← Sentinel query library

Telemetry health · Microsoft Sentinel

Investigate historically active sources that are now quiet

Find subscriptions and resource-log sources with sustained prior activity but no recent event-time rows, without treating quiet traffic as a proven ingestion outage.

Before you run

Data & scope

Required tables
AzureActivityAZKVAuditLogsStorageBlobLogs
Time window
Preceding 7-day activity baseline followed by a 4-hour quiet window
  • AzureActivity, AZKVAuditLogs and StorageBlobLogs must all exist and be queryable in this workspace; Analytics plans must support the union.
  • At least eight days of retained source coverage and an independent inventory of expected subscriptions and resources; this query only learns sources already observed.

Documentation review by Codex (AI). No live-tenant execution or detection-performance claim. Validate schema, data coverage and expected results in your workspace.

Kusto Query Language
Download .kql
// Hunting starter. Documentation checked 2026-11-13; not tenant-tested.
let observationEnd = now();
let quietStart = observationEnd - 4h;
let baselineStart = quietStart - 7d;
union
    (AzureActivity
    | where TimeGenerated >= baselineStart and TimeGenerated <= observationEnd
    | project EventTime = TimeGenerated, LogSource = "AzureActivity", ScopeId = tolower(SubscriptionId)),
    (AZKVAuditLogs
    | where TimeGenerated >= baselineStart and TimeGenerated <= observationEnd
    | project EventTime = TimeGenerated, LogSource = "AZKVAuditLogs", ScopeId = tolower(_ResourceId)),
    (StorageBlobLogs
    | where TimeGenerated >= baselineStart and TimeGenerated <= observationEnd
    | project EventTime = TimeGenerated, LogSource = "StorageBlobLogs", ScopeId = tolower(_ResourceId))
| where isnotempty(ScopeId)
| summarize PriorRows = countif(EventTime < quietStart),
    PriorActiveHours = dcount(iff(EventTime < quietStart, bin(EventTime, 1h), datetime(null))),
    PriorActiveDays = dcount(iff(EventTime < quietStart, startofday(EventTime), datetime(null))),
    RecentObservedRows = countif(EventTime >= quietStart), LastObservedEvent = max(EventTime)
    by LogSource, ScopeId
| where PriorRows >= 100 and PriorActiveHours >= 24 and PriorActiveDays >= 4
| where RecentObservedRows == 0
| extend HoursSinceObservedEvent = round(toreal(datetime_diff("minute", observationEnd, LastObservedEvent)) / 60.0, 2),
    Finding = "Historically active; no event-time rows observed in last 4 hours"
| project LogSource, ScopeId, Finding, LastObservedEvent, HoursSinceObservedEvent,
    PriorRows, PriorActiveHours, PriorActiveDays, RecentObservedRows
| order by HoursSinceObservedEvent desc, PriorRows desc
| take 200

Read-only query. Run in your own authorized workspace after checking prerequisites.

This diagnostic learns a candidate source population from the preceding seven days, then identifies members with no event-time rows in the last four hours. Candidates must have at least 100 earlier rows across at least 24 distinct hourly bins and four distinct UTC dates. These thresholds reduce obviously sporadic sources, but they do not establish an expected service-level objective.

Understand what “quiet” means

The comparison uses TimeGenerated, not ingestion time. A source can send old records today and still appear quiet for the recent event-time window. Conversely, a recent event does not establish complete delivery of all expected records. Source-specific buffering and processing delays need separate analysis. Azure Monitor ingestion latency.

The baseline spans seven elapsed days, which can intersect eight UTC calendar dates at its edges. PriorActiveDays and PriorActiveHours count occupied bins approximately; they are activity measures rather than uptime measurements. No source is labelled failed automatically.

This learned population cannot reveal a resource that has never logged, a newly expected source with no history, or a source absent for the entire scan. Maintain an authoritative coverage inventory and compare it separately. Do not treat the returned list as the complete monitoring estate.

Scope the evidence properly

AzureActivity is grouped by subscription, while Key Vault and Blob logs use their recorded resource identifiers. One busy resource cannot hide silence from another resource in those two tables. However, one active category or producer can still hide a missing category within the same scope.

All three tables must exist and be accessible. The query deliberately avoids fuzzy resolution that could silently omit an unavailable table. The table references define the subscription field and resource identifiers for Key Vault and Blob Storage.

Investigate and validate

Check workload activity, maintenance windows, export settings, destination permissions, retention and platform health. A quiet overnight workload and disabled logging can look identical until external evidence resolves the cause. Preserve the last observed timestamp and the actual gap in investigative coverage.

Validate with a naturally quiet source, a known continuously active source and a test dataset whose recent rows are removed. Confirm that never-observed sources remain outside the result and are caught by your separate inventory process. Choose thresholds for each source’s normal cadence before operational use.

This recipe returns up to 200 candidates from a bounded scan. It is documentation-checked, has not been executed in a tenant, and does not declare connector availability or prove the absence of underlying business activity.

Verify the source

Microsoft references

  1. Microsoft Learn — AzureActivity table ↗
  2. Microsoft Learn — AZKVAuditLogs table ↗
  3. Microsoft Learn — StorageBlobLogs table ↗
  4. Microsoft Learn — Log ingestion latency ↗
Before turning a hunt into an alert

Measure normal activity, validate known test cases, account for ingestion delays, choose a threshold and entity mapping, and assign an owner. Results require investigation.

Follow the detection learning path →