A developer monitors a cloud database dashboard with rising metrics and a red alert in a moonlit office.
Microsoft's Ignite announcement of Azure HorizonDB's general availability came with a quieter companion entry in the Azure Updates feed. That entry is "Generally Available: Azure HorizonDB Monitoring via Alerts". It is the sort of item that doesn't make keynote slides, but it matters to anyone who will be paged at 3 a.m. when a database misbehaves.

A developer monitors a cloud database dashboard with rising metrics and a red alert in a moonlit office. What Microsoft announced​

The entry is dated November 17, 2026 and carries the "Launched" status. On the Azure Updates page, that label means fully released, production-ready and available to all Azure customers.

The title says HorizonDB now supports monitoring through alerts. The entry text I could retrieve was only the page chrome, so it gives no walkthrough, thresholds, notification options, pricing or limits. A third-party tracker summarizes the feature as Azure Monitor alert rules built on HorizonDB metrics with configurable thresholds. That is consistent with the title, but it is secondary and not confirmed from the Microsoft entry itself.

The same tracker also shows a "generally available" date of November 30, 2026. That conflicts with the "Launched" status on November 17. I can't resolve the conflict from the available material. Don't plan around either date until you see the feature in your own subscription.

The context: HorizonDB just went GA​

Microsoft announced at Ignite on November 17, 2026 the general availability of Azure HorizonDB and the public preview of Microsoft Fabric HorizonDB. The service is generally available in 13 Azure regions across the Americas, Europe and Asia Pacific, with more coming online soon. Microsoft's blog also lists metrics, wait events, top queries and active sessions in its developer tooling for investigating slow queries and blocking.

Alerting fits naturally into that launch. A production database service needs a way to tell you something is wrong without someone watching a dashboard.

What you can plausibly alert on​

Azure Monitor's own guidance is that metric alerts are the right choice when the data you want to monitor is available in metric data. Microsoft Learn's supported-metrics reference for HorizonDB clusters lists the metrics available. Microsoft hasn't said which of them are alertable, so treat the following as candidates, not confirmed alert targets.

  • Errors: Failed Connections (FailedConnections) counts failed connection attempts.
  • Health: PostgreSQL Instance Is Alive (PostgreSqlInstanceIsAlive) shows whether the PostgreSQL process is running.
  • Compute: CPU Utilization and Memory Utilization report host-level percentages.
  • Concurrency: these metrics cover pressure and long-running work.
    • Active Connections and Maximum Connections.
    • Oldest Query, Oldest Transaction and Oldest Backend, all reported in seconds.
    • Sessions by State and Sessions by Wait Event Type.
    • Deadlocks.
  • Wraparound risk: Maximum Used Transaction IDs is described as a way to monitor proximity to XID wraparound.
  • Storage and replication: the list includes Storage Used, Backup Used, Max Logical Replication Lag, Commit Latency and Write Latency.
  • Vacuum health: Bloat Percent and Estimated Dead Rows in User Tables are listed.

Most entries are sampled at one-minute grain (PT1M). Others, including Database Size, Maximum Connections and the vacuum-related metrics, are only available at 30-minute and coarser grains. That matters for alert design. A one-minute metric suits "is it down right now?" rules. A 30-minute metric will not catch a fast-moving problem.

Many metrics expose dimensions such as Pool, Node and Role, and some add Database, State or WaitEventType. Whether alert rules can split on these dimensions in HorizonDB isn't stated in the material I have. Azure Monitor metric alerts generally support dimensions, but check this in the portal before relying on it.

Any threshold you pick is your own call. Microsoft hasn't published recommended values for HorizonDB.

Metrics versus logs​

These two signal types are easy to confuse:

  • Metric alerts evaluate numeric time series. Azure's alert-type guidance says data available in logs, or needing advanced logic, can use KQL through log search alerts.
  • Log search alerts are a separate route for HorizonDB's server logs. Microsoft Learn documents example Log Analytics queries covering errors, fatal messages, deadlocks, restarts and connection events. Some of those queries are annotated as usable for alert rules.

The Learn metrics page also shows "DS Export: No" for almost every listed metric. The exception I found is Deadlocks, which shows "Yes". So don't assume you can pull most HorizonDB metrics into Log Analytics through diagnostic settings and then alert on them with KQL. Use the metric alert path for metrics, and use server-log queries for log events.

What's still unknown​

Several practical details are unconfirmed:

  • Portal navigation and any API or template requirements.
  • Eligible configurations and regions. HorizonDB is in 13 regions, but nothing says every region supports this feature.
  • Evaluation frequency and aggregation options per rule.
  • Notification channels. These are normally Azure Monitor action groups, but that's general Azure Monitor behavior, not something this entry confirms.
  • Billing. Azure Monitor alert rules are generally billed, but I have no HorizonDB-specific pricing.

Practical takeaways​

  1. Check your HorizonDB cluster's alerts experience in Azure Monitor and confirm the metrics you care about are selectable.
  2. Start with a short list: process health, failed connections, connection saturation against Maximum Connections, and long-running query or transaction age.
  3. Watch Maximum Used Transaction IDs on busy clusters, since wraparound is a classic PostgreSQL failure mode.
  4. Pair metric alerts with log-based alerts for events like restarts and deadlocks.
  5. Test your action group by triggering a deliberately low threshold in a non-production cluster before trusting it.

Bottom line​

This is a modest but useful addition. HorizonDB now has a Microsoft-labeled GA path to Azure Monitor alerting, and the documented metric catalog is broad. The thin announcement text, the conflicting date on a secondary tracker, and the lack of threshold or pricing guidance mean you should verify the details in your own tenant before building on-call rotations around it.

 

References

  1. (Launched) Generally Available: Azure HorizonDB Monitoring via Alerts Azure Updates 2026-11-17T20:00:32Z
  2. Supported metrics - Microsoft.HorizonDB/clusters - Azure Monitor | Microsoft Learn learn.microsoft.com
  3. HorizonDB: PostgreSQL for enterprises, developers, and agents on Azure and Fabric techcommunity.microsoft.com