Tunable Configuration Properties

Tunable properties are cluster properties that are specifically designed for tuning a cluster.

For information on how to edit tunable properties, see Configuring Cluster Properties.


Admin properties

enable_pid_file

Enables creation of a process ID lock file. Recommended to be set true.

Type: boolean

Default: true

Restart required: yes


features_auto_enable

Flag to auto-enable new feature flags after upgrades.

Type: boolean

Default: true

Restart required: no


Object storage properties

cloud_storage_backend

Optional cloud storage backend variant used to select API capabilities. If unknown, will be inferred from other configuration parameters.

Default: unknown

Valid values:

  • unknown

  • aws

  • google_s3_compat

  • azure

  • minio

Restart required: yes

Supported versions: Redpanda v23.1 or later


cloud_storage_cache_check_interval

Minimum time between trims of Tiered Storage cache. If a fetch operation requires trimming the cache, and the time interval between the fetch and the most recent trim is less than this property, then trimming will be delayed until this property’s interval has elapsed.

Units: milliseconds

Default: 5000 (5 seconds)

Restart required: yes


cloud_storage_cache_chunk_size

Size of chunks of segments downloaded into object (cloud) storage cache. Reduces space usage by only downloading the necessary chunk from a segment.

Units: bytes

Default: 16777216 (16 MiB)

Restart required: yes

Related topics:


cloud_storage_chunk_eviction_strategy

Strategy for evicting unused cache chunks. Available eviction strategies:

  • eager: evicts all chunks which are not in use by any data source.

  • capped: evicts chunks up to a preconfigured amount.

  • predictive: uses statistics from readers to determine which chunks to evict.

Type: eviction strategy

Default: eager

Valid values:

  • eager

  • capped

  • predictive

Restart required: no

Related topics:


cloud_storage_credentials_host

Hostname to connect to for retrieving role-based credentials. If null, hostname is derived from cloud_storage_credentials_source. Only required when using IAM role based access.

Type: string

Default: null

Restart required: yes

Supported versions: Redpanda v23.1 or later


cloud_storage_credentials_source

The source of credentials to connect to cloud services. If cloud_storage_credentials_host is null, this is used to derive the hostname to connect to retrieve role-based credentials.

Valid values:

  • If deploying on an EC2 VM: aws_instance_metadata

  • If deploying on EKS: sts (AWS service token service)

  • If deploying on a GCE VM: gcp_instance_metadata

  • If deploying on GKE: gcp_instance_metadata

  • If using fixed credentials from a configuration file: config_file

Default: config_file

Valid values:

  • config_file

  • aws_instance_metadata

  • sts

  • gcp_instance_metadata

Restart required: yes


cloud_storage_disable_chunk_reads

Flag for turning off chunk-based reads and enabling full-segment downloads.

Type: boolean

Default: false

Restart required: no

Related topics:


cloud_storage_enable_compacted_topic_reupload

Enable re-upload of data for compacted topics.

Type: boolean

Default: true

Restart required: no


cloud_storage_enable_remote_read

Default remote read configuration value for new topics.

Type: boolean

Default: false

Restart required: no


cloud_storage_enable_remote_write

Default remote write configuration value for new topics.

Type: boolean

Default: false

Restart required: no


cloud_storage_enable_segment_merging

Flag to enable adjacent segment merging in cloud object storage. When this property is enabled, Redpanda merges adjacent segments in object storage that are smaller than the threshold, cloud_storage_segment_size_min.

Type: boolean

Default: true

Restart required: no

Related properties:

Supported versions: Redpanda v23.1 or later

Related topics:


cloud_storage_graceful_transfer_timeout_ms

Maximum duration to wait for uploads to complete before a leadership transfer. If null, leadership transfers proceed with no delay.

Units: milliseconds

Default: 5000 (5 sec)

Restart required: no

Supported versions: Redpanda v23.1 or later


cloud_storage_housekeeping_interval_ms

Interval for object storage housekeeping tasks.

Units: milliseconds

Default: 300000 (5 minutes)

Restart required: no

Related topics:


cloud_storage_hydrated_chunks_per_segment_ratio

The desired ratio of hydrated to non-hydrated chunks for each segment, where a current ratio above this value results in unused chunks being trimmed (evicted).

Units: ratio of hydrated to non-hydrated chunks for a segment

Default: 0.7

Restart required: no

Related topics:


cloud_storage_idle_threshold_rps

Request rate threshold for idle state detection of cloud object storage. If the average object storage API request rate is lower than this threshold, then the object storage is considered to be idle.

