bind_addr = "[::]:3902"
root_domain = ".web.garage"
add_host_to_metrics = true

[admin]
api_bind_addr = "0.0.0.0:3903"
metrics_token = "BCAdFjoa9G0KJR0WXnHHm7fs1ZAbfpI8iIZ+Z/a2NgI="
metrics_require_token = true
admin_token = "UkLeGWEvHnXBqnueR3ISEMWpOnm40jH2tM2HnnL/0F4="
trace_sink = "http://localhost:4317"

Top-Level Configuration Options

Replication and Consistency

replication_factor

integer

required

Number of copies of data to store across the cluster. Must be a positive integer ≤ node count.Recommendations:

1: Single node, no redundancy (testing only)
2: Tolerates one node failure, read-only mode when one node is down
3: Recommended - tolerates two node failures, maintains read/write with one failure
5, 7, etc.: Higher fault tolerance with uneven numbers preferred

Important: All nodes must have the same replication_factor. Changing this requires deleting the cluster layout and rebalancing all data.

consistency_mode

string

default:"consistent"

Controls read and write quorum behavior.Options:

consistent: Default. Guarantees read-after-write consistency
degraded: Lowers read quorum to 1. Allows reads when multiple nodes are down, but no read-after-write consistency
dangerous: Lowers both read and write quorums to 1. Least consistent mode, use with extreme caution

Storage Directories

metadata_dir

string

required

Directory for storing metadata (node ID, network config, peer list, bucket/key info, object indexes).Recommendation: Use fast SSD storage for best performance.

metadata_dir = "/var/lib/garage/meta"

data_dir

string | array

required

Directory for storing data blocks. Can be a single path or multiple paths with attributes.Single directory:

data_dir = "/var/lib/garage/data"

Multiple directories (since v0.9.0):

data_dir = [
    { path = "/path/to/old_data", read_only = true },
    { path = "/path/to/new_hdd1", capacity = "2T" },
    { path = "/path/to/new_hdd2", capacity = "4T" },
]

metadata_snapshots_dir

string

Directory for metadata snapshots. By default, snapshots are stored in <metadata_dir>/snapshots/.Set this to a location with more space if snapshots fill your metadata storage.

metadata_snapshots_dir = "/var/lib/garage/snapshots"

Database Configuration

db_engine

string

default:"lmdb"

Database engine for metadata storage.Options:

lmdb: Recommended for high-performance clusters (default since v0.9.0)
sqlite: Alternative backend, slower but more portable
fjall: Experimental LSM-tree based engine (testing only)

LMDB Notes:

Prone to corruption after unclean shutdown (use metadata_auto_snapshot_interval)
Database format not portable between architectures
Not recommended for 32-bit systems
Keys limited to 511 bytes (affects S3 object keys)

lmdb_map_size

string

Maximum size for LMDB database. This is virtual memory mapping size, not physical RAM.Defaults:

1GiB on 32-bit systems
1TiB on 64-bit systems

lmdb_map_size = "1T"

Data Integrity and Durability

metadata_fsync

boolean

default:"false"

Enable synchronous writes for metadata database. Reduces corruption risk during power failures but significantly impacts write performance.Impact by database engine:

SQLite: PRAGMA synchronous = NORMAL (vs OFF)
LMDB: MDB_NOMETASYNC only (vs MDB_NOMETASYNC + MDB_NOSYNC)
Fjall: Not supported

Not usually necessary with multi-site replication.

data_fsync

boolean

default:"false"

Call fsync() after writing data blocks to disk. Moderately impacts write performance.Reduces risk of data loss during simultaneous power failures on multiple nodes.

disable_scrub

boolean

default:"false"

Disable automatic monthly data scrubbing. Garage normally reads all data files to verify integrity and rebuilds corrupted files.Set to true if scrubbing at filesystem level. If you find corrupted files, delete them and run garage repair blocks.

metadata_auto_snapshot_interval

string

Automatically snapshot metadata database at regular intervals. Accepts duration strings like "6h", "1d", etc.Garage keeps the two most recent snapshots. Snapshots enable recovery from metadata corruption.

metadata_auto_snapshot_interval = "6h"

Note: Snapshot operations are intensive and may impact performance during heavy write loads.

