Cluster Configuration Properties
Cluster configuration properties are the same for all nodes in a cluster. They can be set at the cluster level.
For information on how to edit cluster properties, see Configuring Cluster Properties. Some properties require the cluster to be restarted in order to be applied; see a specific property’s reference for whether restart is required.
Cloud storage properties
cloud_storage_bucket
AWS or GCS bucket that should be used to store data.
Default: null
Restart required: yes
cloud_storage_cache_size
Max size of archival cache.
Units: bytes
Default: 21474836480 (20 GiB)
Restart required: yes
cloud_storage_credentials_source
The source of credentials used to connect to cloud services.
Default: config_file
Valid values: config_file, aws_instance_metadata, sts, gcp_instance_metadata
Restart required: yes
cloud_storage_disable_tls
Disable TLS for all S3 or GCS connections.
Type: boolean
Default: false
Restart required: yes
cloud_storage_enabled
Enable archival storage. Must be enabled true in order to use Tiered Storage or Remote Read Replicas.
Type: boolean
Default: false
Restart required: yes
cloud_storage_max_connections
Max number of simultaneous connections to S3 per shard. Includes connections used for both uploads and downloads.
Units: number of simultaneous connections
Default: 20
Restart required: yes
Cluster management properties
Kafka API properties
enable_sasl
Enable SASL authentication for Kafka connections.
Type: boolean
Default: false
Restart required: no
fetch_max_bytes
Maximum number of bytes returned in a fetch request.
Units: bytes
Default: 57671680 (55 MiB)
Restart required: no
group_max_session_timeout_ms
The maximum allowed session timeout for registered consumers. Longer timeouts give consumers more time to process messages in between heartbeats at the cost of a longer time to detect failures.
Units: milliseconds
Default: 300000 (300 sec)
Restart required: no
group_min_session_timeout_ms
The minimum allowed session timeout for registered consumers. Shorter timeouts result in quicker failure detection at the cost of more frequent consumer heartbeating which can overwhelm broker resources.
Units: milliseconds
Default: 6000 (6 sec)
Restart required: no
kafka_connection_rate_limit
Maximum connections per second for one core. If null (the default), the number of connections per second is unlimited.
Units: number of connections per second, per core
Default: null
Range: [1, …]
Restart required: yes
kafka_connection_rate_limit_overrides
Overrides the maximum connections per second for one core for the specified IP addresses (for example, ['127.0.0.1:90', '50.20.1.1:40'])
Type: string
Default: null
Restart required: no
kafka_connections_max
Maximum number of Kafka client connections per broker. If null, the property is disabled.
Units: number of Kafka client connections per broker
Default: null
Restart required: no
kafka_connections_max_overrides
A list of IP addresses for which Kafka client connection limits are overridden and don’t apply. For example, (['127.0.0.1:90', '50.20.1.1:40']).
Default: {} (empty list)
Restart required: no
kafka_connections_max_per_ip
Maximum number of Kafka client connections per IP address, per broker. If null, the property is disabled.
Units: number of Kafka client connections per IP address, per broker
Default: null
Restart required: no
kafka_group_recovery_timeout_ms
Kafka group recovery timeout.
Units: milliseconds
Default: 30000 (30 sec)
Restart required: no
kafka_qdc_enable
Enable Kafka queue depth control.
Type: boolean
Default: false
Restart required: yes
kafka_qdc_max_latency_ms
Maximum latency threshold for Kafka queue depth control depth tracking.
Units: milliseconds
Default: 80
Restart required: yes
rm_sync_timeout_ms
Resource manager’s synchronization timeout. Maximum time for this node to wait for internal state machine to catch up with all events written by previous leaders before rejecting a request.
Units: milliseconds
Default: 10000 (10 sec)
Restart required: yes
rpc_server_listen_backlog
Maximum TCP connection queue length for Kafka server and internal RPC server. If null (the default value), no queue length is set.
Units: number of queue entries
Default: null
Range: [1, …]
Restart required: yes
rpc_server_tcp_recv_buf
Internal RPC TCP receive buffer size. If null (the default value), no buffer size is set by Redpanda.
Units: bytes
Default: null
Range: [32 KiB, …], aligned to 4096 bytes
Restart required: yes
Metrics properties
aggregate_metrics
Enable aggregation of metrics returned by the /metrics endpoint. Metric aggregation is performed by summing the values of samples by labels and is done when it makes sense by the shard and/or partition labels.
Type: boolean
Default: false
Restart required: yes
disable_metrics
Disable registering metrics exposed on the internal metrics endpoint.
