Telemetry table reference

Beta

This feature is in Beta. Workspace admins can control access to this feature from the Previews page. See Manage Databricks previews.

This page documents the tables Lakebase writes to your Unity Catalog schema once you configure observability for a project. These are the tables described in How it works. Insights and Genie read this same data, and because it lands in your own Unity Catalog, you can query it directly with any Databricks tool.

note

This is a Beta feature. Table and column names might change before general availability.

Columns common to every table

Every table below includes these columns, which identify where a row came from:

Column	Type	Description
project_id	STRING	The Lakebase project the row came from.
branch_id	STRING	The branch the row came from.
endpoint_id	STRING	The compute endpoint the row came from.
compute_id	STRING	The specific compute instance. This changes when a compute scales to zero and resumes.
ts	TIMESTAMP	When the row was recorded.
compute_mode	STRING	The compute's mode at the time, for example read-write or read-only.

Column	Type	Description
project_id	STRING	The Lakebase project the row came from.
branch_id	STRING	The branch the row came from.
endpoint_id	STRING	The compute endpoint the row came from.
compute_id	STRING	The specific compute instance. This changes when a compute scales to zero and resumes.
ts	TIMESTAMP	When the row was recorded.
compute_mode	STRING	The compute's mode at the time, for example read-write or read-only.

Tables with names ending in _counters also include a previous_ts column, marking the start of the interval a row's values accumulated over. These are cumulative values for that interval, not point-in-time readings. Tables ending in _gauges are point-in-time readings instead, and have no previous_ts.

The tables below list only the columns beyond these common ones.

Session activity

A snapshot of every active backend, sampled continuously. This is the same pattern as AWS Performance Insights or Oracle's Active Session History (ASH): if a session is doing anything, including waiting, it shows up here.

active_session_history

Column	Type	Description
sample_seq	LONG	Ordinal position of this sample within its batch.
pid	LONG	The backend's process ID.
datid	LONG	The database's object ID.
userid	LONG	The connecting role's object ID.
queryid	LONG	Identifies the query being run. Matches queryid in pg_stat_statements_counters.
wait_event_info	LONG	Internal encoding of the wait event.
wait_event_type	STRING	The wait event's category, for example Timeout or Lock.
wait_event	STRING	The specific wait event, for example PgSleep.
backend_state	LONG	The backend's state, for example active or idle.
backend_type	LONG	The type of backend process.
xact_age_ms	LONG	How long the current transaction has been open.
query_age_ms	LONG	How long the current query has been running.
blocking_pid	LONG	The PID of a session blocking this one, if any.
leader_pid	LONG	The PID of the parallel query leader, if this backend is a parallel worker.
flags	LONG	Internal status flags.
plan_fingerprint	LONG	Links to a specific plan shape in plan_history.
appname_hash	LONG	A hash of the connecting application's name.

Column	Type	Description
sample_seq	LONG	Ordinal position of this sample within its batch.
pid	LONG	The backend's process ID.
datid	LONG	The database's object ID.
userid	LONG	The connecting role's object ID.
queryid	LONG	Identifies the query being run. Matches queryid in pg_stat_statements_counters.
wait_event_info	LONG	Internal encoding of the wait event.
wait_event_type	STRING	The wait event's category, for example Timeout or Lock.
wait_event	STRING	The specific wait event, for example PgSleep.
backend_state	LONG	The backend's state, for example active or idle.
backend_type	LONG	The type of backend process.
xact_age_ms	LONG	How long the current transaction has been open.
query_age_ms	LONG	How long the current query has been running.
blocking_pid	LONG	The PID of a session blocking this one, if any.
leader_pid	LONG	The PID of the parallel query leader, if this backend is a parallel worker.
flags	LONG	Internal status flags.
plan_fingerprint	LONG	Links to a specific plan shape in plan_history.
appname_hash	LONG	A hash of the connecting application's name.

