added peti hetzner

admin-system/wan-monitor.yaml

@@ -154,63 +154,87 @@ data:
     set -u
     SHARED=/shared
     HDR=/scripts/metrics-header.prom
-    HETZNER="${HETZNER_HOST:?set HETZNER_HOST}"
+
+    # Space-separated endpoint labels. For each LABEL there must be an env var
+    # <LABEL>_HOST (uppercased) and optionally <LABEL>_HMAC.
+    ENDPOINTS="${ENDPOINTS:-hetzner}"
+    # Subset of ENDPOINTS that also get iperf3 throughput tests (the data-heavy part).
+    TPUT_ENDPOINTS="${TPUT_ENDPOINTS:-$ENDPOINTS}"
+
     IRTT_PORT="${IRTT_PORT:-2112}"
     IPERF_PORT="${IPERF_PORT:-5201}"
     IRTT_INTERVAL="${IRTT_INTERVAL:-20ms}"
     IRTT_DURATION="${IRTT_DURATION:-60}"   # seconds (numeric, for timeout math)
     TPUT_EVERY="${TPUT_EVERY:-900}"        # seconds between throughput tests
     TPUT_TIME="${TPUT_TIME:-10}"           # iperf3 seconds per direction
-    IRTT_TARGET="${IRTT_TARGET:-hetzner}"
+    IPERF_PARALLEL="${IPERF_PARALLEL:-4}"  # parallel streams: one can't fill the pipe over the RTT
-    TPUT_TARGET="${TPUT_TARGET:-hetzner}"
+
-    HMAC_OPT=""
+    upper()    { echo "$1" | tr '[:lower:]' '[:upper:]'; }
-    [ -n "${IRTT_HMAC:-}" ] && HMAC_OPT="--hmac=${IRTT_HMAC}"
+    host_for() { U="$(upper "$1")"; eval "printf '%s' \"\${${U}_HOST:-}\""; }
+    hmac_for() { U="$(upper "$1")"; eval "printf '%s' \"\${${U}_HMAC:-}\""; }
 
     mkdir -p "$SHARED"
-    : > "$SHARED/.irtt.prom"; : > "$SHARED/.irttload.prom"; : > "$SHARED/.tput.prom"
+    FRAGMENTS=""
+    for ep in $ENDPOINTS; do
+      : > "$SHARED/.irtt.$ep.prom"
+      : > "$SHARED/.irttload.$ep.prom"
+      : > "$SHARED/.tput.$ep.prom"
+      FRAGMENTS="$FRAGMENTS $SHARED/.irtt.$ep.prom $SHARED/.irttload.$ep.prom $SHARED/.tput.$ep.prom"
+    done
     cp "$HDR" "$SHARED/metrics"   # serve header immediately so first scrapes don't 404
 
-    # Concatenate fragments into the served file via temp + atomic rename.
+    # Concatenate header + all fragments into the served file via temp + atomic rename.
     assemble() {
-      cat "$HDR" "$SHARED/.irtt.prom" "$SHARED/.irttload.prom" "$SHARED/.tput.prom" \
+      cat "$HDR" $FRAGMENTS > "$SHARED/.metrics.tmp" 2>/dev/null
-        > "$SHARED/.metrics.tmp" 2>/dev/null
       mv "$SHARED/.metrics.tmp" "$SHARED/metrics"
     }
 
-    # Each fragment is written to <file>.tmp then renamed, so assemble() never
+    # $1 condition  $2 label  $3 host  $4 hmac  $5 outfile  $6 duration(seconds)
-    # cats a partially written file (the cause of the impossible loss spikes).
+    # Writes <outfile>.tmp then renames, so assemble() never cats a partial file.
-    run_irtt() {  # $1 condition  $2 outfile  $3 duration(seconds)
+    run_irtt() {
-      timeout "$(( $3 + 25 ))" irtt client -i "$IRTT_INTERVAL" -d "${3}s" -q $HMAC_OPT \
+      hmopt=""
-        -o - "${HETZNER}:${IRTT_PORT}" 2>/dev/null \
+      [ -n "$4" ] && hmopt="--hmac=$4"
-        | python3 /scripts/irtt_to_prom.py "$1" "$IRTT_TARGET" > "$2.tmp"
+      timeout "$(( $6 + 25 ))" irtt client -i "$IRTT_INTERVAL" -d "${6}s" -q $hmopt \
-      mv "$2.tmp" "$2"
+        -o - "$3:${IRTT_PORT}" 2>/dev/null \
+        | python3 /scripts/irtt_to_prom.py "$1" "$2" > "$5.tmp"
+      mv "$5.tmp" "$5"
     }
 
