SQL linked queries: test beyond restart
Practical perspective from an IT leader working across operations, security, automation, and change.
8 minute read with practical, decision-oriented guidance.
Leaders and operators looking for concise, actionable takeaways.
Topics covered
SQL linked queries need to work throughout the period your business relies on them. If a report succeeds immediately after a restart but slows down as it runs repeatedly, the first success does not establish that its normal schedule is dependable. Agree the workload and evidence needed with your application or database provider before accepting a return to normal operation.
That distinction matters for the connection issue Microsoft confirmed in November 2026. Its notice describes a problem that can emerge over prolonged operation. For a business owner, the useful decision is whether the affected report or application can sustain the work it must do, with an explicit restriction where that remains unproven.
Check whether the dependency applies
SQL Server is Microsoft's database platform. A configured linked server lets it request data from another data source. An OLE DB provider is software used for that connection. A reporting application might use this arrangement to combine information held in SQL Server with records in an external database.
You do not need to operate the database to identify the business dependency. Ask the existing application or database owner which reports use that route, how often they run and which decisions depend on their results. They should then check the actual platform, update history and provider against Microsoft's notice. Using SQL Server alone does not establish that the issue applies.
In the Windows Server release-health notice, Microsoft says that after Windows updates released from 13 October 2026, connections created by SQL Server linked-server queries using OLE DB providers can remain open after queries finish. Oracle and PostgreSQL connections are among those affected. Repeated queries can accumulate connections and degrade performance after prolonged operation.
The notice remained marked Confirmed when checked on 13 November. Microsoft described an out-of-band update as planned; it was not shown as released. The Server 2025 originating update details do not automatically apply to other platforms. Your provider needs to check the entry relevant to your environment rather than treating one update number as universal.
This is also not a diagnosis of every slow report. Workload changes, other dependencies and configuration may need investigation. The provider must interpret observations from your system rather than infer the cause from a similar symptom.
Separate temporary relief from acceptance
Microsoft describes closing identified idle connections or restarting the SQL service as temporary relief. Those actions can interrupt active queries or transactions, and the issue can recur after sufficient operation. They are decisions for the responsible technical team under a controlled plan, not instructions for a business user to try during the working day.
Microsoft directs affected enterprise customers to Microsoft Support for Business for a mitigation. The public notice does not supply a universal implementation that this article can reproduce. Ask your provider to establish applicability, obtain the appropriate advice and explain the scope of any proposed change.
A restart followed by one successful report answers a narrow question: did that report complete then? Acceptance needs to answer a different question: can the workflow deliver usable results at the frequency, volume and duration the business requires?
Define those requirements before the check. A daily planning process may need repeated reports before allocation deadlines. A continuous integration may need evidence beyond a working day. Expected data size and concurrent work also matter. The technical owner should choose a safe, representative test arrangement and explain what it does not reproduce. Do not create a production load test merely to see whether the service fails.
A fictional report that passes once and fails later
Consider a fictional workshop whose capacity report combines scheduling records in SQL Server with information in an external PostgreSQL database. Priya, the planning manager, decides whether staff can rely on the report. Sam, its existing application and database provider, owns the technical investigation and any authorised change.
All names, timings, findings and criteria in this example are invented. They illustrate a decision, not a client result, a Microsoft test or recommended database thresholds.
The report runs every 30 minutes during an eight-hour planning day. A representative rehearsal therefore contains 16 scheduled executions. Priya requires each result within five minutes so planners have time to check it before their next allocation decision. Sam defines separate connection observations appropriate to this test environment. Meeting the business deadline and meeting the technical criteria are both required.
After an already authorised temporary restart in this fictional incident, the first report completes in two minutes. Priya does not approve the normal schedule from that result. Sam continues with the agreed repeated workload in an isolated test environment.
The first 12 reports meet the five-minute deadline. Runs 13 to 16 each take eight minutes. Sam's observations show connections associated with the tested dependency accumulating across the repeated executions. Those findings support the investigation in this example. They do not mean that every open connection in another system is faulty.
Four of the 16 reports missed the planning deadline. Priya withholds unrestricted acceptance even though the first run looked satisfactory. A record that simply said "report works after restart" would omit the failure that affects the business.
Keep the failed decision owned
While Sam seeks applicable support advice, Priya suspends automatic publication of the combined report. Planners use an already checked capacity view in the underlying operational system for the limited work it covers.