Units: object storage API requests per second

Default: 1.0

Restart required: no

Related topics:

Supported versions: Redpanda v23.1 or later


cloud_storage_idle_timeout_ms

Timeout used to detect idle state of the object storage API. If no object storage API requests are made for at least this timeout’s duration, the object storage is considered idle.

Units: milliseconds

Default: 10000 (10 sec)

Restart required: no

Related topics:

Supported versions: Redpanda v23.1 or later


cloud_storage_initial_backoff_ms

Initial backoff time for exponential backoff algorithm.

Units: milliseconds

Default: 100

Restart required: yes


cloud_storage_manifest_upload_timeout_ms

Manifest upload timeout.

Units: milliseconds

Default: 10000 (10 seconds)

Restart required: yes

Supported versions: Redpanda v23.1 or later


cloud_storage_max_connection_idle_time_ms

Maximum HTTPS connection idle time.

Units: milliseconds

Default: 5000

Restart required: yes


cloud_storage_max_materialized_segments_per_shard

Maximum number of concurrent readers of remote data per shard (CPU core). If null, the value of topic_partitions_per_shard multiplied by two is used.

Type: unsigned integer

Units: number of concurrent readers

Default: null

Restart required: no


cloud_storage_max_readers_per_shard

Maximum concurrent readers of remote data per shard (CPU core). If null, the value of topic_partitions_per_shard is used, with one reader per partition if the shard is at its maximum partition capacity.

Type: unsigned integer

Units: number of concurrent readers

Default: null

Restart required: no


cloud_storage_max_segments_pending_deletion_per_partition

The maximum limit per partition for the number of segments pending deletion from object storage. Segments can be deleted due to retention or compaction. If this limit is breached and deletion fails, then segments are orphaned in object storage and must be removed manually.

Type: unsigned integer

Units: number of segments

Default: 5000

Restart required: no

Supported versions: Redpanda v23.1 or later


cloud_storage_metadata_sync_timeout_ms

Timeout for shadow indexing (SI) metadata synchronization.

Units: milliseconds

Default: 10000 (10 seconds)

Restart required: no


cloud_storage_min_chunks_per_segment_threshold

The minimum number of chunks per segment for trimming to be enabled. If the number of chunks in a segment is below this threshold, the segment is small enough that all chunks in it can be hydrated at any given time.

Units: number of chunks per segment

Default: 5

Restart required: no

Related topics:


cloud_storage_readreplica_manifest_sync_timeout_ms

Timeout to check if new data is available for a partition in S3 for read replica.

Units: milliseconds

Default: 30000 (30 seconds)

Restart required: no


cloud_storage_reconciliation_interval_ms

 Deprecated red

Interval at which the archival service runs reconciliation.

Units: milliseconds

Default: 1000

Restart required: yes


cloud_storage_recovery_temporary_retention_bytes_default

The number of bytes of size-based retention for topics that are created during automated recovery of topics from object storage.

Units: bytes

Default: 1073741824 (1 GiB)

Restart required: no

Supported versions: Redpanda v23.1 or later


cloud_storage_roles_operation_timeout_ms

Timeout for IAM role related operations. While connecting to object storage, failed tasks that refresh the IAM roles token are retried after this timeout elapses.

Units: milliseconds

Default: 30000 (30 sec)

Restart required: yes

Supported versions: Redpanda v23.1 or later


cloud_storage_segment_max_upload_interval_sec

Duration that a segment can be kept in local storage without uploading it to object storage. If null, defaults to infinite duration.

Units: seconds

Default: null (infinite duration)

Restart required: yes


cloud_storage_segment_size_target

The desired segment size in object storage. If null, the property is disabled, and the segment size in object storage defaults to segment.bytes.

Units: bytes

Default: null

Restart required: no

Related properties:

Related topics:

Supported versions: Redpanda v23.1 or later


cloud_storage_segment_size_min

The minimum segment size in object storage. If null, the property is disabled, and the minimum segment size defaults to cloud_storage_segment_size_target/2.

Units: bytes

Default: null

Restart required: no

Related properties:

Related topics:

Supported versions: Redpanda v23.1 or later


cloud_storage_segment_upload_timeout_ms

Log segment upload timeout.

Units: milliseconds

Default: 30000 (30 seconds)

Restart required: no


cloud_storage_upload_ctrl_d_coeff

Derivative coefficient of the upload PID controller.

