Fix potential deadlock + use-after-free in WatchForExitedProcesses (#…

Welcome to F# Weekly,

A roundup of F# content from this past week:

News

• F# in .NET 11 RC 2 – Release Notes
• Performance improvements in the F# 11 compiler

Microsoft News

• Introducing dotnetup: A toolchain manager for .NET
• Announcing .NET 11 Release Candidate 2
• Durable Task in the Real World: What Problems Does It Actually Solve?
• Stop waiting on a failed build – Visual Studio Blog
• What’s new in Microsoft Foundry | September 2026 | Microsoft Foundry Blog

Videos

• .NET Unboxed: .NET 11 Release Candidate 2
• Houston Haynes – The Clef Programming Language – A Ship of Theseus Story – YouTube
• .NET + Foundry, better together
• .NET AI Community Standup: Build agentic experiences with the Blazor AI Components – YouTube

Blogs

• Event Sourcing: A close look at the projector for values – planetgeek.ch

Highlighted projects

• AngelMunoz/MVUish.Avalonia – lightweight, performant Model-View-Update (MVU) library for Avalonia, designed for both C# and F#.
• EluciusFTW/SpectreTuff: A thin, opinionated wrapper around Spectre.Tui in F#.
• WillEhrendreich/SageFs: Sage Mode for F# development — REPL with solution or project loading, Live Testing for FREE, Hot Reload, and session management.
• TheAngryByrd/fsharp-skills – Reusable instructions for AI agents working on F# code

New Releases

• Fable 5.22.0
• NBomber 6.7.1
• SageFs 0.6.908
• FsUnit 8.0.0
• Expecto 11.1.1
• Thoth.Json 10.5.2
• Farmer 1.9.33
• WebSharper 10.1.6.678
• MVUish.Avalonia 1.0.1

That’s all for now. Have a great week.

The Security Update for SQL Server 2025 RTM CU9 is now available for download at the Microsoft Download Center and Microsoft Update Catalog sites. This package cumulatively includes all previous security fixes for SQL Server 2025 RTM CUs, plus it includes the new security fixes detailed in the KB Article.

• Security Bulletins:
• CVE-2026-32176 - Security Update Guide - Microsoft - Microsoft SQL Server Denial of Service Vulnerability
• Security Update of SQL Server 2025 RTM CU9 KB Article: KB5126967
• Microsoft Download Center: https://www.microsoft.com/download/details.aspx?familyid=6d582614-a5b0-4cfa-b1f5-ef5c74f5d252
• Microsoft Update Catalog: https://www.catalog.update.microsoft.com/Search.aspx?q=5126967
• Latest Updates for Microsoft SQL Server: https://learn.microsoft.com/en-us/troubleshoot/sql/releases/download-and-install-latest-updates

The Security Update for SQL Server 2025 RTM GDR is now available for download at the Microsoft Download Center and Microsoft Update Catalog sites. This package cumulatively includes all previous security fixes for SQL Server 2025 RTM, plus it includes the new security fixes detailed in the KB Article.

• Security Bulletins:
• CVE-2026-32176 - Security Update Guide - Microsoft - Microsoft SQL Server Denial of Service Vulnerability
• Security Update of SQL Server 2025 RTM GDR KB Article: KB5128492
• Microsoft Download Center: https://www.microsoft.com/download/details.aspx?familyid=05d38ba3-f609-4011-b045-d56a9eb0bccb
• Microsoft Update Catalog: https://www.catalog.update.microsoft.com/Search.aspx?q=5128492
• Latest Updates for Microsoft SQL Server: https://learn.microsoft.com/en-us/troubleshoot/sql/releases/download-and-install-latest-updates

As AI projects move from pilots to production, model deployments often spread across subscriptions, regions, and application teams. A practical operations dashboard gives platform, engineering, governance, and finance teams one shared view of what is deployed, how it is used, what it costs, and where quota or capacity risk is forming. 

The operating problem 

Microsoft Foundry brings agents, models, tools, monitoring, tracing, evaluations, RBAC, networking, and policy controls under a unified platform surface. That is a strong foundation, but enterprise operations still need a view that cuts across the way Azure is usually organized: multiple subscriptions, resource groups, regions, business units, and application teams. 

The goal is not to replace product dashboards. The goal is to add a management-group or tenant-wide operating layer that answers simple but urgent questions: Which models are deployed? Which teams own them? Which deployments use Standard, priority, batch, or provisioned throughput? Where are token usage and PTU utilization rising? Which subscriptions are approaching quota pressure? Which costs should be allocated to which product or environment? 

Design principle: keep the dashboard about deployment and operations metadata. Do not collect prompts, completions, customer data, or other application payloads unless a separate privacy, security, and compliance design explicitly requires it. 

A practical reference architecture 

Use three data planes.  

