Fix stress pipeline: rename variables to avoid env var injection, simplify runner (#4329)

Author: paulmedynski

.github/instructions/ado-pipelines.instructions.md

@@ -110,6 +110,20 @@ Test timeout — `--blame-hang-timeout 10m` (configured in `build.proj` and thre
 - `localFeedPath` = `$(Build.SourcesDirectory)/packages` — local NuGet feed for inter-package deps
 - `packagePath` = `$(Build.SourcesDirectory)/output` — NuGet pack output
 
+## Variable Naming — Avoid `{COMMAND}ARGUMENTS` Names
+
+The dotnet CLI (via System.CommandLine) reads environment variables named `{COMMAND}ARGUMENTS` and silently injects their content into the parsed arguments for that subcommand. Because Azure DevOps automatically exposes all pipeline variables as uppercased environment variables, a pipeline variable named `runArguments` becomes `RUNARGUMENTS`, which `dotnet run` reads and injects into the application's `args[]` — bypassing the `--` separator.
+
+**Forbidden variable names** (any casing):
+- `runArguments` — injected into `dotnet run`
+- `buildArguments` — injected into `dotnet build`
+- `testArguments` — injected into `dotnet test`
+- Any name matching `{dotnet-subcommand}Arguments`
+
+**Use instead**: `dotnetBuildOpts`, `dotnetRunOpts`, `stressTestArgs`, or other names that do not match the `{COMMAND}ARGUMENTS` pattern.
+
+This affects ALL .NET SDK versions (8.0+). The injection is invisible in `[command]` log lines, making it extremely hard to diagnose. The only symptom is the application receiving unexpected arguments.
+
 ## Conventions When Editing Pipelines
 
 - Always use templates for reusable logic — do not inline complex steps

eng/pipelines/stress/stress-tests-job.yml

@@ -24,7 +24,6 @@ parameters:
   # True to enable debugging steps.
   - name: debug
     type: boolean
-    default: false
 
   # The prefix to prepend to the job's display name:
   #
@@ -36,7 +35,6 @@ parameters:
   # The verbosity level for the dotnet CLI commands.
   - name: dotnetVerbosity
     type: string
-    default: normal
     values:
     - quiet
     - minimal
@@ -48,20 +46,18 @@ parameters:
   - name: jobNameSuffix
     type: string
 
-  # True to fail the job when stress tests fail.  When false, test failures produce warnings
-  # (SucceededWithIssues) but do not fail the job.
-  - name: failOnTestFailure
+  # When true, test failures produce warnings (SucceededWithIssues) but do not fail the job.
+  # When false, test failures fail the job.  All test steps always run regardless of this setting.
+  - name: warnOnTestFailure
     type: boolean
 
   # The list of .NET Framework runtimes to test against.
   - name: netFrameworkTestRuntimes
     type: object
-    default: []
 
   # The list of .NET runtimes to test against.
   - name: netTestRuntimes
     type: object
-    default: []
 
   # The name of the Azure Pipelines pool to use.
   - name: poolName
@@ -105,18 +101,6 @@ jobs:
       - name: project
         value: $(Build.SourcesDirectory)/src/Microsoft.Data.SqlClient/tests/StressTests/SqlClient.Stress.Runner/SqlClient.Stress.Runner.csproj
 
-      # dotnet CLI arguments for build.
-      - name: buildArguments
-        value: >-
-          --verbosity ${{ parameters.dotnetVerbosity }}
-          -p:Configuration=${{ parameters.buildConfiguration }}
-
-      # dotnet CLI arguments for run.
-      - name: runArguments
-        value: >-
-          --verbosity ${{ parameters.dotnetVerbosity }}
-          --configuration ${{ parameters.buildConfiguration }}
-
       # The contents of the config file to use for all tests.  We will write this to a JSON file and
       # then point to it via the STRESS_CONFIG_FILE environment variable.
       - name: configContent
@@ -137,9 +121,29 @@ jobs:
             }
           ]
 