Type: double

Default: 0.0

Restart required: yes


cloud_storage_upload_ctrl_max_shares

Maximum number of IO and CPU shares that the archival upload can use.

Units: number of IO and CPU shares

Default: 1000

Restart required: yes


cloud_storage_upload_ctrl_min_shares

Minimum number of IO and CPU shares that the archival upload can use.

Units: number of IO and CPU shares

Default: 100

Restart required: yes


cloud_storage_upload_ctrl_p_coeff

Proportional coefficient of the upload PID controller.

Type: double

Default: -2.0

Restart required: yes


cloud_storage_upload_ctrl_update_interval_ms

Sampling interval of the upload PID controller.

Units: milliseconds

Default: 60000 (1 minute)

Restart required: yes


cloud_storage_upload_loop_initial_backoff_ms

Initial backoff interval when there is nothing to upload for a partition.

Units: milliseconds

Default: 100

Restart required: yes


cloud_storage_upload_loop_max_backoff_ms

Maximum backoff interval when there is nothing to upload for a partition.

Units: milliseconds

Default: 10000 (10 seconds)

Restart required: yes


Cluster management properties

controller_backend_housekeeping_interval_ms

Interval between iterations of the controller backend’s housekeeping loop.

Units: milliseconds

Default: 1000

Restart required: yes


controller_log_accummulation_rps_capacity_acls_and_users_operations

Maximum capacity of accumulated requests for the cluster controller’s access control list (ACL) operations and user operations. If null, no maximum capacity is applied.

Type: unsigned integer

Units: number of requests

Default: null

Restart required: no


controller_log_accummulation_rps_capacity_configuration_operations

Maximum capacity of accumulated requests for the cluster controller’s configuration operations. If null, no maximum capacity is applied.

Type: unsigned integer

Units: number of requests

Default: null

Restart required: no


controller_log_accummulation_rps_capacity_move_operations

Maximum capacity of accumulated requests for the cluster controller’s move operations. If null, no maximum capacity is applied.

Type: unsigned integer

Units: number of requests

Default: null

Restart required: no

Related properties: rps_limit_move_operations


controller_log_accummulation_rps_capacity_node_management_operations

Maximum capacity of accumulated requests for the cluster controller’s node management operations. If null, no maximum capacity is applied.

Type: unsigned integer

Units: number of requests

Default: null

Restart required: no


controller_log_accummulation_rps_capacity_topic_operations

Maximum capacity of accumulated requests for the cluster controller’s topic management operations. If null, no maximum capacity is applied.

Type: unsigned integer

Units: number of requests

Default: null

Restart required: no


health_manager_tick_interval

Period at which the health manager runs.

Units: milliseconds

Default: 180000 (3 minutes)

Restart required: yes


health_monitor_max_metadata_age

Maximum duration that metadata is cached in the health monitor of a non-controller node.

Units: milliseconds

Default: 10000 (10 seconds)

Restart required: no


health_monitor_tick_interval

Period at which the health monitor refreshes cluster state.

Units: milliseconds

Default: 10000 (10 seconds)

Restart required: no


id_allocator_batch_size

Size of one batch message (one log record) of an ID allocator log.

Type: integer

Units: bytes

Default: 1000

Restart required: yes


id_allocator_log_capacity

Maximum number of messages (batches) of an ID allocator log. Once the log reaches this limit, it is compacted.

Type: integer

Units: number of log messages (batches)

Default: 100

Restart required: yes


join_retry_timeout_ms

Duration between cluster join retries.

Units: milliseconds

Default: 5000 (5 seconds)

Restart required: yes


leader_balancer_idle_timeout

Timeout at which to run the leader balancer when in an idle state.

Units: milliseconds

Default: 120000 (2 minutes)

Restart required: no


leader_balancer_mute_timeout

Timeout used to mute groups, where muted groups are removed from consideration as leadership rebalancing targets.

Units: milliseconds

Default: 300000 (5 minutes)

Restart required: no


leader_balancer_node_mute_timeout

Timeout used to mute a node, where a muted node is removed from consideration as leadership rebalancing target.

Units: milliseconds

Default: 20000 (20 seconds)

Restart required: no


leader_balancer_transfer_limit_per_shard

Maximum limit per shard (CPU core) on the number of in-progress leadership transfers.

Type: unsigned integer

Units: number of in-progress leadership transfers per shard

Default: 512

Range: [1, 2048]

Restart required: no


max_concurrent_producer_ids

Maximum number of producer IDs that the resource manager state machine caches in internal state.

