fix(core): fix opentelemetry upgrade issues

compose/cluster-viz.sh

@@ -0,0 +1,252 @@
+#!/bin/bash
+# Visualize Dgraph cluster topology from docker-compose.yml
+
+FILE="${1:-docker-compose.yml}"
+
+if [[ ! -f ${FILE} ]]; then
+	echo "Usage: $0 [docker-compose.yml]"
+	exit 1
+fi
+
+# Flatten the file (join continuation lines) for easier parsing
+FLAT=$(tr '\n' ' ' <"${FILE}" | sed 's/  */ /g' || true)
+
+# Extract zero nodes (find --my=zeroN:port patterns near "zero")
+zeros=$(echo "${FLAT}" | grep -oE '\-\-my=zero[a-zA-Z0-9_-]*:[0-9]+' | sed -E 's/--my=([^:]+):.*/\1/' | sort -u || true)
+
+# Extract alpha nodes with their group assignments
+TMPFILE=$(mktemp)
+
+# Extract all alphas by their --my flag. Group assignment is optional in flags.
+# If no group= is found near the alpha's command, put it under UNASSIGNED.
+alphas=$(echo "${FLAT}" | grep -oE '\-\-my=alpha[a-zA-Z0-9_-]*:[0-9]+' | sed -E 's/--my=([^:]+):.*/\1/' | sort -u || true)
+for alpha in ${alphas}; do
+	# Look for group= near this alpha occurrence in the flattened compose.
+	# Many compose files won't specify it; in that case group is assigned dynamically by Zero.
+	# Use awk (portable on macOS) instead of BSD sed regex intervals.
+	group=$(awk -v s="${FLAT}" -v a="--my=${alpha}:" '
+		BEGIN { pos = index(s, a); if (pos == 0) exit }
+		END {
+			win = substr(s, pos, 600)
+			if (match(win, /group=[0-9]+/)) {
+				g = substr(win, RSTART, RLENGTH)
+				split(g, parts, "=")
+				print parts[2]
+			}
+		}
+	' </dev/null || true)
+	if [[ -z ${group} ]]; then
+		group="UNASSIGNED"
+	fi
+	echo "${group} ${alpha}"
+done >"${TMPFILE}"
+
+# Get unique groups (numeric groups first, UNASSIGNED last, never duplicated)
+numeric_groups=$(grep -E '^[0-9]+ ' "${TMPFILE}" | cut -d' ' -f1 | sort -n | uniq || true)
+unique_groups="${numeric_groups}"
+if grep -q '^UNASSIGNED ' "${TMPFILE}"; then
+	unique_groups="${unique_groups} UNASSIGNED"
+fi
+
+# Extract ratel if present
+has_ratel=$(grep -q 'ratel' "${FILE}" && echo "yes" || echo "no")
+
+# Get replica count
+replicas=$(grep -oE '\-\-replicas=[0-9]+' "${FILE}" | head -1 | cut -d= -f2 || true)
+replicas=${replicas:-1}
+
+# Calculate max alphas per group for width calculation
+max_per_group=0
+for group in ${unique_groups}; do
+	cnt=$(grep -c "^${group} " "${TMPFILE}" 2>/dev/null || echo 0)
+	if [[ ${cnt} -gt ${max_per_group} ]]; then
+		max_per_group=${cnt}
+	fi
+done
+
+# Count zeros
+zero_count=$(echo "${zeros}" | wc -w | tr -d ' ' || true)
+
+# Box width: 14 chars per node (┌──────────┐ + 2 spaces) + margins
+# Use the larger of zeros or max alphas per group
+max_nodes=${zero_count}
+if [[ ${max_per_group} -gt ${max_nodes} ]]; then
+	max_nodes=${max_per_group}
+fi
+[[ ${max_nodes} -lt 3 ]] && max_nodes=3
+
+inner_width=$((max_nodes * 14 + 4))
+[[ ${inner_width} -lt 60 ]] && inner_width=60
+outer_width=$((inner_width + 4))
+
+# Total line width (display columns between outer left and right border)
+W=$((outer_width))
+
+# Print N space characters
+sp() {
+	local i
+	for ((i = 0; i < $1; i++)); do printf " "; done
+}
+
+# Print N copies of a character
+rp() {
+	local i
+	for ((i = 0; i < $1; i++)); do printf "%s" "$2"; done
+}
+
+# Finish outer row: pad from current column to W, then print |
+fin() {
+	sp "$((W - $1))"
+	printf "|\n"
+}
+
+# Finish inner row (inside group box): pad then print | |