-      # Stress test command-line arguments.
-      - name: testArguments
-        value: -a SqlClient.Stress.Tests -console
+      # IMPORTANT: Do NOT name pipeline variables "runArguments", "buildArguments", or
+      # "testArguments".  ADO exposes all pipeline variables as environment variables (uppercased),
+      # and the dotnet CLI's System.CommandLine reads env vars matching {COMMAND}ARGUMENTS (e.g.
+      # RUNARGUMENTS, BUILDARGUMENTS) and silently injects their content into the parsed arguments —
+      # bypassing the "--" separator.  This causes app arguments to contain SDK options and triggers
+      # unintended behavior.
+
+      # dotnet CLI options for build.
+      - name: dotnetBuildOpts
+        value: >-
+          --verbosity ${{ parameters.dotnetVerbosity }}
+          -p:Configuration=${{ parameters.buildConfiguration }}
+
+      # dotnet run options shared by all test steps (framework is appended per-step).
+      - name: dotnetRunOpts
+        value: >-
+          --no-build
+          --verbosity ${{ parameters.dotnetVerbosity }}
+          --configuration ${{ parameters.buildConfiguration }}
+
+      # Stress test options passed after the "--" separator.
+      - name: stressTestOpts
+        value: --assembly SqlClient.Stress.Tests --console
 
     steps:
 
@@ -178,7 +182,7 @@ jobs:
         inputs:
           command: build
           projects: $(project)
-          arguments: $(buildArguments)
+          arguments: ${{ variables.dotnetBuildOpts }}
 
       # Set a flag so test steps can distinguish a build failure from a test failure.
       - pwsh: Write-Host "##vso[task.setvariable variable=buildSucceeded]true"
@@ -194,29 +198,33 @@ jobs:
       #     - eq(variables['buildSucceeded'], 'true') gates on the flag set above, so tests are
       #       skipped entirely if the build or any setup step failed (since there's nothing to run).
       #
-      #   continueOnError: ${{ not(parameters.failOnTestFailure) }}
-      #     - When failOnTestFailure is false, continueOnError is true: a test failure marks the
+      #   continueOnError: ${{ parameters.warnOnTestFailure }}
+      #     - When warnOnTestFailure is true, continueOnError is true: a test failure marks the
       #       step and job as SucceededWithIssues (orange warning) rather than Failed.
-      #     - When failOnTestFailure is true, continueOnError is false: a test failure fails the
+      #     - When warnOnTestFailure is false, continueOnError is false: a test failure fails the
       #       job (red), though subsequent runtimes still run due to the condition above.
       #
       - ${{ each runtime in parameters.netTestRuntimes }}:
         - task: DotNetCoreCLI@2
           displayName: Test [${{ runtime }}]
           condition: and(succeededOrFailed(), eq(variables['buildSucceeded'], 'true'))
-          continueOnError: ${{ not(parameters.failOnTestFailure) }}
+          continueOnError: ${{ parameters.warnOnTestFailure }}
+          env:
+            STRESS_CONFIG_FILE: config.json
           inputs:
             command: run
             projects: $(project)
-            arguments: $(runArguments) --no-build -f ${{ runtime }} -e STRESS_CONFIG_FILE=config.jsonc -- $(testArguments)
+            arguments: ${{ variables.dotnetRunOpts }} -f ${{ runtime }} -- ${{ variables.stressTestOpts }}
 
       # Run the stress tests for each .NET Framework runtime.
       - ${{ each runtime in parameters.netFrameworkTestRuntimes }}:
         - task: DotNetCoreCLI@2
           displayName: Test [${{ runtime }}]
           condition: and(succeededOrFailed(), eq(variables['buildSucceeded'], 'true'))
-          continueOnError: ${{ not(parameters.failOnTestFailure) }}
+          continueOnError: ${{ parameters.warnOnTestFailure }}
+          env:
+            STRESS_CONFIG_FILE: config.json
           inputs:
             command: run
             projects: $(project)
-            arguments: $(runArguments) --no-build -f ${{ runtime }} -e STRESS_CONFIG_FILE=config.jsonc -- $(testArguments)
+            arguments: ${{ variables.dotnetRunOpts }} -f ${{ runtime }} -- ${{ variables.stressTestOpts }}

eng/pipelines/stress/stress-tests-pipeline.yml

@@ -14,7 +14,7 @@
 # ADO.Net project:
 #   Triggering pipeline:  MDS Main CI (branch internal/main only)
 #   Pipeline name:        sqlclient-stress
-#   Pipeline URL:         TODO: add URL when pipeline is created
+#   Pipeline URL:         https://dev.azure.com/SqlClientDrivers/ADO.Net/_build?definitionId=2284
 