Type: 64-bit unsigned integer

Units: number of IDs

Default: maximum value of a 64-bit unsigned integer

Range: [1, maximum value of a 64-bit unsigned integer]

Restart required: no

Supported versions: Redpanda v23.1 or later


members_backend_retry_ms

Interval at which brokers check and retry partition reallocation requests.

When brokers (members) of a Redpanda cluster make a request to reallocate partition replicas but they don’t succeed immediately or are in progress, the Redpanda cluster management controller (backend) checks on the status of reallocations at an interval set by members_backend_retry_ms until the current cluster state is reconciled with the requested partition allocation state.

Units: milliseconds

Default: 5000

Restart required: yes


metadata_dissemination_interval_ms

Interval for metadata dissemination batching.

Units: milliseconds

Default: 3000

Restart required: yes


metadata_dissemination_retries

Number of attempts looking up a topic’s metadata, like shard, before failing a request.

Units: number of retries

Default: 30

Restart required: yes


metadata_dissemination_retry_delay_ms

Delay before retrying a topic-lookup in a shard or other meta-tables.

Units: milliseconds

Default: 500

Restart required: yes


metadata_status_wait_timeout_ms

Maximum duration to wait in metadata request for cluster health to be refreshed.

Units: milliseconds

Default: 2000

Restart required: yes


node_isolation_heartbeat_timeout

If the duration since the last received response to a node’s heartbeat request is greater than or equal to this timeout, then the node is considered to be isolated. Redpanda prevents isolated nodes from handling Kafka API requests.

Units: milliseconds

Default: 3000

Range: [100, 10000]

Restart required: no

Supported versions: Redpanda v23.1 or later


node_management_operation_timeout_ms

Timeout for executing node management operations.

Units: milliseconds

Default: 5000

Restart required: yes


node_status_interval

Time interval between two node status messages, which establish liveness status outside of the Raft protocol.

Units: milliseconds

Default: 100

Restart required: no


partition_autobalancing_concurrent_moves

Number of partitions that can be reassigned at once by the partition auto-balancer.

Type: unsigned integer

Units: number of partitions

Default: 50

Restart required: no


partition_autobalancing_tick_interval_ms

Tick interval of the partition auto-balancer.

Units: milliseconds

Default: 30000 (30 seconds)

Restart required: no


partition_autobalancing_movement_batch_size_bytes

 Deprecated red

Total size of partitions that the partition auto-balancer moves in one batch.

Type: unsigned integer

Units: bytes

Default: 5368709120 (5 GiB)

Restart required: no


rps_limit_acls_and_users_operations

Request-per-second (rps) rate limit for the cluster controller’s access control list (ACL) operations and user operations.

Type: unsigned integer

Units: requests per second

Default: 1000

Restart required: no


rps_limit_configuration_operations

Request-per-second (rps) rate limit for the cluster controller’s configuration operations.

Type: unsigned integer

Units: requests per second

Default: 1000

Restart required: no


rps_limit_move_operations

Request-per-second (rps) rate limit for the cluster controller’s move operations.

Type: unsigned integer

Units: requests per second

Default: 1000

Restart required: no


rps_limit_node_management_operations

Request-per-second (rps) rate limit for the cluster controller’s node management operations.

Type: unsigned integer

Units: requests per second

Default: 1000

Restart required: no


rps_limit_topic_operations

Request-per-second (rps) rate limit for the cluster controller’s topic operations.

Type: unsigned integer

Units: requests per second

Default: 1000

Restart required: no


Kafka API properties

default_num_windows

Default number of quota tracking windows.

Units: number of windows

Default: 10

Restart required: no


default_window_sec

Default duration of quota tracking window.

Units: milliseconds

Default: 1000

Restart required: no

Related properties:


fetch_reads_debounce_timeout

Duration to wait for the next read in fetch request when minimum number of requested bytes wasn’t read.

Units: milliseconds

Default: 1

Restart required: no


fetch_session_eviction_timeout_ms

Minimum time before an unused session gets evicted. The maximum time after which an inactive session is deleted is twice the given configuration value cache.

Units: milliseconds

Default: 60000 (1 minute)

Restart required: yes


group_initial_rebalance_delay

A delay at the start of a consumer group rebalancing event to wait for new members.

Units: milliseconds

Default: 3000

Range: [0, …​]

Restart required: no


group_new_member_join_timeout

Timeout to wait for new members to join a consumer group.

