Fibril demo · settings

Settings

Runtime settings are persisted and, in a cluster, replicated to brokers. Each worker adopts changes at its next supported boundary. Startup-only values require a restart.

Persisted settings are applied live where supported. Broker runtime settings and Stroma-owned dead-letter defaults are saved separately, each with its own version check.

Broker Runtime Settings

Live broker behavior such as delivery timing and idle queue cleanup.

Loading settings authority…

Installation is observed on this node at the time of loading. Existing operations can retain an older snapshot. Replication stream buffer changes apply to the next stream. Controller checkpoint and eager failover settings, and drain handoff timing, are not tracked here. Reload to refresh adoption status.

Inputs show the saved document. A saved revision does not report adoption by every worker. Startup seeds below come from this node’s resolved configuration and initialize a missing settings document. They do not override saved values.

Delivery

Controls how long messages can stay leased and how aggressively delayed or expired work is polled.

Inflight TTL How long a delivered message can remain unacknowledged before it becomes eligible again. Expiry poll minimum Minimum wait between checks for delayed messages becoming ready. Expiry batch max Maximum delayed or expired offsets processed in one pass. Delivery poll max Maximum wait used by broker delivery polling when no message is immediately ready.

Connections

Controls how the broker treats short TCP disconnects from clients that can resume.

Reconnect grace Blank disables grace. Otherwise, a disconnected client can resume within this window before its subscriptions are cleaned up. Drain handoff timeout Maximum wait for partition ownership handoff during a coordinated drain. Blank uses 30 seconds. Expiry leaves recovery running. Restart session TTL Maximum age of persisted session metadata accepted when a client resumes after broker restart. Blank or zero disables restart resumption. Live disconnects use reconnect grace.

Replication

Controls replica-durable publish confirms and follower catch-up work.

Confirm timeout How long a replica-durable publish confirm can wait for enough durable follower progress. Caught-up poll Follower pull interval while already caught up with the owner.

Retry poll Follower retry interval after a partial pull or transient replication error. Checkpoint retry poll Follower retry interval while it needs a checkpoint before it can continue.

Messages per read Maximum message records a follower asks the owner for in one pull. Events per read Maximum event records a follower asks the owner for in one pull. Bytes per read Approximate byte budget for one owner replication response.

Iterations per tick Maximum pull/apply iterations before the follower yields. Minimum in-sync replicas Fail replica-durable publishes quickly when fewer replicas are recently in sync. Use 1 to disable the floor.

ISR timeout How recently a follower must report durable progress to count as in sync.

Owner read timeout slack Extra time allowed beyond a replication read’s long-poll window before reconnecting. Owner connect timeout Time allowed for TCP connection and authentication to the owner before retrying.

Streaming replication Use credit-based streaming on followers. When disabled, followers fall back to polling pulls.

Stream apply linger Microseconds a streaming follower spends gathering contiguous frames before one fsynced apply. Higher trades apply latency for fsync amortization. Zero drains only. Stream apply max merge bytes Byte cap on a single coalesced streaming apply (peak memory vs fsync amortization). Stream buffer batches In-flight batch buffer depth (credit window) for the streaming follower. Applied on the next stream.

Agreed checkpoint interval (ms) Opt-in background checkpoints on all admitted queue replicas. 0 disables new attempts. 1000–86400000 sets the periodic interval. Work thresholds can start an attempt earlier, at most once per second. Existing attempts finish. Snapshot and payload verification uses bounded background work.

Checkpoint event threshold Start earlier after this many events beyond the agreed cut. 0 disables this trigger. The checkpoint interval must be enabled. Checkpoint append-byte threshold Approximate local append content bytes since observing the last certificate. 0 disables this trigger. Resets on reopen. Periodic checkpoints remain the fallback.

Eager failover Opt in to faster reassignment after repeated explicit peer connection failures. Safe recovery still runs. Heartbeat expiry handles silent failures. Brief network interruptions can cause extra recovery work. Eager failover grace (ms) Require failed reconnects across this interval. Zero reacts after an immediate failed reconnect check. Changing the policy restarts pending suspicion; Raft timers stay unchanged.

Partitioning and Consumer Groups

Defaults for new partitioned queues and exclusive consumer-group health signals.

Default partition count Partition count used when a new queue is declared without an explicit count. Target partitions per consumer Blank disables the exclusive-consumer-group under-provisioned signal.

Idle Queue Cleanup

Lets unused queues leave memory after they have no active publishers or subscribers for long enough.

Enabled When disabled, idle queues remain materialized after first use.

Evict after How long a queue must have no publishers or subscribers before it can be removed from memory. Sweep interval How often the broker checks for queues that have become idle. Publisher idle timeout Blank disables publisher idle expiry. Otherwise, unused publisher leases stop keeping the queue active after this long.

Plexus Streams

Cursor acknowledgment batching and idle channel eviction. Changes are adopted on the next commit batch or idle sweep. An active wait can finish using its previous interval.

Cursor commit window (µs) Time to gather cursor acknowledgments into one durable batch. Zero flushes each drained batch immediately. Cursor commit batch size Maximum distinct cursors in one commit batch. Effective minimum: 1.

Evict idle stream channels Release in-memory channels with no live subscriptions. Durable data remains and the channel is loaded again on use.

Stream idle timeout How long an unused stream channel without subscriptions must remain idle before eviction. Stream sweep interval Time between idle stream eviction sweeps. Effective minimum: 1 ms.

Applies the Broker Runtime Settings above without a restart. Existing waits or sessions may retain earlier values until their next update boundary. Startup Config below is read-only and set before the broker starts.

Desktop Notifications

Notify this browser when a new attention condition appears - a quarantined partition, a backlog with no consumer, a certificate near expiry. Stored in this browser only. The broker is not involved.

Notify on

Users

Broker accounts for client authentication. The built-in fibril/fibril pair works from loopback only, so create a user for remote access. Passwords are stored as argon2 hashes and never shown.

User	Created	Updated

Node-local Storage

Changes affect only , the node serving this dashboard. To edit another node, open its dashboard. Overrides survive restart and take precedence over its startup configuration.

Override startup preallocation Segment preallocation (bytes) Zero disables preallocation. Applies to message, event and internal metadata logs when they next roll over or reopen. Existing segments keep their policy. No rollover is forced. Filesystem allocation failures fall back to extending writes and appear below.

Uncheck the override and save to use this node's startup value again. Idle logs may remain pending until they roll over or reopen. Refresh to check adoption.

Log	Segment	Applied revision	Effective preallocation	Status

Queue Dead Letter Policy

Declare how one queue handles messages that still fail after the retry limit.

Queue topic Source queue to configure. Queue group Optional group name for the source queue.

Policy Discard, use the global DLQ, or route to a queue-specific target. Max retries Blank leaves the current value for this queue unchanged.

Target topic Queue topic that receives this queue's dead-lettered messages. Target group Optional group name for the custom DLQ target.