Add host temperature telemetry and history

2026-10-13 23:39:04 -04:00
parent 09396ab584
commit 171f657876
28 changed files with 1466 additions and 32 deletions
@@ -1,5 +1,24 @@
 # Agent changelog

+## 2026-10-13
+
+### 1.10.0 — Host temperature telemetry
+
+- Read supported Linux hwmon CPU package/core, AMD physical die/CCD, NVMe and
+  Intel PCH temperatures with the Python standard library. Sensor failures,
+  faults and disabled inputs are skipped independently; unsupported hosts report
+  an empty list. AMD Tctl control offsets are excluded from physical readings.
+- Include labeled, bounded sensors in heartbeats with the `host.temperatures`
+  capability. Physical device/channel identities survive hwmon renumbering.
+- The controller shows current CPU readings and expandable sensor details, and
+  adds temperature history to Capacity. Fleet history has one CPU line per host;
+  individual hosts can add NVMe composite and chipset lines. History retains
+  minute peaks using the host-resource retention setting; missing data stays
+  unavailable and no thermal alerts or placement rules are introduced.
+- Adds database migration 23. Existing agents remain compatible. Release
+  artifacts are packaged and the stale signature is removed. Offline signing
+  and host rollout are still required; neither was performed here.
+
 ## 2026-10-12

 ### 1.9.3 — Protect private bank configuration (EC #152)
@@ -35,7 +35,7 @@ import urllib.parse
 import urllib.request
 import uuid

-VERSION = "1.9.3"
+VERSION = "1.10.0"
 LABEL = "io.electric-cherry"
 MAX_FILE = 32 * 1024 * 1024
 MANIFEST_MAX_ENTRIES = 10000
@@ -63,6 +63,7 @@ UPDATE_TIMER_ENABLED = Path("/etc/systemd/system/timers.target.wants/electric-ch
 UPDATE_JOURNAL = Path("/etc/electric-cherry/update/journal.json")
 UPDATE_STATUS = Path("/etc/electric-cherry/update-status.json")
 AGENT_HEALTH = Path("/etc/electric-cherry/agent-health.json")
+HWMON_ROOT = Path("/sys/class/hwmon")
 ROOT_MARKER = ".electric-cherry.json"


@@ -1308,6 +1309,59 @@ class Agent:
                return None
            return round(100 * (total - deltas[3] - deltas[4]) / total, 1)

+    def temperatures(self):
+        """Best-effort, read-only samples from known millidegree-Celsius drivers."""
+        samples = []
+        try:
+            chips = sorted(HWMON_ROOT.glob("hwmon*"))[:64]
+        except OSError:
+            return samples
+        for chip in chips:
+            try:
+                device = (chip / "name").read_text().strip()
+                if device not in ("coretemp", "k10temp", "nvme") and not device.startswith("pch_"):
+                    continue
+                # hwmon numbers can change on reboot. The physical sysfs path
+                # and channel identify separate CPU sockets and NVMe devices.
+                identity = re.sub(r"/(?:hwmon/)?hwmon\d+$", "", str(chip.resolve()))
+                inputs = sorted(chip.glob("temp[0-9]*_input"))[:128]
+            except (OSError, ValueError):
+                continue
+            for source in inputs:
+                try:
+                    channel = source.name.removesuffix("_input")
+                    label_path = chip / (channel + "_label")
+                    label = label_path.read_text().strip() if label_path.exists() else channel
+                    if not label or len(label) > 120:
+                        continue
+                    fault = chip / (channel + "_fault")
+                    enabled = chip / (channel + "_enable")
+                    if (fault.exists() and int(fault.read_text()) != 0) or (enabled.exists() and int(enabled.read_text()) == 0):
+                        continue
+                    if device == "coretemp":
+                        kind = "cpu-package" if label.startswith("Package id ") else "cpu-core" if label.startswith("Core ") else None
+                    elif device == "k10temp":
+                        # Tctl can contain a fan-control offset. Never present
+                        # it as the physical die temperature or a fleet metric.
+                        kind = "cpu-die" if label == "Tdie" or re.fullmatch(r"Tccd\d+", label) else None
+                    else:
+                        kind = "nvme" if device == "nvme" else "chipset"
+                    if kind is None:
+                        continue
+                    celsius = int(source.read_text()) / 1000
+                    if not -100 <= celsius <= 200:
+                        continue
+                    sensor_id = hashlib.sha256((identity + "/" + channel).encode()).hexdigest()[:24]
+                    samples.append({"id": sensor_id, "device": device, "label": label,
+                                    "kind": kind, "celsius": round(celsius, 1)})
+                    if len(samples) >= 128:
+                        return samples
+                except (OSError, ValueError):
+                    # Read permission alone does not imply data is available
+                    # (e.g. ENODATA), and a device can disappear during a read.
+                    continue
+        return samples
+
    def instance_resources(self, observations=None):
        """One bounded Docker stats batch; unavailable samples are never zero load."""
        ids = self.docker.call("container", "ls", "--all", "--filter", "label=" + LABEL + ".managed=true",
@@ -1392,6 +1446,7 @@ class Agent:

    def inventory(self):
        cpu_usage = self.cpu_usage_percent()
+        temperatures = self.temperatures()
        usage = shutil.disk_usage(self.root)
        memory = {}
        try:
@@ -1516,9 +1571,9 @@ class Agent:
                if line.startswith("PRETTY_NAME="):
                    os_name = line.split("=", 1)[1].strip('"')
        update = self.update_inventory()
-        capabilities = ["library.remove", "plutonium.cache", "instance.cancel-create", "instance.safety-override", "instance.unhealthy-recovery", "instance.rebuild-files", "instance.autoscaling-remove", "operations.read-lane", "operations.health-lane", "rcon.status-batch", "instance.heartbeat-state", "files.manifest", "files.blobs", "logs.read-offset", "bank.settlement"] + (["host.decommission"] if self.can_uninstall() else []) + (["agent.update"] if update["enabled"] else [])
+        capabilities = ["host.temperatures", "library.remove", "plutonium.cache", "instance.cancel-create", "instance.safety-override", "instance.unhealthy-recovery", "instance.rebuild-files", "instance.autoscaling-remove", "operations.read-lane", "operations.health-lane", "rcon.status-batch", "instance.heartbeat-state", "files.manifest", "files.blobs", "logs.read-offset", "bank.settlement"] + (["host.decommission"] if self.can_uninstall() else []) + (["agent.update"] if update["enabled"] else [])
        plutonium_cache = self.plutonium_cache_inventory()
-        return {"hostname": socket.gethostname(), "agentVersion": VERSION, "protocolVersion": 1, "capabilities": capabilities, "agentUpdate": update, "agentLanes": self.lane_inventory(), "os": os_name, "osVersion": platform.release(), "architecture": platform.machine(), "cpu": platform.processor() or platform.machine(), "cpuCount": os.cpu_count() or 0, **resources, **({"cpuUsagePercent": cpu_usage} if cpu_usage is not None else {}), "memoryTotal": memory.get("MemTotal", 0), "memoryAvailable": memory.get("MemAvailable", 0), "storageTotal": usage.total, "storageAvailable": usage.free, "dockerVersion": docker_version, "dockerRunning": docker_running, "directoriesWritable": os.access(self.root, os.W_OK), "runtimeImages": sorted(set(images)), "preparedRuntimeImages": sorted(set(prepared_images)), **({"plutoniumCache": plutonium_cache} if plutonium_cache else {}), "libraries": libraries, "libraryCaches": caches, "currentOperation": self.current_operation, "portRange": self.config.get("portRange", {"min": 28960, "max": 29060})}
+        return {"hostname": socket.gethostname(), "agentVersion": VERSION, "protocolVersion": 1, "capabilities": capabilities, "agentUpdate": update, "agentLanes": self.lane_inventory(), "os": os_name, "osVersion": platform.release(), "architecture": platform.machine(), "cpu": platform.processor() or platform.machine(), "cpuCount": os.cpu_count() or 0, "temperatures": temperatures, **resources, **({"cpuUsagePercent": cpu_usage} if cpu_usage is not None else {}), "memoryTotal": memory.get("MemTotal", 0), "memoryAvailable": memory.get("MemAvailable", 0), "storageTotal": usage.total, "storageAvailable": usage.free, "dockerVersion": docker_version, "dockerRunning": docker_running, "directoriesWritable": os.access(self.root, os.W_OK), "runtimeImages": sorted(set(images)), "preparedRuntimeImages": sorted(set(prepared_images)), **({"plutoniumCache": plutonium_cache} if plutonium_cache else {}), "libraries": libraries, "libraryCaches": caches, "currentOperation": self.current_operation, "portRange": self.config.get("portRange", {"min": 28960, "max": 29060})}

    def safe_to_disrupt(self, instance_id, generation, *, allow_unhealthy=False):
        health = self.inspect(instance_id, generation)
@@ -778,6 +778,12 @@ class AgentTests(unittest.TestCase):
            else:
                self.assertEqual(inventory["cpuUsagePercent"], usage)

