Installer UX: drop PriorityClass, fix namespace, one-per-machine guard, surface version

client/tests/mysql_test.yaml

@@ -61,8 +61,18 @@ tests:
           path: spec.template.spec.containers[0].livenessProbe.failureThreshold
           value: 5
 
-  - it: should reference the data-plane PriorityClass when enabled
+  - it: should NOT set priorityClassName by default (PriorityClass dropped)
     template: templates/mysql-deployment.yaml
+    asserts:
+      - notExists:
+          path: spec.template.spec.priorityClassName
+
+  - it: should reference the PriorityClass when a name is set
+    template: templates/mysql-deployment.yaml
+    set:
+      priorityClass:
+        create: true
+        name: tracebloc-data-plane
     asserts:
       - equal:
           path: spec.template.spec.priorityClassName

client/tests/priority_class_pdb_test.yaml

@@ -7,8 +7,18 @@ set:
   clientId: "test-id"
   clientPassword: "test"
 tests:
-  - it: should create the data-plane PriorityClass by default
+  - it: should NOT render the PriorityClass by default (dropped)
     template: templates/priority-class.yaml
+    asserts:
+      - hasDocuments:
+          count: 0
+
+  - it: should render the PriorityClass when explicitly enabled
+    template: templates/priority-class.yaml
+    set:
+      priorityClass:
+        create: true
+        name: tracebloc-data-plane
     asserts:
       - isKind:
           of: PriorityClass

client/values.yaml

@@ -313,22 +313,23 @@ resources:
       memory: "512Mi"
 
 # -- PriorityClass for the data-plane (mysql).
-# Cluster-scoped resource. Created with helm.sh/resource-policy: keep so a
-# release uninstall does not yank it from sibling releases that share it.
-# Value 1,000,000 sits well above default (0) so the scheduler will preempt
-# noisy training jobs to keep mysql scheduled, but well below
-# system-cluster-critical (2,000,000,000).
+# DROPPED BY DEFAULT (create: false, name: ""). mysql doesn't need it: its
+# memory requests==limits (so it's among the last evicted under memory
+# pressure), its data lives on a PVC (eviction = a transient restart, never
+# data loss), and a PodDisruptionBudget guards voluntary disruptions. Dropping
+# this cluster-scoped, fixed-name object also lets multiple tracebloc
+# namespaces coexist in one cluster (BYO) and removes the one-client-per-
+# cluster install collision.
 #
-# create: false  — chart does not template the PriorityClass; you (or your
-#                  GitOps tool / shared platform) manage it out-of-band.
-#                  The mysql pod still references `name`, so make sure the
-#                  PriorityClass exists at install time.
-# name: ""       — disable the priorityClassName reference entirely. mysql
-#                  falls back to the cluster default priority (0) and loses
-#                  the OOM-protection this chart's mysql tuning relies on.
+# Opt back in ONLY on a heavily-contended cluster where you want the scheduler
+# to preempt noisy training jobs to keep mysql scheduled:
+#   create: true, name: <some-name>, value: ~1000000  (above default 0, below
+#   system-cluster-critical 2,000,000,000).
+# For a PriorityClass your platform manages out-of-band, use create: false +
+# name: <existing-name> (mysql references it without the chart templating it).
 priorityClass:
-  create: true
-  name: tracebloc-data-plane
+  create: false
+  name: ""
   value: 1000000
 
 # -- PodDisruptionBudgets.

scripts/install-k8s.ps1

@@ -112,6 +112,15 @@ function ConvertTo-WorkspaceName {
   return $sanitized
 }
 
+# Best-effort chart version of the installed client release (e.g. "1.4.4");
+# empty if not found / cluster unreachable. Greps helm's CHART column.
+function Get-ChartVersion {
+  param([string]$Namespace = "tracebloc")
+  $out = (helm list -n $Namespace 2>$null) | Out-String
+  if ($out -match 'client-([0-9][^\s]*)') { return $Matches[1] }
+  return ""
+}
+
 # =============================================================================
 #  CONFIGURATION
 # =============================================================================
@@ -1006,7 +1015,6 @@ function Install-ClientHelm {
   }
   $valuesFile = Join-Path $HOST_DATA_DIR "values.yaml"
 
-  $defaultNamespace = "default"
   $defaultClientId = ""
   $defaultClientPassword = ""
 