That view omits cross-site capacity. Cross-site commitments therefore wait for confirmation from the responsible planner. If the underlying view is unavailable or too old for the allocation decision, the decision waits too. Staff do not substitute an unverified stale report because it is easier to obtain.
This arrangement takes staff time and delays some commitments. Priya accepts those costs for the interim period; she has not declared the alternative equivalent to the combined report. The business continuity guide explains how to define reduced operation and the point at which its limits become unacceptable.
Reducing the report frequency would be another proposal to assess, but it would also give planners older information. It cannot be assumed to fix the underlying issue. Repeated restarts would bring their own disruption and could hide recurrence between checks. Waiting for the planned update would still leave Priya needing a workable interim decision.
Retest the agreed workload and limit the conclusion
In this fictional example, Sam obtains applicable support advice and implements a scoped mitigation in the controlled test environment. Its mechanism is deliberately unspecified because the inspected public notice does not describe it. The following results are invented possible outcomes, not evidence that a supplied Microsoft mitigation passed a real test.
Sam repeats the agreed 16-run rehearsal. Every scheduled report finishes within three minutes. The connection observations meet the criteria agreed before the rehearsal, and Priya checks the expected test results. There is no restart or manual connection cleanup between runs that could conceal recurring behaviour.
The second rehearsal supports a narrower conclusion than "the database is fixed". It demonstrates this report's eight-hour schedule under the recorded conditions. It does not establish that a higher query frequency, different data volume or continuous overnight operation will behave the same way.
Priya accepts a monitored return for the demonstrated schedule, subject to Sam's separate production change approval. Sam owns observation of the actual agreed workload and reporting of recurrence. Priya owns the operational decision if the report again becomes unreliable. The limited planning arrangement remains available while that return is assessed.
A report missing the agreed five-minute deadline, a failed result check or connection behaviour outside Sam's agreed criteria reopens the decision. Priya restores the relevant operating restriction while Sam investigates and obtains further advice. If the second rehearsal had failed, the original hold would have remained. A planned completion date would not have converted the failed check into acceptance.
A future released update also needs its own applicability assessment and evidence. Neither a successful rehearsal nor a monitored return proves indefinite stability. The acceptance record should say which workload was demonstrated, which conditions remain untested and who acts when the evidence changes.
Bring the dependency into the system brief
Ask your provider for the affected workflow, its normal schedule, the observed failure and the evidence supporting the proposed return. Add the business deadline and the consequence of missing it. This gives both sides a concrete decision to resolve rather than a general request to make the database faster.
My web development and business systems service covers agreed integrations, database design and performance checks matched to the expected data and workload, with launch checks and documented handover. A consultation can help scope the workflow dependency, acceptance checks and application responsibilities alongside your existing provider. Hosting, maintenance and ongoing support require an explicit agreement. This is not an emergency SQL Server support service or a promise of a particular incident remedy.
Discuss your business-system dependency and acceptance checks with a short description of the report, its schedule, the business deadline and your provider's findings. Keep credentials, customer records and detailed diagnostic evidence out of the initial enquiry. We can establish the appropriate scope and responsibilities before any work begins.
Share this post
About the author
Daniel J Glover
IT Leader with experience spanning IT management, compliance, development, automation, AI, and project management. I write about technology, leadership, and building better systems.
Continue exploring
Keep building context around this topic
Jump to closely related posts and topic hubs to deepen understanding and discover connected ideas faster.
Explore topic hubs
Related article
Order reconciliation after a shop fix
Order reconciliation after a shop security fix means checking payments, order changes and dispatch evidence before deciding what staff may fulfil or correct.
Related article
Remote Desktop: restore staff access
A Remote Desktop connection that works for an administrator may still fail your staff. Check the business task before accepting a security workaround.
Related article
AI ads phishing: check before connecting
AI ads phishing makes account connections a business decision. Verify the supplier, challenge requested access and agree a narrower reporting workflow.
Related article
Plan your website enquiry workflow
Plan your website enquiry workflow before the build. Follow a fictional quotation request through ownership, failed receipt checks and an acceptance decision.
Ready to improve your IT operations?
Request a free 30-minute consultation to discuss your IT challenges. Send a short outline and I will reply to arrange a suitable time. No obligation.
Request a free 30-minute consultationGet Occasional IT Leadership Insights
IT leadership insights, occasionally. No fluff. Unsubscribe any time.
No spam. Unsubscribe any time.