Skip to main content

Server Configuration Flags

Dragonfly can be tuned and configured by a set of config flags. These flags can be:

  1. Passed to a Dragonfly as command line arguments. E.g, dragonfly --port=6379.
  2. Loaded from a file. E.g. dragonfly --flagfile=path_to_flags/flags.txt
  3. Can be set via env variables by adding the prefix DFLY_ followed by the flag name. E.g. export DFLY_port=6379 (note it's case sensitive)
  4. At runtime via CONFIG SET command. Not all flags can be configured at runtime.

You can try dragonfly --helpfull to get a list of all flags or --help=substring shows help for flags which include specified substring in either in the name, description or path.

Available Flags​

--port​

Redis port. 0 disables the port, -1 will bind on a random available port.

default: 6379

--cache_mode​

If true, the backend behaves like a cache, by evicting entries when getting close to maxmemory limit

default: false

--cf_bucket_size​

Default cuckoo filter bucket size, in slots per bucket, when a filter is created without an explicit bucket size.

default: 2

--cf_expansion_factor​

Default growth factor for cuckoo filter sub-filters.

default: 1

--cf_initial_size​

Default initial capacity for a cuckoo filter.

default: 1024

--cf_max_expansions​

Maximum number of sub-filters a cuckoo filter can grow to through CF.ADD, CF.ADDNX, CF.INSERT, or CF.INSERTNX before those commands start failing.

default: 32

--cf_max_iterations​

Default maximum number of cuckoo filter relocation iterations before an insert fails.

default: 20

--cluster_mode​

Set cluster mode. Available options are: yes, emulated and empty "".

default: ""

--maxmemory​

Limit on maximum-memory that is used by the database. 0 - means the program will automatically determine its maximum memory usage.

default: 0B

--dbnum​

Number of databases.

default: 16

--bind​

Bind address. If empty - binds on all interfaces. It's not advised due to security implications.

default: ""

--requirepass​

Password for AUTH authentication. If empty, it can also be set with the DFLY_PASSWORD environment variable.

default: ""

--dbfilename​

The filename to save/load the DB. In addition to {timestamp}, the {Y}, {m}, and {d} macros can be used.

default: "dump-{timestamp}"

--dir​

Working directory

default: ""

--snapshot_cron​

Cron expression for the time to save a snapshot, crontab style.

default:

--snapshot_egress_limit_bytes​

Per-shard-thread socket egress bandwidth budget in bytes per second. Each shard throttles its snapshot traversal loop to stay under this rate. Accepts human-readable sizes such as 100mb and 1gb. A value of 0 disables throttling.

default: 0B

--admin_bind​

If set, the admin console TCP connection would be bind to the given address. This supports both HTTP and RESP protocols.

default: ""

--admin_port​

If set, would enable admin access to console on the assigned port. This supports both HTTP and RESP protocols.

default: 0

--max_client_iobuf_len​

Maximum io buffer length that is used to read client requests.

default: 32.0KiB

--iobuf_min_shrink_interval_sec​

Minimum time, in seconds, before a client input buffer may shrink. A value of 0 disables shrinking.

default: 30

--max_multi_bulk_len​

Maximum multi-bulk (array) length that is allowed to be accepted when parsing RESP protocol

default: 65536

--migrate_connections​

When enabled, Dragonfly will try to migrate connections to the target thread on which they operate. Currently this is only supported for Lua script invocations, and can happen at most once per connection.

default: true

--no_tls_on_admin_port​

Allow non-tls connections on admin port.

default: false

--publish_buffer_limit​

Amount of memory to use for storing publish commands per IO thread. This is the soft Pub/Sub back-pressure limit; publishers are only parked once the per-thread subscriber memory reaches this value times the internal hard-limit multiplier.

default: 196.00MiB

--pubsub_slow_subscriber_timeout_ms​

If a subscriber connection keeps a Pub/Sub socket write blocked for at least this many milliseconds, Dragonfly closes it when either its queued Pub/Sub memory reaches one-sixteenth of publish_buffer_limit or the per-IO-thread subscriber memory is above publish_buffer_limit. A value of 0 disables this slow-subscriber protection. This is a startup-only flag, like publish_buffer_limit.

