SQL Server: follow the connection beyond query completion
A correct query result does not close a capacity assessment. Observe both ends of the flow and separate business continuity from the decision about a fix.
Ler em portuguêsBy Pentesty
A query result is only one measure
Microsoft’s Windows 11 dashboard confirms a regression following October 13, 2026 updates and later: linked-server queries using OLE DB providers can finish while external connections remain open. Oracle and PostgreSQL are among the reported destinations. Accumulation can degrade performance over time.
Consider a planning scenario in which billing queries an external database for every batch. The first run may complete, while the team still needs to observe what remains before the next run. Measure job completion and residual occupancy separately, with owners of both databases involved in the decision.
Map the path between both databases
In Microsoft’s documented linked-server architecture, SQL Server sends requests through an OLE DB provider to an external source. The definition includes a provider and data source, with product-specific requirements. Map the destination and intermediary component alongside the local database instance.
For triage, connect each business process to its destination, peak demand period, and the owner who can reduce its load. Preserve the updated Windows identity and the provider in use. That map supports a selective pause instead of interrupting every job sharing the same instance.
Observe the destination without creating another workload
PostgreSQL pg_stat_activity distinguishes active, idle, and idle-in-transaction sessions. backend_start, query_start, and state_change help relate connection and activity; visibility depends on permissions. That supports destination-side observation without assuming that an idle connection proves this Windows regression.
Define an observation interval and an authorized workload representing the real process. Compare periods before, during, and after a batch, including user-visible errors. Prefer aggregate measurements approved by the database owner. Avoid distributing query text or customer data in an executive availability report.
Treat temporary relief as an operational decision
The Windows Server 2025 dashboard says closing identified connections or restarting SQL can provide temporary relief while interrupting queries or transactions. Enterprise mitigation is available through Microsoft support. At our November 12 check, an out-of-band fix was still described as forthcoming.
Before intervening, agree who decides about billing already in progress, how to recognize an incomplete run, and who authorizes resumption. Record the vendor case and the conditions attached to its mitigation. A temporary reduction in occupancy should be followed by observation during a comparable business window.
Report the obligation that remains open
Track three decisions separately: keeping the service running, controlling observed accumulation, and evaluating a future fix. Progress on one does not close the others. Show the business owner dependent processes, observed capacity, the chosen intervention, and the next review point rather than only a list of updated machines.
When an applicable package becomes available, compare behavior before and after across more than one process cycle. Keep an explanation of workload differences and observations from both ends of the connection. Resuming normal operations can then depend on the business flow that needed to be preserved.
Translate observations into a business choice about upcoming batches: what can wait, which customers depend on processing, and when the team will inspect the destination again. Also agree who may pause automatic retries after an incomplete run. Establish that decision before an intervention, so pressure to finish the job does not determine resumption on its own. Give the application owner enough context to explain that choice to affected users.
Frequently asked questions
Does every idle connection confirm the problem?
No. Correlate linked-server workload, provider, Windows, and behavior over time; a single session does not establish the cause.
Does restarting SQL Server permanently resolve it?
The vendor describes temporary relief and interruption risk. Resumption needs observation, and mitigation should follow support guidance.
Does monitoring require unrestricted access?
Agree on required measurements and permissions with the database owner. This proposal does not require publishing queries or customer data.
Should every job be canceled?
Map affected flows and agree on priorities with their owners. A broad interruption needs its own justification beyond the existence of an advisory.
Sources and references
- Microsoft: connection regression
- Microsoft: mitigation and follow-up
- Microsoft: linked-server architecture
- PostgreSQL: session observation
Original Pentesty analysis, published November 12, 2026. The proposed scenarios are editorial guidance, with scope and validation to be defined for each environment.