fix(traefik): count each Traefik process only against its own last scrape (#5844)

One metrics URL can reach several Traefik processes: a swarm service's
replicas behind the routing mesh or its VIP, a Kubernetes Service's pods,
a keepalived pair after failover, or the same Traefik after a restart.
Each keeps its own counters, so comparing one with another read the second
one's lifetime totals as new traffic: retired routers flashed 100% 502 and
an idle router showed 834 req/s whenever the scrape moved to the other
replica.

Each scrape is now keyed by Traefik's process_start_time_seconds and
compared only with the same process's previous scrape. A process seen for
the first time (or not in the last hour) sets its baseline and counts
nothing. Without a start time every scrape is one process, as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

plugins/traefik/scraper.py

@@ -39,6 +39,53 @@ def parse_prometheus(text: str) -> ParsedMetrics:
     return metrics
 
 
+# ──────────────────────────────────────────────────────────────────────────────
+# Which Traefik answered (#5844)
+# ──────────────────────────────────────────────────────────────────────────────
+#
+# One metrics URL can reach more than one Traefik: a swarm service's replicas
+# behind the routing mesh or its VIP, a Kubernetes Service's pods, a keepalived
+# pair after failover, or the same Traefik after a restart. Each process keeps
+# its own counters, so comparing one process's scrape with another's reads the
+# second one's lifetime totals as new traffic (a retired router flashing 100%
+# 502, an idle one at 800 req/s). Each process is told apart by its start time,
+# which Traefik has exposed since v2.0, and is only ever compared with itself.
+
+# A process not heard from for this long has restarted or gone; its baseline
+# is dropped rather than kept forever.
+BASELINE_TTL_SECONDS = 3600
+
+
+def instance_of(current: ParsedMetrics) -> str:
+    """The Traefik process a scrape came from: its process_start_time_seconds,
+    or "" when the exposition doesn't carry one (then every scrape is taken to
+    be the same process, as before #5844)."""
+    start = current.get("process_start_time_seconds", {})
+    return f"{next(iter(start.values())):.2f}" if start else ""
+
+
+class Baselines:
+    """Each Traefik process's last scrape and when it was taken."""
+
+    def __init__(self) -> None:
+        self._seen: dict[str, tuple[ParsedMetrics, float]] = {}
+
+    def swap(self, current: ParsedMetrics, now: float) -> tuple[ParsedMetrics | None, float]:
+        """(the previous scrape of the process `current` came from, seconds
+        since it), and keep `current` as that process's newest. A process seen
+        for the first time has no previous scrape: (None, 0.0), so nothing is
+        counted until it has a baseline of its own."""
+        key = instance_of(current)
+        prev = self._seen.get(key)
+        self._seen[key] = (current, now)
+        for k, (_, at) in list(self._seen.items()):
+            if now - at > BASELINE_TTL_SECONDS:
+                del self._seen[k]
+        if prev is None or now - prev[1] > BASELINE_TTL_SECONDS:
+            return None, 0.0
+        return prev[0], now - prev[1]
+
+
 # ──────────────────────────────────────────────────────────────────────────────
 # Per-router metric computation
 # ──────────────────────────────────────────────────────────────────────────────