Skip to main content
Version: 22.3

Tunable Configuration Properties

Tunable properties are cluster properties that are specifically designed for tuning a cluster. For information on how to edit tunable properties, refer to Configuring Cluster Properties.

The following table lists the available tunable properties:

PropertyDescription
abort_index_segment_sizeCapacity (in number of txns) of an abort index segment.

Default: 50000
abort_timed_out_transactions_interval_msHow often to look for the inactive transactions and abort them.

Default: 60000
alter_topic_cfg_timeout_msTime to wait for entries replication in controller log when executing alter configuration request.

Default: 5000
append_chunk_sizeSize of direct write operations to disk.

Default: 16384 (16KB)
cloud_storage_cache_check_intervalTimeout to check if cache eviction should be triggered.

Default: 30000
cloud_storage_enable_remote_readEnable remote read for all topics.

Default: false
cloud_storage_enable_remote_writeEnable remote write for all topics.

Default: false
cloud_storage_initial_backoff_msInitial backoff time for exponential backoff algorithm (ms).

Default: 100
cloud_storage_manifest_upload_timeout_msManifest upload timeout (ms).

Default: 10000
cloud_storage_max_connection_idle_time_msMax https connection idle time (ms).

Default: 5000
cloud_storage_reconciliation_interval_msInterval at which the archival service runs reconciliation (ms).

Default: 10000
cloud_storage_segment_max_upload_interval_secTime that segment can be kept locally without uploading it to the remote storage (sec).

Default: Null
cloud_storage_segment_upload_timeout_msLog segment upload timeout (ms).

Default: 30000
cloud_storage_upload_ctrl_d_coeffDerivative coefficient for upload PID controller.

Default: 0.0
cloud_storage_upload_ctrl_max_sharesMaximum number of I/O and CPU shares that archival upload can use.

Default: 1000
cloud_storage_upload_ctrl_min_sharesMinimum number of I/O and CPU shares that archival upload can use.

Default: 100
cloud_storage_upload_ctrl_p_coeffProportional coefficient for upload PID controller.

Default: -2.0
cloud_storage_upload_ctrl_update_interval_msNo description.

Default: 60000
compacted_log_segment_sizeHow large in bytes each compacted log segment should be (default 256 MB).

Default: 268435456 (256 MB)
Minimum = 1048576 (1 MB)
compaction_ctrl_backlog_sizeTarget backlog size for compaction controller. If not set, compaction target that the compaction backlog would be equal to.

Default: Null
compaction_ctrl_d_coeffDerivative coefficient for compaction PID controller.

Default: 0.2
compaction_ctrl_i_coeffIntegral coefficient for compaction PID controller.

Default: 0.0
compaction_ctrl_max_sharesMaximum number of IO and CPU shares that compaction process can use.

Default: 1000
compaction_ctrl_min_sharesMinimum number of IO and CPU shares that compaction process can use.

Default: 10
compaction_ctrl_p_coeffProportional coefficient for compaction PID controller. This has to be negative since compaction backlog should decrease when number of compaction shares increases.

Default: -12.5
compaction_ctrl_update_interval_msNo description.

Default: 30000
controller_backend_housekeeping_interval_msInterval between iterations of controller backend housekeeping loop.

Default: 1000
coproc_max_batch_sizeMaximum amount of bytes to read from one topic read.

Default: 32768 (32 KB)
coproc_max_inflight_bytesMaximum amountt of inflight bytes when sending data to wasm engine.

Default: 10485760 (10 MB)
coproc_max_ingest_bytesMaximum amount of data to hold from input logs in memory.

Default: 655360 (640 KB)
coproc_offset_flush_interval_msInterval for which all coprocessor offsets are flushed to disk.

Default: 300000
create_topic_timeout_msTimeout (ms) to wait for new topic creation.

Default: 2000
default_num_windowsDefault number of quota tracking windows.

Default: 10
Minimum = 1
default_window_secDefault quota tracking window size in milliseconds.

Default: 1000
developer_modeSkips most of the checks performed at startup are not recommended for production use.

Default: false
disable_batch_cacheDisable batch cache in log manager.