Wait events

Aggregated wait-event counts and time, complementing the per-sample view in active_session_history.

wait_event_counters

Column	Type	Description
wait_event_id	LONG	Internal encoding of the wait event.
wait_class_name	STRING	The wait event's category.
wait_event_name	STRING	The specific wait event.
wait_event_count	LONG	How many times this wait event occurred in the interval.
wait_event_time	DOUBLE	Total time spent in this wait event during the interval.

Column	Type	Description
wait_event_id	LONG	Internal encoding of the wait event.
wait_class_name	STRING	The wait event's category.
wait_event_name	STRING	The specific wait event.
wait_event_count	LONG	How many times this wait event occurred in the interval.
wait_event_time	DOUBLE	Total time spent in this wait event during the interval.

Query plans

A per-execution record of query plans, including the full plan and Neon-specific storage stats. This is the most detailed table, and typically where a slow-query investigation ends up.

plan_history

Column	Type	Description
row_position	LONG	Ordinal position of this row within its batch.
analyze_enabled	BOOLEAN	Whether the plan includes actual execution statistics (EXPLAIN ANALYZE), not just estimates.
buffers_enabled	BOOLEAN	Whether the plan includes buffer usage statistics.
query_hash	STRING	Hash identifying the query text.
queryid	LONG	Identifies the query. Matches queryid in pg_stat_statements_counters and active_session_history.
duration_ms	DOUBLE	How long the query took to run.
total_cost	DOUBLE	The planner's estimated cost for the query.
plan_rows	LONG	The planner's estimated row count.
actual_rows	LONG	The actual row count returned.
plan_fingerprint	LONG	Identifies this specific plan shape. Matches plan_fingerprint in active_session_history.
lock_wait_time_ms	DOUBLE	Time spent waiting on locks.
planning_time_ms	DOUBLE	Time spent planning the query, separate from execution.
shared_blks_hit / shared_blks_read / shared_blks_dirtied / shared_blks_written	LONG	Shared buffer cache activity for this execution.
local_blks_hit / local_blks_read / local_blks_dirtied / local_blks_written	LONG	Local buffer activity, for temporary tables.
temp_blks_read / temp_blks_written	LONG	Temporary file activity, for example from a large sort or hash.
neon_getpage_count	LONG	Number of page requests sent to the pageserver.
neon_file_cache_hits	LONG	Number of pages served from the local file cache instead of the pageserver.
neon_getpage_wait_us	LONG	Time spent waiting on pageserver page requests, in microseconds.
backend_pid	LONG	The backend process ID that ran this query.
lock_wait_count	LONG	Number of times this execution waited on a lock.
userid / dbid	LONG	The connecting role's and database's object IDs.
query_text	STRING	The query text.
plan_json	STRING	The full query plan, as JSON.
usename / datname	STRING	The connecting role's and database's names.
application_name / client_addr	STRING	The connecting application's name and client address.
wait_events	STRING	Wait events observed during this execution.
parameters	STRING	Bind parameters used in the query, if any.
blocking_pids	STRING	PIDs of sessions that blocked this execution, if any.
neon_branch_id / neon_endpoint_id / neon_timeline_id	STRING	Internal Neon identifiers for the branch, endpoint, and timeline.
cpu_user_time_ms / cpu_sys_time_ms	DOUBLE	CPU time consumed by this execution, split into user and system time.
voluntary_csw / involuntary_csw	LONG	Context switches during this execution.
trace_id / span_id / trace_flags / service_name	STRING / STRING / LONG / STRING	OpenTelemetry-style tracing identifiers, if the query was part of a traced request.
timeline_id	LONG	Internal timeline identifier.
truncated	BOOLEAN	Whether this row's data (for example, a large plan) was truncated before being written.

