USHIFT-706: Add CSI Deployment Flag and Config

cmd/generate-config/config/config-openapi-spec.json

@@ -9,7 +9,8 @@
     "kubelet",
     "manifests",
     "network",
-    "node"
+    "node",
+    "storage"
   ],
   "properties": {
     "apiServer": {
@@ -267,6 +268,34 @@
           "type": "string"
         }
       }
+    },
+    "storage": {
+      "description": "Storage represents a subfield of the MicroShift config data structure. Its purpose to provide a user\nfacing interface to control whether MicroShift should deploy LVMS on startup.",
+      "type": "object",
+      "properties": {
+        "driver": {
+          "description": "Driver is a user defined string value matching one of the above CSIStorageDriver values. MicroShift uses this\nvalue to decide whether to deploy the LVMS operator. An unset field defaults to \"\" during yaml parsing, and thus\ncould mean that the cluster has been upgraded. In order to support the existing out-of-box behavior, MicroShift\nassumes an empty string to mean the storage driver should be deployed.\nAllowed values are: unset or one of [\"\", \"lvms\", \"none\"]",
+          "type": "string",
+          "enum": [
+            "",
+            "none",
+            "lvms"
+          ]
+        },
+        "optionalCsiComponents": {
+          "description": "OptionalCSIComponents is a user defined slice of CSIComponent values. These value tell MicroShift which\nadditional, non-driver, CSI controllers to deploy on start. MicroShift will deploy snapshot controller\nand webhook when no components are specified. This preserves the current deployment behavior of existing\nclusters. Users must set `.storage.optionalCsiComponents: []` to explicitly tell MicroShift not to deploy any CSI\ncomponents. The CSI Driver is excluded as it is typically deployed via the same manifest as the accompanying\nstorage driver. Like CSIStorageDriver, uninstallation is not supported as this can lead to orphaned storage\nobjects.\nAllowed values are: unset, [], or one or more of [\"snapshot-controller\", \"snapshot-webhook\"]",
+          "type": "array",
+          "items": {
+            "description": "OptionalCsiComponent values determine which CSI components MicroShift should deploy. Currently only csi snapshot components\nare supported.",
+            "type": "string",
+            "enum": [
+              "none",
+              "snapshot-controller",
+              "snapshot-webhook"
+            ]
+          }
+        }
+      }
     }
   }
 }

docs/user/howto_config.md

@@ -50,6 +50,10 @@ node:
     hostnameOverride: ""
     nodeIP: ""
     nodeIPv6: ""
+storage:
+    driver: ""
+    optionalCsiComponents:
+        - ""
 
 ```
 <!---
@@ -110,6 +114,10 @@ node:
     hostnameOverride: ""
     nodeIP: ""
     nodeIPv6: ""
+storage:
+    driver: ""
+    optionalCsiComponents:
+        - ""
 
 ```
 <!---

packaging/greenboot/functions.sh

@@ -294,3 +294,95 @@ function log_script_exit() {
         echo "FINISHED"
     fi
 }
+
+# lvmsDriverShouldExist shallowly wraps some python yaml parsing to determine if the LVMS-Operator is expected to be
+# deployed.
+# args:
+#   none
+# return:
+#   0:  "true", expect the driver to be running
+#   1:  "false", do not expect the driver to be running
+function lvmsDriverShouldExist() {
+    # The python script checks the config's .storage.driver value. Parsing errors are ignored, these will be caught
+    # by MicroShift during startup, before this greenboot script is run. The script favors verbosity over simplified
+    # code.
+    # args:
+    #   None
+    # return:
+    #   None
+    # exit code:
+    #   0: The lvms components should be checked for. This case occurs for any one of:
+    #       - .storage.driver == one of ["", "lvms"]
+    #       - .storage.driver is unset (indicates default config)
+    #       - .storage is unset (indicates default config)
+    #       - the file cannot be opened (indicates default config)
+    #   1: The lvms components should not be checked for. This case occurs for:
+    #       - .storage.driver == "none"
+    if python -c \
+'import yaml, sys
+try:
+    with open("/etc/microshift/config.yaml") as stream:
+        try:
+            config = yaml.safe_load(stream)
+            if config["storage"]["driver"] == "none":
+                sys.exit(1)
+            elif config["storage"]["driver"] == "lvms":
+                sys.exit(0)
+            elif config["storage"]["driver"] == "":
+                sys.exit(0)
+            else:
+                sys.exit(1)
+        except KeyError as e:
+            sys.exit(0)
+        finally:
+            stream.close()
+except Exception as e:
+    print(f"error opening config: {e}", file=sys.Stderr)
+    '; then
+         return 0
+    fi
+    return 1
+}
+
+# csiComponentsShouldBeDeployed checks for a given csi component string identifier in the config's
+# .storage.optionalCsiComponents array.
+# args:
+#   $1    "some-csi-component"
+# return:
+#   0:    the component should be checked for. This case occurs for any one of:
+#           - .storage is unset
+#           - .storage.optionalCsiComponents is unset
+#           - "some-csi-component" matches an element in .storage.optionalCsiComponents
+#   1:    the component should not be checked for. This case occurs for:
+#           - .storage.optionalCsiComponents is set and does not contain "some-csi-component"
+#           - .storage.optionalCsiComponents is set and len(.storage.optionalCsiComponents) == 0
+function csiComponentShouldBeDeployed() {
+    local -r component="${1?csiComponentsShouldBeDeployed expects a single string argument, got none}"
+
+    if python -c \
+'import sys
+import yaml
+try:
+    with open("/etc/microshift/config.yaml") as stream:
+        try:
+            config = yaml.safe_load(stream)
+            if len(config["storage"]["optionalCsiComponents"]) == 0:
+                sys.exit(0)
+            elif "none" in config["storage"]["optionalCsiComponents"]:
+                sys.exit(1)
+            elif sys.argv[1] in config["storage"]["optionalCsiComponents"]:
+                sys.exit(0)
+            else:
+                sys.exit(1)
+        except KeyError as e:
+            sys.exit(0)
+        finally:
+            stream.close()
+except Exception as e:
+    print(f"error opening config: {e}", file=sys.Stderr)
+' "${component}";
+    then
+        return 0
+    fi
+    return 1
+}