Units: milliseconds

Default: 30000

Range: [0, …​]

Restart required: no


group_offset_retention_check_ms

Period at which Redpanda checks for expired consumer group offsets.

Units: milliseconds

Default: 600000 (10 min)

Restart required: no

Related properties:

Supported versions: Redpanda v23.1 or later


group_offset_retention_sec

The retention duration of consumer group offsets. Redpanda identifies offsets that are expired, based on this retention duration, and removes them to reclaim storage. For a consumer group, the retention timeout starts from when the group becomes empty by losing all its consumers; for a standalone consumer, the retention timeout starts from the time of the last commit. Once elapsed, an offset is considered to be expired and is discarded.

Redpanda periodically checks for expired offsets at the rate set by group_offset_retention_check_ms.

If null, the automatic offset retention feature is disabled.

Units: seconds

Default: 604800 (7 days)

Restart required: no

Related properties:

Supported versions: Redpanda v23.1 or later


group_topic_partitions

Number of partitions in the internal group membership topic.

Type: integer

Units: number of partitions

Default: 16

Restart required: no


kafka_batch_max_bytes

Maximum size of a batch processed by the server. If the batch is compressed, this limit applies to the compressed batch size.

The topic property max.message.bytes overrides kafka_batch_max_bytes.

Type: unsigned integer

Units: bytes

Default: 1048576 (1 MiB)

Restart required: no


kafka_max_bytes_per_fetch

Maximum amount of data a consumer receives when fetching a record.

Type: integer

Units: bytes

Default: 67108864 (64 MiB)

Restart required: no


kafka_qdc_depth_alpha

Smoothing factor for queue depth control depth tracking.

Type: double

Default: 0.8

Restart required: yes


kafka_qdc_depth_update_ms

The update frequency of the queue depth control algorithm.

Units: milliseconds

Default: 7000 (7 seconds)

Restart required: yes


kafka_qdc_idle_depth

Queue depth of the queue depth control algorithm when in an idle state.

Units: number of queue entries

Default: 10

Restart required: no


kafka_qdc_latency_alpha

Smoothing parameter for queue depth control algorithm.

Type: double

Default: 0.002

Restart required: yes


kafka_qdc_max_depth

Maximum queue depth of the queue depth control algorithm.

Units: number of queue entries

Default: 100

Restart required: yes


kafka_qdc_min_depth

Minimum queue depth of the queue depth control algorithm.

Units: number of queue entries

Default: 1

Restart required: yes


kafka_qdc_window_count

Number of time windows used by the queue depth control algorithm.

Units: number of time windows

Default: 12

Restart required: yes


kafka_qdc_window_size_ms

Duration of a time window used by the queue depth control algorithm.

Units: milliseconds

Default: 1500

Restart required: yes


kafka_request_max_bytes

Maximum size of a request processed by a Kafka API.

Type: unsigned integer

Units: bytes

Default: 104857600 (100 MiB)

Restart required: no


kafka_rpc_server_stream_recv_buf

Maximum size of the user-space receive buffer. If null, this limit is not applied.

Type: unsigned integer

Units: bytes

Default: null

Range: [512, 512 KiB]. Must be 4096-byte aligned.

Restart required: yes


legacy_group_offset_retention_enabled

With group offset retention enabled by default starting in Redpanda version 23.1, this flag enables group offset retention for deployments of Redpanda upgraded from earlier versions.

For Redpanda versions 23.1 and later, this flag has no effect.

For Redpanda versions earlier than 23.1:

  • In upgraded clusters, to enable group offset retention, this flag must be true and group_offset_retention_check_ms must be non-null.

  • When legacy group offset retention is enabled:

    • Only offsets that were committed or updated following the upgrade to 23.1 or later will be automatically collected. Offsets that were committed prior to upgrading to 23.1 or later will never be automatically deleted, and instead need to be manually removed using the Offset Delete API.

    • The age of an offset is calculated based on when it was committed rather than when this flag was enabled.

Type: boolean

Default: false

Restart required: no

Related properties:

Supported versions: Redpanda v23.1 or later


max_kafka_throttle_delay_ms

The maximum delay inserted in the data path of Kafka API requests to throttle them down. Configuring this to be less than the Kafka client timeout can ensure that the delay that’s inserted won’t be long enough to cause a client timeout by itself.

Units: milliseconds

Default: 60000 (1 minute)

Restart required: no

Related topics:


quota_manager_gc_sec

Period of the quota manager’s GC timer.

Units: milliseconds

Default: 30000 (30 seconds)

Restart required: yes


wait_for_leader_timeout_ms

Timeout to wait for leader election.

