Add flag to enable show tech collect

Signed-off-by: Sergei Lukianov <me@slukjanov.name>

.github/workflows/ci-hw.yaml

@@ -69,11 +69,19 @@ jobs:
       - name: hhfab init and vlab up
         env:
           HHFAB_REG_REPO: 127.0.0.1:30000
+          HHFAB_VLAB_COLLECT: true
         run: |
           source "./lab-ci/envs/$KUBE_NODE/source.sh"
           bin/hhfab init -v --dev --include-onie=${{ matrix.includeonie }} -w "./lab-ci/envs/$KUBE_NODE/wiring.yaml"
           bin/hhfab vlab up -v --ready switch-reinstall ${{ inputs.skip_ready && '' || '--ready setup-vpcs --ready test-connectivity' }} ${{ inputs.keep && '' || '--ready exit' }} --mode=${{ matrix.buildmode }}
 
+      - name: Upload show-tech artifacts
+        uses: actions/upload-artifact@v4
+        if: ${{ always() }}
+        with:
+          name: show-tech-hlab-${{ matrix.fabricmode }}-${{ matrix.includeonie }}-${{ matrix.buildmode }}
+          path: show-tech-output
+
       - name: Dump local registry logs
         if: ${{ always() }}
         run: |

.github/workflows/ci.yaml

@@ -239,6 +239,7 @@ jobs:
       - name: hhfab init and vlab up
         env:
           HHFAB_REG_REPO: 127.0.0.1:30000
+          HHFAB_VLAB_COLLECT: true
         run: |
           bin/hhfab init -v --dev -m ${{ matrix.fabricmode }} --include-onie=${{ matrix.includeonie }}
           bin/hhfab vlab gen -v
@@ -304,6 +305,7 @@ jobs:
       - name: Initial VLAB with old version
         env:
           HHFAB_REG_REPO: 127.0.0.1:30000
+          HHFAB_VLAB_COLLECT: true
         run: |
           mkdir old
           curl -fsSL https://i.hhdev.io/hhfab | USE_SUDO=false INSTALL_DIR=./old VERSION="${{ matrix.fromversion }}" bash
@@ -326,6 +328,7 @@ jobs:
       - name: hhfab vlab up --upgrade
         env:
           HHFAB_REG_REPO: 127.0.0.1:30000
+          HHFAB_VLAB_COLLECT: true
         run: |
           bin/hhfab vlab up -v --ready setup-vpcs --ready test-connectivity --ready exit --upgrade
 
@@ -412,6 +415,7 @@ jobs:
       - name: hhfab init and vlab up
         env:
           HHFAB_REG_REPO: 127.0.0.1:30000
+          HHFAB_VLAB_COLLECT: true
         run: |
           source "./lab-ci/envs/$KUBE_NODE/source.sh"
           bin/hhfab init -v --dev --include-onie=${{ matrix.includeonie }} -w "./lab-ci/envs/$KUBE_NODE/wiring.yaml"

cmd/hhfab/main.go

@@ -56,6 +56,7 @@ const (
 	FlagNameControlUpgrade        = "control-upgrade"
 	FlagNameFailFast              = "fail-fast"
 	FlagNameReady                 = "ready"
+	FlagNameCollectShowTech       = "collect-show-tech"
 )
 
 func main() {
@@ -636,6 +637,12 @@ func Run(ctx context.Context) error {
 								Aliases: []string{"r"},
 								Usage:   "run commands on all VMs ready (one of: " + strings.Join(onReadyCommands, ", ") + ")",
 							},
+							&cli.BoolFlag{
+								Name:    FlagNameCollectShowTech,
+								Aliases: []string{"collect"},
+								Usage:   "collect show-tech from all devices at exit or error",
+								EnvVars: []string{"HHFAB_VLAB_COLLECT"},
+							},
 						),
 						Before: before(false),
 						Action: func(c *cli.Context) error {
@@ -651,6 +658,7 @@ func Run(ctx context.Context) error {
 									ControlUpgrade:     c.Bool(FlagNameControlUpgrade),
 									FailFast:           c.Bool(FlagNameFailFast),
 									OnReady:            c.StringSlice(FlagNameReady),
+									CollectShowTech:    c.Bool(FlagNameCollectShowTech),
 								},
 							}); err != nil {
 								return fmt.Errorf("running VLAB: %w", err)

pkg/hhfab/vlabrunner.go

@@ -86,6 +86,7 @@ type VLABRunOpts struct {
 	ControlUpgrade     bool
 	FailFast           bool
 	OnReady            []string
+	CollectShowTech    bool
 }
 
 type OnReady string
@@ -375,10 +376,13 @@ func (c *Config) VLABRun(ctx context.Context, vlab *VLAB, opts VLABRunOpts) erro
 
 			if err := execCmd(ctx, true, vmDir, VLABCmdQemuSystem, args, "vm", vm.Name); err != nil {
 				slog.Warn("Failed running VM", "vm", vm.Name, "type", vm.Type, "err", err)
-				if err := c.VLABShowTech(ctx, vlab); err != nil {
-					slog.Warn("Failed to collect show-tech diagnostics", "err", err)
 