Column	Type	Description
row_position	LONG	Ordinal position of this row within its batch.
analyze_enabled	BOOLEAN	Whether the plan includes actual execution statistics (EXPLAIN ANALYZE), not just estimates.
buffers_enabled	BOOLEAN	Whether the plan includes buffer usage statistics.
query_hash	STRING	Hash identifying the query text.
queryid	LONG	Identifies the query. Matches queryid in pg_stat_statements_counters and active_session_history.
duration_ms	DOUBLE	How long the query took to run.
total_cost	DOUBLE	The planner's estimated cost for the query.
plan_rows	LONG	The planner's estimated row count.
actual_rows	LONG	The actual row count returned.
plan_fingerprint	LONG	Identifies this specific plan shape. Matches plan_fingerprint in active_session_history.
lock_wait_time_ms	DOUBLE	Time spent waiting on locks.
planning_time_ms	DOUBLE	Time spent planning the query, separate from execution.
shared_blks_hit / shared_blks_read / shared_blks_dirtied / shared_blks_written	LONG	Shared buffer cache activity for this execution.
local_blks_hit / local_blks_read / local_blks_dirtied / local_blks_written	LONG	Local buffer activity, for temporary tables.
temp_blks_read / temp_blks_written	LONG	Temporary file activity, for example from a large sort or hash.
neon_getpage_count	LONG	Number of page requests sent to the pageserver.
neon_file_cache_hits	LONG	Number of pages served from the local file cache instead of the pageserver.
neon_getpage_wait_us	LONG	Time spent waiting on pageserver page requests, in microseconds.
backend_pid	LONG	The backend process ID that ran this query.
lock_wait_count	LONG	Number of times this execution waited on a lock.
userid / dbid	LONG	The connecting role's and database's object IDs.
query_text	STRING	The query text.
plan_json	STRING	The full query plan, as JSON.
usename / datname	STRING	The connecting role's and database's names.
application_name / client_addr	STRING	The connecting application's name and client address.
wait_events	STRING	Wait events observed during this execution.
parameters	STRING	Bind parameters used in the query, if any.
blocking_pids	STRING	PIDs of sessions that blocked this execution, if any.
neon_branch_id / neon_endpoint_id / neon_timeline_id	STRING	Internal Neon identifiers for the branch, endpoint, and timeline.
cpu_user_time_ms / cpu_sys_time_ms	DOUBLE	CPU time consumed by this execution, split into user and system time.
voluntary_csw / involuntary_csw	LONG	Context switches during this execution.
trace_id / span_id / trace_flags / service_name	STRING / STRING / LONG / STRING	OpenTelemetry-style tracing identifiers, if the query was part of a traced request.
timeline_id	LONG	Internal timeline identifier.
truncated	BOOLEAN	Whether this row's data (for example, a large plan) was truncated before being written.

Query statistics

Aggregated per-query statistics, sourced from the standard pg_stat_statements extension.

pg_stat_statements_counters

Column	Type	Description
userid / dbid	LONG	The connecting role's and database's object IDs.
queryid	LONG	Identifies the query. Matches queryid in plan_history and active_session_history.
toplevel	BOOLEAN	Whether this row is for a top-level statement, as opposed to one nested inside a PL/pgSQL function or procedure. See the PL/pgSQL limitation.
query	STRING	The query text.
calls	LONG	Number of times this query was executed in the interval.
plans	LONG	Number of times this query was planned in the interval.
rows	LONG	Total rows returned or affected.
total_exec_time / total_plan_time	DOUBLE	Total time spent executing and planning this query in the interval.
shared_blks_hit / shared_blks_read / shared_blks_dirtied / shared_blks_written	LONG	Shared buffer cache activity.
local_blks_hit / local_blks_read / local_blks_dirtied / local_blks_written	LONG	Local buffer activity, for temporary tables.
temp_blks_read / temp_blks_written	LONG	Temporary file activity.
wal_bytes / wal_fpi / wal_records	LONG	Write-ahead log volume generated by this query.
jit_emission_count / jit_emission_time / jit_functions / jit_generation_time / jit_inlining_count / jit_inlining_time / jit_optimization_count / jit_optimization_time	Mixed	JIT (just-in-time) compilation statistics, if JIT was used for this query.

