High pulsar_transaction_aborted_count relative to pulsar_transaction_active indicates transactions are failing to commit, potentially due to coordinator failures, network issues, or application-side timeout problems.
Elevated pulsar_function_user_exceptions_count, pulsar_source_source_exceptions_count, or pulsar_sink_sink_exceptions_count signals that Pulsar Functions or connectors are failing to process messages, potentially causing data loss or duplicate processing.
Growing message backlog indicates consumers are falling behind producers or have stopped processing messages entirely. This can lead to memory pressure, disk exhaustion, and eventually message loss if retention policies expire unacknowledged messages.
High storage write latency (>1s) signals that BookKeeper ledgers cannot persist messages fast enough, creating a bottleneck that cascades to producer publish latency and overall throughput degradation.
High latency in pulsar_metadata_store_ops_latency_ms_bucket indicates ZooKeeper is struggling with metadata operations (topic creation, subscription management, ownership changes). This blocks broker operations and causes cascading failures across the cluster.
High pulsar_ml_cache_evictions rate indicates the managed ledger cache is undersized, forcing repeated cache misses that degrade consumer read performance and increase disk I/O load on BookKeeper.
Active pulsar_broker_throttled_backend_global_limit signals that the broker is applying backpressure due to resource constraints or configured quotas, blocking producers and causing publish failures or timeouts.
Divergence between pulsar_replication_rate_in and pulsar_replication_rate_out, or sustained pulsar_replication_rate_expired messages, indicates geo-replication is failing or falling behind, risking data loss in disaster recovery scenarios.
Elevated pulsar_compaction_failed combined with low pulsar_compaction_write_throughput indicates topic compaction is failing or not running, preventing cleanup of superseded messages and causing unbounded storage growth for compacted topics.