-					return fmt.Errorf("getting show-tech: %w", err)
+				if opts.CollectShowTech {
+					if err := c.VLABShowTech(ctx, vlab); err != nil {
+						slog.Warn("Failed to collect show-tech diagnostics", "err", err)
+
+						return fmt.Errorf("getting show-tech: %w", err)
+					}
 				}
 
 				if opts.FailFast {
@@ -394,10 +398,13 @@ func (c *Config) VLABRun(ctx context.Context, vlab *VLAB, opts VLABRunOpts) erro
 			group.Go(func() error {
 				if err := c.vmPostProcess(ctx, vlab, d, vm, opts); err != nil {
 					slog.Warn("Failed to post-process VM", "vm", vm.Name, "type", vm.Type, "err", err)
-					if err := c.VLABShowTech(ctx, vlab); err != nil {
-						slog.Warn("Failed to collect show-tech diagnostics", "err", err)
 
-						return fmt.Errorf("getting show-tech: %w", err)
+					if opts.CollectShowTech {
+						if err := c.VLABShowTech(ctx, vlab); err != nil {
+							slog.Warn("Failed to collect show-tech diagnostics", "err", err)
+
+							return fmt.Errorf("getting show-tech: %w", err)
+						}
 					}
 
 					if opts.FailFast {
@@ -445,10 +452,13 @@ func (c *Config) VLABRun(ctx context.Context, vlab *VLAB, opts VLABRunOpts) erro
 						WaitReady:   false,
 					}); err != nil {
 						slog.Warn("Failed to reinstall switches", "err", err)
-						if err := c.VLABShowTech(ctx, vlab); err != nil {
-							slog.Warn("Failed to collect show-tech diagnostics", "err", err)
 
-							return fmt.Errorf("getting show-tech: %w", err)
+						if opts.CollectShowTech {
+							if err := c.VLABShowTech(ctx, vlab); err != nil {
+								slog.Warn("Failed to collect show-tech diagnostics", "err", err)
+
+								return fmt.Errorf("getting show-tech: %w", err)
+							}
 						}
 
 						return fmt.Errorf("reinstalling switches: %w", err)
@@ -465,10 +475,13 @@ func (c *Config) VLABRun(ctx context.Context, vlab *VLAB, opts VLABRunOpts) erro
 						TimeServers:       []string{"219.239.35.0"},
 					}); err != nil {
 						slog.Warn("Failed to setup VPCs", "err", err)
-						if err := c.VLABShowTech(ctx, vlab); err != nil {
-							slog.Warn("Failed to collect show-tech diagnostics", "err", err)
 
-							return fmt.Errorf("getting show-tech: %w", err)
+						if opts.CollectShowTech {
+							if err := c.VLABShowTech(ctx, vlab); err != nil {
+								slog.Warn("Failed to collect show-tech diagnostics", "err", err)
+
+								return fmt.Errorf("getting show-tech: %w", err)
+							}
 						}
 
 						return fmt.Errorf("setting up VPCs: %w", err)
@@ -482,20 +495,26 @@ func (c *Config) VLABRun(ctx context.Context, vlab *VLAB, opts VLABRunOpts) erro
 						CurlsCount:        3,
 					}); err != nil {
 						slog.Warn("Failed to test connectivity", "err", err)
-						if err := c.VLABShowTech(ctx, vlab); err != nil {
-							slog.Warn("Failed to collect show-tech diagnostics", "err", err)
 
-							return fmt.Errorf("getting show-tech: %w", err)
+						if opts.CollectShowTech {
+							if err := c.VLABShowTech(ctx, vlab); err != nil {
+								slog.Warn("Failed to collect show-tech diagnostics", "err", err)
+
+								return fmt.Errorf("getting show-tech: %w", err)
+							}
 						}
 
 						return fmt.Errorf("testing connectivity: %w", err)
 					}
 				case OnReadyExit:
-					if err := c.VLABShowTech(ctx, vlab); err != nil {
-						slog.Warn("Failed to collect show-tech diagnostics", "err", err)
+					if opts.CollectShowTech {
+						if err := c.VLABShowTech(ctx, vlab); err != nil {
+							slog.Warn("Failed to collect show-tech diagnostics", "err", err)
 
-						return fmt.Errorf("getting show-tech: %w", err)
+							return fmt.Errorf("getting show-tech: %w", err)
+						}
 					}
+
 					// TODO seems like some graceful shutdown logic isn't working in CI and we're getting stuck w/o this
 					if os.Getenv("GITHUB_ACTIONS") == "true" {
 						slog.Warn("Immediately exiting b/c running in GHA")