 # Set the pipeline run name to the day-of-year and the daily run counter.
 name: $(DayOfYear)$(Rev:rr)
@@ -73,10 +73,10 @@ parameters:
     type: boolean
     default: false
 
-  # True to fail the pipeline when stress tests fail.  When false (default), test failures produce
-  # warnings but do not fail the overall pipeline run.
-  - name: failOnTestFailure
-    displayName: Fail pipeline on test failure
+  # When true, test failures produce warnings (SucceededWithIssues) but do not fail the pipeline.
+  # When false (default), test failures fail the pipeline.
+  - name: warnOnTestFailure
+    displayName: Warn (not fail) on test failure
     type: boolean
     default: false
 
@@ -105,5 +105,5 @@ stages:
     parameters:
       buildConfiguration: ${{ parameters.buildConfiguration }}
       debug: ${{ parameters.debug }}
-      failOnTestFailure: ${{ parameters.failOnTestFailure }}
+      warnOnTestFailure: ${{ parameters.warnOnTestFailure }}
       dotnetVerbosity: ${{ parameters.dotnetVerbosity }}

eng/pipelines/stress/stress-tests-stage.yml

@@ -30,16 +30,15 @@ parameters:
   # True to enable debugging steps.
   - name: debug
     type: boolean
-    default: false
 
-  # True to fail the job when stress tests fail.  When false, test failures produce warnings.
-  - name: failOnTestFailure
+  # When true, test failures produce warnings (SucceededWithIssues) but do not fail the job.
+  # When false, test failures fail the job.  All test steps always run regardless of this setting.
+  - name: warnOnTestFailure
     type: boolean
 
   # The verbosity level for the dotnet CLI commands.
   - name: dotnetVerbosity
     type: string
-    default: normal
     values:
     - quiet
     - minimal
@@ -52,11 +51,11 @@ parameters:
     type: object
     default: [net462]
 
-  # The list of .NET runtimes to test against.  These must align with the TargetFrameworks defined
-  # in the stress test Directory.Build.props and the TFMs that SqlClient ships.
+  # The list of .NET runtimes to test against.  These should include the TFMs that SqlClient ships
+  # as well as any upcoming runtimes being validated (e.g. net10.0 is tested but not yet shipped).
   - name: netTestRuntimes
     type: object
-    default: [net8.0, net9.0]
+    default: [net8.0, net9.0, net10.0]
 
 stages:
   - stage: stress_tests_stage
@@ -69,6 +68,15 @@ stages:
       - name: saPassword
         value: $[stageDependencies.secrets_stage.secrets_job.outputs['SaPassword.Value']]
 
+      # The 1ES pool name, determined automatically by ADO project:
+      #   - ADO.Net  -> ADO-1ES-Pool
+      #   - public   -> ADO-CI-1ES-Pool
+      - name: poolName
+        ${{ if eq(variables['System.TeamProject'], 'ADO.Net') }}:
+          value: ADO-1ES-Pool
+        ${{ else }}:
+          value: ADO-CI-1ES-Pool
+
     jobs:
 