default: 0

--pipeline_squash​

Number of queued pipelined commands above which squashing is enabled, 0 means disabled.

default: 1

--primary_port_http_enabled​

If true allows accessing http console on main TCP port.

default: true

--request_cache_limit​

Amount of memory to use for request cache in bytes per IO thread.

default: 64.00MiB

--tcp_nodelay​

Configures dragonfly connections with socket option TCP_NODELAY.

default: true

--conn_io_thread_start​

Starting thread id for handling server connections.

default: 0

--conn_io_threads​

Number of threads used for handing server connections.

default: 0

--conn_use_incoming_cpu​

If true uses incoming cpu of a socket in order to distribute incoming connections.

default: false

--container_iteration_yield_interval_usec​

Yield the fiber every N microseconds during container iteration. 0 disables yielding.

default: 500

--tcp_keepalive​

The period in seconds of inactivity after which keep-alives are triggerred, the duration until an inactive connection is terminated is twice the specified time.

default: 300

--timeout​

Close the connection after it is idle for N seconds (0 to disable).

default: 0

--tls​

Enable tls.

default: false

--tls_ca_cert_dir​

Certified authority signed certificates directory. Use c_rehash on the directory before specifying this flag.

default: ""

--tls_ca_cert_file​

Certified authority signed certificate to validate tls connections.

default: ""

--tls_cert_file​

Cert file(public key) for tls connections.

default: ""

--tls_key_file​

Private Key file for tls connections.

default: ""

--restricted_commands​

Commands restricted to connections on the admin port.

default:

--lock_on_hashtags​

When true, locks are done in the {hashtag} level instead of key level. Only use this with --cluster_mode=emulated|yes.

default: false

--enable_heartbeat_eviction​

Enable eviction during heartbeat when memory is under pressure.

default: true

--max_eviction_per_heartbeat​

The maximum number of key-value pairs that will be deleted in each eviction when heartbeat based eviction is triggered under memory pressure.

default: 100

--max_segment_to_consider​

The maximum number of dashtable segments to scan in each eviction when heartbeat based eviction is triggered under memory pressure.

default: 4

--force_epoll​

If true - uses linux epoll engine underneath. Can fit for kernels older than 5.10.

default: false

--pidfile​

If not empty - server writes its pid into the file).

default: ""

--unixsocket​

If not empty - specifies path for the Unix socket that will be used for listening for incoming connections.

default: ""

--unixsocketperm​

Set permissions for unixsocket, in octal value.

default: ""

--version_check​

If true, Will monitor for new releases on Dragonfly servers once a day.

default: true

--hz​

Base frequency at which the server performs other background tasks. Warning: not advised to decrease in production.

default: 100

--masterauth​

Password for authentication with master.

default: ""

--replicaof​

Specifies a host and port which point to a target master to replicate. Format should be <IPv4>:<PORT> or host:<PORT> or [<IPv6>]:<PORT>.

default:

--mem_defrag_page_utilization_threshold​

Memory page under utilization threshold. Ratio between used and committed size, below this, memory in this page will defragmented.

default: 0.8

--mem_defrag_threshold​

Minimum percentage of used memory relative to maxmemory cap before running defragmentation.

default: 0.7

--mem_defrag_waste_threshold​

The ratio of wasted/committed memory above which we run defragmentation.

default: 0.2

--mem_defrag_check_sec_interval​

Number of seconds between every defragmentation necessity check.

default: 60

--mem_defrag_max_burst_duration_us​

Maximum real time, measured in microseconds, that DefragTask() may run at the proactor's highest on-idle priority before forcing a cooldown. A value of 0 disables the duty-cycle cap.

default: 0

--mem_defrag_backoff_duration_us​

Duration in microseconds to drop to low on-idle priority after mem_defrag_max_burst_duration_us is reached, before resuming at high priority.

default: 0

--shard_round_robin_prefix​

Deprecated and will be removed.

