poc - etcd analysis uses UBI, installs required tools at runtime

Author: cadenmarchese

cmd/aro/mirror.go

@@ -159,9 +159,6 @@ func mirror(ctx context.Context, _log *logrus.Entry) error {
 		// https://gitlab.cee.redhat.com/service/app-interface/-/blob/master/data/services/osd-operators/cicd/saas/saas-managed-upgrade-operator.yaml?ref_type=heads
 		"quay.io/app-sre/managed-upgrade-operator:v0.1.1297-ge922e64",
 
-		// https://quay.io/repository/redhat_emp1/octosql-etcd
-		"quay.io/redhat_emp1/octosql-etcd:latest",
-
 		// https://registry.redhat.io/openshift4/ose-must-gather (standard OpenShift must-gather)
 		"registry.redhat.io/openshift4/ose-must-gather:latest",
 

pkg/frontend/admin_openshiftcluster_etcdanalysis.go

@@ -29,7 +29,7 @@ import (
 )
 
 const (
-	etcdAnalysisImage        = "arosvc.azurecr.io/redhat_emp1/octosql-etcd:latest"
+	etcdAnalysisImage        = "arosvc.azurecr.io/ubi9/ubi-minimal:latest"
 	etcdDataDir              = "/var/lib/etcd"
 	etcdAnalysisSAPrefix     = "etcd-analysis-privileged"
 	etcdAnalysisSuffixLength = 8
@@ -175,8 +175,10 @@ func etcdAnalysisStream(ctx context.Context, log *logrus.Entry, k adminactions.K
 	runJobStream(ctx, log, k, job, nopWriteCloser{w}, defaultJobRetryDelay)
 }
 
-// buildEtcdAnalysisJob builds the Job that runs octosql-etcd against the snapshot.
+// buildEtcdAnalysisJob builds the Job that runs OctoSQL analysis against the snapshot.
 func buildEtcdAnalysisJob(vmName, filename, saName string) *batchv1.Job {
+	snapshotPath := "/snapshot/" + filename
+
 	return &batchv1.Job{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      "etcd-analysis-" + utilrand.String(5),
@@ -196,10 +198,13 @@ func buildEtcdAnalysisJob(vmName, filename, saName string) *batchv1.Job {
 					},
 					Containers: []corev1.Container{
 						{
-							Name:    "analyzer",
-							Image:   etcdAnalysisImage,
-							Command: []string{"/usr/local/bin/analyze-snapshot.sh"},
-							Args:    []string{"--delete", "/snapshot/" + filename},
+							Name:  "analyzer",
+							Image: etcdAnalysisImage,
+							Command: []string{
+								"/bin/bash", "-c",
+								etcdAnalysisScript,
+							},
+							Args: []string{"--delete", snapshotPath},
 							SecurityContext: &corev1.SecurityContext{
 								Privileged: pointerutils.ToPtr(true),
 							},

pkg/frontend/admin_openshiftcluster_etcdanalysis_test.go

@@ -419,14 +419,18 @@ func TestBuildEtcdAnalysisJob(t *testing.T) {
 	if c.Image != etcdAnalysisImage {
 		t.Errorf("container.Image = %q, want %q", c.Image, etcdAnalysisImage)
 	}
-	wantCommand := []string{"/usr/local/bin/analyze-snapshot.sh"}
+	wantCommand := []string{"/bin/bash", "-c", etcdAnalysisScript}
 	if len(c.Command) != len(wantCommand) {
-		t.Errorf("container.Command = %v, want %v", c.Command, wantCommand)
+		t.Errorf("container.Command length = %d, want %d", len(c.Command), len(wantCommand))
 	} else {
-		for i, v := range wantCommand {
-			if c.Command[i] != v {
-				t.Errorf("container.Command[%d] = %q, want %q", i, c.Command[i], v)
-			}
+		if c.Command[0] != wantCommand[0] {
+			t.Errorf("container.Command[0] = %q, want %q", c.Command[0], wantCommand[0])
+		}
+		if c.Command[1] != wantCommand[1] {
+			t.Errorf("container.Command[1] = %q, want %q", c.Command[1], wantCommand[1])
+		}
+		if c.Command[2] != etcdAnalysisScript {
+			t.Errorf("container.Command[2] does not match embedded etcdAnalysisScript")
 		}
 	}
 	if len(c.Args) < 2 {