     # ----------------------------------------------------------------------------------------------
@@ -80,10 +88,12 @@ stages:
         debug: ${{ parameters.debug }}
         displayNamePrefix: Linux
         dotnetVerbosity: ${{ parameters.dotnetVerbosity }}
-        failOnTestFailure: ${{ parameters.failOnTestFailure }}
+        warnOnTestFailure: ${{ parameters.warnOnTestFailure }}
         jobNameSuffix: linux
+        # No .NET Framework runtimes on Linux.
+        netFrameworkTestRuntimes: []
         netTestRuntimes: ${{ parameters.netTestRuntimes }}
-        poolName: ADO-CI-1ES-Pool
+        poolName: ${{ variables.poolName }}
         saPassword: $(saPassword)
         sqlSetupStep:
           template: /eng/pipelines/common/templates/steps/configure-sql-server-linux-step.yml@self
@@ -100,12 +110,12 @@ stages:
         debug: ${{ parameters.debug }}
         displayNamePrefix: Win
         dotnetVerbosity: ${{ parameters.dotnetVerbosity }}
-        failOnTestFailure: ${{ parameters.failOnTestFailure }}
+        warnOnTestFailure: ${{ parameters.warnOnTestFailure }}
         jobNameSuffix: windows
         # Note that we include the .NET Framework runtimes for test runs on Windows.
         netFrameworkTestRuntimes: ${{ parameters.netFrameworkTestRuntimes }}
         netTestRuntimes: ${{ parameters.netTestRuntimes }}
-        poolName: ADO-CI-1ES-Pool
+        poolName: ${{ variables.poolName }}
         saPassword: $(saPassword)
         sqlSetupStep:
           template: /eng/pipelines/common/templates/steps/configure-sql-server-win-step.yml@self
@@ -124,8 +134,10 @@ stages:
         debug: ${{ parameters.debug }}
         displayNamePrefix: macOS
         dotnetVerbosity: ${{ parameters.dotnetVerbosity }}
-        failOnTestFailure: ${{ parameters.failOnTestFailure }}
+        warnOnTestFailure: ${{ parameters.warnOnTestFailure }}
         jobNameSuffix: macos
+        # No .NET Framework runtimes on macOS.
+        netFrameworkTestRuntimes: []
         netTestRuntimes: ${{ parameters.netTestRuntimes }}
         # We don't have any 1ES Hosted Pool images for macOS, so we use a generic one from Azure
         # Pipelines.

src/Microsoft.Data.SqlClient/tests/StressTests/Directory.Packages.props

@@ -9,4 +9,8 @@
     <CentralPackageTransitivePinningEnabled>true</CentralPackageTransitivePinningEnabled>
   </PropertyGroup>
 
+  <ItemGroup>
+    <PackageVersion Include="System.CommandLine" Version="2.0.8" />
+  </ItemGroup>
+
 </Project>

src/Microsoft.Data.SqlClient/tests/StressTests/IMonitorLoader/IMonitorLoader.cs

@@ -4,7 +4,7 @@
 
 using System.Collections.Generic;
 
-namespace Monitoring
+namespace Microsoft.Data.SqlClient.Test.Stress
 {
     public interface IMonitorLoader
     {

src/Microsoft.Data.SqlClient/tests/StressTests/IMonitorLoader/MonitorMetrics.cs

@@ -4,7 +4,7 @@
 
 using System;
 
-namespace Monitoring
+namespace Microsoft.Data.SqlClient.Test.Stress
 {
     public class MonitorMetrics
     {

src/Microsoft.Data.SqlClient/tests/StressTests/SqlClient.Stress.Common/Attributes/GlobalExceptionHandlerAttribute.cs

@@ -4,7 +4,7 @@
 
 using System;
 
-namespace DPStressHarness
+namespace Microsoft.Data.SqlClient.Test.Stress
 {
     [AttributeUsage(AttributeTargets.Method, Inherited = true, AllowMultiple = false)]
     public class GlobalExceptionHandlerAttribute : Attribute

src/Microsoft.Data.SqlClient/tests/StressTests/SqlClient.Stress.Common/Attributes/GlobalTestCleanupAttribute.cs

@@ -4,7 +4,7 @@
 
 using System;
 
-namespace DPStressHarness
+namespace Microsoft.Data.SqlClient.Test.Stress
 {
     [AttributeUsage(AttributeTargets.Method, Inherited = true, AllowMultiple = false)]
     public class GlobalTestCleanupAttribute : Attribute

src/Microsoft.Data.SqlClient/tests/StressTests/SqlClient.Stress.Common/Attributes/GlobalTestSetupAttribute.cs

@@ -4,7 +4,7 @@
 
 using System;
 
-namespace DPStressHarness
+namespace Microsoft.Data.SqlClient.Test.Stress
 {
     [AttributeUsage(AttributeTargets.Method, Inherited = true, AllowMultiple = false)]
     public class GlobalTestSetupAttribute : Attribute