When non-empty, keys which start with this prefix are not distributed across shards based on their value but instead via round-robin. Use cautiously! This can efficiently support up to a few hundreds of prefixes. Note: the prefix is inspected inside hashtags when cluster mode is enabled.

default: ""

--tiered_prefix​

Enables tiered storage if set. The string denotes the path and prefix of the files associated with tiered storage. For example, tiered_prefix=/path/to/file-prefix.

default: ""

--tiered_offload_threshold​

Ratio of free memory (free/max memory) below which offloading starts

default: 0.5

--tiered_upload_threshold​

Ratio of free memory (free/max memory) below which uploading stops.

default: 0.1

--tiered_min_value_size​

Minimum size of values eligible for offloading. Must be at least 64

default: 64

--tiered_min_ttl_to_offload_ms​

Minimum remaining TTL in milliseconds for a value to be eligible for offloading.

default: 5000

--tiered_max_pending_stash_bytes​

Maximum bytes in-flight to disk before rejecting new stashes or applying client backpressure. Allows batching writes to saturate disk I/O even with few clients.

default: 256.0KiB

--tiered_offload_scan_budget_us​

Base CPU time slice, in microseconds, granted to a single background offloading scan. A value of 0 disables offloading scans.

default: 100

--tiered_defrag_scan_budget_us​

Base CPU time slice, in microseconds, granted to a single background defragmentation scan. It scales up to three times this value with the amount of fragmentation found. A value of 0 disables defragmentation scans.

default: 2

--tiered_max_pending_defrags​

Maximum number of concurrent defragmentation read operations.

default: 50

--keys_output_limit​

Maximum number of keys output by keys command.

default: 8192

--backing_file_direct​

If true uses O_DIRECT to open backing files.

default: true

--tiering_disk_storage_initial_size​

Initial disk storage size.

default: 256.00MiB

--list_compress_depth​

Compress depth of the list. Default is no compression.

default: 0

--list_compress_dict_threshold​

Minimum list malloc usage in bytes before attempting ZSTD dictionary compression. A value of 0 disables it. Compression is synchronous and may block the thread.

default: 0

--list_compress_level​

Compression level for QList ZSTD dictionaries. A value of -1 uses ZSTD's default tuning.

default: -1

--list_max_listpack_size​

Maximum listpack size, default is 8kb.

default: -2

--listpack_max_bytes​

Maximum total bytes of a hash in listpack encoding before converting to a hash table.

default: 1024

--listpack_max_field_len​

Maximum length of a hash field or value to be stored in listpack encoding.

default: 64

--admin_nopass​

If set, would enable open admin access to console on the assigned port, without authorization needed.

default: false

--memcached_port​

Memcached port:

default: 0

--multi_eval_squash_buffer​

Max buffer for squashed commands per script:

default: 8096

--multi_exec_squash​

Whether multi exec will squash single shard commands to optimize performance.

default: true

--num_shards​

Number of database shards, 0 - to choose automatically.

default: 0

--tls_replication​

Enable TLS on replication.

default: false

--compression_level​

The compression level to use on zstd/lz4 compression.

default: 2

--compression_mode​

Select the snapshot compression mode.

default: 3

Where:

  • 0 — NONE — no compression.
  • 1 — SINGLE_ENTRY — single-entry LZF compression.
  • 2 — MULTI_ENTRY_ZSTD — multi-entry ZSTD compression for Dragonfly snapshots and single-entry compression for RDB snapshots.
  • 3 — MULTI_ENTRY_LZ4 — multi-entry LZ4 compression for Dragonfly snapshots and single-entry compression for RDB snapshots.

--master_connect_timeout_ms​

Timeout for establishing connection to a replication master.

default: 20000

--master_reconnect_timeout_ms​

Timeout for re-establishing connection to a replication master.

default: 1000

--replica_partial_sync​

Use partial sync to reconnect when a replica connection is interrupted.

default: true

--replication_acks_interval​

Interval between acks in milliseconds.

default: 1000

--default_lua_flags​

Configure default flags for running Lua scripts:

  • Use allow-undeclared-keys to allow accessing undeclared keys.
  • Use disable-atomicity to allow running scripts non-atomically.
  • Use legacy-float to return floats as integers.

