Metrics reference

You should use caution when interpreting unfamiliar metrics. Reading the Performance section is recommended to better understand the metrics.

Types of metrics

Neo4j has the following types of metrics:

  • Global — covers the whole Neo4j DBMS.

  • Per database — covers an individual database.

The metrics fall into one of the following categories:

  • Gauge — shows an instantaneous reading of a particular value.

  • Counter — shows an accumulated value.

  • Histogram — shows the distribution of values.

Global metrics

Global metrics cover the whole database management system and represent the system’s status as a whole.

Global metrics have the following name format:

  • <metrics-prefix>.dbms.<metric-name>, where the <user-configured-prefix> can be configured with the server.metrics.prefix configuration setting.

Metrics of this type are reported as soon as the database management system is available. For example, all JVM-related metrics are global. In particular, the neo4j.dbms.vm.thread.count metric has a default user-configured-prefix neo4j and the global metric name is vm.thread.count.

By default, global metrics include:

  • Bolt metrics

  • Page cache metrics

  • GC metrics

  • Thread metrics

  • Database operation metrics

  • Web Server metrics

  • JVM metrics

Database metrics

Each database metric is reported for a particular database only. Database metrics are only available during the lifetime of the database. When a database becomes unavailable, all of its metrics become unavailable also.

Database metrics have the following name format:

  • <user-configured-prefix>.database.<database-name>.<metric-name>, where the <user-configured-prefix> can be configured with the server.metrics.prefix configuration setting.

For example, any transaction metric is a database metric. In particular, the neo4j.database.mydb.transaction.started metric has a default user-configured-prefix neo4j and is a metric for the mydb database.

By default, database metrics include:

  • Transaction metrics

  • Checkpoint metrics

  • Log rotation metrics

  • Database data metrics

  • Cypher metrics

  • Clustering metrics

General-purpose metrics

Table 1. Bolt metrics
Name Description

<prefix>.bolt.connections_opened

The total number of Bolt connections opened since startup. This includes both succeeded and failed connections. Useful for monitoring load via the Bolt drivers in combination with other metrics. (counter)

<prefix>.bolt.connections_closed

The total number of Bolt connections closed since startup. This includes both properly and abnormally ended connections. Useful for monitoring load via Bolt drivers in combination with other metrics. (counter)

<prefix>.bolt.connections_running

The total number of Bolt connections that are currently executing Cypher and returning results. Useful to track the overall load on Bolt connections. This is limited to the number of Bolt worker threads that have been configured via dbms.connector.bolt.thread_pool_max_size. Reaching this maximum indicated the server is running at capacity. (gauge)

<prefix>.bolt.connections_idle

The total number of Bolt connections that are not currently executing Cypher or returning results. (gauge)

<prefix>.bolt.messages_received

The total number of messages received via Bolt since startup. Useful to track general message activity in combination with other metrics. (counter)

<prefix>.bolt.messages_started

The total number of messages that have started processing since being received. A received message may have begun processing until a Bolt worker thread becomes available. A large gap observed between bolt.messages_received and bolt.messages_started could indicate the server is running at capacity. (counter)

<prefix>.bolt.messages_done

The total number of Bolt messages that have completed processing whether successfully or unsuccessfully. Useful for tracking overall load. (counter)

<prefix>.bolt.messages_failed

The total number of messages that have failed while processing. A high number of failures may indicate an issue with the server and further investigation of the logs is recommended. (counter)

<prefix>.bolt.accumulated_queue_time

(unsupported feature) When internal.server.bolt.thread_pool_queue_size is enabled, the total time in milliseconds that a Bolt message waits in the processing queue before a Bolt worker thread becomes available to process it. Sharp increases in this value indicate that the server is running at capacity. If internal.server.bolt.thread_pool_queue_size is disabled, the value should be 0, meaning that messages are directly handed off to worker threads. (counter)

<prefix>.bolt.accumulated_processing_time

The total amount of time in milliseconds that worker threads have been processing messages. Useful for monitoring load via Bolt drivers in combination with other metrics. (counter)
Table 2. Database checkpointing metrics
Name Description

