Milvus Insight

Updated at:
Copy as MD

Milvus Insight is a built-in diagnostic feature in Milvus Manager that consolidates instance overview, operation audit, query details, and async task tracking into a single entry point. Use Milvus Insight to locate slow queries, failed requests, hot collections, and async task execution progress for routine operations and troubleshooting.

Features

TabDescription
OverviewDisplays key metrics and anomaly trends for the instance over the last 24 hours, including DDL/DCL operations, DML calls, DQL calls, slow queries, active collections, active users, and security alerts.
Operation AuditQuery operation records by time range, collection, database, user, operation type, operation, and status. View execution status, duration, and request details.
Query DetailsView the filter conditions, duration, nq, slow query levels, and execution status of Query, Search, and HybridSearch requests.
Async TasksView the status, progress, duration, and failure reasons of async tasks such as LoadCollection, IndexBuild, Compaction, Import, and Sync.

Prerequisites

  • The instance version must meet the minimum version requirements for Milvus Insight: version 2.5.12-0.7.0_3.11.0 or later for the 2.5 series, and version 2.6.18-0.1.6_3.11.0 or later for the 2.6 series. If the page indicates that the current instance version does not support Milvus Insight, click Upgrade to upgrade the instance to the latest version, and then view the observability and diagnostic data after the upgrade is complete. For upgrade instructions, see Upgrade versions.

After upgrading the instance to a version that supports Milvus Insight, Milvus Insight only displays observability data generated after the version that supports this feature.

Access Milvus Insight

  1. Log on to the Alibaba Cloud Milvus console.

  1. In the left-side navigation pane, click Instances.

  1. In the top navigation bar, select the target region.

  1. In the instance list, click the name of the target instance.

  1. At the top of the instance details page, click Milvus Manager and connect to the target instance.

  1. In the left-side navigation pane of Milvus Manager, click Milvus Insight.

The page displays the Overview, Operation Audit, Query Details, and Async Tasks tabs. If a version upgrade guide page is displayed, the current instance version does not support Milvus Insight. Upgrade the instance version first.

View instance overview

On the Overview tab, view the operating status of the instance over the last 24 hours.

MetricDescription
DDL/DCL OperationsNumber of calls for structure and permission operations such as collections, indexes, and permissions.
DML CallsNumber of calls and affected rows for write operations, broken down by Insert, Upsert, Delete, and Bulk.
DQL CallsNumber of calls for query operations, broken down by Search, Query, and HybridSearch.
Slow QueriesNumber of requests that reached the slow query threshold. For threshold classification, see the Slow query levels section in this document.
Active CollectionsNumber of collections that have been accessed in the last 24 hours.
Active UsersNumber of users who have had access records in the last 24 hours.
Security AlertsRisk alerts generated based on access behavior and anomalies. These serve as troubleshooting aids and do not constitute comprehensive security audit or intrusion detection results.

The page also displays TOP 5 Active Users, TOP 5 Hot Collections, and Anomaly/Slow Query Trend charts. The legend in the trend chart includes Slow, Very Slow, Critical, and Error, which can be used to identify the time periods when anomalies are concentrated.

Query operation audit

On the Operation Audit tab, query the operation records of the instance by conditions. Set the time range, collection, database, and user in the query bar at the top of the page. Set the operation type, operation, and status through the filter icon in the table header. After the settings are complete, click Query.

FilterDescription
Time RangeThe time range when the operation occurred. Default is the last 1 hour.
CollectionFilter by collection name.
DatabaseFilter by database name.
UserFilter by the username that performed the operation.
Operation TypeFilter by DDL, DCL, DML, DQL, or Other. Multiple selections are supported.
OperationFilter by specific operation name, such as CreateCollection, Insert, Flush, or LoadCollection. You can toggle between Include and Exclude at the top of the filter panel. Selecting Exclude means the selected operations are excluded from the query results, for example, excluding Flush.
StatusFilter by Success or Failed.