@@ -1027,21 +1035,15 @@ function Install-ClientHelm {
     }
   }
 
-  # -- Workspace name prompt --
-  PromptHeader "Choose a workspace name"
-  Hint "This identifies your tracebloc client on this machine."
-  Write-Host ""
-  Hint "Examples: myteam, vision-lab, lukas"
-  Write-Host ""
-  $nsInput = Read-Host "  Workspace name [$defaultNamespace]"
-  $rawName = if ($nsInput) { $nsInput } else { $defaultNamespace }
-  $TB_NAMESPACE = ConvertTo-WorkspaceName -Input_ $rawName
+  # -- Namespace (fixed; not prompted) --
+  # The on-prem client is one-per-machine and is identified to the backend by
+  # its credentials (clientId), not by this name -- so we don't ask the user to
+  # invent one. It's just the local k8s namespace / Helm release name.
+  # Advanced / GitOps setups can override with TB_NAMESPACE=<name>.
+  $rawNs = if ($env:TB_NAMESPACE) { $env:TB_NAMESPACE } else { "tracebloc" }
+  $TB_NAMESPACE = ConvertTo-WorkspaceName -Input_ $rawNs
   $script:TB_NAMESPACE = $TB_NAMESPACE   # share with Wait-ForClientReady / Print-Summary
 
-  if ($TB_NAMESPACE -ne $rawName) {
-    Info "Using workspace: $TB_NAMESPACE"
-  }
-
   # -- Step 4/4: Connect to tracebloc network --
   Step 4 4 "Connect to tracebloc network"
 
@@ -1101,6 +1103,43 @@ function Install-ClientHelm {
     $defaultClientId = ""; $defaultClientPassword = ""
   }
 
+  # -- One-client-per-machine guard --
+  # A machine runs exactly one tracebloc client: it shares this cluster and the
+  # host's CPU/RAM/GPU, and the platform counts each client as separate
+  # capacity. If a DIFFERENT client is already installed here, a re-install
+  # would silently re-point the machine -- so we stop and let the operator
+  # decide. The same clientId is a normal re-run/upgrade and passes through.
+  # Check ANY namespace: a fresh install lands in 'tracebloc', but an install
+  # from an older installer version may be in a different namespace. Enumerate
+  # client-chart releases and read each clientId (ConvertFrom-Json -- no jq).
+  $existingId = ""; $existingNs = ""
+  $listJson = (helm list -A -o json 2>$null) | Out-String
+  if ($LASTEXITCODE -eq 0 -and $listJson.Trim()) {
+    try {
+      foreach ($rel in ($listJson | ConvertFrom-Json)) {
+        if ($rel.chart -and $rel.chart.StartsWith("client-")) {
+          $vals = (helm get values $rel.name -n $rel.namespace 2>$null) | Out-String
+          if ($vals -match 'clientId:\s*"([^"]+)"') { $existingId = $Matches[1].Trim(); $existingNs = $rel.namespace; break }
+        }
+      }
+    } catch { }
+  }
+  if ($existingId -and $existingId -ne $TB_CLIENT_ID) {
+    Write-Host ""
+    Warn "This machine already runs the tracebloc client '$existingId' (namespace '$existingNs')."
+    Hint "tracebloc runs one client per machine -- it shares this cluster and host"
+    Hint "resources, and the platform counts each client as separate capacity."
+    Write-Host ""
+    Hint "You entered a different Client ID ('$TB_CLIENT_ID'). Pick one:"
+    Hint "  - Repair / update '$existingId'  -> re-run with that same Client ID"
+    Hint "  - Switch to '$TB_CLIENT_ID'       -> remove the current client first:"
+    Hint "        k3d cluster delete $CLUSTER_NAME   (wipes this client + its local data)"
+    Hint "      then re-run this installer"
+    Hint "  - Run both clients                -> install on a separate machine"
+    Write-Host ""
+    Err "Refusing to replace the existing client. See the options above."
+  }
+
   $passwordEscaped = $TB_CLIENT_PASSWORD -replace "'", "''"
 
