Progress - Complete the bars of 13 more commands in finally blocks (tier 3)

[andreasjordan]

Part of RFC #10548: the try/finally leak fixes, tier 3 of 3 (the commands with few or no Stop-Function calls inside the bar's lifetime). Same pattern as #10773 and #10774, one commit per command with its test. Independent of both.

Problem

These 13 functions complete their progress bar only on the success path, so a throw, a stopped pipeline (Ctrl+C) or Select-Object -First left the bar on screen for the rest of the session. Three of them are internal functions nested in a public command: Get-SqlFileStructure in Copy-DbaDatabase, Get-InferredSchema in Import-DbaCsv (the full scan of -DetectColumnTypes) and Test-SqlInstance in Find-DbaInstance.

Change

Each bar's completion moves into a finally around that bar's lifetime.

• Stop-DbaDbEncryption has two bars, the sequential one (Id 0) and the one that waits for the -Parallel threads (Id 1). Each gets its own finally, as in Start-DbaDbEncryption in Progress - Complete the bars of 13 long-running commands in finally blocks (tier 1) #10773.
• Sync-DbaLoginPermission already ran its login loop in try/finally to restore the database context; the completion moves into that finally.
• Install-DbaParquet completed its bar by hand in five places, Save-DbaDiagnosticQueryScript in two. Those completions go away.
• Save-DbaKbUpdate: each of its two downloads gets its own try/finally.
• The others (Disable-DbaDbEncryption, Install-DbaSqlWatch, Remove-DbaNetworkCertificate, New-DbatoolsSupportPackage, New-DbaComputerCertificateSigningRequest and the three nested functions): one try/finally around the loop or the steps that draw the bar.

Most of the diff is re-indentation: git diff -w shows 45 lines added and 11 removed for the 13 commands.

Test

Same boundary as in #10773: the real command runs in a runspace of its own, created by the PowerShell API without a host, where Write-Progress puts every record into Streams.Progress. The test asserts that the command drew its bar and that no progress Id is left whose last record is not a completed one.

Command	How the test ends the run
Stop-DbaDbEncryption	Select-Object -First 1 at the first database, which the earlier contexts left not encrypted
Sync-DbaLoginPermission	Select-Object -First 1 at the result of the login
Install-DbaSqlWatch	pipeline stopped at the first record, before the dacpac is published (5.1 only, like the existing tests)
Remove-DbaNetworkCertificate	new integration Describe on InstanceRestart, Select-Object -First 1 at the result; a certificate that was configured is put back
Save-DbaDiagnosticQueryScript	new integration Describe, Select-Object -First 1 at the first downloaded script
Install-DbaParquet	-Force into the test drive, pipeline stopped at the first record (while NuGet is resolved)
Save-DbaKbUpdate	a download link under .invalid and -ErrorAction Stop, so the failed download throws out of the command; needs neither the catalog nor the internet. Without -ErrorAction Stop the old code completed the bar as well: Invoke-WebRequest reports the failure as an error that ends only its own statement, so the next line still ran
New-DbatoolsSupportPackage	new integration Describe, pipeline stopped at the first record while the information is collected
New-DbaComputerCertificateSigningRequest	Select-Object -First 1 at the first file; the request is removed from LocalMachine\REQUEST like the others
Import-DbaCsv	a 200,000-row file with -DetectColumnTypes, pipeline stopped at the first record of the type detection
Find-DbaInstance	the existing scan of InstanceSingle, pipeline stopped at the first record of the computer
Disable-DbaDbEncryption	no new test, see below
Copy-DbaDatabase	no new test, see below

No automated test for two of them:

• Disable-DbaDbEncryption draws its bar only while it waits for the decryption, and the decryption of a test database is done within the first second, before the first record. Trace flag 5004 holds the scan (checked in the lab: state 5, 0 percent, even after TRACEOFF until the scan is restarted), but it works for the whole instance, and a run that dies between TRACEON and TRACEOFF would leave every later TDE test waiting. ALTER DATABASE ... SET ENCRYPTION SUSPEND per database does not help: the decryption is done before it can be suspended, and SMO issues the ALTER again, which SQL Server refuses for a database whose encryption is already off. If a test with trace flag 5004 is wanted, I can add it.
• Copy-DbaDatabase: the loop of Get-SqlFileStructure takes milliseconds per database and throws nothing that a test could provoke, so neither a stop nor a throw can be placed inside it.

The existing tests of both pass.

The -Parallel bar of Stop-DbaDbEncryption runs its threads in a runspace pool on $Host, like Start-DbaDbEncryption in #10773, so the API host does not see its records either. I checked it the same way, in a real console window after Stop-DbaDbEncryption -Parallel | Select-Object -First 1 against one unencrypted test database: the old command leaves "Disabling encryption" (Id 1) behind on 5.1 and 7.6, the new one does not.

Results (lab, InstanceSingle SQL Server 2019, Multi1 2025, Multi2/Restart 2022):

• New command: the 11 changed test files and the existing tests of Copy-DbaDatabase and Disable-DbaDbEncryption pass on PowerShell 5.1 and 7.6, without warnings. The integration Describe of Install-DbaSqlWatch is skipped on pwsh, as before.
• Old command: in every file exactly the new progress test fails, on both editions (Install-DbaSqlWatch on 5.1), with one exception: Install-DbaParquet fails on 5.1 only. On 7.x its cleanup of the temp folder (Remove-Item -Recurse) writes a completed record of its own for Id 0, which clears the leftover bar by accident; on 5.1 the bar stays.
• Once, the existing install context of Install-DbaSqlWatch failed with a deadlock (Msg 1205) inside SqlWatch's own post-deployment script; the three runs after it passed. Not related to this change.
• TestTestfiles.Tests.ps1 and TestTestLayout.ps1 report nothing for the 11 changed test files.

Left open

• The sequential Stop-DbaDbEncryption calls Disable-DbaDbEncryption for each encrypted database, and both draw their bar through Write-ProgressHelper with Id 0. The completion of the inner bar therefore also removes the outer one until the next database redraws it. That is the shared-Id question of the RFC, not a leak, so it is not changed here.

🤖 Generated with Claude Code