Type: boolean
Default: false
Restart required: yes
disable_public_metrics
Disable registering metrics exposed on the public metrics endpoint.
Type: boolean
Default: false
Restart required: yes
enable_metrics_reporter
Enable the cluster metrics reporter. If true, the metrics reporter collects and exports to Redpanda Data a set of customer usage metrics at the interval set by metrics_reporter_report_interval.
|
The cluster metrics of the metrics reporter are different from monitoring metrics.
|
Type: boolean
Default: true
Restart required: no
Raft properties
raft_learner_recovery_rate
Raft learner recovery rate limit. Throttles the rate of data communicated to nodes (learners) that need to catch up to leaders.
This rate limit is placed on a node sending data to a recovering node. Each sending node will be limited to this rate. The recovering node will accept data as fast as possible according to the combined limits of all healthy nodes in the cluster. For example, if two nodes are sending data to the recovering node, and raft_learner_recovery_rate is 100 MB/sec, then the recovering node will recover at a rate of 200 MB/sec.
Units: bytes per second
Default: 104857600 (100 MB/sec)
Restart required: no
Storage properties
delete_retention_ms
Delete segments that are older than this age.
Units: milliseconds
Default: 604800000 (1 week)
Restart required: no
log_cleanup_policy
Bit flags setting the default topic cleanup policy: compact and/or delete.
Default: delete
Valid Values: compact, delete, compact,delete, none
Restart required: no
log_compaction_interval_ms
How often to trigger background compaction.
Units: milliseconds
Default: 10000 (10 sec)
Restart required: no
log_compression_type
Default topic compression type (gzip, snappy, lz4, zstd, producer, or none).
Default: producer
Valid values: gzip, snappy, lz4, zstd, producer, none
Restart required: no
log_message_timestamp_type
Default timestamp type for topic messages (CreateTime or LogAppendTime).
Default: CreateTime
Valid values: CreateTime, LogAppendTime
Restart required: no
storage_strict_data_init
Requires that an empty file named .redpanda_data_dir be present in the data directory. If set true, Redpanda will refuse to start if the file is not found in the data directory.
Default: false
Restart required: no
storage_ignore_timestamps_in_future_sec
The maximum number of seconds that a record’s timestamp can be ahead of a Redpanda broker’s clock and still be used when deciding whether to clean up the record for data retention. This property makes possible the timely cleanup of records from clients with clocks that are drastically unsynchronized relative to Redpanda.
When determining whether to clean up a record with timestamp more than storage_ignore_timestamps_in_future_sec seconds ahead of the broker, Redpanda ignores the record’s timestamp and instead uses a valid timestamp of another record in the same segment, or (if another record’s valid timestamp is unavailable) the timestamp of when the segment file was last modified (mtime).
By default, storage_ignore_timestamps_in_future_sec is disabled (null).
|
To figure out whether to set
|
Units: seconds
Default: null
Restart required: no
Supported versions: Redpanda v22.3 or later
Topic and partition properties
auto_create_topics_enabled
Allow automatic topic creation.
Type: boolean
Default: false
Restart required: no
default_topic_partitions
Default number of partitions per topic.
Units: number of partitions per topic
Default: 1
Restart required: no
default_topic_replications
Default replication factor for new topics.
Units: number of replicas per topic
Default: 1
Range: [1, …], must be odd
Restart required: no
internal_topic_replication_factor
Target replication factor for internal topics.
Units: number of replicas per topic
Default: 3
Restart required: no
Transaction properties
enable_transactions
Enable transactions (atomic writes).
Type: boolean
Default: true
Restart required: yes
seq_table_min_size
The minimum threshold number of sessions to keep in the seq table. Not affected by compaction.
Default: 1000
Restart required: yes
tm_sync_timeout_ms
Transaction manager’s synchronization timeout. Maximum time to wait for internal state machine to catch up before rejecting a request.
Units: milliseconds
Default: 10000 (10 sec)
Restart required: yes
tm_violation_recovery_policy
Describes how to recover from an invariant violation at the transaction coordinator level.
Default: crash
Restart required: yes
transaction_coordinator_cleanup_policy
Cleanup policy for a transaction coordinator topic.
Default: delete
Valid Values: compact, delete, compact,delete, none
Restart required: no
transaction_coordinator_delete_retention_ms
Delete segments older than this age. To ensure transaction state is retained as long as the longest-running transaction, make sure this is no less than transactional_id_expiration_ms.
Units: milliseconds
Default: 604800000 (1 week)
Restart required: no
Suggested reading
-
Fast distributed transactions with Redpanda article