The list displays the time, operation type, operation, collection, database, user, status, and duration of each operation. When an operation does not involve a collection (such as Flush), the Collection column displays -.

Click Details in the Operation column to view the complete context of the operation in the Audit Log Details dialog box, including client address, SDK version, error code, error message, trace ID, response size, and request parameter details. The client address reflects the call source: operations initiated through SDK display the client IP and port, while operations initiated in Milvus Manager display as Milvus Manager Console.

To analyze data offline, click Export CSV to export all records matching the current filter criteria.

The column headers in the exported file are the original API field names, such as traceId, logTime, timeCostMs, and operationType, which differ from the column names displayed on the interface.

View query details

On the Query Details tab, view the execution status of Query, Search, and HybridSearch requests. This tab is suitable for locating slow queries, failed queries, and hot access patterns.

In addition to the same filter criteria as Operation Audit, the query bar also supports Minimum Duration (ms) to view only requests with a duration exceeding the specified milliseconds. The table header also supports filtering by Slow Query Level.

The list displays the time, operation, collection, database, user, filter conditions, duration, nq, slow query level, and status of each request.

Click Details to view the complete query context in the Query Details dialog box, including partition, number of query vectors, whether it is a slow query, whether the result is empty, filter expression, request parameters, SDK version, client address, trace ID, and error message. To analyze data offline, click Export CSV to export records matching the current filter criteria.

View async tasks

On the Async Tasks tab, view the execution status and progress of instance async tasks. The list displays the task ID, task type, collection, database, start time, end time, duration, progress, and status. The Data Update Time is displayed at the bottom of the page.

Task TypeDescription
ImportData import task.
LoadCollectionCollection loading task.
SyncData synchronization task.
CompactionData compaction task.
IndexBuildIndex building task.

You can filter by Task Type and Status. Status options include Running, Completed, Failed, Timeout, and Canceled. Click Details to view the task type name, collection ID, current status, whether it is a final state, failure reason, instance ID, and original task details in the Async Task Details dialog box. This helps troubleshoot task failures or long-running tasks. To analyze data offline, click Export CSV to export records matching the current filter criteria.

Different task types may provide different fields. If a task does not have collection, database, end time, or progress information, the corresponding field displays -.

Slow query levels

Milvus Insight classifies slow query levels based on the request processing duration (time_cost) recorded in the Milvus server ACCESS logs.

Slow Query LevelServer Request Processing Duration
NORMALLess than 1,000 ms.
SLOWGreater than or equal to 1,000 ms and less than 3,000 ms.
VERY_SLOWGreater than or equal to 3,000 ms and less than 10,000 ms.
CRITICALGreater than or equal to 10,000 ms.
  • The Slow Query Level filter only provides the SLOW, VERY_SLOW, and CRITICAL options. NORMAL indicates that the slow query threshold has not been reached and only appears as a value in the list.

  • The duration and slow query level displayed on the page are based on the Milvus server records and are not equivalent to the end-to-end duration of the client SDK call. Factors such as network transmission, client deserialization, and large result set transmission may cause the SDK-side duration to be longer.

Limits

  • Maximum time span per query: Query Details supports a maximum of 1 hour, Operation Audit and Async Tasks support a maximum of 24 hours. Queries exceeding the limit will fail. Narrow the time range and try again.

  • All three tabs query data from the last 1 hour by default. Operation Audit and Async Tasks can be manually adjusted to a longer time range within the limit.

  • Operation Audit, Query Details, and Async Tasks can export a maximum of 10,000 records per export. If the number of records matching the filter criteria exceeds the export limit, a prompt is displayed on the page. Narrow the time range or add filter criteria before exporting again.

  • Overview displays aggregated data for the last 24 hours. The Data Snapshot Time is displayed at the bottom of the page. For newly created instances or instances that have just been upgraded to a version that supports Milvus Insight, a period of data accumulation is required before complete aggregation results are available.