Default: false
election_timeout_msElection timeout expressed in milliseconds.

Default: 1500
enable_pid_fileEnable pid file. You probably don't want to change this.

Default: true
features_auto_enableWhether new feature flags may auto-activate after upgrades (true) or must wait for manual activation via the admin API (false).

Default: true
fetch_reads_debounce_timeoutTime to wait for next read in fetch request when requested min bytes wasn't reached.

Default: 1
fetch_session_eviction_timeout_msMinimum time before which unused session will get evicted from sessions. Maximum time after which inactive session will be deleted is two time given configuration value cache.

Default: 60000
full_raft_configuration_recovery_patternRecover raft configuration on start for NTPs matching pattern.

Default: Null
group_initial_rebalance_delayExtra delay (ms) added to rebalance phase to wait for new members.

Default: 300
group_new_member_join_timeoutTimeout for new member joins.

Default: 30000
group_topic_partitionsNumber of partitions in the internal group membership topic.

Default: 1
health_manager_tick_intervalHow often the health manager runs.

Default: 180000
health_monitor_max_metadata_ageMax age of metadata cached in the health monitor of non controller node.

Default: 10000
health_monitor_tick_intervalHow often health monitor refreshes cluster state.

Default: 10000
id_allocator_batch_sizeId allocator allocates messages in batches (each batch is a one log record) and then serves requests from memory without touching the log until the batch is exhausted.

Default: 1000
id_allocator_log_capacityCapacity of the id_allocator log in number of messages. Once it reached id_allocator_stm should compact the log.

Default: 100
join_retry_timeout_msTime between cluster join retries in milliseconds.

Default: 5000
kafka_max_bytes_per_fetchLimit fetch responses to this many bytes even if total of partition bytes limits is higher.

Default: 67108864 (64 MB)
kafka_qdc_depth_alphaSmoothing factor for kafka queue depth control depth tracking.

Default: 0.8
kafka_qdc_depth_update_msUpdate frequency for kafka queue depth control.

Default: 7000
kafka_qdc_idle_depthQueue depth when idleness is detected in kafka queue depth control.

Default: 10
kafka_qdc_latency_alphaSmoothing parameter for kafka queue depth control latency tracking.

Default: 0.002
kafka_qdc_max_depthMaximum queue depth used in kafka queue depth control.

Default: 100
kafka_qdc_min_depthMinimum queue depth used in kafka queue depth control.

Default: 1
kafka_qdc_window_countNumber of windows used in kafka queue depth control latency tracking.

Default: 12
kafka_qdc_window_size_msWindow size for kafka queue depth control latency tracking.

Default: 1500
kvstore_flush_intervalKey-value store flush interval (ms).

Default: 10
kvstore_max_segment_sizeKey-value maximum segment size (bytes).

Default: 16777216 (16 MB)
leader_balancer_idle_timeoutLeadership rebalancing idle timeout.

Default: 120000
leader_balancer_mute_timeoutLeadership rebalancing mute timeout.

Default: 300000
leader_balancer_mute_timeoutLeadership rebalancing node mute timeout.

Default: 20000
log_segment_sizeHow large in bytes should each log segment be (default 1 GB).

Default: 1048576000 (1 GB)
Minimum = 1048576 (1 MB)
max_compacted_log_segment_sizeMax compacted segment size after consolidation.

Default: 5242880000 (5 GB)
max_kafka_throttle_delay_msFail-safe maximum throttle delay on kafka requests.

Default: 60000
members_backend_retry_msTime between members backend reconciliation loop retries.

Default: 5000
metadata_dissemination_interval_msInterval for metadata dissemination batching.

Default: 3000
metadata_dissemination_retriesNumber of attempts to look up a topic's metadata like shard before failing a request.

Default: 30
metadata_dissemination_retry_delay_msDelay before retrying a topic lookup in a shard or other meta tables.

Default: 0500
metadata_status_wait_timeout_msMaximum time to wait in metadata request for cluster health to be refreshed.

Default: 2000
metrics_reporter_report_intervalCluster metrics reporter report interval.

Default: 864000000
metrics_reporter_tick_intervalCluster metrics reporter tick interval.