<prefix>.check_point.events

The total number of checkpoint events executed so far. (counter)

<prefix>.check_point.total_time

The total time, in milliseconds, spent in checkpointing so far. (counter)

<prefix>.check_point.duration

The duration, in milliseconds, of the last checkpoint event. Checkpoints should generally take several seconds to several minutes. Long checkpoints can be an issue, as these are invoked when the database stops, when a hot backup is taken, and periodically as well. Values over 30 minutes or so should be cause for some investigation. (gauge)

<prefix>.check_point.flushed_bytes

Introduced in 5.10The accumulated number of bytes flushed during the last checkpoint event. (gauge)

<prefix>.check_point.pages_flushed

The number of pages that were flushed during the last checkpoint event. (gauge)

<prefix>.check_point.io_performed

The number of IOs from Neo4j perspective performed during the last check point event. (gauge)

<prefix>.check_point.io_limit

The IO limit used during the last checkpoint event. (gauge)
Table 3. Cypher metrics
Name Description

<prefix>.cypher.replan_events

The total number of times Cypher has decided to re-plan a query. Neo4j caches 1000 plans by default. Seeing sustained replanning events or large spikes could indicate an issue that needs to be investigated. (counter)

<prefix>.cypher.replan_wait_time

The total number of seconds waited between query replans. (counter)
Table 4. Database data count metrics
Name Description

<prefix>.neo4j.count.relationship

The total number of relationships in the database. (gauge)

<prefix>.neo4j.count.node

The total number of nodes in the database. A rough metric of how big your graph is. And if you are running a bulk insert operation you can see this tick up. (gauge)
Table 5. Database neo4j pools metrics
Name Description

<prefix>.pool.<pool>.<database>.used_heap

Used or reserved heap memory in bytes. (gauge)

<prefix>.pool.<pool>.<database>.used_native

Used or reserved native memory in bytes. (gauge)

<prefix>.pool.<pool>.<database>.total_used

Sum total used heap and native memory in bytes. (gauge)

<prefix>.pool.<pool>.<database>.total_size

Sum total size of capacity of the heap and/or native memory pool. (gauge)

<prefix>.pool.<pool>.<database>.free

Available unused memory in the pool, in bytes. (gauge)
Table 6. Database operation count metrics
Name Description

<prefix>.db.operation.count.create

Count of successful database create operations. (counter)

<prefix>.db.operation.count.start

Count of successful database start operations. (counter)

<prefix>.db.operation.count.stop

Count of successful database stop operations. (counter)

<prefix>.db.operation.count.drop

Count of successful database drop operations. (counter)

<prefix>.db.operation.count.failed

Count of failed database operations. (counter)

<prefix>.db.operation.count.recovered

Count of database operations that failed previously but have recovered. (counter)
Table 7. Database state metrics
Name Description

<prefix>.db.state.count.hosted

Databases hosted on this server. Databases in states started, store copying, or draining are considered hosted. (gauge)

<prefix>.db.state.count.failed

Databases in a failed state on this server. (gauge)

<prefix>.db.state.count.desired_started

Databases that desire to be started on this server. (gauge)
Table 8. Database data metrics
Name Description

<prefix>.ids_in_use.relationship_type

The total number of different relationship types stored in the database. Informational, not an indication of any issue. Spikes or large increases indicate large data loads, which could correspond with some behavior you are investigating. (gauge)

<prefix>.ids_in_use.property

The total number of different property names used in the database. Informational, not an indication of any issue. Spikes or large increases indicate large data loads, which could correspond with some behavior you are investigating. (gauge)

<prefix>.ids_in_use.relationship

The total number of relationships stored in the database. Informational, not an indication of any issue. Spikes or large increases indicate large data loads, which could correspond with some behavior you are investigating. (gauge)

<prefix>.ids_in_use.node

The total number of nodes stored in the database. Informational, not an indication of any issue. Spikes or large increases indicate large data loads, which could correspond with some behavior you are investigating. (gauge)
Table 9. Global neo4j pools metrics
Name Description