Units: milliseconds

Default: 5000

Restart required: no


Metrics properties

metrics_reporter_report_interval

Interval of reports sent to clients from the metrics reporter.

Units: milliseconds

Default: 86400000 (24 hours)

Restart required: no


metrics_reporter_tick_interval

Tick timer interval for capturing metrics.

Units: milliseconds

Default: 60000 (1 minute)

Restart required: no


Raft properties

election_timeout_ms

Raft election timeout.

Units: milliseconds

Default: 1500

Restart required: yes


full_raft_configuration_recovery_pattern

Recover raft configurations on start for NTPs that match this pattern.

Type: string, or list of strings

Default: {} (empty)

Restart required: yes


raft_heartbeat_disconnect_failures

Number of consecutive failed heartbeat responses before forcibly closing an unresponsive TCP connection. Disabled when set to 0.

Type: unsigned integer

Units: number of consecutive failed heartbeat responses

Default: 3

Range: [0, …​]

Restart required: yes


raft_heartbeat_interval_ms

Period of raft leader heartbeat.

Units: milliseconds

Default: 150

Restart required: yes


raft_heartbeat_timeout_ms

Timeout duration of raft leader heartbeat RPC.

Units: milliseconds

Default: 3000

Restart required: yes


raft_io_timeout_ms

Timeout duration for raft I/O.

Units: milliseconds

Default: 10000 (10 seconds)

Restart required: yes


raft_max_concurrent_append_requests_per_follower

Maximum number of concurrent requests per follower.

Type: unsigned integer

Units: number of concurrent requests per follower

Default: 16

Restart required: yes


raft_max_recovery_memory

Maximum amount of memory available for raft follower recovery. If null, defaults to 10% of total memory.

Type: unsigned integer

Units: bytes

Default: null

Range: [32 MiB, …​]

Restart required: no


raft_recovery_default_read_size

Default size of a read issued during raft follower recovery.

Type: unsigned integer

Units: bytes

Default: 524288 (512 KiB)

Range: [128, 5 MiB]

Restart required: no


raft_replicate_batch_window_size

Maximum size of a batch for raft replication.

Type: unsigned integer

Units: bytes

Default: 1048576 (1 MiB)

Restart required: yes


raft_smp_max_non_local_requests

Maximum number of non-local requests that can execute concurrently on a CPU core (seastar 'shard'). If null, defaults to the value determined by the following formula:

topic_partitions_per_shard * (266) * (<number of CPU cores> - 1)

Type: unsigned integer

Units: number of requests

Default: null

Restart required: yes


raft_timeout_now_timeout_ms

Timeout of the raft timeout now RPC request.

Units: milliseconds

Default: 1000

Restart required: no


raft_transfer_leader_recovery_timeout_ms

Timeout waiting for follower recovery when transferring leadership.

Units: milliseconds

Default: 10000 (10 seconds)

Restart required: no


recovery_append_timeout_ms

Timeout for raft’s append entries RPC.

Units: milliseconds

Default: 5000

Restart required: yes


Schema properties

kafka_schema_id_validation_cache_capacity

Cache capacity per shard for validating schema IDs.

Units: number of cache entries

Default: 128

Restart required: no


Storage properties

abort_index_segment_size

Capacity of an abort index segment.

Type: unsigned integer

Units: number of transactions

Default: 50000

Restart required: yes


append_chunk_size

Size of direct-write operations to disk.

Type: unsigned integer

Units: bytes

Default: 16384 (16 KiB)

Range: [4096, 32 MiB]. Must be 4096-byte aligned.

Restart required: yes


compacted_log_segment_size

Size of a compacted log segment.

Type: unsigned 64-bit integer

Units: bytes

Default: 268435456 (256 MiB)

Range: [1 MiB, …​]

Restart required: no


compaction_ctrl_backlog_size

Size of the compaction backlog of the backlog controller. If null (empty), defaults to 10% of disk capacity.

Type: integer

Units: bytes

Default: null

Range: [1 MiB, …​]

Restart required: yes


compaction_ctrl_d_coeff

Derivative coefficient for compaction PID controller.

Type: double

Default: 0.2

Restart required: yes


compaction_ctrl_i_coeff

Integral coefficient for compaction PID controller.

Type: double

Default: 0.0

Range: [0.0, …​]

Restart required: yes


compaction_ctrl_max_shares

Maximum number of I/O and CPU shares that the compaction process can use.