   $gpuVal = ""
@@ -1257,6 +1296,8 @@ function Print-Summary {
       Write-Host "  " -NoNewline; Write-Host "$([char]0x2714) Connected to tracebloc" -ForegroundColor Green
       Write-Host ""
       Write-Host "  Workspace : " -NoNewline; Write-Host $ns -ForegroundColor Cyan
+      $cver = Get-ChartVersion -Namespace $ns; if (-not $cver) { $cver = "unknown" }
+      Write-Host "  Version   : " -NoNewline; Write-Host $cver -ForegroundColor Cyan
       Write-Host "  Mode      : " -NoNewline; Write-Host $mode -ForegroundColor Cyan
       Write-Host ""
       Write-Host "  Your client is live. Confirm it shows as Online:"
@@ -1460,8 +1501,6 @@ function Invoke-DiagnoseBundle {
   $d = Join-Path $work "tracebloc-diagnose-$ts"
   New-Item -ItemType Directory -Path (Join-Path $d "logs") -Force | Out-Null
 
-  Info "Collecting diagnostics -- this is safe; credentials are redacted before the file is written."
-
   # Namespace discovery (TB_NAMESPACE isn't set on a standalone diagnose run).
   $ns = $TB_NAMESPACE
   if (-not $ns) {
@@ -1470,9 +1509,14 @@ function Invoke-DiagnoseBundle {
   }
   if (-not $ns) { $ns = "default" }
 
+  # Surface the client version first -- the #1 thing support needs to know.
+  $cver = Get-ChartVersion -Namespace $ns; if (-not $cver) { $cver = "unknown" }
+  Info "tracebloc client version: $cver   (namespace: $ns)"
+  Info "Collecting diagnostics -- this is safe; credentials are redacted before the file is written."
+
   # host / versions
   $h = @("# tracebloc diagnose ($ts)", "OS: Windows  ARCH: $(Get-WindowsArch)",
-         "CLIENT_ENV: $($env:CLIENT_ENV)  CLUSTER_NAME: $cn  NAMESPACE: $ns", "## versions",
+         "CLIENT_ENV: $($env:CLIENT_ENV)  CLUSTER_NAME: $cn  NAMESPACE: $ns", "CLIENT VERSION: $cver", "## versions",
          (k3d version 2>&1 | Out-String), (kubectl version --client 2>&1 | Out-String),
          (helm version --short 2>&1 | Out-String), (docker version 2>&1 | Out-String))
   try { $cs = Get-CimInstance Win32_ComputerSystem -ErrorAction Stop; $h += "CPUs=$($cs.NumberOfLogicalProcessors)  MemBytes=$($cs.TotalPhysicalMemory)" } catch {}

scripts/install-k8s.sh

@@ -15,6 +15,8 @@
 #
 #  Environment variable overrides (optional):
 #    CLUSTER_NAME=myapp          default: tracebloc
+#    TB_NAMESPACE=myns           default: tracebloc  (k8s namespace + local label;
+#                                not prompted — the client is identified by its credentials)
 #    SERVERS=1                   default: 1  (control-plane nodes)
 #    AGENTS=1                    default: 1  (worker nodes)
 #    K8S_VERSION=v1.29.4-k3s1   default: latest stable k3s

scripts/lib/common.sh

@@ -35,6 +35,15 @@ hint()           { echo -e "  ${DIM}$*${RESET}"; }
 # ── Utility ──────────────────────────────────────────────────────────────────
 has() { command -v "$1" &>/dev/null; }
 
+# Best-effort chart version of the installed client release in namespace $1
+# (e.g. "1.4.4"); empty if not found / cluster unreachable. Greps helm's CHART
+# column ("client-<ver>"), so it needs no jq.
+_chart_version() {
+  local ns="${1:-${TB_NAMESPACE:-tracebloc}}"
+  has helm || return 0
+  helm list -n "$ns" 2>/dev/null | grep -oE 'client-[0-9][^[:space:]]*' | head -1 | sed 's/^client-//'
+}
+
 # ── macOS: Docker Desktop architecture vs machine (for wrong-arch UX) ────────
 #  Call early on macOS to fail fast with clear instructions if Docker.app
 #  is for the wrong architecture (e.g. Intel Docker on Apple Silicon).
@@ -361,6 +370,7 @@ Commands:
 
 Advanced configuration (environment variables):
   CLUSTER_NAME   Cluster name                   (default: tracebloc)
+  TB_NAMESPACE   Namespace / workspace label    (default: tracebloc)
   SERVERS        Control-plane nodes             (default: 1)
   AGENTS         Worker nodes                    (default: 1)
   K8S_VERSION    k3s image tag                   (default: v1.29.4-k3s1)

scripts/lib/diagnose.sh

@@ -49,8 +49,6 @@ run_diagnose() {
   if [[ -z "$outdir" || ! -d "$outdir" ]]; then echo "  diagnose: cannot create a temp directory" >&2; return 1; fi
   d="$outdir/tracebloc-diagnose-$ts"; mkdir -p "$d/logs"
 