Specify multiple values separated by space, for example allow-undeclared-keys disable-atomicity runs scripts non-atomically and allows accessing undeclared keys.

default: ""

--lua_auto_async​

If enabled, call/pcall with discarded values are automatically replaced with acall/apcall.

default: false

--df_snapshot_format​

If true, save in dragonfly-specific snapshotting format

default: true

--epoll_file_threads​

Thread size for file workers when running in epoll mode, default is hardware concurrent threads.

default: 0

--maxclients​

Maximum number of concurrent clients allowed.

default: 64000

--s3_ec2_metadata​

Whether to load credentials and configuration from EC2 metadata.

default: false

--s3_endpoint​

Endpoint for s3 snapshots, default uses aws regional endpoint.

default: ""

--s3_sign_payload​

Whether to sign the s3 request payload when uploading snapshots.

default: true

--s3_use_https​

Whether to use https for s3 endpoints.

default: true

--s3_use_helio_client​

If true, use helio's native S3 client; if false, use aws-sdk-cpp.

default: true

--slowlog_log_slower_than​

Add commands slower than this threshold to slow log. The value is expressed in microseconds and if it's negative disables the slowlog.

default: 10000

--slowlog_max_len​

Slow log maximum length.

default: 20

--interpreter_per_thread​

Lua interpreters per thread

default: 10

--serialization_max_chunk_size​

Maximum size of a value that may be serialized at once during snapshotting or full sync. Values bigger than this threshold will be serialized using streaming serialization. 0 - to disable streaming mode.

default: 65536

--serialization_tagged_chunks​

Allow serializer output to be split into tagged chunks and reassembled by receiver.

default: true

--aclfile​

Path and name to aclfile.

default: ""

--acllog_max_len​

Specify the number of log entries. Logs are kept locally for each thread and therefore the total number of entries are acllog_max_len * threads

default: 32

--jwt_validate​

Master switch for JWT-mode authentication. When true, every AUTH credential is sent to --jwt_validate_url instead of being checked against the local ACL password store. The URL must be configured at startup. This flag can be changed at runtime with CONFIG SET.

default: false

--jwt_validate_timeout_ms​

Deadline in milliseconds for the complete JWT validation HTTP call. A validator that does not respond before the deadline causes AUTH to be rejected.

default: 300

--jwt_validate_url​

HTTP endpoint used to validate AUTH credentials in JWT mode. The endpoint must return JSON containing valid, username, and optionally exp. The named ACL user must already be provisioned. Dragonfly does not cache validation results, and the endpoint uses plain HTTP without TLS.

default: ""

--cluster_announce_ip​

Ip that cluster commands announce to the client.

default: ""

--cluster_coordinator_connect_timeout_ms​

Timeout in milliseconds for coordinator to connect to remote shards.

default: 3000

--cluster_coordinator_response_timeout_ms​

Timeout in milliseconds for coordinator to read responses from remote shards.

default: 3000

--shard_repl_backlog_len​

Legacy maximum number of entries retained by each shard's replication backlog. A nonzero value disables time- and byte-based eviction unless either newer backlog limit flag is explicitly configured.

default: 0

--shard_repl_backlog_max_bytes​

Maximum bytes retained by each shard's replication backlog. A value of 0 uses maxmemory / shard count / 200.

default: 0B

--shard_repl_backlog_time_ms​

Target retention age, in milliseconds, for entries in each shard's replication backlog. Older entries are evicted on later journal writes. A value of 0 disables time-based eviction.

default: 5000

--proactor_affinity_mode​

Can be on, off or auto.

default: "on"

--proactor_threads​

Number of io threads in the pool. If zero is specified, it will use as many as there are CPU cores.