Units: short

Units: number of shares

Default: 1000

Restart required: yes


compaction_ctrl_min_shares

Minimum number of I/O and CPU shares that the compaction process can use.

Type: short

Units: number of shares

Default: 10

Restart required: yes


compaction_ctrl_p_coeff

Proportional coefficient for compaction PID controller. Must be negative to decrease the compaction backlog when compaction shares increase.

Type: double

Default: -12.5

Range: […​, 0.0)

Restart required: yes


compaction_ctrl_update_interval_ms

Sampling interval of the compaction PID controller.

Units: milliseconds

Default: 30000 (30 seconds)

Restart required: yes


disable_batch_cache

Flag to disable the batch cache in log manager.

Type: boolean

Default: false

Restart required: yes


kvstore_flush_interval

Flush interval of the key-value store.

Units: milliseconds

Default: 10

Restart required: no


kvstore_max_segment_size

Maximum size of a segment of the key-value store.

Type: unsigned integer

Units: bytes

Default: 16777216 (16 MiB)

Restart required: yes


log_segment_ms_min

The lower bound for the topic property segment.ms and the cluster property log_segment_ms, where lower values are clamped to this.

Units: milliseconds

Default: 60000 (60 sec)

Range: [60000 (60 sec), …​]

Restart required: no

Related properties:

Supported versions: Redpanda v23.1 or later


log_segment_ms_max

The upper bound for the topic property segment.ms and the cluster property log_segment_ms, where higher values are clamped to this.

Units: milliseconds

Default: 31536000000 (365 days)

Range: [60000 (60 sec), …​]

Restart required: no

Related properties:

Supported versions: Redpanda v23.1 or later


log_segment_size

Default size of log segment. Per topic, it is superseded by topic property segment.bytes.

Type: unsigned 64-bit integer

Units: bytes

Default: 134217728 (128 MiB)

Range: [1 MiB, …​]

Restart required: no


log_segment_size_max

Upper limit of topic segment.bytes. Higher values are clamped to this limit. If null, no limit is applied.

Type: unsigned 64-bit integer

Units: bytes

Default: null (no limit applied)

Restart required: no


log_segment_size_min

Lower limit of topic segment.bytes. Lower values are clamped to this limit. If null, no limit is applied.

Type: unsigned 64-bit integer

Units: bytes

Default: 1048576 (1 MiB)

Restart required: no


log_segment_size_jitter_percent

Configures the amount of size variation added to a segment used for each partition: a percentage value in the range [0, log_segment_size_jitter_percent] is randomly chosen and that percentage of the original segment size limit is added.

Applying segment size jitter is a good defensive measure to space out the overhead created from segment rolls over time.

Type: unsigned short

Units: percent of segment size that

Default: 5

Range: [0, 99]

Restart required: yes


max_compacted_log_segment_size

Max compacted segment size after consolidation.

Type: unsigned integer

Units: bytes

Default: 5368709120 (5 GiB)

Restart required: no


memory_abort_on_alloc_failure

Flag to enable termination of the Redpanda process when a memory allocation fails. If false, a failed memory allocation throws an exception.

Type: boolean

Default: true

Restart required: no

Supported versions: Redpanda v23.1 or later


readers_cache_eviction_timeout_ms

Duration after which inactive readers are evicted from the cache of readers.

Units: milliseconds

Default: 30000 (30 seconds)

Restart required: yes


reclaim_batch_cache_min_free

Minimum amount of free memory maintained by the batch cache.

Type: unsigned integer

Units: bytes

Default: 67108864 (64 MiB)

Restart required: yes


reclaim_growth_window

Starting from the last point in time when memory was reclaimed from the batch cache, this is the duration during which the amount of memory to reclaim grows at a significant rate, based on heuristics about the amount of available memory.

Units: milliseconds

Default: 3000

Restart required: yes


reclaim_max_size

Maximum batch cache reclaim size.

Units: bytes

Default: 4194304 (4 MiB)

Restart required: yes


reclaim_min_size

Minimum batch cache reclaim size.

Units: bytes

Default: 131072 (128 KiB)

Restart required: yes


reclaim_stable_window

If the duration since the last time memory was reclaimed has been longer than this property, the memory usage of the batch cache is considered stable, so only the minimum size (reclaim_min_size) is set to be reclaimed.

Units: milliseconds

Default: 10000 (10 seconds)

Restart required: yes


release_cache_on_segment_roll

Flag for whether to release cache when a full segment is rolled.