<prefix>.dbms.pool.<pool>.used_heap

Used or reserved heap memory in bytes. (gauge)

<prefix>.dbms.pool.<pool>.used_native

Used or reserved native memory in bytes. (gauge)

<prefix>.dbms.pool.<pool>.total_used

Sum total used heap and native memory in bytes. (gauge)

<prefix>.dbms.pool.<pool>.total_size

Sum total size of the capacity of the heap and/or native memory pool. (gauge)

<prefix>.dbms.pool.<pool>.free

Available unused memory in the pool, in bytes. (gauge)
Table 10. Database page cache metrics
Name Description

<prefix>.page_cache.eviction_exceptions

The total number of exceptions seen during the eviction process in the page cache. (counter)

<prefix>.page_cache.flushes

The total number of page flushes executed by the page cache. (counter)

<prefix>.page_cache.merges

The total number of page merges executed by the page cache. (counter)

<prefix>.page_cache.unpins

The total number of page unpins executed by the page cache. (counter)

<prefix>.page_cache.pins

The total number of page pins executed by the page cache. (counter)

<prefix>.page_cache.evictions

The total number of page evictions executed by the page cache. (counter)

<prefix>.page_cache.evictions.cooperative

The total number of cooperative page evictions executed by the page cache due to low available pages. (counter)

<prefix>.page_cache.page_faults

The total number of page faults happened in the page cache. If this continues to rise over time, it could be an indication that more page cache is needed. (counter)

<prefix>.page_cache.page_fault_failures

The total number of failed page faults happened in the page cache. (counter)

<prefix>.page_cache.page_cancelled_faults

The total number of canceled page faults happened in the page cache. (counter)

<prefix>.page_cache.hits

The total number of page hits happened in the page cache. (counter)

<prefix>.page_cache.hit_ratio

The ratio of hits to the total number of lookups in the page cache. Performance relies on efficiently using the page cache, so this metric should be in the 98-100% range consistently. If it is much lower than that, then the database is going to disk too often. (gauge)

<prefix>.page_cache.usage_ratio

The ratio of number of used pages to total number of available pages. This metric shows what percentage of the allocated page cache is actually being used. If it is 100%, then it is likely that the hit ratio will start dropping, and you should consider allocating more RAM to page cache. (gauge)

<prefix>.page_cache.bytes_read

The total number of bytes read by the page cache. (counter)

<prefix>.page_cache.bytes_written

The total number of bytes written by the page cache. (counter)

<prefix>.page_cache.iops

The total number of IO operations performed by page cache.

<prefix>.page_cache.throttled.times

The total number of times page cache flush IO limiter was throttled during ongoing IO operations.

<prefix>.page_cache.throttled.millis

The total number of millis page cache flush IO limiter was throttled during ongoing IO operations.

<prefix>.page_cache.pages_copied

The total number of page copies happened in the page cache. (counter)
Table 11. Query execution metrics
Name Description

<prefix>.db.query.execution.success

Count of successful queries executed. (counter)

<prefix>.db.query.execution.failure

Count of failed queries executed. (counter)

<prefix>.db.query.execution.latency.millis

Execution time in milliseconds of queries executed successfully. (histogram)

<prefix>.dbms.routing.query.count.local

Introduced in 5.10 The total number of queries executed locally. (counter)

<prefix>.dbms.routing.query.count.remote_internal

Introduced in 5.10 The total number of queries executed remotely to a member of the same cluster. (counter)

<prefix>.dbms.routing.query.count.remote_external

Introduced in 5.10 The total number of queries executed remotely to a member of a different cluster. (counter)
Table 12. Database store size metrics
Name Description

<prefix>.store.size.total

The total size of the database and transaction logs, in bytes. The total size of the database helps determine how much cache page is required. It also helps compare the total disk space used by the data store and how much is available. (gauge)

<prefix>.store.size.database

The size of the database, in bytes. The total size of the database helps determine how much cache page is required. It also helps compare the total disk space used by the data store and how much is available. (gauge)
Table 13. Database transaction log metrics
Name Description