Block Storage Configuration

block_size

string

default:"1M"

Size of data blocks for splitting stored objects. Accepts human-readable sizes.Recommendations:

Default: 1M (1 MiB)
Fast networks (≥1 Gbps) with large files: 10M or higher

block_size = "1M"

Note: Only affects newly uploaded files. Existing files remain accessible but won’t deduplicate with new files.

block_ram_buffer_max

string

default:"256MiB"

Maximum RAM for buffering data blocks awaiting asynchronous transfer to storage nodes.When limit is reached, backpressure is applied to S3 clients. This is not a hard limit on total RAM consumption.

block_ram_buffer_max = "256MiB"

block_max_concurrent_reads

integer

default:"16"

Maximum number of block files open simultaneously for reading (since v1.3.0/v2.1.0).Provides backpressure mechanism for HDD read speed. Requests wait up to 15 seconds for a read slot; timeouts indicate I/O bottleneck.Monitor block_read_semaphore_timeouts metric in Prometheus for timeout events.

block_max_concurrent_writes_per_request

integer

default:"3"

Maximum parallel block writes per PUT request (since v1.3.1/v2.2.0).Recommendations:

NVMe: 10-30
HDD: 3-10

Higher values improve throughput but increase memory usage.

compression_level

integer | string

default:"1"

Zstd compression level for data blocks.Values:

-99 to -1: Fast compression (smaller file to faster compression)
0: Zstd default (currently 3)
1 to 19: Standard compression (faster to smaller)
20 to 22: Ultra compression (high memory usage)
'none': Disable compression entirely

Compression is synchronous and affects write latency. Values can differ between nodes.

RPC Configuration

rpc_secret

string

required

32-byte hex-encoded secret shared by all cluster nodes for authentication.Generate with: openssl rand -hex 32Can also be set via rpc_secret_file or environment variables (see Environment Variables).

rpc_secret = "4425f5c26c5e11581d3223904324dcb5b5d5dfb14e5e7f35e38c595424f5f1e6"

rpc_bind_addr

string

required

Address and port for inter-cluster RPC communication.Important: Port must be consistent between internal and external addresses, even behind NAT. Multiple nodes behind NAT must use different ports.

rpc_bind_addr = "[::]:3901"

rpc_public_addr

string

Public address and port other nodes use to contact this node. Recommended but optional.Useful when behind NAT with port mapping.

rpc_public_addr = "[fc00:1::1]:3901"

rpc_public_addr_subnet

string

When rpc_public_addr is not set, filter autodiscovered IPs to specific subnet.

rpc_public_addr_subnet = "2001:0db8:f00:b00:/64"

rpc_bind_outgoing

boolean

default:"false"

Bind outgoing RPC connections to the same IP as rpc_bind_addr (since v0.9.2).Necessary when node has multiple IPs but only one can reach other nodes due to firewall or routing rules.

Cluster Discovery

bootstrap_peers

array

default:"[]"

List of peer identifiers for initial cluster discovery.Format: <node-public-key>@<ip-or-hostname>:<port>

bootstrap_peers = [
    "563e1ac825ee3323aa441e72c26d1030d6d4414aeb3dd25287c531e7fc2bc95d@[fc00:1::1]:3901",
    "86f0f26ae4afbd59aaf9cfb059eefac844951efd5b8caeec0d53f4ed6c85f332@[fc00:1::2]:3901",
]

Get node identifier with: garage node id

Security

allow_world_readable_secrets

boolean

default:"false"

Bypass permission checks on secret files. Useful with POSIX ACLs or complex permissions.Can also be set via GARAGE_ALLOW_WORLD_READABLE_SECRETS environment variable.

allow_punycode

boolean

default:"false"

Allow punycode in bucket names. Enables almost any Unicode character in website domain names.

Miscellaneous

use_local_tz

boolean

default:"false"

Run lifecycle worker at midnight in local timezone instead of UTC (since v1.1.0).Ensure all nodes have the same timezone configuration.

Consul Discovery Configuration

The [consul_discovery] section enables automatic node discovery via Consul. Nodes must have rpc_public_addr set.

consul_discovery.api

string

default:"catalog"

Consul API to use for service registration.Options:

catalog: Use /v1/catalog endpoints with optional mTLS
agent: Use /v1/agent endpoints with optional token

consul_discovery.consul_http_addr

string

required

Full HTTP(S) address of Consul server.

consul_http_addr = "https://127.0.0.1:8500"

consul_discovery.service_name

string

required

Service name under which Garage RPC ports are announced.

service_name = "garage-daemon"

consul_discovery.ca_cert

string

Path to CA certificate for TLS communication with Consul.

ca_cert = "/etc/consul/consul-ca.crt"

consul_discovery.client_cert

string

Path to client certificate for mTLS with Consul. Only available with api = "catalog".Both client_cert and client_key must be provided.

consul_discovery.client_key

string

Path to client key for mTLS with Consul. Only available with api = "catalog".

consul_discovery.tls_skip_verify

boolean

default:"false"

Skip TLS hostname verification. When enabled, ca_cert is ignored.

consul_discovery.token

string

Authentication token for Consul. Only available with api = "agent".Required policy:

service "garage" {
  policy = "write"
}
service_prefix "" {
  policy = "read"
}
node_prefix "" {
  policy = "read"
}

consul_discovery.tags

array

default:"[]"

Additional tags for service registration.

tags = ["dns-enabled"]

consul_discovery.meta

object

Additional service metadata.