1. First, build the inventory plane with Azure Resource Graph, because Resource Graph is designed for efficient, large-scale querying across subscriptions and tenants. Combine account-level resource data with deployment-level data from the Azure AI Services deployments API, which lists deployments associated with a Cognitive Services account. Normalize subscription, resource group, region, account, deployment name, model name, model version, deployment type, SKU or capacity, tags, owner, environment, and business unit. 
2. Build the telemetry plane with Azure Monitor. Public Azure OpenAI monitoring guidance describes dashboards for HTTP requests, token-based usage, PTU utilization, and fine-tuning. Azure Monitor supported metrics and resource log references provide the authoritative place to check what is available for each resource type and how exportability works. For exploratory operations views, Azure Workbooks are a good fit because they combine logs, metrics, parameters, Azure Resource Graph, Azure Resource Manager, JSON, and other data sources in one interactive report. 
3. Build the finance plane with Cost Management APIs or exports. Cost Management automation guidance supports retrieving and managing cost data, including for larger datasets that need offline analysis. The Cost Management Query API supports usage queries at scopes such as subscription, resource group, billing account, management group, billing profile, and invoice section, subject to permissions. Power BI can be useful for leadership reporting, but the Cost Management connector has agreement constraints; direct connector support is documented for Enterprise Agreement and direct Microsoft Customer Agreement customers, while unsupported agreements should use exports. 

Separate inventory, usage, cost, and quota 

Do not treat every signal as the same kind of data. Inventory is mostly current-state metadata. Usage is time-series behavior. Cost is billing-scope data that may lag and must be joined carefully by resource identity, meter, and tags. Quota is a capacity-control signal, not a spend signal. Azure OpenAI quota is not enforced at tenant level; Microsoft Learn states that the highest quota restriction is scoped at the Azure subscription level. That means an executive dashboard can summarize quota posture broadly, but engineering views should still preserve subscription, model, region, and deployment-type detail. 

Provisioned throughput deserves a dedicated view. Microsoft documents PTU as dedicated model processing capacity for predictable throughput and latency, billed per PTU per hour or through reservations. That makes PTU utilization, idle capacity, traffic predictability, and latency requirement visible design inputs rather than after-the-fact finance questions. 

Dashboard views that matter 

Start with a few durable views rather than a large reporting program. The most useful first version usually includes: 

• Executive summary: deployment count, model families, regions, monthly cost trend, and quota risk. 
• Inventory: deployments by subscription, resource group, region, model, version, deployment type, owner, and environment. 
• Operations: request volume, token trends, errors, latency indicators, and PTU utilization where available. 
• Cost allocation: spend by owning team, application, environment, and deployment type. 
• Governance hygiene: missing tags, ownerless deployments, old model versions, and regional concentration. 

Decision guidance 

Choose the reporting surface by audience. Platform engineers usually need Workbooks because they support operational drill-down and can combine Azure Monitor and Resource Graph. Finance teams often need Power BI because it supports broader cost modeling and distribution. Application owners need a small scorecard: their deployments, their usage, their costs, their quota posture, and their next required action. 

Production considerations and limitations 

• Require tags before reporting on ownership. Dashboards cannot create accountability if resources do not carry owner, application, environment, and cost-center metadata. 
• Use least-privilege access. Cost, inventory, and telemetry sources may require different permissions and scopes. 
• Check portal-version language. Some public monitoring guidance is explicitly marked for the Foundry classic portal, so avoid assuming identical behavior in every new portal experience. 
• Verify regional availability. Model availability varies by region, and deployment lifecycle settings such as automatic model updates should be reviewed before making standardization claims. 
• Respect metric export limits. Azure Monitor notes that some metrics have export constraints, especially around multidimensional metrics and diagnostic settings. 

Implementation checklist 

• Define the dashboard audience and the decisions it must support. 
• Inventory Foundry and Azure OpenAI resources across subscriptions with Azure Resource Graph. 
• List deployments per account and normalize deployment metadata. 
• Route or query Azure Monitor metrics and logs needed for usage and PTU views. 
• Add Cost Management queries or exports at the correct billing scope. 
• Join by resource ID and tags; flag missing ownership metadata. 
• Publish an Azure Workbook for operations and a Power BI model only where agreement support and permissions fit. 
• Review model versions, quota posture, and PTU utilization on a fixed cadence. 

Conclusion 

A tenant-wide model operations dashboard is a practical bridge between AI experimentation and AI platform management. It does not need to be complex on day one. Start with inventory, usage, cost, quota, and ownership. Keep application content out of scope. Preserve subscription-level quota detail. Then evolve the dashboard into a repeatable operating rhythm for model lifecycle, cost control, capacity planning, and governance.

When it comes to classic Windows games, for me Minesweeper. Now you can pop into some fun mine sweeping in your system tray with Mini Sweeper! This is a faithful recreation of the original minesweeper but with some extra updated visuals and animations build with Windows Composition APIs. If you are really aching for the battleship gray look, you can swap over to the classic style in settings.

As I’ve mentioned on the previous 2 Saturdays, little games in the tray are meant to keep you on the same task when you face a delay. Instead of multi-tasking or grabbing your phone, both of which can lead to complete distraction, Mini Sweeper does not have depth to further distract you. Play a quick round or two, then get back to your task.

Happy Tray-tober!

Joe