Configuration Reference

This chapter provides detailed descriptions of Curvine configuration options.

Cluster Configuration

Configuration Item	Default Value	Description
format_master	true	Controls whether to format Master metadata storage
format_worker	true	Controls whether to format Worker data storage
testing	false	Marks whether in testing mode, data flow is disabled in testing mode
cluster_id	curve	Unique identifier for the cluster

Master Configuration

Configuration Item	Default Value	Description
master.hostname	localhost	Master service hostname, can be overridden by environment variable CURVINE_MASTER_HOSTNAME
master.rpc_port	8995	Master service RPC port
master.web_port	9000	Master service Web port
master.io_threads	32	Number of Master service I/O threads
master.worker_threads	CPU cores * 2	Number of Master service worker threads
master.io_timeout	10m	Master network read/write data timeout
master.io_close_idle	true	Whether to close idle connections
master.meta_dir	./fs-meta	Metadata storage directory
master.meta_disable_wal	true	Whether to disable RocksDB's Write-Ahead Log (WAL)
master.meta_compression_type	none	Compression type for metadata storage
master.meta_db_write_buffer_size	0	Database write buffer size for metadata storage
master.meta_write_buffer_size	64MB	Write buffer size for metadata storage
master.min_block_size	1024 * 1024	Minimum block_size
master.max_block_size	100 * 1024 * 1024 * 1024	Maximum block_size
master.min_replication	1	Minimum number of replicas
master.max_replication	100	Maximum number of replicas
master.max_path_len	8000	Maximum path length
master.max_path_depth	1000	Maximum path depth
master.retry_cache_enable	true	Whether to enable filesystem request retry cache
master.retry_cache_size	100_000	Size of retry cache
master.retry_cache_ttl	10m	Expiration time of retry cache
master.block_report_limit	1000	Block report limit, maximum number of blocks allowed per report
master.worker_policy	local	Worker selection policy, default local, prioritizes workers on the same node as client
master.executor_threads	10	Number of master background task executor threads
master.executor_channel_size	1000	Master background task executor channel size
master.heartbeat_check_interval	5m	Interval for Master to check Worker heartbeats
master.heartbeat_interval	3s	Interval for Worker to send heartbeats
master.audit_logging_enabled	true	Whether to enable audit logging
master.log	{level: "info", log_dir: "stdout"}	Master log configuration, defaults to stdout output

Journal Configuration

Configuration Item	Default Value	Description
enable	true	If raft log sync is enabled, logs will be synchronized; if disabled, equivalent to single-node system
group_name	raft-group	Raft group name
rpc_port	8996	RPC port number
io_threads	8	Number of I/O threads
worker_threads	8	Number of worker threads
message_size	200	Message size
journal_addrs	{id: 1, hostname: "localhost", port: 8996}	List of Master candidate node addresses
journal_dir	./journal	Raft log storage directory
writer_debug	false	Whether to enable debug mode for write operations, prints each log entry when enabled
writer_channel_size	0	Buffer queue size for journal writing, defaults to unbounded queue
writer_flush_batch_size	1000	Number of entries for write operation batch flush
writer_flush_batch_ms	100	Time interval for write operation batch flush (milliseconds)
snapshot_interval	6h	Snapshot creation interval
snapshot_entries	100_000	Number of entries generated after creating snapshot
snapshot_read_chunk_size	1024 * 1024	Snapshot read chunk size
conn_retry_max_duration_ms	0	Maximum duration for connection retry (milliseconds)
conn_retry_min_sleep_ms	10 * 1000	Minimum sleep time for connection retry (milliseconds)
conn_retry_max_sleep_ms	10 * 1000	Maximum sleep time for connection retry (milliseconds)
rpc_close_idle	false	Whether to close idle RPC connections
rpc_retry_max_duration_ms	60 * 1000	Maximum duration for RPC request retry (milliseconds)
rpc_retry_min_sleep_ms	20 * 1000	Minimum sleep time for RPC request retry (milliseconds)
rpc_retry_max_sleep_ms	20 * 1000	Maximum sleep time for RPC request retry (milliseconds)
conn_timeout_ms	30 * 1000	Connection timeout (milliseconds), default 30 seconds
io_timeout_ms	60 * 1000	Socket data read/write timeout (milliseconds), default 60 seconds
conn_size	1	Number of available connections when sharing connections
raft_poll_interval_ms	100	Raft polling interval (milliseconds)
raft_tick_interval_ms	1000	Raft heartbeat interval (milliseconds)
raft_election_tick	10	Raft election timeout cycles
raft_heartbeat_tick	3	Raft heartbeat timeout cycles
raft_min_election_ticks	10	Raft minimum election timeout cycles
raft_max_election_ticks	30	Raft maximum election timeout cycles
raft_check_quorum	true	Whether to check quorum
raft_max_size_per_msg	1024 * 1024	Maximum size per message
raft_max_inflight_msgs	256	Maximum number of unacknowledged messages
raft_max_committed_size_per_ready	16 * 1024 * 1024	Maximum commit size per ready state
raft_retry_cache_size	100_000	Raft request retry cache size, prevents duplicate requests
raft_retry_cache_ttl	10m	Expiration time for raft request retry cache
retain_checkpoint_num	3	Number of checkpoints to retain
ignore_replay_error	false	Whether to ignore errors during log replay