Column	Type	Description
userid / dbid	LONG	The connecting role's and database's object IDs.
queryid	LONG	Identifies the query. Matches queryid in plan_history and active_session_history.
toplevel	BOOLEAN	Whether this row is for a top-level statement, as opposed to one nested inside a PL/pgSQL function or procedure. See the PL/pgSQL limitation.
query	STRING	The query text.
calls	LONG	Number of times this query was executed in the interval.
plans	LONG	Number of times this query was planned in the interval.
rows	LONG	Total rows returned or affected.
total_exec_time / total_plan_time	DOUBLE	Total time spent executing and planning this query in the interval.
shared_blks_hit / shared_blks_read / shared_blks_dirtied / shared_blks_written	LONG	Shared buffer cache activity.
local_blks_hit / local_blks_read / local_blks_dirtied / local_blks_written	LONG	Local buffer activity, for temporary tables.
temp_blks_read / temp_blks_written	LONG	Temporary file activity.
wal_bytes / wal_fpi / wal_records	LONG	Write-ahead log volume generated by this query.
jit_emission_count / jit_emission_time / jit_functions / jit_generation_time / jit_inlining_count / jit_inlining_time / jit_optimization_count / jit_optimization_time	Mixed	JIT (just-in-time) compilation statistics, if JIT was used for this query.

Schema changes

Schema changes: what changed, when, and by whom. This is what Genie reads to trace an incident back to a schema change, whether from a deploy or a manual edit.

ddl_history

Column	Type	Description
row_position	LONG	Ordinal position of this row within its batch.
command_tag	STRING	The type of DDL command, for example ALTER TABLE.
object_type / object_name	STRING	The kind and name of the object that changed.
schema_name	STRING	The Postgres schema the object belongs to.
query_text	STRING	The full DDL statement.
duration_ms	DOUBLE	How long the DDL statement took to run.
userid / usename	LONG / STRING	The connecting role's object ID and name.
application_name	STRING	The connecting application's name.
backend_pid	LONG	The backend process ID that ran the statement.
dbid	LONG	The database's object ID.
search_path	STRING	The session's search path at the time.
ddl_json	STRING	A structured, parsed representation of the DDL statement.

Column	Type	Description
row_position	LONG	Ordinal position of this row within its batch.
command_tag	STRING	The type of DDL command, for example ALTER TABLE.
object_type / object_name	STRING	The kind and name of the object that changed.
schema_name	STRING	The Postgres schema the object belongs to.
query_text	STRING	The full DDL statement.
duration_ms	DOUBLE	How long the DDL statement took to run.
userid / usename	LONG / STRING	The connecting role's object ID and name.
application_name	STRING	The connecting application's name.
backend_pid	LONG	The backend process ID that ran the statement.
dbid	LONG	The database's object ID.
search_path	STRING	The session's search path at the time.
ddl_json	STRING	A structured, parsed representation of the DDL statement.

note

DDL capture has a real gap: a change made right around a compute resuming from scale-to-zero can be missed. If a schema change doesn't appear here even though you know it happened, that's the most likely reason. See Limitations.

Postgres logs

Raw Postgres log messages, including the specific error behind a failure.

postgres_logs

Column	Type	Description
sample_seq	LONG	Ordinal position of this row within its batch.
elevel	LONG	Internal log level encoding.
severity	STRING	The log severity, for example ERROR or LOG.
sqlstate	STRING	The Postgres SQLSTATE error code, for example 42703 for an undefined column.
backend_pid / leader_backend_pid	LONG	The backend process ID, and its parallel query leader's PID if applicable.
userid / usename	LONG / STRING	The connecting role's object ID and name.
datid / datname	LONG / STRING	The database's object ID and name.
application_name / client_addr	STRING	The connecting application's name and client address.
schema_name / table_name / column_name / constraint_name / datatype_name	STRING	The schema, table, column, constraint, or data type referenced by the error, when Postgres has one.
funcname	STRING	The function the error occurred in, if applicable.
filename / lineno	STRING / LONG	The Postgres source file and line the error originated from.
message	STRING	The log message text.
detail / hint / context	STRING	Additional detail, a suggested fix, and the error's context, when Postgres has them.
internalquery	STRING	An internally generated query related to the error, if applicable.
statement	STRING	The statement that triggered this log entry.
truncated_mask	LONG	Indicates which fields, if any, were truncated before being written.

