Struct PartitionMetrics 

pub struct PartitionMetrics { /* private fields */ }

Partition monitoring for detecting hot partitions and skew (SierraDB pattern)

Design Pattern

Uses atomic operations for lock-free metric updates per partition. Tracks event counts, write latencies, and error rates per partition.

SierraDB Context

SierraDB uses fixed partitions (32 for single-node, 1024+ for clusters). Detecting hot partitions is critical for:

• Load balancing decisions
• Identifying skewed hash functions
• Capacity planning
• Performance troubleshooting

Imbalance Detection

A partition is considered imbalanced if it has >2x the average load. This threshold is based on SierraDB’s experience with production workloads.

Example

let partition_metrics = PartitionMetrics::new(32);

// Record write to partition 5
let start = Instant::now();
// ... write operation ...
partition_metrics.record_write(5, start.elapsed());

// Check for imbalances
let alerts = partition_metrics.detect_partition_imbalance();
for alert in alerts {
    tracing::warn!("Partition imbalance: {}", alert.message);
}

Implementations

impl PartitionMetrics

pub fn new(partition_count: u32) -> Self

Create a new partition metrics tracker

Arguments

• partition_count - Number of partitions (default: 32 for single-node)

pub fn with_default_partitions() -> Self

Create partition metrics with default partition count (32)

pub fn record_write(&self, partition_id: u32, latency: Duration)

Record a successful write to a partition

Arguments

• partition_id - The partition ID (0 to partition_count-1)
• latency - The write latency

pub fn record_error(&self, partition_id: u32)

Record an error for a partition

Arguments

• partition_id - The partition ID (0 to partition_count-1)

pub fn record_batch_write( &self, partition_id: u32, count: u64, latency: Duration, )

Record a batch write to a partition

Arguments

• partition_id - The partition ID (0 to partition_count-1)
• count - Number of events in the batch
• latency - Total latency for the batch write

pub fn get_partition_stats(&self, partition_id: u32) -> Option<PartitionStats>

Get statistics for a specific partition

pub fn get_all_partition_stats(&self) -> Vec<PartitionStats>

Get statistics for all partitions

pub fn detect_partition_imbalance(&self) -> Vec<PartitionImbalanceAlert>

Detect partition imbalance (hot partitions)

Returns alerts for any partition with >2x average event count. This is the SierraDB pattern for detecting skew and hot partitions.

Returns

Vector of alerts for imbalanced partitions

pub fn partition_count(&self) -> u32

Get partition count

pub fn total_events(&self) -> u64

Get total events across all partitions

pub fn total_errors(&self) -> u64

Get total errors across all partitions

pub fn uptime(&self) -> Duration

Get uptime since metrics collection started

pub fn registry(&self) -> &Registry

Get the Prometheus registry for this partition metrics

pub fn encode(&self) -> Result<String, Box<dyn Error>>

Encode metrics in Prometheus text format

pub fn get_distribution(&self) -> HashMap<u32, u64>

Get partition distribution as a map

pub fn reset(&self)

Reset all partition metrics

Trait Implementations

impl Default for PartitionMetrics

fn default() -> Self

Returns the “default value” for a type.