Worker Configuration

Configuration Item	Default Value	Description
worker.dir_reserved	0	Directory reserved space configuration
worker.data_dir	vec![]	Worker data directory list
worker.io_slow_threshold	300ms	Worker I/O operation slow threshold, operations exceeding this time are considered slow
worker.io_threads	32	Number of Worker network I/O threads
worker.worker_threads	CPU cores * 2	Number of Worker worker threads
worker.io_timeout	10m	Worker network read/write data timeout
worker.io_close_idle	false	Whether to close idle connections, defaults to false since client doesn't implement heartbeat mechanism; uses keepalive to check connection validity
worker.scheduler_threads	2	Number of background scheduling task threads
worker.executor_threads	10	Number of background asynchronous task threads
worker.executor_channel_size	1000	Background asynchronous task queue size
worker.log	{level: "info", log_dir: "stdout"}	Worker log configuration, defaults to stdout output

Worker Data Loading Configuration

Configuration Item	Default Value	Description
worker.load.task_status_report_interval_ms	5000	Progress report interval (milliseconds)
worker.load.task_read_chunk_size_bytes	1024 * 1024	Default chunk size (bytes)
worker.load.task_transfer_buffer_count	16	Default buffer cache count
worker.load.task_timeout_seconds	3600	Task timeout (seconds), -1 means no timeout
worker.load.task_transfer_max_concurrent_tasks	5	Maximum concurrent tasks