Column	Type	Description
sample_seq	LONG	Ordinal position of this row within its batch.
elevel	LONG	Internal log level encoding.
severity	STRING	The log severity, for example ERROR or LOG.
sqlstate	STRING	The Postgres SQLSTATE error code, for example 42703 for an undefined column.
backend_pid / leader_backend_pid	LONG	The backend process ID, and its parallel query leader's PID if applicable.
userid / usename	LONG / STRING	The connecting role's object ID and name.
datid / datname	LONG / STRING	The database's object ID and name.
application_name / client_addr	STRING	The connecting application's name and client address.
schema_name / table_name / column_name / constraint_name / datatype_name	STRING	The schema, table, column, constraint, or data type referenced by the error, when Postgres has one.
funcname	STRING	The function the error occurred in, if applicable.
filename / lineno	STRING / LONG	The Postgres source file and line the error originated from.
message	STRING	The log message text.
detail / hint / context	STRING	Additional detail, a suggested fix, and the error's context, when Postgres has them.
internalquery	STRING	An internally generated query related to the error, if applicable.
statement	STRING	The statement that triggered this log entry.
truncated_mask	LONG	Indicates which fields, if any, were truncated before being written.

note

The structured schema_name, table_name, and column_name fields aren't populated for every error type. A plain "column does not exist" parse error (SQLSTATE 42703), for example, leaves them empty even though message names the column directly. Don't rely on these fields being populated. Check message first.

Compute metrics

Compute-level resource usage: CPU, memory, disk and network I/O, and the local file cache (LFC), Lakebase's compute-side cache in front of the pageserver.

compute_counters

Column	Type	Description
backpressure_throttling_seconds	DOUBLE	Time the storage layer spent throttling writes in this interval. A non-zero value means write volume exceeded what storage can absorb.
host_busiest_cpu_seconds / host_cpu_seconds	DOUBLE	CPU time consumed, for the busiest core and in total.
host_disk_read_bytes / host_disk_written_bytes	DOUBLE	Disk I/O volume.
host_network_receive_bytes / host_network_transmit_bytes	DOUBLE	Network I/O volume.
lfc_evictions / lfc_hits / lfc_misses / lfc_writes	LONG	Local file cache activity for this interval.
replica_lfc_redo_evictions	LONG	Local file cache evictions caused by replica redo, if this compute is a read replica.

Column	Type	Description
backpressure_throttling_seconds	DOUBLE	Time the storage layer spent throttling writes in this interval. A non-zero value means write volume exceeded what storage can absorb.
host_busiest_cpu_seconds / host_cpu_seconds	DOUBLE	CPU time consumed, for the busiest core and in total.
host_disk_read_bytes / host_disk_written_bytes	DOUBLE	Disk I/O volume.
host_network_receive_bytes / host_network_transmit_bytes	DOUBLE	Network I/O volume.
lfc_evictions / lfc_hits / lfc_misses / lfc_writes	LONG	Local file cache activity for this interval.
replica_lfc_redo_evictions	LONG	Local file cache evictions caused by replica redo, if this compute is a read replica.

compute_gauges

Column	Type	Description
current_lsn	LONG	The current write-ahead log position.
replica_wal_replay_lsn / replica_wal_receive_lsn	LONG	WAL replay and receive positions, if this compute is a read replica.
max_connections	LONG	The configured connection limit.
lfc_working_set_size_5m / lfc_working_set_size_15m / lfc_working_set_size_60m	LONG	Estimated working set size over the last 5, 15, and 60 minutes.
lfc_size_limit / lfc_allocated / lfc_used	LONG	Local file cache capacity and current usage.
total_size_limit / total_size	LONG	Overall storage capacity and current usage.
replication_lag_bytes / replication_lag_time_ms	LONG / DOUBLE	Replication lag, if this compute is a read replica.
host_logical_cpus / host_memory_total_bytes	DOUBLE	The compute's provisioned CPU and memory.