default: 0

--proactor_busy_poll_usec​

How long a proactor busy-polls for completions before falling back to a blocking wait. Raising this value trades CPU usage for wakeup latency.

default: 20

--proactor_irq_cpus​

Explicit list of CPU IDs or ranges that handle IRQs, such as 1,4,6,7 or 0-47. These CPUs are pinned to the highest proactor thread indices. This does not affect the pool size and is ignored if it names an offline CPU or more CPUs than the pool has threads.

default: ""

--flagfile​

Comma-separated list of files to load flags from.

default:

--alsologtostderr​

Log messages go to stderr in addition to logfiles.

default: false

--log_dir​

If specified, logfiles are written into this directory instead of the default logging directory.

default: ""

--logbuflevel​

Buffer log messages logged at this level or below. (-1 means don't buffer; 0 means buffer INFO only).

default: 0

--logbufsecs​

Buffer log messages for at most this many seconds.

default: 30

--logtostderr​

Log messages go to stderr instead of logfiles.

default: false

--max_log_size​

Approx. maximum log file size (in MB). A value of 0 will be silently overridden to 1.

default: 1800

--minloglevel​

Messages logged at a lower level than this don't actually get logged anywhere.

default: 0

--stderrthreshold​

Log messages at or above this level are copied to stderr in addition to logfiles. This flag obsoletes --alsologtostderr

default: 2

--command_alias​

Add an alias for given commands, format is: <alias>=<original>, <alias>=<original>. Aliases must be set identically on replicas, if applicable.

default:

--allocation_tracker​

Logs stack trace of memory allocation within these ranges. Format is min:max,min:max,....

default: ""

--always_flush_pipeline​

If true, will flush pipeline response after each pipeline squashing.

default: false

--announce_port​

Port that Dragonfly announces to cluster clients and replication master.

default: 0

--async_dispatch_quota​

Maximum number of consecutive async dispatch messages to process before either yielding to I/O when the pipeline appears empty or forcibly processing a queued pipelined command to prevent starvation. Set to 0 to disable this mechanism.

default: 100

--availability_zone​

Server availability zone, used by clients to read from local-zone replicas.

default: ""

--background_debug_jobs​

Use background fibers for debug jobs.

default: false

--background_heartbeat​

Whether to run heartbeat as a background fiber.

default: false

--background_snapshotting​

Whether to run snapshot as a background fiber.

default: false

--break_replication_on_master_restart​

When in replica mode, and master restarts, break replication from master to avoid flushing the replica's data.

default: false

--cluster_flush_decommit_memory​

Decommit memory after flushing slots.

default: false

--cluster_node_id​

ID within a cluster, used for slot assignment. MUST be unique. If empty, uses master replication ID (random string).

default: ""

--cluster_search​

Enable search commands for cross-shard search. Turned off by default for safety.

default: false

--deserialize_hnsw_index​

Deserialize HNSW vector index graph structure.

default: false

--disable_json_defragmentation​

If true, disable json object defragmentation.

default: false

--enable_heartbeat_rss_eviction​

Enable eviction during heartbeat when rss memory is under pressure. Eviction based on used_memory will still be enabled.

default: true

--enable_memcache_io_loop_v2​

Enable the event-driven IoLoopV2 for non-TLS Memcache connections.

default: true

--enable_pipeline_squashing_v2​

Enable vectorized pipeline squashing for the V2 dispatch loop. It groups consecutive single-shard pipeline commands by shard and executes them in parallel.

default: true

--enable_resp_io_loop_v2​

Enable the event-driven IoLoopV2 for non-TLS RESP connections.

default: false

--enable_shared_read_buffer​

Use one read buffer per proactor for eligible RESP IoLoopV2 connections.

default: false

--shared_read_buffer_len​

Capacity of the per-proactor RESP IoLoopV2 shared read buffer.

default: 32.0KiB

--enable_tcp_defer_accept​

Enable TCP_DEFER_ACCEPT option on server sockets.

default: true

--eviction_memory_budget_threshold​

Eviction starts when the free memory (including RSS memory) drops below eviction_memory_budget_threshold * max_memory_limit.