-  info "Collecting diagnostics — this is safe; credentials are redacted before the file is written."
-
   # Namespace discovery — TB_NAMESPACE isn't set on a standalone diagnose run,
   # so find the namespace of the jobs-manager pod (falls back to "default").
   ns="${TB_NAMESPACE:-}"
@@ -59,12 +57,18 @@ run_diagnose() {
   fi
   [[ -z "$ns" ]] && ns="default"
 
+  # Surface the client version first — the #1 thing support needs to know.
+  local cver; cver="$(_chart_version "$ns")"
+  info "tracebloc client version: ${cver:-unknown}   (namespace: $ns)"
+  info "Collecting diagnostics — this is safe; credentials are redacted before the file is written."
+
   # ── host / versions ──
   {
     echo "# tracebloc diagnose ($ts)"
     echo "OS:   $(uname -s) $(uname -r)"
     echo "ARCH: $(uname -m)"
     echo "CLIENT_ENV: ${CLIENT_ENV:-<unset>}   CLUSTER_NAME: $cn   NAMESPACE: $ns"
+    echo "CLIENT VERSION: ${cver:-unknown}"
     echo; echo "## versions"
     has k3d     && k3d version
     has kubectl && kubectl version --client 2>/dev/null

scripts/lib/install-client-helm.sh

@@ -153,13 +153,12 @@ install_client_helm() {
   if [[ -n "${TRACEBLOC_VALUES_FILE:-}" ]]; then
     [[ -f "$TRACEBLOC_VALUES_FILE" ]] || error "TRACEBLOC_VALUES_FILE not found: $TRACEBLOC_VALUES_FILE"
     values_file="$TRACEBLOC_VALUES_FILE"
-    TB_NAMESPACE="${TB_NAMESPACE:-default}"
+    TB_NAMESPACE="${TB_NAMESPACE:-tracebloc}"
     info "Dev mode: using caller-provided values file"
     log "Using values file: $values_file (namespace: $TB_NAMESPACE)"
   else
 
   local use_existing=""
-  local default_namespace="default"
   local default_client_id=""
   local default_client_password=""
 
@@ -179,19 +178,12 @@ install_client_helm() {
     fi
   fi
 
-  # ── Workspace name prompt ────────────────────────────────────────────────
-  prompt_header "Choose a workspace name"
-  hint "This identifies your tracebloc client on this machine."
-  echo ""
-  hint "Examples: berlin-team, vision-lab, ml-mardan"
-  echo ""
-  read -r -p "  Workspace name [${default_namespace}]: " TB_NAMESPACE_INPUT
-  local raw_name="${TB_NAMESPACE_INPUT:-$default_namespace}"
-  TB_NAMESPACE=$(_sanitize_workspace_name "$raw_name")
-
-  if [[ "$TB_NAMESPACE" != "$raw_name" ]]; then
-    info "Using workspace: ${BOLD}${TB_NAMESPACE}${RESET}"
-  fi
+  # ── Namespace (fixed; not prompted) ──────────────────────────────────────
+  # The on-prem client is one-per-machine and is identified to the backend by
+  # its credentials (clientId), not by this name — so we don't ask the user to
+  # invent one. It's just the local k8s namespace / Helm release name.
+  # Advanced / GitOps setups can override with TB_NAMESPACE=<name>.
+  TB_NAMESPACE=$(_sanitize_workspace_name "${TB_NAMESPACE:-tracebloc}")
 
   # ── Step 4/4: Connect to tracebloc network ──────────────────────────────
   step 4 4 "Connect to tracebloc network"
@@ -255,6 +247,46 @@ install_client_helm() {
     default_client_id=""; default_client_password=""
   done
 