+# Total should be W+1 cols: content($1) + spaces + 3(| |) = W+1
+fini() {
+	sp "$((W - $1 - 2))"
+	printf "| |\n"
+}
+
+# Separator using full width
+sep() {
+	printf "%s" "$1"
+	rp "${W}" "$2"
+	printf "%s\n" "$3"
+}
+
+# Print a row that starts with '|' and does NOT yet include the final right border.
+# Pads with spaces so the final '|' is always aligned.
+outln() {
+	local s="$1"
+	local l=${#s}
+	# Total line length must be W + 2 (left border + interior + right border).
+	# If s already includes the left border '|', it must be padded to length W+1,
+	# then we print the final right border.
+	printf "%s" "${s}"
+	local pad=$((W + 1 - l))
+	if [[ ${pad} -lt 0 ]]; then
+		pad=0
+	fi
+	sp "${pad}"
+	printf "|\n"
+}
+
+# Render a row inside the group box.
+# The group box top uses: "|  +" + GW*"-" + "+".
+# For interior rows we use: "|  |" + <payload> + <pad-to-GW> + "|".
+out_group() {
+	local payload="$1"
+	local plen=${#payload}
+	local pad=$((GW - plen))
+	if [[ ${pad} -lt 0 ]]; then
+		pad=0
+	fi
+	local s="|  |${payload}"
+	# pad inside the group box up to width GW
+	for ((i = 0; i < pad; i++)); do
+		s+=" "
+	done
+	s+="|"
+	outln "${s}"
+}
+
+echo ""
+
+# === HEADER ===
+sep "+" "=" "+"
+outln "|  DGRAPH CLUSTER TOPOLOGY"
+sep "+" "=" "+"
+echo ""
+
+# === ZERO LAYER ===
+sep "+" "-" "+"
+outln "|  ZERO LAYER (Cluster Coordination)"
+sep "+" "-" "+"
+
+# Zero boxes: each is 14 cols (box 12 + 2 spaces)
+line="|  "
+for z in ${zeros}; do line+="+----------+  "; done
+outln "${line}"
+line="|  "
+for z in ${zeros}; do line+=$(printf "| %-8s |  " "${z}"); done
+outln "${line}"
+line="|  "
+for z in ${zeros}; do line+="|  (zero)  |  "; done
+outln "${line}"
+line="|  "
+for z in ${zeros}; do line+="+----------+  "; done
+outln "${line}"
+
+sep "+" "-" "+"
+echo ""
+
+# === ALPHA LAYER ===
+sep "+" "-" "+"
+outln "|  ALPHA LAYER (Data Storage) - Replicas: ${replicas}"
+sep "+" "-" "+"
+
+# Group box inner width
+GW=$((inner_width - 2))
+
+for group in ${unique_groups}; do
+	alphas=$(grep "^${group} " "${TMPFILE}" | cut -d' ' -f2 || true)
+	count=$(echo "${alphas}" | wc -l | tr -d ' ' || true)
+
+	# Empty row
+	outln "|"
+
+	# Group box top
+	line="|  +"
+	for ((i = 0; i < GW; i++)); do line+="-"; done
+	line+="+"
+	outln "${line}"
+
+	# Group title row
+	if [[ ${group} == "UNASSIGNED" ]]; then
+		gtitle="  GROUP UNASSIGNED (Raft Replicas: ${count})"
+	else
+		gtitle="  GROUP ${group}  (Raft Replicas: ${count})"
+	fi
+	out_group "${gtitle}"
+
+	# Alpha boxes
+	line="  "
+	for a in ${alphas}; do line+="+----------+  "; done
+	out_group "${line}"
+	line="  "
+	for a in ${alphas}; do line+=$(printf "| %-8s |  " "${a}" || true); done
+	out_group "${line}"
+	line="  "
+	for a in ${alphas}; do line+="| (alpha)  |  "; done
+	out_group "${line}"
+	line="  "
+	for a in ${alphas}; do line+="+----------+  "; done
+	out_group "${line}"
+
+	# Group box bottom
+	line="|  +"
+	for ((i = 0; i < GW; i++)); do line+="-"; done
+	line+="+"
+	outln "${line}"
+done
+
+outln "|"
+sep "+" "-" "+"
+
+# === RATEL ===
+if [[ ${has_ratel} == "yes" ]]; then
+	echo ""
+	sep "+" "-" "+"
+	outln "|  UI LAYER"
+	sep "+" "-" "+"
+	outln "|  +----------+"
+	outln "|  |  ratel   |  (Web UI on :8000)"
+	outln "|  +----------+"
+	sep "+" "-" "+"
+fi
+
+rm -f "${TMPFILE}"
+
+echo ""
+echo "Legend: Alphas in same GROUP replicate data via Raft consensus"
+echo ""