default: 0.1

--experimental_cascaded_partial_sync​

Enable experimental cascaded partial synchronization.

default: false

--experimental_cluster_shard_by_slot​

If true, cluster mode is enabled and sharding is done by slot. Otherwise, sharding is done by hash tag.

default: false

--experimental_flat_json​

If true, uses flat json implementation.

default: false

--expose_http_api​

If set, will expose a POST /api handler for sending redis commands as json array.

default: false

--fiber_run_warning_threshold_ms​

If greater than 0, will warn if a fiber runs longer than this threshold (in milliseconds).

default: 0

--fiber_safety_margin​

If > 0, ensures the stack each fiber has at least this margin. The check is done at fiber destruction time.

default: 1024

--gcs_auth_token​

Authentication token for Google Cloud Storage access.

default: ""

--gcs_dry_upload​

If true, performs a dry run upload to Google Cloud Storage without actually writing data.

default: false

--get_zero_copy​

If true, GET returns a borrowed view into the CompactObj raw payload for large raw strings. If false, it uses the materializing path. This flag can be used to compare the zero-copy and materializing GET paths.

default: true

--info_replication_valkey_compatible​

When true, output valkey compatible values for info-replication.

default: true

--journal_omit_redundant_writes​

If true, omit journal writes for keys during full sync that are yet to be reached by the serialization loop. Reduces full sync overhead.

default: true

--jsonpathv2​

If true, uses Dragonfly jsonpath implementation, otherwise uses legacy jsoncons implementation.

default: true

--keep_legacy_memory_metrics​

When true, keeps the legacy metrics format for memory-related info fields.

default: false

--latency_tracking​

If true, track latency for commands.

default: false

--list_tiering_threshold​

Tiering threshold for lists. Default - no tiering.

default: 0

--list_tiering_prefetch_depth​

Before loading a tiered list node, scan up to this many neighboring nodes and issue asynchronous load requests for any that are offloaded. A value of 0 disables prefetching.

default: 0

--locktag_delimiter​

If set, this char is used to extract a lock tag by looking at delimiters, like hash tags. If unset, regular hashtag extraction is done (with {}). Must be used with --lock_on_hashtags.

default: ""

--locktag_prefix​

Only keys with this prefix participate in tag extraction.

default: ""

--locktag_skip_n_end_delimiters​

How many closing tag delimiters should we skip when extracting lock tags. 0 for no skipping. For example, when delimiter is : and this flag is 2, the locktag for :a:b:c:d:e will be a:b:c.

default: 0

--log_squash_info_threshold_usec​

Threshold in microseconds above which to log squashing timings.

default: 2147483648

--lua_allow_undeclared_auto_correct​

If enabled, when a script that is not allowed to run with undeclared keys is trying to access undeclared keys, automatically set the script flag to be able to run with undeclared key.

default: false

--lua_enable_redis_log​

Enable redis.log to write logs from lua script.

default: false

--lua_float_as_int_shas​

Comma-separated list of Lua script SHAs which should return floats as integers. SHAs are only looked at when loading the script.

default:

--lua_mem_gc_threshold​

Specifies Lua interpreter's per thread memory limit in bytes after which the GC will be called forcefully. 0 value removes forced GC calls.

default: 10000000

--lua_resp2_legacy_float​

Return rounded down integers instead of floats for lua scripts with RESP2.

default: false

--lua_undeclared_keys_shas​

Comma-separated list of Lua script SHAs which are allowed to access undeclared keys. SHAs are only looked at when loading the script, and new values do not affect already-loaded scripts.

default:

--luagc​

Specifies Lua garbage collector preferences. By default uses default lua GC parameters. Format should be inc/200/100/13 or gen/20/100 where inc and gen are types of GC, numbers are parameters. For more information check https://www.lua.org/manual/5.4/manual.html#2.5

default:

--managed_service_info​

Hides some implementation details from users when true (i.e. in managed service env).

default: false

--masteruser​

Username for authentication with master.

