Restart SQL Server on Ubuntu with a maintenance checklist
Restarting a shared database service can affect several applications at the same time. A convenient button should make the target and result clearer, while keeping the maintenance decision deliberate.
The local SSH mode in Server App Manager can monitor and restart the canonical mssql-server.service unit when the account is permitted to do so. This guide describes a controlled maintenance exercise on a disposable SQL Server lab, with separate checks for service state, database readiness and application recovery.
Start with an actual maintenance reason
Write down why a restart is being considered and who owns the decision. A database restart should follow a specific maintenance or recovery need rather than a vague desire to clear an unexplained application error.
Record what success will look like. That should include the expected database availability and the behavior of the applications that depend on it.
If you are investigating an incident, collect the current service state and relevant evidence before starting another lifecycle. Preserve the distinction between an observed database problem and an application symptom that merely mentions a database connection.
Identify the dependent applications
List the applications, workers and scheduled tasks that use the instance. Consider who needs to be notified and which activity should be paused under the team's existing procedure.
An HTTP application can remain running while its database is unavailable. A worker can repeatedly retry a failed connection. Those process states do not tell you that all business operations are safe to resume.
For the lab exercise, use a sample dependent application with a read-only readiness check. Keep its expected response and normal recovery behavior documented before the restart, so the result is not decided after the fact.
Verify the supported lab installation
Use a SQL Server and Ubuntu combination supported by Microsoft for the version under test. Record the actual versions rather than assuming that a generic Ubuntu label establishes compatibility.
Confirm the service identity. The dashboard's database page refers to the fixed canonical SQL Server unit, not a service name entered freely from a browser form.
The app is not a database client and does not modify databases through this management page. The database owner remains responsible for database-level readiness checks, backup policies and any recovery procedure beyond the systemd service operation.
Read the initial database service view
Open Database after signing in to the verified SSH endpoint. Check whether the SQL Server unit exists and read its current state.
Where available, note memory, CPU time and service uptime. Unavailable metrics should remain N/A. These fields describe the Linux service and accounting interfaces; they are not a complete picture of query workload, transaction health or database recovery.
Record the sample time. If the page reports a connection problem, resolve that before interpreting the visible service state as a current reading from Ubuntu.
Confirm that the account may perform the action
The restart uses the registered SSH account's existing sudo permissions. Logging in successfully does not automatically authorize changing a database service.
If the command is denied, capture the sanitized result and involve the server owner. Do not substitute a broader account or change policy without an explicit administrative decision.
A permitted service operation is still subject to maintenance ownership. Review access and recovery arrangements before the planned window so that a permission problem does not create pressure to make an unreviewed change during downtime.
Read the impact warning
Choose Restart SQL Server only when the intended database service is the correct target. Read the warning about dependent applications and confirm when the maintenance window is ready.
During a lab exercise, cancel a confirmation first and verify that no restart occurred. This makes the cancellation path familiar before the actual operation.
When confirming the real test, submit once and wait. The dashboard prevents conflicting operations through its coordinator and reports the command result. Repeated clicks or requests from other tabs do not make database recovery complete sooner.
Interpret the command result narrowly
A successful restart command is evidence about the service management operation. It does not establish that every database is online, that all application connections recovered, or that every dependent job resumed.
If the operation fails, read the actual error and current service state. Differentiate permission denial, a command failure and a client timeout. Those conditions suggest different follow-up work.
The Microsoft SQL Server Linux troubleshooting guide provides database-operator diagnostics. Use its version-appropriate guidance for the installation instead of treating the dashboard's service status as a complete recovery report.
Verify database readiness separately
After the service operation, the authorized database operator should perform the team's established readiness check. That check may involve a known connection and a small read-only query through existing tools.
Keep that activity outside the dashboard's service-management claim. The app does not acquire application database credentials or issue readiness queries on your behalf.
Record the actual result and time. A process can exist before a particular database is ready for the application's use. Allow the expected recovery procedure to complete before declaring the maintenance successful or deciding a new restart is required.
Check each dependent application
Use Website Health for appropriate public endpoints, then verify a representative application function through its approved operational check.
A homepage that does not access SQL Server may answer while a database-backed feature still fails. A protected endpoint can intentionally return an authentication response. Choose checks whose expected behavior you understand.
For workers, use their own read-only operational signals. A running process is not proof that its last job completed. Keep the test tied to the dependent function that the maintenance could have interrupted.
Investigate a timeout before retrying
A bounded client request can time out while a remote systemd job continues. Inspect the current unit state and relevant operator diagnostics before submitting another restart.
Record the time and result of the first request. Check whether the service is still transitioning or whether database recovery is progressing under the existing process.
If the evidence remains unclear, hand the case to the database owner with the observed state and sanitized error. Repeating a restart can interrupt recovery and make the sequence harder to understand. The next action should follow a new decision based on current evidence.
Keep the maintenance record precise
Open Audit History and verify the recorded target, action, account, timestamp and result. The dashboard's audit entry is a local record of the management workflow.
Add the separate database and dependent-application checks to the team's maintenance note. This connects the service action to the actual outcome users care about.
Protect the note and any log excerpts appropriately. A public issue needs only the minimum details required to reproduce a product defect. Connection strings, credentials, private database names and complete error logs often exceed that need.
Plan rollback and escalation before the window
A restart does not itself supply a database rollback mechanism. The database owner's recovery plan should already define what to do if the service cannot start or a dependent application fails after maintenance.
Identify who can inspect database-level diagnostics and who can approve further action. Record when the team should stop an exercise and restore the known lab state.
This preparation keeps the dashboard in its intended role: a clear, bounded way to request a known service operation and inspect its result. The broader recovery decision stays with the responsible operators.
Complete the exercise with all three outcomes
A successful lab record covers the Linux service operation, database readiness and dependent application recovery separately. It includes a cancelled confirmation and a failure or denied-permission scenario if those are part of the evaluation.
Do not compress a partially completed recovery into a single success label. State which checks passed and which remain pending.
Read the SQL Server guide, audit guide and application security notes for the current dashboard scope and operational boundaries.