Monitor Microsoft Fabric HorizonDB

Monitor Fabric HorizonDB by combining Fabric item and capacity state with PostgreSQL diagnostic views. No single signal describes database, mirroring, federation, and Copy job health.

Important

Fabric HorizonDB is in public preview. Monitoring controls and status can vary by environment. Use the information exposed by your database item, mirroring experience, and Copy job run history; don't assume dashboard parity with another service. For more information, see Microsoft Fabric preview features.

Monitor separate health dimensions

Review these dimensions independently:

Dimension Examples
Database lifecycle Provisioning, ready, scaling, pausing, paused, resuming, maintenance, failed, or degraded.
Compute Current, target, requested maximum, effective maximum, and whether a transition is pending or constrained.
PostgreSQL workload Connections, transactions, waits, blocking, temporary I/O, cache behavior, and table size.
Mirroring Initial copy, catching up, healthy, suspended, lag, or reseed required.
Federation Remote authorization, foreign-server metadata, bounded query behavior, and remote endpoint health.
Copy job Source read, target write, row validation, retries, and publish status.
Capacity Item and operation consumption in the Fabric Capacity Metrics experience.

A ready PostgreSQL endpoint doesn't prove that an integration is healthy. A healthy SQL analytics endpoint can still be behind the operational database.

Review item and operation state

Open the HorizonDB database and review the current lifecycle and most recent operation. For an asynchronous operation, preserve:

  • Operation ID.
  • Requested and effective state.
  • Current stage.
  • Last update time.
  • Retry guidance.
  • Request or correlation ID.

Don't treat a client timeout as operation failure. Reopen the same item or operation before submitting a duplicate mutation.

Monitor database activity

Use PostgreSQL catalog views to inspect activity while connected as the database creator. Other users and principals can't connect in the current preview.

Database statistics

SELECT
    datname AS database_name,
    numbackends AS active_connections,
    xact_commit,
    xact_rollback,
    blks_read,
    blks_hit,
    temp_files,
    temp_bytes,
    deadlocks,
    stats_reset
FROM pg_catalog.pg_stat_database
WHERE datname = current_database();

Use deltas over a known interval for cumulative counters. Check stats_reset before comparing two observations.

Current sessions

SELECT
    pid,
    usename,
    application_name,
    client_addr,
    state,
    wait_event_type,
    wait_event,
    xact_start,
    query_start,
    age(clock_timestamp(), query_start) AS query_duration,
    left(query, 200) AS query_preview
FROM pg_catalog.pg_stat_activity
WHERE datname = current_database()
  AND pid <> pg_backend_pid()
ORDER BY query_start NULLS LAST;

Visibility depends on PostgreSQL privileges. Query text can contain sensitive data.

Blocked sessions

SELECT
    activity.pid AS blocked_pid,
    activity.usename AS blocked_user,
    activity.wait_event_type,
    activity.wait_event,
    pg_blocking_pids(activity.pid) AS blocking_pids,
    age(clock_timestamp(), activity.query_start) AS blocked_duration,
    left(activity.query, 200) AS blocked_query
FROM pg_catalog.pg_stat_activity AS activity
WHERE activity.datname = current_database()
  AND cardinality(pg_blocking_pids(activity.pid)) > 0
ORDER BY blocked_duration DESC;

Don't cancel or terminate another user's session without authorization and an understanding of the transaction impact.

Largest customer tables

SELECT
    statistics.schemaname AS schema_name,
    statistics.relname AS table_name,
    pg_total_relation_size(statistics.relid) AS total_bytes,
    pg_size_pretty(pg_total_relation_size(statistics.relid)) AS total_size,
    pg_size_pretty(pg_relation_size(statistics.relid)) AS table_size,
    pg_size_pretty(
        pg_total_relation_size(statistics.relid)
        - pg_relation_size(statistics.relid)
    ) AS related_size
FROM pg_catalog.pg_statio_user_tables AS statistics
ORDER BY total_bytes DESC
LIMIT 20;

Interpret telemetry carefully

  • PostgreSQL statistics can reset or evict older entries.
  • Service samples can arrive late.
  • A paused database can't answer a live catalog query.
  • Missing values aren't zero values.
  • Elapsed execution time isn't exact CPU time.
  • Buffer and temporary-file counts aren't exact memory attribution.
  • The performance view isn't the billing ledger.

Use the Fabric Capacity Metrics experience for authoritative capacity consumption. Correlate it with lifecycle and workload evidence by item, operation, and time range.

Optional statement diagnostics

Use statement-level views only when the released service has installed the required extension and your identity has permission:

SELECT
    extname,
    extversion
FROM pg_catalog.pg_extension
WHERE extname = 'pg_stat_statements';

Don't try to install a diagnostic extension merely because it appears in pg_available_extensions. Fabric HorizonDB uses a curated extension catalog.

Build a troubleshooting record

For an issue, collect:

  • UTC timestamp and time range.
  • Database lifecycle state.
  • Available compute controls and the state displayed in your preview.
  • Sanitized session, blocking, or table-size evidence.
  • Relevant integration state.
  • Capacity metric time range.
  • Operation, request, and correlation IDs.
  • Expected and actual behavior.

Remove tokens, complete connection strings, real data values, and unreviewed query text.