<prefix>.log.rotation_events

The total number of transaction log rotations executed so far. (counter)

<prefix>.log.rotation_total_time

The total time, in milliseconds, spent in rotating transaction logs so far. (counter)

<prefix>.log.rotation_duration

The duration, in milliseconds, of the last log rotation event. (gauge)

<prefix>.log.appended_bytes

The total number of bytes appended to the transaction log. (counter)

<prefix>.log.flushes

The total number of transaction log flushes. (counter)

<prefix>.log.append_batch_size

The size of the last transaction append batch. (gauge)
Table 14. Database transaction metrics
Name Description

<prefix>.transaction.started

The total number of started transactions. (counter)

<prefix>.transaction.peak_concurrent

The highest peak of concurrent transactions. This is a useful value to understand. It can help you with the design for the highest load scenarios and whether the Bolt thread settings should be altered. (counter)

<prefix>.transaction.active

The number of currently active transactions. Informational, not an indication of any issue. Spikes or large increases could indicate large data loads or just high read load. (gauge)

<prefix>.transaction.active_read

The number of currently active read transactions. (gauge)

<prefix>.transaction.active_write

The number of currently active write transactions. (gauge)

<prefix>.transaction.committed

The total number of committed transactions. Informational, not an indication of any issue. Spikes or large increases indicate large data loads or just high read load. (counter)

<prefix>.transaction.committed_read

The total number of committed read transactions. Informational, not an indication of any issue. Spikes or large increases indicate high read load. (counter)

<prefix>.transaction.committed_write

The total number of committed write transactions. Informational, not an indication of any issue. Spikes or large increases indicate large data loads, which could correspond with some behavior you are investigating. (counter)

<prefix>.transaction.rollbacks

The total number of rolled back transactions. (counter)

<prefix>.transaction.rollbacks_read

The total number of rolled back read transactions. (counter)

<prefix>.transaction.rollbacks_write

The total number of rolled back write transactions. Seeing a lot of writes rolled back may indicate various issues with locking, transaction timeouts, etc. (counter)

<prefix>.transaction.terminated

The total number of terminated transactions. (counter)

<prefix>.transaction.terminated_read

The total number of terminated read transactions. (counter)

<prefix>.transaction.terminated_write

The total number of terminated write transactions. (counter)

<prefix>.transaction.last_committed_tx_id

The ID of the last committed transaction. Track this for each instance. (Cluster) Track this for each primary, and each secondary. Might break into separate charts. It should show one line, ever increasing, and if one of the lines levels off or falls behind, it is clear that this member is no longer replicating data, and action is needed to rectify the situation. (counter)

<prefix>.transaction.last_closed_tx_id

The ID of the last closed transaction. (counter)

<prefix>.transaction.tx_size_heap

The transactions' size on heap in bytes. (histogram)

<prefix>.transaction.tx_size_native

The transactions' size in native memory in bytes. (histogram)
Table 15. Database index metrics
Name Description

<prefix>.index.fulltext.queried

The total number of times fulltext indexes have been queried. (counter)

<prefix>.index.fulltext.populated

The total number of fulltext index population jobs that have been completed. (counter)

<prefix>.index.lookup.queried

The total number of times lookup indexes have been queried. (counter)

<prefix>.index.lookup.populated

The total number of lookup index population jobs that have been completed. (counter)

<prefix>.index.text.queried

The total number of times text indexes have been queried. (counter)

<prefix>.index.text.populated

The total number of text index population jobs that have been completed. (counter)

<prefix>.index.range.queried

The total number of times range indexes have been queried. (counter)

<prefix>.index.range.populated

The total number of range index population jobs that have been completed. (counter)

<prefix>.index.point.queried

The total number of times point indexes have been queried. (counter)

<prefix>.index.point.populated

The total number of point index population jobs that have been completed. (counter)

<prefix>.index.vector.queried

The total number of times vector indexes have been queried. (counter)

<prefix>.index.vector.populated

The total number of vector index population jobs that have been completed. (counter)
Table 16. Server metrics
Name Description

<prefix>.server.threads.jetty.idle

The total number of idle threads in the jetty pool. (gauge)

