views,*: Improved NVidia support + fixes

Improve the NVidia parser-file, add stub list- and info-views for NVIDIADriver, fix a bug in the logparser key_value rule (hopefully without introducing new issues), and add a bit of highliting for some error states. Signed-off-by: David Weinehall <[email protected]>
intel · Oct 26, 2024 · a063d51 · a063d51
1 parent b66df71
commit a063d51
Show file tree

Hide file tree

Showing 10 changed files with 210 additions and 4 deletions.
diff --git a/clustermanagementtoolkit/logparser.py b/clustermanagementtoolkit/logparser.py
@@ -2075,8 +2075,8 @@ def key_value(message: str, **kwargs: Any) -> Tuple[str, LogLevel, str,
     collector_bullets = deep_get(options, DictPath("collector_bullets"), False)
     is_event: bool = deep_get(options, DictPath("is_event"), False)
 
-    # Split all key=value pairs.
-    key_value_regex = re.compile(r"^(.*?)=(.*)")
+    # Split all key=value pairs. Make sure not to process "=="
+    key_value_regex = re.compile(r"^(.*?[^=])=($|[^=].*$)")
     tmp = re.findall(r"(?:\".*?\"|\S)+", message.replace("\\\"", "<<<quote>>>"))
     # pylint: disable-next=too-many-nested-blocks
     if tmp is not None:

diff --git a/parsers/nvidia.yaml b/parsers/nvidia.yaml
@@ -1,13 +1,68 @@
+- name: "nvidia"
+  show_in_selector: false
+  matchkeys:
+  - image_name: "/nvidia/k8s-device-plugin"
+  - image_name: "/nvidia/cloud-native/k8s-driver-manager"
+    container_type: "init_container"
+  parser_rules:
+  - name: "glog"
+  - name: "custom_line"
+    options:
+      block_start:
+      - matchtype: "exact"
+        matchkey: "Running with config:"
+        matchline: "any"
+        format_block_start: true
+      block_end:
+      - matchtype: "exact"
+        matchkey: "}"
+        matchline: "any"
+        format_block_end: true
+      loglevel: "info"
 - name: "nvidia"
   show_in_selector: false
   matchkeys:
   - image_name: "/nvidia/k8s/container-toolkit"
+  - image_name: "/nvidia/k8s/dcgm-exporter"
   - image_name: "/nvidia/cloud-native/gpu-operator-validator"
     container_type: "init_container"
   parser_rules:
   - name: "key_value"
   - name: "override_severity"
     overrides:
+    - matchtype: "startswith"
+      matchkey: "version: "
+      loglevel: "notice"
+    - matchtype: "startswith"
+      matchkey: "NVIDIA-SMI has failed"
+      loglevel: "error"
     - matchtype: "startswith"
       matchkey: "CRITICAL:"
       loglevel: "critical"
+  - name: "custom_line"
+    options:
+      block_start:
+      - matchtype: "exact"
+        matchkey: "Using config:"
+        matchline: "any"
+        format_block_start: true
+      block_end:
+      - matchtype: "startswith"
+        matchkey: "time"
+        matchline: "any"
+        format_block_end: false
+        process_block_end: false
+      loglevel: "info"
+  - name: "custom_line"
+    options:
+      block_start:
+      - matchtype: "startswith"
+        matchkey: "+---------------"
+        matchline: "any"
+        format_block_start: true
+      block_end:
+      - matchtype: "startswith"
+        matchkey: "+---------------"
+        matchline: "any"
+        format_block_end: true
+      loglevel: "info"
diff --git a/views/Challenge.acme.cert-manager.io.yaml b/views/Challenge.acme.cert-manager.io.yaml
@@ -6,5 +6,46 @@ listview:
   group: "Certificate Management"
   field_indexes:
     Normal:
-      fields: ["namespace", "name", "age"]
+      fields: ["namespace", "name", "state", "age"]
       sortcolumn: "namespace"
+  fields:
+    state:
+      header: "State:"
+      path: "status#state"
+      type: "str"
+      formatter: "value_mapper"
+      formatting:
+        mapping:
+          mappings:
+            errored:
+              field_colors:
+              - context: "main"
+                type: "status_not_ok"
+            expired:
+              field_colors:
+              - context: "main"
+                type: "status_warning"
+            invalid:
+              field_colors:
+              - context: "main"
+                type: "status_not_ok"
+            pending:
+              field_colors:
+              - context: "main"
+                type: "status_pending"
+            processing:
+              field_colors:
+              - context: "main"
+                type: "status_pending"
+            ready:
+              field_colors:
+              - context: "main"
+                type: "status_done"
+            valid:
+              field_colors:
+              - context: "main"
+                type: "status_ok"
+            __default:
+              field_colors:
+              - context: "main"
+                type: "status_unknown"
diff --git a/views/ClusterPolicy.nvidia.com.yaml b/views/ClusterPolicy.nvidia.com.yaml
@@ -11,7 +11,7 @@ command:
 - "nvclrpol"
 listview:
   name: "Cluster Policies"
-  group: "API & Extendability"
+  group: "Accelerators"
   field_indexes:
     Normal:
       fields: ["name", "state", "age"]
@@ -24,10 +24,18 @@ listview:
       formatting:
         mapping:
           mappings:
+            ignored:
+              field_colors:
+              - context: "main"
+                type: "status_ok"
             notReady:
               field_colors:
               - context: "main"
                 type: "status_not_ok"
+            ready:
+              field_colors:
+              - context: "main"
+                type: "status_ok"
 infoview:
   name: "Cluster Policy Info"
   infopad:

diff --git a/views/Event.events.k8s.io.yaml b/views/Event.events.k8s.io.yaml
@@ -699,6 +699,10 @@ listview:
               field_colors:
               - context: "main"
                 type: "status_ok"
+            GPUDriverUpgrade:
+              field_colors:
+              - context: "main"
+                type: "status_ok"
             HcoUpdateError:
               field_colors:
               - context: "main"
@@ -2466,6 +2470,10 @@ infoview:
                 field_colors:
                 - context: "main"
                   type: "status_ok"
+              GPUDriverUpgrade:
+                field_colors:
+                - context: "main"
+                  type: "status_ok"
               HcoUpdateError:
                 field_colors:
                 - context: "main"

diff --git a/views/Event.yaml b/views/Event.yaml
@@ -694,6 +694,10 @@ listview:
               field_colors:
               - context: "main"
                 type: "status_ok"
+            GPUDriverUpgrade:
+              field_colors:
+              - context: "main"
+                type: "status_ok"
             HcoUpdateError:
               field_colors:
               - context: "main"
@@ -2461,6 +2465,10 @@ infoview:
                 field_colors:
                 - context: "main"
                   type: "status_ok"
+              GPUDriverUpgrade:
+                field_colors:
+                - context: "main"
+                  type: "status_ok"
               HcoUpdateError:
                 field_colors:
                 - context: "main"

diff --git a/views/NVIDIADriver.nvidia.com.yaml b/views/NVIDIADriver.nvidia.com.yaml
@@ -0,0 +1,43 @@
+kind: "NVIDIADriver"
+api_family: "nvidia.com"
+default_command: "nvidiadrivers"
+command:
+- "nvidiadrvs"
+- "nvidiadrv"
+- "nvdrvs"
+- "nvdrv"
+listview:
+  name: "NVIDIA Drivers"
+  group: "Accelerators"
+  field_indexes:
+    Normal:
+      fields: ["name", "driver_type", "state", "age"]
+  fields:
+    driver_type:
+      header: "Driver Type:"
+      path: "spec#driverType"
+      default: "gpu"
+      type: "str"
+    state:
+      header: "State:"
+      path: "status#state"
+      type: "str"
+      formatter: "value_mapper"
+      formatting:
+        mapping:
+          mappings:
+            ignored:
+              field_colors:
+              - context: "main"
+                type: "status_ok"
+            notReady:
+              field_colors:
+              - context: "main"
+                type: "status_not_ok"
+            ready:
+              field_colors:
+              - context: "main"
+                type: "status_ok"
+infoview:
+  name: "NVIDIA Driver Info"
+  infopad:
diff --git a/views/NodeFeatureGroup.nfd.k8s-sigs.io.yaml b/views/NodeFeatureGroup.nfd.k8s-sigs.io.yaml
@@ -10,3 +10,6 @@ listview:
   field_indexes:
     Normal:
       fields: ["namespace", "name", "age"]
+infoview:
+  name: "Node Feature Group Info"
+  infopad:
diff --git a/views/Order.acme.cert-manager.io.yaml b/views/Order.acme.cert-manager.io.yaml
@@ -21,3 +21,39 @@ listview:
       header: "State:"
       path: "status#state"
       type: "str"
+      formatter: "value_mapper"
+      formatting:
+        mapping:
+          mappings:
+            errored:
+              field_colors:
+              - context: "main"
+                type: "status_not_ok"
+            expired:
+              field_colors:
+              - context: "main"
+                type: "status_warning"
+            invalid:
+              field_colors:
+              - context: "main"
+                type: "status_not_ok"
+            pending:
+              field_colors:
+              - context: "main"
+                type: "status_pending"
+            processing:
+              field_colors:
+              - context: "main"
+                type: "status_pending"
+            ready:
+              field_colors:
+              - context: "main"
+                type: "status_done"
+            valid:
+              field_colors:
+              - context: "main"
+                type: "status_ok"
+            __default:
+              field_colors:
+              - context: "main"
+                type: "status_unknown"
diff --git a/views/variables/event_reasons.var b/views/variables/event_reasons.var
@@ -628,6 +628,10 @@ Generated:
   field_colors:
   - context: "main"
     type: "status_ok"
+GPUDriverUpgrade:
+  field_colors:
+  - context: "main"
+    type: "status_ok"
 HcoUpdateError:
   field_colors:
   - context: "main"