fix(metrics): correct cpu/disk% chart scaling, evict stale collector state, lighten nic query, drop polling & dead RRD code

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

backend/app/api/proxmox/vms.py

@@ -21,6 +21,7 @@ from ...ipam_service import IPAMService
 from ._helpers import (check_vm_access, require_vm_access, _get_proxmox_client,
                         get_next_vmid, archive_and_delete_snapshots,
                         save_vm_instance, get_vm_instance, soft_delete_vm_instance)
+from ...services.metrics_history import query_instance_metrics, timeframe_to_range
 
 router = APIRouter()
 
@@ -1848,7 +1849,6 @@ def get_container_rrddata(
 # ==================== Metrics History (DB-only) ====================
 
 def _resolve_metrics_window(timeframe, from_ts, to_ts):
-    from ...services.metrics_history import timeframe_to_range
     from ...config import settings
     now = int(time_lib.time())
     if from_ts is not None and to_ts is not None:
@@ -1874,7 +1874,6 @@ def get_vm_metrics(
     current_user: User = Depends(PermissionChecker("vm:view")),
 ):
     """Retrieve VM metrics time-series from the local DB (no Proxmox call)."""
-    from ...services.metrics_history import query_instance_metrics
     f, t = _resolve_metrics_window(timeframe, from_ts, to_ts)
     return query_instance_metrics(db, server_id, vmid, f, t, nic)
 
@@ -1892,7 +1891,6 @@ def get_container_metrics(
     current_user: User = Depends(PermissionChecker("vm:view")),
 ):
     """Retrieve container metrics time-series from the local DB (no Proxmox call)."""
-    from ...services.metrics_history import query_instance_metrics
     f, t = _resolve_metrics_window(timeframe, from_ts, to_ts)
     return query_instance_metrics(db, server_id, vmid, f, t, nic)
 

backend/app/services/metrics_collector.py

@@ -26,6 +26,8 @@ def collect_once(db, items, clients, prev, _now=None):
 
     # netstat is one call per (server_id, node); cache per tick.
     netstat_cache: dict[tuple, list] = {}
+    # Track which prev keys are touched this tick so stale entries can be evicted.
+    seen: set = set()
 
     for it in items:
         if it.get("status") != "running":
@@ -76,6 +78,7 @@ def collect_once(db, items, clients, prev, _now=None):
                 iops_write=rate("wr_ops") if inst_type == "qemu" else None,
             ))
             prev[key] = (now, cur)
+            seen.add(key)
 
             # per-NIC
             ck = (sid, node)
@@ -93,9 +96,15 @@ def collect_once(db, items, clients, prev, _now=None):
                     out_rate=compute_rate(npc.get("out"), float(dout), ndt),
                 ))
                 prev[nkey] = (now, {"in": float(din), "out": float(dout)})
+                seen.add(nkey)
         except Exception as e:
             logger.debug(f"[metrics_collector] {sid}:{vmid} error: {e}")
 
+    # Evict stale keys from prev (VMs/NICs not seen this tick).
+    for k in list(prev):
+        if k not in seen:
+            del prev[k]
+
 
 async def metrics_collector_loop(db_factory) -> None:
     from ..config import settings

backend/app/services/metrics_history.py

@@ -2,6 +2,7 @@
 from __future__ import annotations
 
 import re
+from datetime import timezone as _tz
 
 _RD_OPS = re.compile(r"rd_operations=(\d+)")
 _WR_OPS = re.compile(r"wr_operations=(\d+)")
@@ -93,8 +94,6 @@ def aggregate_series(rows, from_ts, bucket, fields):
     return out
 
 
-from datetime import timezone as _tz
-
 _AGG_FIELDS = ["cpu", "mem", "maxmem", "netin", "netout",
                "diskread", "diskwrite", "diskpct", "iops_read", "iops_write"]
 
@@ -119,16 +118,24 @@ def query_instance_metrics(db, server_id, vmid, from_ts, to_ts, nic="all"):
               .order_by(InstanceMetric.ts.asc()).all())
 
     # Distinct NICs seen in the window (for the selector).
-    nic_rows = (db.query(InstanceNicMetric)
-                  .filter(InstanceNicMetric.server_id == server_id,
-                          InstanceNicMetric.vmid == vmid,
-                          InstanceNicMetric.ts >= start, InstanceNicMetric.ts <= end)
-                  .order_by(InstanceNicMetric.ts.asc()).all())
-    nics = sorted({r.dev for r in nic_rows})
-
-    # Optional per-NIC override of net rates, keyed by exact timestamp.
-    nic_by_ts: dict[int, tuple] = {}
-    if nic != "all":
+    if nic == "all":
+        # Only fetch distinct dev names — avoids loading all NIC rows.
+        dev_rows = (db.query(InstanceNicMetric.dev)
+                      .filter(InstanceNicMetric.server_id == server_id,
+                              InstanceNicMetric.vmid == vmid,
+                              InstanceNicMetric.ts >= start, InstanceNicMetric.ts <= end)
+                      .distinct().all())
+        nics = sorted({r.dev for r in dev_rows})
+        nic_by_ts: dict[int, tuple] = {}
+    else:
+        # Load full NIC rows for the requested dev to build per-ts override map.
+        nic_rows = (db.query(InstanceNicMetric)
+                      .filter(InstanceNicMetric.server_id == server_id,
+                              InstanceNicMetric.vmid == vmid,
+                              InstanceNicMetric.ts >= start, InstanceNicMetric.ts <= end)
+                      .order_by(InstanceNicMetric.ts.asc()).all())
+        nics = sorted({r.dev for r in nic_rows})
+        nic_by_ts = {}
         for r in nic_rows:
             if r.dev == nic:
                 nic_by_ts[_to_ts(r.ts)] = (r.in_rate, r.out_rate)