pkg/frontend/scripts.go

@@ -10,3 +10,6 @@ var backupOrFixEtcd string
 
 //go:embed scripts/etcdkeycount.sh
 var etcdKeyCountScript string
+
+//go:embed scripts/etcdanalysis.sh
+var etcdAnalysisScript string

pkg/frontend/scripts/etcdanalysis.sh

@@ -0,0 +1,71 @@
+#!/bin/bash
+# Copyright (c) Microsoft Corporation.
+# Licensed under the Apache License 2.0.
+#
+# Installs OctoSQL and analyzes an etcd snapshot.
+# Usage: etcdanalysis.sh [--delete] /path/to/snapshot.snapshot
+set -euo pipefail
+
+DELETE_SNAPSHOT=false
+if [ "${1:-}" = "--delete" ]; then
+    DELETE_SNAPSHOT=true
+    shift
+fi
+
+SNAPSHOT_PATH="${1:-}"
+if [ -z "$SNAPSHOT_PATH" ]; then
+    echo "Error: snapshot path required" >&2
+    echo "Usage: $0 [--delete] /path/to/snapshot.snapshot" >&2
+    exit 1
+fi
+
+OCTOSQL_VERSION="0.13.0"
+
+echo "Installing dependencies..."
+microdnf install -y curl tar gzip jq ca-certificates 2>&1
+
+echo "Downloading OctoSQL ${OCTOSQL_VERSION}..."
+curl -sL "https://github.com/cube2222/octosql/releases/download/v${OCTOSQL_VERSION}/octosql_${OCTOSQL_VERSION}_linux_amd64.tar.gz" | \
+    tar -xz -C /usr/local/bin octosql
+chmod +x /usr/local/bin/octosql
+
+echo "Installing etcdsnapshot plugin..."
+export HOME=/tmp
+export OCTOSQL_NO_TELEMETRY=1
+octosql plugin repository add https://raw.githubusercontent.com/tjungblu/octosql-plugin-etcdsnapshot/main/plugin_repository.json 2>&1
+octosql plugin install etcdsnapshot/etcdsnapshot 2>&1
+
+echo "Analyzing snapshot: ${SNAPSHOT_PATH}"
+SNAPSHOT_DIR=$(dirname "$SNAPSHOT_PATH")
+SNAPSHOT_FILE=$(basename "$SNAPSHOT_PATH")
+cd "$SNAPSHOT_DIR"
+
+# Get metadata
+echo ""
+echo "=== Snapshot Metadata ==="
+octosql -ojson "SELECT * FROM ${SNAPSHOT_FILE}?meta=true" 2>/dev/null || echo "Warning: metadata query failed"
+
+# Top 20 namespaces by total size
+echo ""
+echo "=== Top 20 Namespaces by Size ==="
+octosql -ojson "SELECT namespace, COUNT(*) as key_count, SUM(valueSize) as total_bytes FROM ${SNAPSHOT_FILE} WHERE namespace IS NOT NULL GROUP BY namespace ORDER BY total_bytes DESC LIMIT 20" 2>/dev/null || echo "Warning: namespace query failed"
+
+# Top 20 largest individual resources
+echo ""
+echo "=== Top 20 Largest Resources ==="
+octosql -ojson "SELECT namespace, resourceType, name, valueSize FROM ${SNAPSHOT_FILE} WHERE namespace IS NOT NULL ORDER BY valueSize DESC LIMIT 20" 2>/dev/null || echo "Warning: largest resources query failed"
+
+# Resource type distribution
+echo ""
+echo "=== Resource Type Distribution ==="
+octosql -ojson "SELECT resourceType, COUNT(*) as count, SUM(valueSize) as total_bytes FROM ${SNAPSHOT_FILE} WHERE resourceType IS NOT NULL GROUP BY resourceType ORDER BY total_bytes DESC LIMIT 30" 2>/dev/null || echo "Warning: resource type query failed"
+
+# Clean up snapshot if requested
+if [ "$DELETE_SNAPSHOT" = "true" ]; then
+    echo ""
+    echo "Deleting snapshot file: ${SNAPSHOT_PATH}"
+    rm -f "$SNAPSHOT_PATH"
+fi
+
+echo ""
+echo "Analysis complete."