<prefix>.server.threads.jetty.all

The total number of threads (both idle and busy) in the jetty pool. (gauge)

Metrics specific to clustering

Table 17. CatchUp Metrics
Name Description

<prefix>.cluster.catchup.tx_pull_requests_received

TX pull requests received from secondaries. (counter)
Table 18. Discovery database primary metrics
Name Description

<prefix>.cluster.discovery.replicated_data

Size of replicated data structures. (gauge)

<prefix>.cluster.discovery.cluster.members

Discovery cluster member size. (gauge)

<prefix>.cluster.discovery.cluster.unreachable

Discovery cluster unreachable size. (gauge)

<prefix>.cluster.discovery.cluster.converged

Discovery cluster convergence. (gauge)
Table 19. Raft database primary metrics
Name Description

<prefix>.cluster.raft.append_index

The append index of the Raft log. Each index represents a write transaction (possibly internal) proposed for commitment. The values mostly increase, but sometimes they can decrease as a consequence of leader changes. The append index should always be less than or equal to the commit index. (gauge)

<prefix>.cluster.raft.commit_index

The commit index of the Raft log. Represents the commitment of previously appended entries. Its value increases monotonically if you do not unbind the cluster state. The commit index should always be bigger than or equal to the appended index. (gauge)

<prefix>.cluster.raft.applied_index

The applied index of the Raft log. Represents the application of the committed Raft log entries to the database and internal state. The applied index should always be bigger than or equal to the commit index. The difference between this and the commit index can be used to monitor how up-to-date the follower database is. (gauge)

<prefix>.cluster.raft.term

The Raft Term of this server. It increases monotonically if you do not unbind the cluster state. (gauge)

<prefix>.cluster.raft.tx_retries

Transaction retries. (counter)

<prefix>.cluster.raft.is_leader

Is this server the leader? Track this for each database primary in the cluster. It reports 0 if it is not the leader and 1 if it is the leader. The sum of all of these should always be 1. However, there are transient periods in which the sum can be more than 1 because more than one member thinks it is the leader. Action may be needed if the metric shows 0 for more than 30 seconds. (gauge)

<prefix>.cluster.raft.in_flight_cache.total_bytes

In-flight cache total bytes. (gauge)

<prefix>.cluster.raft.in_flight_cache.max_bytes

In-flight cache max bytes. (gauge)

<prefix>.cluster.raft.in_flight_cache.element_count

In-flight cache element count. (gauge)

<prefix>.cluster.raft.in_flight_cache.max_elements

In-flight cache maximum elements. (gauge)

<prefix>.cluster.raft.in_flight_cache.hits

In-flight cache hits. (counter)

<prefix>.cluster.raft.in_flight_cache.misses

In-flight cache misses. (counter)

<prefix>.cluster.raft.raft_log_entry_prefetch_buffer.lag

Raft Log Entry Prefetch Lag. (gauge)

<prefix>.cluster.raft.raft_log_entry_prefetch_buffer.bytes

Raft Log Entry Prefetch total bytes. (gauge)

<prefix>.cluster.raft.raft_log_entry_prefetch_buffer.size

Raft Log Entry Prefetch buffer size. (gauge)

<prefix>.cluster.raft.raft_log_entry_prefetch_buffer.async_put

Raft Log Entry Prefetch buffer async puts. (gauge)

<prefix>.cluster.raft.raft_log_entry_prefetch_buffer.sync_put

Raft Log Entry Prefetch buffer sync puts. (gauge)

<prefix>.cluster.raft.message_processing_delay

The time the Raft message stays in the queue after being received and before being processed, i.e. append and commit to the store. The higher the value, the longer the messages wait to be processed. This metric is closely linked to disk write performance. (gauge)

<prefix>.cluster.raft.message_processing_timer

Timer for Raft message processing. (counter, histogram)

<prefix>.cluster.raft.replication_new

The total number of Raft replication requests. It increases with write transactions (possibly internal) activity. (counter)

<prefix>.cluster.raft.replication_attempt

The total number of Raft replication requests attempts. It is bigger or equal to the replication requests. (counter)