default: ""

--max_bulk_len​

Maximum bulk length that is allowed to be accepted when parsing RESP protocol.

default: 2147483648

--max_busy_read_usec​

Maximum time we read and parse from a socket without yielding, in microseconds.

default: 200

--max_busy_squash_usec​

Maximum time in microseconds to execute squashed commands before yielding.

default: 1000

--max_rdb_save_serialize_buffer_capacity​

Maximum buffer capacity that the RDB serializer retains after a flush.

default: 4194304

--max_squashed_cmd_num​

Max number of commands squashed in a single shard during squash optimization.

default: 100

--mget_dedup_keys​

If true, MGET will deduplicate keys.

default: false

--migration_buckets_cpu_budget​

How much CPU budget to use for migration buckets serialization.

default: 0.2

--migration_buckets_serialization_threshold​

The number of buckets to serialize on each iteration before yielding.

default: 10

--migration_buckets_sleep_usec​

Sleep time in microseconds after each time we reach migration_buckets_serialization_threshold.

default: 500

--migration_finalization_timeout_ms​

Timeout for migration finalization operation.

default: 30000

--notify_keyspace_events​

notify-keyspace-events. Only Ex is supported for now.

default: ""

--omit_basic_usage​

Omit printing basic usage info.

default: false

--pause_wait_timeout​

Timeout in seconds, to set up the pause for all connections for CLIENT PAUSE command and cluster slot migration finalization procedure.

default: 1

--pipeline_buffer_limit​

Amount of memory to use for storing pipeline requests per IO thread. Please note that clients that send excessively huge pipelines may deadlock themselves. See https://github.com/dragonflydb/dragonfly/discussions/3997 for details.

default: 128.00MiB

--pipeline_parse_in_proactor​

V2 only: parse newly arrived bytes from the proactor OnRecv callback while the fiber is parked waiting for parallel work, so the next batch has already grown when execution resumes.

default: true

--pipeline_prioritize_large_batches​

V2 only: in ParseLoop, defer executing a parsed pipeline batch and return to the read loop to accumulate more already-available input, so the squasher sees one large batch instead of many small ones.

default: true

--pipeline_queue_limit​

Pipeline queue max length. The server will stop reading from the client socket once its pipeline queue crosses this limit, and will resume once it processes excessive requests. This is to prevent OOM states. Users of huge pipeline sizes may require increasing this limit to prevent the risk of deadlocking. See https://github.com/dragonflydb/dragonfly/discussions/3997 for details.

default: 10000

--pipeline_squash_limit​

Limit on the size of a squashed pipeline.

default: 1073741824

--pipeline_wait_batch_usec​

If non-zero, waits for this time for more I/O events to come for the connection in case there is only one command in the pipeline.

default: 0

--rdb_ignore_expiry​

Ignore Key Expiry when loading from RDB snapshot.

default: false

--rdb_load_dry_run​

Dry run RDB load without applying changes.

default: false

--rdb_sbf_chunked​

Enable new save format for saving SBFs in chunks.

default: true

--registered_buffer_size​

Size of registered buffer for IoUring fixed read/writes.

default: 524288

--replica_announce_ip​

IP address that Dragonfly announces to replication master.

default: ""

--replica_delete_expired​

If true, replicas proactively delete expired keys on the read path.

default: true

--replica_priority​

Published by info command for sentinel to pick replica based on score during a failover.

default: 100

--replicaof_no_one_start_journal​

When set, preserves journal offsets after REPLICAOF NO ONE.

default: true

--replication_dispatch_threshold​

Number of bytes to aggregate before replication.

default: 1500

--replication_stream_output_limit​

Limit on the replication output buffer size (in bytes). The streamer throttles when this limit is reached.

default: 1048576

--replication_timeout​

Time in milliseconds to wait for the replication writes being stuck.

default: 30000

--rss_oom_deny_ratio​

When the ratio between maxmemory and RSS memory exceeds this value, commands marked as DENYOOM will fail with OOM error and new connections to non-admin port will be rejected. Negative value disables this feature.