Type: boolean

Default: false

Restart required: no


replicate_append_timeout_ms

Timeout for append entry’s RPCs issued while replicating entries.

Units: milliseconds

Default: 3000

Restart required: yes


segment_appender_flush_timeout_ms

Maximum duration until buffered data is written.

Units: milliseconds

Default: 1000

Restart required: no


segment_fallocation_step

Size of a file allocation (f allocation) step of a segment.

Type: unsigned integer

Units: bytes

Default: 33554432 (32 MiB)

Range: [4096, 1 GiB]. Must be 4096-byte aligned.

Restart required: no


storage_compaction_index_memory

Maximum number of bytes that may be used on each shard (CPU core) by compaction index writers.

Type: unsigned integer

Units: bytes

Default: 134217728 (128 MiB)

Range: [16 MiB, 100 GiB]

Restart required: no


storage_max_concurrent_replay

Maximum number of partition logs that are replayed concurrently on startup or flushed concurrently on shutdown.

Type: unsigned integer

Units number of concurrently replayed logs

Default: 1024

Range: [128, …​]

Restart required: no


storage_min_free_bytes

Threshold of minimum amount of free space before rejecting producers.

Type: unsigned integer

Units: bytes

Default: 5368709120 (5 GiB)

Range: [10 MiB, …​]

Restart required: no


storage_read_buffer_size

Size of a read buffer (one per in-flight read, one per log segment).

Type: unsigned integer

Units: bytes

Default: 131072 (128 KiB)

Restart required: yes


storage_read_readahead_count

Number of read-ahead reads when reading from a segment.

Type: short

Units: number of reads

Default: 10

Restart required: yes


storage_space_alert_free_threshold_bytes

The threshold of free storage (in bytes of free storage) below which a low storage-space alert is logged.

Type: unsigned integer

Units: bytes of free storage

Default: 0

Range: [0, …​]

Restart required: no


storage_space_alert_free_threshold_percent

The threshold of free storage (in percentage of total storage that is free) below which a low storage-space alert is logged.

Type: unsigned integer

Units: percentage of total storage that is free

Default: 5

Range: [0, 50]

Restart required: no


storage_target_replay_bytes

Target bytes to replay from disk on startup after a clean shutdown. Controls the frequency of snapshots and checkpoints.

Type: unsigned integer

Units: bytes

Default: 10737418240 (10 GiB)

Range: [128 MiB, 1 TiB]

Restart required: no


tx_log_stats_interval_s

Period for which to log transmit (tx) statistics per partition. Requires the tx logger’s log-level to be at least debug.

Units: seconds

Default: 10

Restart required: no

Supported versions: Redpanda v23.1 or later


zstd_decompress_workspace_bytes

Size of the Zstandard (zstd) compression algorithm’s memory buffer (workspace) in which a decompressed object is emplaced.

Type: unsigned integer

Units: bytes

Default: 8388608 (8 MiB)

Restart required: yes


Topic and partition properties

alter_topic_cfg_timeout_ms

Time to wait for entries replication in controller log when executing alter topic configuration request.

Units: milliseconds

Default: 5000

Restart required: no


create_topic_timeout_ms

Timeout to wait for new topic creation.

Units: milliseconds

Default: 2000

Restart required: no


topic_fds_per_partition

The number of file descriptors reserved for topics per partition.

Type: integer

Units: number of file descriptors

Default: 5

Range: [1, 1000]

Restart required: no


topic_memory_per_partition

The amount of memory reserved for topics per partition.

Type: unsigned integer

Units: bytes

Default: 1048576 (1 MiB)

Range: [1, 100 MiB]

Restart required: no


topic_partitions_per_shard

Maximum number of partitions which may be allocated per shard (CPU core).

Type: unsigned integer

Units: number of partitions

Default: 1000

Range: [16, 131072]

Restart required: no


topic_partitions_reserve_shard0

Reserved partition slots on shard (CPU core) 0 on each node. If this is >= topic_partitions_per_shard, no data partitions are scheduled on shard 0.

Type: unsigned integer

Units: number of partitions

Default: 2

Range: [0, 131072]

Restart required: no


Transaction properties

abort_timed_out_transactions_interval_ms

Timer timeout to check for and abort inactive transactions.

Units: milliseconds

Default: 10000 (10 seconds)

Restart required: yes


transaction_coordinator_log_segment_size

Size of the log segment of the Kafka broker’s transaction coordinator.

Type: unsigned integer

Units: bytes

Default: 1073741824 (1 GiB)

Restart required: no