+  # ── One-client-per-machine guard ─────────────────────────────────────────
+  # A machine runs exactly one tracebloc client: it shares this cluster and the
+  # host's CPU/RAM/GPU, and the platform counts each client as separate
+  # capacity. If a DIFFERENT client is already installed here, a re-install
+  # would silently re-point the machine — so we stop and let the operator
+  # decide. The same clientId is a normal re-run/upgrade and passes through.
+  # Check ANY namespace: a fresh install lands in `tracebloc`, but an install
+  # from an older installer version may be in a different namespace. Enumerate
+  # client-chart releases and read each clientId. jq is already used elsewhere
+  # in the installer; if it's somehow absent, fall back to the `tracebloc` ns.
+  local existing_id="" existing_ns="" _gvf _rel _ns _id
+  _gvf="$(mktemp)"
+  if has jq; then
+    while IFS=$'\t' read -r _rel _ns; do
+      [[ -z "$_rel" ]] && continue
+      if helm get values "$_rel" -n "$_ns" > "$_gvf" 2>/dev/null; then
+        _id="$(_extract_yaml_value "$_gvf" clientId)"
+        [[ -n "$_id" ]] && { existing_id="$_id"; existing_ns="$_ns"; break; }
+      fi
+    done < <(helm list -A -o json 2>/dev/null | jq -r '.[] | select((.chart // "") | startswith("client-")) | "\(.name)\t\(.namespace)"')
+  elif helm get values "$TB_NAMESPACE" -n "$TB_NAMESPACE" > "$_gvf" 2>/dev/null; then
+    existing_id="$(_extract_yaml_value "$_gvf" clientId)"; existing_ns="$TB_NAMESPACE"
+  fi
+  rm -f "$_gvf"
+  if [[ -n "$existing_id" && "$existing_id" != "$TB_CLIENT_ID" ]]; then
+    echo ""
+    warn "This machine already runs the tracebloc client '${existing_id}' (namespace '${existing_ns}')."
+    hint "tracebloc runs one client per machine — it shares this cluster and host"
+    hint "resources, and the platform counts each client as separate capacity."
+    echo ""
+    hint "You entered a different Client ID ('${TB_CLIENT_ID}'). Pick one:"
+    hint "  • Repair / update '${existing_id}'  →  re-run with that same Client ID"
+    hint "  • Switch to '${TB_CLIENT_ID}'        →  remove the current client first:"
+    hint "        k3d cluster delete ${CLUSTER_NAME:-tracebloc}   (wipes this client + its local data)"
+    hint "      then re-run this installer"
+    hint "  • Run both clients                   →  install on a separate machine"
+    echo ""
+    error "Refusing to replace the existing client. See the options above."
+  fi
+
   TB_CLIENT_PASSWORD_ESCAPED="${TB_CLIENT_PASSWORD//\'/\'\'}"
 
   # ── GPU limits ──────────────────────────────────────────────────────────

scripts/lib/summary.sh

@@ -89,6 +89,7 @@ print_summary() {
   [[ "$GPU_VENDOR" == "nvidia" ]] && mode="NVIDIA GPU"
   [[ "$GPU_VENDOR" == "amd" ]] && mode="AMD GPU"
   local ns="${TB_NAMESPACE:-default}"
+  local cver; cver="$(_chart_version "$ns")"
   local line="━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
 
   echo ""
@@ -99,6 +100,7 @@ print_summary() {
       echo -e "  ${BOLD}${GREEN}✔ Connected to tracebloc${RESET}"
       echo ""
       echo -e "  ${BOLD}Workspace${RESET} : ${CYAN}${ns}${RESET}"
+      echo -e "  ${BOLD}Version${RESET}   : ${CYAN}${cver:-unknown}${RESET}"
       echo -e "  ${BOLD}Mode${RESET}      : ${CYAN}${mode}${RESET}"
       echo ""
       echo -e "  Your client is live. Confirm it shows as ${BOLD}🟢 Online${RESET}:"

scripts/tests/diagnose.bats

@@ -104,3 +104,14 @@ setup() {
   # Finding 2 (security review): `helm get manifest` (base64 Secrets) is NOT collected
   ! tar -xzOf "$tgz" 2>/dev/null | grep -q 'get manifest'
 }
+
+@test "run_diagnose: surfaces + records the client version" {
+  has() { case "$1" in helm) return 0 ;; *) return 1 ;; esac; }   # only helm present
+  helm() { echo "tracebloc tracebloc 1 now deployed client-1.4.4 1.4.4"; }
+  run run_diagnose
+  [ "$status" -eq 0 ]
+  [[ "$output" == *"client version: 1.4.4"* ]]
+  tgz="$(ls "$HOST_DATA_DIR"/tracebloc-diagnose-*.tgz 2>/dev/null | head -1)"
+  [ -n "$tgz" ]
+  tar -xzOf "$tgz" 2>/dev/null | grep -q 'CLIENT VERSION: 1.4.4'
+}