default: 1.25

--save_schedule​

Deprecated. Please use --snapshot_cron instead.

default: ""

--scheduler_background_budget​

Background fiber budget in nanoseconds.

default: 50000

--scheduler_background_sleep_prob​

Sleep probability of background fibers on reaching budget.

default: 50

--scheduler_background_warrant​

Percentage of guaranteed cpu time for background fibers.

default: 5

--search_query_string_bytes​

Maximum number of bytes in search query string.

default: 10240

--search_reject_legacy_field​

FT.AGGREGATE: Reject legacy field names.

default: true

--send_timeout​

Close the connection after it is stuck on send for N seconds (0 to disable).

default: 0

--serialize_hnsw_index​

Serialize HNSW vector index graph structure.

default: false

--shard_thread_busy_polling_usec​

If non-zero, overrides the busy polling parameter for shard threads.

default: 0

--slot_migration_connection_timeout_ms​

Connection creating timeout for migration operations.

default: 2000

--slot_migration_throttle_us​

Incoming migration throttle time in microseconds, throttled every 100us of migration commands processing. 0 to disable. Recommended value is 20. Values more than 50 can significantly reduce migration speed.

default: 0

--squashed_reply_size_limit​

Max bytes allowed for squashing_current_reply_size. If this limit is reached, connections dispatching pipelines won't squash them.

default: 0

--subset_knn_search_threshold​

If prefilter results are below this threshold, exact subset search will be used instead of HNSW graph search.

default: 8192

--table_growth_margin​

Prevents table from growing if number of free slots x average object size x this ratio is larger than memory budget.

default: 0.15

--tcp_backlog​

TCP listen(2) backlog parameter.

default: 256

--tcp_user_timeout​

The maximum period in milliseconds that transmitted data may stay unacknowledged before TCP aborts the connection. 0 means OS default timeout.

default: 0

--tiered_experimental_cooling​

If true, uses intermediate cooling layer when offloading values to storage.

default: true

--tiered_experimental_hash_support​

Experimental hash datatype offloading.

default: false

--tiered_experimental_list_support​

Experimental list node offloading.

default: false

--tiered_max_file_size​

Limit on maximum file size that is used by the database for tiered storage. 0 means the program will automatically determine its maximum file size.

default: 0B

--tls_cipher_suites​

TLS ciphers configuration for TLS 1.3.

default: ""

--tls_ciphers​

TLS ciphers configuration for TLS 1.2.

default: "DEFAULT:!MEDIUM"

--tls_prefer_server_ciphers​

If true, prefer server ciphers over client ciphers.

default: false

--tls_session_cache_size​

Size of the cache for tls sessions.

default: 20480

--tls_session_cache_timeout​

Timeout for each session/ticket.

default: 300

--tls_session_caching​

If true, enables session caching and tickets.

default: false

--tx_queue_warning_len​

Length threshold for warning about long transaction queue.

default: 96

--disable_scope_based_mem_track​

Turn off scope-based memory metrics.

default: true

--unlink_experimental_async​

If true, runs unlink command asynchronously.

default: true

--uring_direct_table_len​

If positive, create direct fd table of this length.

default: 0

--uring_disable_iowait​

If true, disables iowait cpu accounting.

default: false

--uring_recv_buffer_cnt​

How many buffer ring entries to allocate per thread for io_uring receive operations. Relevant only for modern kernels with io_uring enabled.

default: 0

--use_numeric_range_tree​

Use range tree for numeric index. If false, use a simple implementation with btree_set. Range tree is more memory efficient and faster for range queries, but slower for single value queries.

default: true

--use_oah_set​

If true, store SET values in OAHSet instead of StringSet.

default: false

--user​

If not empty - drop privileges to this user (and their primary group) after binding ports. Accepts username or numeric uid. If --dir is set, chowns the data directory to this user.

default: ""

--write_connection_throttling_sleep_usec​

Sleep period for a write connection in microseconds. A value of 0 disables yielding.

default: 0