<prefix>.cluster.raft.replication_fail

The total number of Raft replication attempts that have failed. (counter)

<prefix>.cluster.raft.replication_maybe

Raft Replication maybe count. (counter)

<prefix>.cluster.raft.replication_success

The total number of Raft replication requests that have succeeded. (counter)

<prefix>.cluster.raft.last_leader_message

The time elapsed since the last message from a leader in milliseconds. Should reset periodically. (gauge)
Table 20. Database secondary Metrics
Name Description

<prefix>.cluster.store_copy.pull_updates

The total number of pull requests made by this instance. (counter)

<prefix>.cluster.store_copy.pull_update_highest_tx_id_requested

The highest transaction id requested in a pull update by this instance. (counter)

<prefix>.cluster.store_copy.pull_update_highest_tx_id_received

The highest transaction id that has been pulled in the last pull updates by this instance. (counter)

Java Virtual Machine Metrics

The JVM metrics show information about garbage collections (for example, the number of events and time spent collecting), memory pools and buffers, and the number of active threads running. They are environment dependent and therefore, may vary on different hardware and with different JVM configurations. The metrics about the JVM’s memory usage expose values that are provided by the MemoryPoolMXBeans and BufferPoolMXBeans. The memory pools are memory managed by the JVM, for example, neo4j.dbms.vm.memory.pool.g1_survivor_space. Therefore, if necessary, you can tune them using the JVM settings. The buffer pools are space outside of the memory managed by the garbage collector. Neo4j allocates buffers in those pools as it needs them. You can limit this memory using JVM settings, but there is never any good reason for you to set them.

Table 21. JVM file descriptor metrics.
Name Description

<prefix>.vm.file.descriptors.count

The current number of open file descriptors. (gauge)

<prefix>.vm.file.descriptors.maximum

(OS setting) The maximum number of open file descriptors. It is recommended to be set to 40K file handles, because of the native and Lucene indexing Neo4j uses. If this metric gets close to the limit, you should consider raising it. (gauge)
Table 22. GC metrics.
Name Description

<prefix>.vm.gc.time.<gc>

Accumulated garbage collection time in milliseconds. Long GCs can be an indication of performance issues or potential instability. If this approaches the heartbeat timeout in a cluster, it may cause unwanted leader switches. (counter)

<prefix>.vm.gc.count.<gc>

Total number of garbage collections. (counter)
Table 23. JVM Heap metrics.
Name Description

<prefix>.vm.heap.committed

Amount of memory (in bytes) guaranteed to be available for use by the JVM. (gauge)

<prefix>.vm.heap.used

Amount of memory (in bytes) currently used. This is the amount of heap space currently used at a given point in time. Monitor this to identify if you are maxing out consistently, in which case, you should increase the initial and max heap size, or if you are underutilizing, you should decrease the initial and max heap sizes. (gauge)

<prefix>.vm.heap.max

Maximum amount of heap memory (in bytes) that can be used. This is the amount of heap space currently used at a given point in time. Monitor this to identify if you are maxing out consistently, in which case, you should increase the initial and max heap size, or if you are underutilizing, you should decrease the initial and max heap sizes. (gauge)
Table 24. JVM memory buffers metrics.
Name Description

<prefix>.vm.memory.buffer.<bufferpool>.count

Estimated number of buffers in the pool. (gauge)

<prefix>.vm.memory.buffer.<bufferpool>.used

Estimated amount of memory used by the pool. (gauge)

<prefix>.vm.memory.buffer.<bufferpool>.capacity

Estimated total capacity of buffers in the pool. (gauge)
Table 25. JVM memory pools metrics.
Name Description

<prefix>.vm.memory.pool.<pool>

Estimated amount of memory in bytes used by the pool. (gauge)
Table 26. JVM pause time metrics.
Name Description

<prefix>.vm.pause_time

Accumulated detected VM pause time. (counter)
Table 27. JVM threads metrics.
Name Description

<prefix>.vm.threads

The total number of live threads including daemon and non-daemon threads. (gauge)