packaging/greenboot/microshift-running-check.sh

@@ -3,8 +3,8 @@ set -e
 
 SCRIPT_NAME=$(basename "$0")
 SCRIPT_PID=$$
-PODS_NS_LIST=(openshift-ovn-kubernetes openshift-service-ca openshift-ingress openshift-dns kube-system)
-PODS_CT_LIST=(2                        1                    1                 2             2)
+PODS_NS_LIST=(openshift-ovn-kubernetes openshift-service-ca openshift-ingress openshift-dns)
+PODS_CT_LIST=(2 1 1 2)
 LOG_POD_EVENTS=false
 
 # Source the MicroShift health check functions library
@@ -25,19 +25,21 @@ function forced_termination() {
     exit 1
 }
 
-# Check preconditions for existence of topolvm deployment.
+# Check preconditions for existence of lvms deployment.
 # Adapted from MicroShift code.
 #
 # args: None
-# return: 0 if topolvm readiness should be checked, 1 otherwise
-function topolvmShouldBeDeployed() {
+# return: 0 if lvms readiness should be checked, 1 otherwise
+function lvmsShouldBeDeployed() {
     if ! hash vgs 2>/dev/null; then
         return 1
     fi
-
     if [ -f /etc/microshift/lvmd.yaml ]; then
         return 0
     fi
+    if ! lvmsDriverShouldExist; then
+        return 1
+    fi
 
     local -r volume_groups=$(vgs --readonly --options=name --noheadings)
     local -r volume_groups_count=$(echo "${volume_groups}" | wc -w)
@@ -139,10 +141,21 @@ if ! wait_for "${WAIT_TIMEOUT_SECS}" microshift_health_endpoints_ok ; then
     exit 1
 fi
 
-if topolvmShouldBeDeployed; then
+if lvmsShouldBeDeployed; then
     PODS_NS_LIST+=(openshift-storage)
     PODS_CT_LIST+=(2)
 fi
+declare -a csi_components=('csi-snapshot-controller' 'csi-snapshot-webhook')
+csi_pods_ct=0
+for csi_c in "${csi_components[@]}"; do
+    if csiComponentShouldBeDeployed "${csi_c}"; then
+        (( csi_pods_ct += 1 ))
+    fi
+done
+if [ ${csi_pods_ct} -gt 0 ]; then
+    PODS_NS_LIST+=(kube-system)
+    PODS_CT_LIST+=("${csi_pods_ct}")
+fi
 
 # Starting pod-specific checks
 # Log list of pods and their events on failure
@@ -160,7 +173,7 @@ for i in "${!PODS_NS_LIST[@]}"; do
     CHECK_PODS_NS=${PODS_NS_LIST[${i}]}
 
     echo "Waiting ${WAIT_TIMEOUT_SECS}s for pod image(s) from the '${CHECK_PODS_NS}' namespace to be downloaded"
-    if ! wait_for "${WAIT_TIMEOUT_SECS}" namespace_images_downloaded ; then
+    if ! wait_for "${WAIT_TIMEOUT_SECS}" namespace_images_downloaded; then
         echo "Error: Timed out waiting for pod image(s) from the '${CHECK_PODS_NS}' namespace to be downloaded"
         exit 1
     fi
@@ -172,7 +185,7 @@ for i in "${!PODS_NS_LIST[@]}"; do
     CHECK_PODS_CT=${PODS_CT_LIST[${i}]}
 
     echo "Waiting ${WAIT_TIMEOUT_SECS}s for ${CHECK_PODS_CT} pod(s) from the '${CHECK_PODS_NS}' namespace to be in 'Ready' state"
-    if ! wait_for "${WAIT_TIMEOUT_SECS}" namespace_pods_ready ; then
+    if ! wait_for "${WAIT_TIMEOUT_SECS}" namespace_pods_ready; then
         echo "Error: Timed out waiting for ${CHECK_PODS_CT} pod(s) in the '${CHECK_PODS_NS}' namespace to be in 'Ready' state"
         exit 1
     fi