Column	Type	Description
current_lsn	LONG	The current write-ahead log position.
replica_wal_replay_lsn / replica_wal_receive_lsn	LONG	WAL replay and receive positions, if this compute is a read replica.
max_connections	LONG	The configured connection limit.
lfc_working_set_size_5m / lfc_working_set_size_15m / lfc_working_set_size_60m	LONG	Estimated working set size over the last 5, 15, and 60 minutes.
lfc_size_limit / lfc_allocated / lfc_used	LONG	Local file cache capacity and current usage.
total_size_limit / total_size	LONG	Overall storage capacity and current usage.
replication_lag_bytes / replication_lag_time_ms	LONG / DOUBLE	Replication lag, if this compute is a read replica.
host_logical_cpus / host_memory_total_bytes	DOUBLE	The compute's provisioned CPU and memory.

Database metrics

Per-database activity: rows changed, transactions, sessions, and vacuum health. This is where table bloat and autovacuum signals live.

database_counters

Column	Type	Description
datid / datname	LONG / STRING	The database's object ID and name.
active_time_ms / idle_in_transaction_time_ms / session_time_ms	DOUBLE	Time spent active, idle in an open transaction, and connected overall.
deadlocks	LONG	Number of deadlocks detected.
rows_deleted / rows_fetched / rows_inserted / rows_returned / rows_updated	LONG	Row-level activity for the interval.
sessions_abandoned / sessions_established / sessions_fatal / sessions_killed	LONG	Session lifecycle counts.
shared_blks_hit / shared_blks_read	LONG	Shared buffer cache activity.
xact_commit / xact_rollback	LONG	Committed and rolled-back transactions.

Column	Type	Description
datid / datname	LONG / STRING	The database's object ID and name.
active_time_ms / idle_in_transaction_time_ms / session_time_ms	DOUBLE	Time spent active, idle in an open transaction, and connected overall.
deadlocks	LONG	Number of deadlocks detected.
rows_deleted / rows_fetched / rows_inserted / rows_returned / rows_updated	LONG	Row-level activity for the interval.
sessions_abandoned / sessions_established / sessions_fatal / sessions_killed	LONG	Session lifecycle counts.
shared_blks_hit / shared_blks_read	LONG	Shared buffer cache activity.
xact_commit / xact_rollback	LONG	Committed and rolled-back transactions.

database_gauges

Column	Type	Description
datid / datname	LONG / STRING	The database's object ID and name.
numbackends	LONG	Current number of connections to this database.
size	LONG	The database's on-disk size.
oldest_frozen_xid_age / oldest_mxid_age	LONG	Transaction ID and multixact ID age of the oldest unfrozen row. Rising values are the earliest signal of vacuum falling behind, well before bloat becomes visible in table size.

Column	Type	Description
datid / datname	LONG / STRING	The database's object ID and name.
numbackends	LONG	Current number of connections to this database.
size	LONG	The database's on-disk size.
oldest_frozen_xid_age / oldest_mxid_age	LONG	Transaction ID and multixact ID age of the oldest unfrozen row. Rising values are the earliest signal of vacuum falling behind, well before bloat becomes visible in table size.

Next steps

• AI-assisted troubleshooting overview — how this telemetry powers Insights and Genie.
• Find and resolve issues with Insights — see this data surfaced as a diagnosis, not raw rows.
• Diagnose and fix issues with Genie — investigate a problem conversationally.