Default: 60000
node_management_operation_timeout_msTimeout for executing node management operations.

Default: 5000
quota_manager_gc_secQuota manager GC frequency in milliseconds.

Default: 30000
raft_heartbeat_disconnect_failuresAfter how many failed heartbeats to forcibly close an unresponsive TCP connection. Set to 0 to disable force disconnection.

Default: 3
raft_heartbeat_interval_msThe interval in ms between raft leader heartbeats.

Default: 150
raft_heartbeat_timeout_msRaft heartbeat RPC timeout.

Default: 3000
Minimum = milliseconds(1)
raft_io_timeout_msRaft I/O timeout.

Default: 10000
raft_max_concurrent_append_requests_per_followerMaximum number of concurrent append entries requests sent by leader to one follower.

Default: 16
raft_max_recovery_memoryMaximum memory that can be used for reads in the raft recovery process.

Default: 15% of total memory
Minimum = 33554432 (32 MB)
raft_recovery_default_read_sizeDefault size of read issued during raft follower recovery.

Default: 524288 (512 KB)
Minimum = 128
Maximum = 5242880 (5 MB)
raft_replicate_batch_window_sizeMaximum size of requests cached for replication.

Default: 1048576 (1 MB)
raft_smp_max_non_local_requestsMaximum number of x-core requests pending in Raft seastar::smp group. (for more details look at seastar::smp_service_group documentation).

Default: The maximum number of x-core pending in Raft relies on the total number of cores that your environment is executing.
raft_timeout_now_timeout_msTimeout for a timeout now request.

Default: 1000
raft_transfer_leader_recovery_timeout_msTimeout waiting for follower recovery when transferring leadership.

Default: 10000
readers_cache_eviction_timeout_msDuration after which inactive readers will be evicted from cache.

Default: 30000
reclaim_batch_cache_min_freeFree memory limit that will be kept by batch cache background reclaimer.

Default: 67108864 (64 MB)
reclaim_growth_windowLength of time in which reclaim sizes grow.

Default: 3000
reclaim_max_sizeMaximum batch cache reclaim size.

Default: 4194304 (4 MB)
reclaim_min_sizeMinimum batch cache reclaim size.

Default: 131072 (128 KB)
reclaim_stable_windowLength of time above which growth is reset.

Default: 10000
recovery_append_timeout_msTimeout for append entries requests issued while updating stale follower.

Default: 5000
release_cache_on_segment_rollFree cache when segments roll.

Default: false
replicate_append_timeout_msTimeout for append entries requests issued while replicating entries.

Default: 3000
segment_appender_flush_timeout_msMaximum delay until buffered data is written.

Default: 1000
segment_fallocation_stepSize for segments fallocation.

Default: 33554432 (32 MB)
storage_read_buffer_sizeSize of each read buffer (one per in-flight read, per log segment).

Default: 131072 (128 KB)
storage_read_readahead_countHow many additional reads to issue ahead of current read location.

Default: 10
storage_space_alert_free_threshold_bytesThreshold of minimim bytes free space before setting storage space alert.

Default: 1048576000 (1 GB)
Minimum = 1048576000 ( 1 GB)
storage_space_alert_free_threshold_percentThreshold of minimim percent free space before setting storage space alert.

Default: 5
Minimum = 1
Maximum = 50
topic_fds_per_partitionRequired file handles per partition when creating topics.

Default: 10
Minimum = 1
Maximum = 1000
topic_memory_per_partitionRequired memory per partition when creating topics.

Default: 1048576 (1 MB)
Minimum = 1
Maximum = 104857600 (100 MB)
transaction_coordinator_log_segment_sizeHow large in bytes should each log segment be (default 1 GB).

Default: 1048576000 (1 GB)
wait_for_leader_timeout_msTimeout (ms) to wait for leadership in metadata cache.

Default: 5000
zstd_decompress_workspace_bytesSize of the zstd decompression workspace.

Default: 8388608 (8 MB)

Suggested reading

  • Benchmark of Redpanda and Apache Kafka article
  • Fast distributed transactions with Redpanda article