Client Configuration

Configuration Item	Default Value	Description
client.master_addrs	localhost:8995	Master node address list
client.hostname	localhost	Hostname of the machine where client is located, needs to be set in certain scenarios to identify if client and worker nodes are on the same machine
client.io_threads	16	Number of client I/O threads
client.worker_threads	CPU cores * 2	Number of client worker threads
client.replicas	1	Default number of replicas
client.block_size	128MB	Block size (bytes), actual value converted from block_size_str
client.write_chunk_size	128KB	Write buffer size
client.write_chunk_num	8	Write buffer queue length
client.read_chunk_size	128KB	Read buffer size
client.read_chunk_num	32	Read buffer queue length
client.read_parallel	1	Parallelism for reading files
client.read_slice_size	0	Read slice size (bytes), default value is read_chunk_size * read_chunk_num
client.close_reader_limit	20	In random read scenarios, if block switching due to seek operations exceeds this value, all block readers will be cached
client.short_circuit	true	Whether to enable short-circuit functionality
client.storage_type	disk	Default storage type
client.ttl_ms	0	Default data cache time (milliseconds)
client.ttl_action	none	Action after cache expiration
client.auto_cache_enabled	false	Whether to enable auto-cache functionality, when enabled, reading files from external filesystem will automatically submit load requests to master and cache files
client.auto_cache_ttl	7d	Default TTL for auto-cache, format is number plus unit or pure number (seconds)
client.conn_retry_max_duration_ms	2 * 60 * 1000	Maximum duration for connection retry (milliseconds)
client.conn_retry_min_sleep_ms	300	Minimum sleep time for connection retry (milliseconds)
client.conn_retry_max_sleep_ms	10 * 1000	Maximum sleep time for connection retry (milliseconds)
client.rpc_retry_max_duration_ms	5 * 60 * 1000	Maximum duration for RPC request retry (milliseconds)
client.rpc_retry_min_sleep_ms	300	Minimum sleep time for RPC request retry (milliseconds)
client.rpc_retry_max_sleep_ms	30 * 1000	Maximum sleep time for RPC request retry (milliseconds)
client.rpc_close_idle	true	Whether to close idle RPC connections
client.conn_timeout_ms	30 * 1000	Connection timeout (milliseconds)
client.rpc_timeout_ms	120 * 1000	RPC request timeout (milliseconds)
client.data_timeout_ms	5 * 60 * 1000	Data transfer timeout (milliseconds)
client.master_conn_pool_size	1	Master node connection pool size, testing shows 3 connections perform best, default is 1
client.enable_read_ahead	true	Whether to enable read-ahead functionality
client.read_ahead_len	0	Read-ahead length (bytes), default value is read_chunk_size * read_chunk_num
client.drop_cache_len	0	Drop cache length (bytes)
client.failed_worker_ttl_str	10m	Worker node blacklist expiration time

FUSE Configuration

Configuration Item	Default Value	Description
fuse.debug	false	Whether to output request/response logs
fuse.io_threads	32	Number of I/O threads
fuse.worker_threads	CPU cores * 2	Number of worker threads
fuse.mnt_path	/curvine-fuse	Mount path
fuse.fs_path	/	Specifies the root path of filesystem accessed by mount point
fuse.mnt_number	1	Number of mount points
fuse.mnt_per_task	0 (initialized to io_threads value if 0)	Number of tasks that can read/write data per mount point
fuse.clone_fd	true	Whether to enable clone fd functionality
fuse.fuse_channel_size	0	Fuse request queue size, 0 uses unbounded queue
fuse.stream_channel_size	0	Read/write file request queue size, 0 uses unbounded queue
fuse.fuse_opts	[]	Mount configuration, needs to be passed to Linux kernel
fuse.umask	022	Overrides permission bits set by filesystem in st_mode, represented in octal
fuse.uid	sys::get_uid()	User ID
fuse.gid	sys::get_gid()	Group ID
fuse.auto_cache	false	Whether to enable auto-cache
fuse.read_dir_fill_ino	true	Whether to fill fuse node ID when traversing directories
fuse.entry_timeout	1.0	Name lookup cache time (seconds)
fuse.negative_timeout	0.0	Timeout for caching negative lookups (seconds), 0 means disabled
fuse.attr_timeout	1.0	Cache time for file and directory attributes (seconds)
fuse.ac_attr_timeout	1.0	Timeout for file attribute cache during auto-cache refresh (seconds)
fuse.ac_attr_timeout_set	1.0	Timeout for file attribute cache during auto-cache refresh (seconds)
fuse.remember	false	Whether filesystem remembers opened files and directories
fuse.max_background	256	Maximum concurrent execution of background tasks in filesystem
fuse.congestion_threshold	192	Congestion threshold
fuse.node_cache_size	200000	Inode cache capacity
fuse.node_cache_timeout	1h	Inode cache lifetime
fuse.direct_io	true	File and directory related options, whether to enable direct I/O
fuse.kernel_cache	false	File and directory related options, whether to enable kernel cache
fuse.cache_readdir	false	File and directory related options, whether to cache directory read results
fuse.non_seekable	false	File and directory related options, whether in non-seekable mode
fuse.log	{level: "info", log_dir: "stdout"}	FUSE log configuration, defaults to stdout output