compose/compose.go

@@ -79,7 +79,9 @@ type options struct {
 	DataVol        bool
 	TmpFS          bool
 	UserOwnership  bool
-	Jaeger         bool
+	Jaeger         int
+	TraceRatio     string
+	TraceService   bool
 	Metrics        bool
 	PortOffset     int
 	Verbosity      int
@@ -202,8 +204,15 @@ func initService(basename string, idx, grpcPort int) service {
 		svc.Command += fmt.Sprintf(" --cwd=/data/%s", svc.name)
 	}
 	svc.Command += " " + basename
-	if opts.Jaeger {
-		svc.Command += ` --trace "jaeger=http://jaeger:14268;"`
+	if opts.Jaeger > 0 {
+		traceFlag := "jaeger=http://jaeger:4318;"
+		if opts.TraceRatio != "" {
+			traceFlag += fmt.Sprintf(" ratio=%s;", opts.TraceRatio)
+		}
+		if opts.TraceService {
+			traceFlag += fmt.Sprintf(" service=%s;", svc.name)
+		}
+		svc.Command += fmt.Sprintf(` --trace "%s"`, traceFlag)
 	}
 	return svc
 }
@@ -375,23 +384,44 @@ func getVolume(vol string) volume {
 
 }
 
-func getJaeger() service {
-	svc := service{
-		Image:         "jaegertracing/all-in-one:1.18",
+func getJaeger(version int) service {
+	if version == 2 {
+		// Jaeger v2.x uses OpenTelemetry Collector architecture with YAML config.
+		// OTLP receivers bind to localhost by default, so we need JAEGER_LISTEN_HOST=0.0.0.0
+		// for container environments.
+		return service{
+			Image:         "jaegertracing/jaeger:latest",
+			ContainerName: containerName("jaeger"),
+			Ports: []string{
+				toPort(4318),
+				toPort(16686),
+			},
+			Environment: []string{
+				"JAEGER_LISTEN_HOST=0.0.0.0",
+			},
+		}
+	}
+	// Jaeger v1.60 with badger storage
+	return service{
+		Image:         "jaegertracing/all-in-one:1.60",
 		ContainerName: containerName("jaeger"),
-		WorkingDir:    "/working/jaeger",
+		User:          "0",
 		Ports: []string{
-			toPort(14268),
+			toPort(4318),
 			toPort(16686),
 		},
+		Volumes: []volume{{
+			Type:   "volume",
+			Source: "jaeger-volume",
+			Target: "/jaeger",
+		}},
 		Environment: []string{
 			"SPAN_STORAGE_TYPE=badger",
 		},
 		Command: "--badger.ephemeral=false" +
-			" --badger.directory-key /working/jaeger" +
-			" --badger.directory-value /working/jaeger",
+			" --badger.directory-key /jaeger" +
+			" --badger.directory-value /jaeger",
 	}
-	return svc
 }
 
 func getMinio(minioDataDir string) service {
@@ -421,12 +451,12 @@ func getRatel() service {
 		portFlag = fmt.Sprintf(" -port=%d", opts.RatelPort)
 	}
 	svc := service{
-		Image:         opts.Image + ":" + opts.Tag,
+		Image:         "dgraph/ratel:latest",
 		ContainerName: containerName("ratel"),
 		Ports: []string{
 			toPort(opts.RatelPort),
 		},
-		Command: "dgraph-ratel" + portFlag,
+		Command: portFlag,
 	}
 	return svc
 }
@@ -543,8 +573,13 @@ func main() {
 		"run as the current user rather than root")
 	cmd.PersistentFlags().BoolVar(&opts.TmpFS, "tmpfs", false,
 		"store w and zw directories on a tmpfs filesystem")
-	cmd.PersistentFlags().BoolVarP(&opts.Jaeger, "jaeger", "j", false,
-		"include jaeger service")
+	cmd.PersistentFlags().IntVarP(&opts.Jaeger, "jaeger", "j", 0,
+		"include jaeger service (1 for v1.60, 2 for v2.x)")
+	cmd.PersistentFlags().Lookup("jaeger").NoOptDefVal = "1"
+	cmd.PersistentFlags().StringVar(&opts.TraceRatio, "trace_ratio", "",
+		"ratio of queries to trace (e.g., 0.01 for 1%)")
+	cmd.PersistentFlags().BoolVar(&opts.TraceService, "trace_service", false,
+		"use compose service name as trace service name (e.g., alpha1, zero1)")
 	cmd.PersistentFlags().BoolVarP(&opts.Metrics, "metrics", "m", false,
 		"include metrics (prometheus, grafana) services")
 	cmd.PersistentFlags().IntVarP(&opts.PortOffset, "port_offset", "o", 100,
@@ -689,8 +724,11 @@ func main() {
 		cfg.Volumes["data"] = stringMap{}
 	}
 
-	if opts.Jaeger {
-		services["jaeger"] = getJaeger()
+	if opts.Jaeger > 0 {
+		services["jaeger"] = getJaeger(opts.Jaeger)
+		if opts.Jaeger == 1 {
+			cfg.Volumes["jaeger-volume"] = stringMap{}
+		}
 	}
 
 	if opts.Ratel {

contrib/jepsen/main.go

@@ -116,7 +116,7 @@ var (
 	testCount = pflag.IntP("test-count", "c", 1, "Test count per Jepsen test.")
 	jaeger    = pflag.StringP("jaeger", "j", "",
 		"Run with Jaeger collector. Set to empty string to disable collection to Jaeger."+
-			" Otherwise set to http://jaeger:14268.")
+			" Otherwise set to http://jaeger:4318.")
 	jaegerSaveTraces = pflag.Bool("jaeger-save-traces", true, "Save Jaeger traces on test error.")
 	deferDbTeardown  = pflag.Bool("defer-db-teardown", false,
 		"Wait until user input to tear down DB nodes")