+    # $1 label  $2 host
     run_tput() {
-      P="${IPERF_PARALLEL:-4}"   # parallel streams: a single stream can't fill the pipe over the RTT
       TO="$(( TPUT_TIME + 20 ))"
-      TMP="$SHARED/.tput.prom.partial"
+      TMP="$SHARED/.tput.$1.partial"
       : > "$TMP"
-      timeout "$TO" iperf3 -c "$HETZNER" -p "$IPERF_PORT" -t "$TPUT_TIME" -P "$P" --connect-timeout 5000 -R -J 2>/dev/null \
+      timeout "$TO" iperf3 -c "$2" -p "$IPERF_PORT" -t "$TPUT_TIME" -P "$IPERF_PARALLEL" --connect-timeout 5000 -R -J 2>/dev/null \
-        | python3 /scripts/tput_to_prom.py download "$TPUT_TARGET" >  "$TMP"
+        | python3 /scripts/tput_to_prom.py download "$1" >  "$TMP"
-      timeout "$TO" iperf3 -c "$HETZNER" -p "$IPERF_PORT" -t "$TPUT_TIME" -P "$P" --connect-timeout 5000 -J 2>/dev/null \
+      timeout "$TO" iperf3 -c "$2" -p "$IPERF_PORT" -t "$TPUT_TIME" -P "$IPERF_PARALLEL" --connect-timeout 5000 -J 2>/dev/null \
-        | python3 /scripts/tput_to_prom.py upload   "$TPUT_TARGET" >> "$TMP"
+        | python3 /scripts/tput_to_prom.py upload   "$1" >> "$TMP"
-      mv "$TMP" "$SHARED/.tput.prom"
+      mv "$TMP" "$SHARED/.tput.$1.prom"
     }
 
     last_tput=0
     while true; do
-      run_irtt idle "$SHARED/.irtt.prom" "$IRTT_DURATION"   # blocks ~IRTT_DURATION = loop cadence
+      # idle irtt to every endpoint in parallel (both paths sampled at the same instant)
+      for ep in $ENDPOINTS; do
+        h="$(host_for "$ep")"; [ -z "$h" ] && continue
+        run_irtt idle "$ep" "$h" "$(hmac_for "$ep")" "$SHARED/.irtt.$ep.prom" "$IRTT_DURATION" &
+      done
+      wait
       assemble
+
       now=$(date +%s)
       if [ $(( now - last_tput )) -ge "$TPUT_EVERY" ]; then
         LOAD_DUR=$(( 2 * TPUT_TIME + 4 ))
-        run_irtt under_load "$SHARED/.irttload.prom" "$LOAD_DUR" &   # concurrent = bufferbloat
+        for ep in $TPUT_ENDPOINTS; do
+          h="$(host_for "$ep")"; [ -z "$h" ] && continue
+          # concurrent under-load irtt to the same endpoint = bufferbloat
+          run_irtt under_load "$ep" "$h" "$(hmac_for "$ep")" "$SHARED/.irttload.$ep.prom" "$LOAD_DUR" &
           LOADPID=$!
-        run_tput
+          run_tput "$ep" "$h"
           wait "$LOADPID" 2>/dev/null
-        last_tput="$now"
           assemble
+        done
+        last_tput="$now"
       fi
     done
 ---
@@ -266,11 +290,28 @@ spec:
           image: gitea.dooplex.hu/admin/wan-probe:0.1.0
           command: ["/bin/sh", "/scripts/probe-loop.sh"]
           env:
+            - name: ENDPOINTS
+              value: "hetzner abonet"          # space-separated endpoint labels
+            # - name: TPUT_ENDPOINTS
+            #   value: "hetzner"               # uncomment to keep iperf3 OFF the abonet VM (saves its egress)
             - name: HETZNER_HOST
-              # MUST be the Hetzner origin: a DNS-only (grey-cloud) record or raw IP.
+              # MUST be the origin (DNS-only record or raw IP), NOT a Cloudflare-proxied
-              # NOT the Cloudflare-proxied jarrs.eu — CF only forwards HTTP/HTTPS, so
+              # name -- CF only forwards HTTP/HTTPS, so UDP 2112 / TCP 5201 never arrive.
-              # UDP 2112 (irtt) / TCP 5201 (iperf3) never reach the origin behind it.
+              value: "metrics.jarrs.eu"
-              value: "metrics.jarrs.eu"        # DNS-only A record -> Hetzner IPv4
+            - name: ABONET_HOST
+              value: "hetzner.abonet.hu"       # colleague's VM (verify it's a direct A record, not CF-proxied)
+            - name: HETZNER_HMAC
+              valueFrom:
+                secretKeyRef:
+                  name: wan-monitor-irtt
+                  key: hetzner
+                  optional: true
+            - name: ABONET_HMAC
+              valueFrom:
+                secretKeyRef:
+                  name: wan-monitor-irtt
+                  key: abonet
+                  optional: true
             - name: IRTT_PORT
               value: "2112"
             - name: IPERF_PORT
@@ -278,19 +319,13 @@ spec:
             - name: IRTT_INTERVAL
               value: "20ms"
             - name: IRTT_DURATION
-              value: "60"                  # seconds (numeric)
+              value: "60"
             - name: TPUT_EVERY
-              value: "900"                 # 15 min
+              value: "900"
             - name: TPUT_TIME
               value: "10"
             - name: IPERF_PARALLEL
               value: "4"
-            - name: IRTT_HMAC               # shared key; apply via secret (see below)
-              valueFrom:
-                secretKeyRef:
-                  name: wan-monitor-irtt
-                  key: hmac
-                  optional: true
           resources:
             requests: { cpu: 20m, memory: 48Mi }
             limits:   { memory: 96Mi }