meta = { dns-acl = "allow trusted" }

consul_discovery.datacenters

array

default:"[]"

List of datacenters for service discovery in WAN federation.

datacenters = ["dc1", "dc2", "dc3"]

Kubernetes Discovery Configuration

The [kubernetes_discovery] section enables automatic node discovery via Kubernetes custom resources.

kubernetes_discovery.namespace

string

required

Kubernetes namespace containing custom resources.

namespace = "garage"

kubernetes_discovery.service_name

string

required

Service name label for filtering resources (allows multiple deployments per namespace).

service_name = "garage-daemon"

kubernetes_discovery.skip_crd

boolean

default:"false"

Skip automatic creation/patching of garagenodes.deuxfleurs.fr CRD. You must create the CRD manually.

S3 API Configuration

The [s3_api] section configures the S3-compatible API endpoint.

s3_api.api_bind_addr

string

IP and port for S3 API. Can be a TCP address or Unix socket path (mode 0222).This endpoint does not support TLS; use a reverse proxy for HTTPS.

api_bind_addr = "[::]:3900"
# or Unix socket:
api_bind_addr = "/var/run/garage-s3.sock"

s3_api.s3_region

string

required

S3 region name. API calls for other regions fail with AuthorizationHeaderMalformed.

s3_region = "garage"

s3_api.root_domain

string

Domain suffix for vhost-style S3 requests. Requires wildcard DNS and optionally wildcard TLS certificate.Path-style requests always work regardless of this setting.

root_domain = ".s3.garage"

With this setting, bucket my-bucket is accessible at my-bucket.s3.garage.

S3 Web Configuration

The [s3_web] section configures static website hosting from S3 buckets.

s3_web.bind_addr

string

required

IP and port for web server. Can be TCP address or Unix socket path (mode 0222).This endpoint does not support TLS; use a reverse proxy for HTTPS.

bind_addr = "[::]:3902"

s3_web.root_domain

string

required

Domain suffix for website hosting.

root_domain = ".web.garage"

Bucket deuxfleurs.fr accessible at both:

deuxfleurs.fr.web.garage
deuxfleurs.fr

s3_web.add_host_to_metrics

boolean

default:"false"

Include HTTP Host header in Prometheus metrics. Disabled by default to prevent cardinality explosion.

Admin API Configuration

The [admin] section configures the administration and monitoring API.

admin.api_bind_addr

string

IP and port for admin API. Can be TCP address or Unix socket path (mode 0220).

api_bind_addr = "0.0.0.0:3903"

admin.admin_token

string

Bearer token for admin API access (except metrics endpoint).Generate with: openssl rand -base64 32Can also be set via admin_token_file or environment variables (see Environment Variables).Since v2.0, additional tokens can be defined dynamically with expiration and scoped permissions. This token acts as full-scope, non-expiring token.

admin.metrics_token

string

Bearer token for /metrics endpoint access.Generate with: openssl rand -base64 32If not set and metrics_require_token is false, metrics are accessible without authentication.Can also be set via metrics_token_file or environment variables (see Environment Variables).

admin.metrics_require_token

boolean

default:"false"

Require authentication token for metrics endpoint (since v2.0.0).Valid tokens include metrics_token and dynamically defined admin tokens with Metrics scope.

admin.trace_sink

string

OpenTelemetry collector address for distributed tracing.

trace_sink = "http://localhost:4317"

Configuration File Conversion

To convert metadata between database engines:

garage convert-db \
  -a <input_engine> -i <input_db_path> \
  -b <output_engine> -o <output_db_path>

Example:

garage convert-db \
  -a sqlite -i /var/lib/garage/meta/db.sqlite \
  -b lmdb -o /var/lib/garage/meta/db.lmdb/

Environment Variables - Environment variable overrides and configuration
Admin API Reference - Administration API endpoints
Cluster Management - Managing your Garage cluster

Documentation Index

​Configuration File Format

​Configuration File Location

​Full Configuration Example

​Top-Level Configuration Options

​Replication and Consistency

​Storage Directories

​Database Configuration

​Data Integrity and Durability

​Block Storage Configuration

​RPC Configuration

​Cluster Discovery

​Security

​Miscellaneous

​Consul Discovery Configuration

​Kubernetes Discovery Configuration

​S3 API Configuration

​S3 Web Configuration

​Admin API Configuration

​Configuration File Conversion

​Related Resources

Configuration File Format

Configuration File Location

Full Configuration Example

Top-Level Configuration Options

Replication and Consistency

Storage Directories

Database Configuration

Data Integrity and Durability

Block Storage Configuration

RPC Configuration

Cluster Discovery

Security

Miscellaneous

Consul Discovery Configuration

Kubernetes Discovery Configuration

S3 API Configuration

S3 Web Configuration

Admin API Configuration

Configuration File Conversion

Related Resources