+    def test_inventory_reports_optional_temperature_capability_and_empty_samples(self):
+        with patch.object(self.agent, "temperatures", return_value=[]):
+            inventory = self.agent.inventory()
+        self.assertEqual(inventory["temperatures"], [])
+        self.assertIn("host.temperatures", inventory["capabilities"])
+
    def resource_container(self):
        return {"Name": self.agent.container_name(INSTANCE, 1),
                "Config": {"Labels": {module.LABEL + ".managed": "true", module.LABEL + ".instance": INSTANCE, module.LABEL + ".generation": "1"}},
@@ -0,0 +1,96 @@
+import errno
+import importlib.util
+from pathlib import Path
+import tempfile
+import unittest
+from unittest.mock import patch
+
+spec = importlib.util.spec_from_file_location("ec_agent_temperatures", Path(__file__).resolve().parents[1] / "ec_agent.py")
+module = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(module)
+
+
+class TemperatureTests(unittest.TestCase):
+    def setUp(self):
+        self.temp = tempfile.TemporaryDirectory()
+        self.addCleanup(self.temp.cleanup)
+        self.root = Path(self.temp.name)
+        self.hwmon = self.root / "class"
+        self.hwmon.mkdir()
+        self.patch = patch.object(module, "HWMON_ROOT", self.hwmon)
+        self.patch.start()
+        self.addCleanup(self.patch.stop)
+        self.agent = object.__new__(module.Agent)
+
+    def chip(self, number, device, readings):
+        physical = self.root / (device + str(number)) / "hwmon" / ("hwmon" + str(number))
+        physical.mkdir(parents=True)
+        (self.hwmon / physical.name).symlink_to(physical, target_is_directory=True)
+        (physical / "name").write_text(device)
+        for channel, label, value in readings:
+            (physical / (channel + "_input")).write_text(str(value))
+            if label is not None:
+                (physical / (channel + "_label")).write_text(label)
+        return physical
+
+    def test_ec2_sensors_and_units_without_guessing_unknown_devices(self):
+        self.chip(0, "nvme", [("temp1", "Composite", 50900), ("temp2", "Sensor 1", 50900)])
+        self.chip(1, "pch_cometlake", [("temp1", None, 62000)])
+        self.chip(2, "coretemp", [("temp1", "Package id 0", 65000), ("temp2", "Core 0", 63000)])
+        self.chip(3, "iwlwifi_1", [("temp1", None, "unavailable")])
+        samples = self.agent.temperatures()
+        self.assertEqual([s["celsius"] for s in samples], [50.9, 50.9, 62.0, 65.0, 63.0])
+        self.assertEqual([s["kind"] for s in samples], ["nvme", "nvme", "chipset", "cpu-package", "cpu-core"])
+        self.assertEqual(len({s["id"] for s in samples}), len(samples))
+
+    def test_fault_disabled_bad_value_and_enodata_do_not_lose_other_sensors(self):
+        chip = self.chip(0, "coretemp", [("temp" + str(i), "Core " + str(i), value)
+                         for i, value in enumerate([65000, 60000, "bad", 500000, 61000, 63000, 0], 1)])
+        (chip / "temp1_fault").write_text("1")
+        (chip / "temp2_enable").write_text("0")
+        original = Path.read_text
+        def read(path, *args, **kwargs):
+            if path.name == "temp5_input":
+                raise OSError(errno.ENODATA, "No data available")
+            return original(path, *args, **kwargs)
+        with patch.object(Path, "read_text", read):
+            self.assertEqual([s["celsius"] for s in self.agent.temperatures()], [63.0, 0.0])
+
+    def test_amd_keeps_physical_die_readings_and_excludes_control_offsets(self):
+        self.chip(0, "k10temp", [("temp1", "Tctl", 85000), ("temp2", "Tdie", 65000), ("temp3", "Tccd1", 64000)])
+        self.assertEqual([s["label"] for s in self.agent.temperatures()], ["Tdie", "Tccd1"])
+
+    def test_sensor_identity_survives_hwmon_renumbering_and_separates_drives(self):
+        physical = self.chip(1, "nvme", [("temp1", "Composite", 50000)])
+        self.chip(2, "nvme", [("temp1", "Composite", 51000)])
+        before = {s["celsius"]: s["id"] for s in self.agent.temperatures()}
+        renamed = physical.with_name("hwmon9")
+        physical.rename(renamed)
+        (self.hwmon / "hwmon1").unlink()
+        (self.hwmon / "hwmon9").symlink_to(renamed, target_is_directory=True)
+        self.assertEqual(before, {s["celsius"]: s["id"] for s in self.agent.temperatures()})
+        self.assertEqual(len(set(before.values())), 2)
+
+    def test_missing_hwmon_and_disappearing_devices_are_optional(self):
+        self.assertEqual(self.agent.temperatures(), [])
+        (self.hwmon / "hwmon0").symlink_to(self.root / "missing")
+        self.assertEqual(self.agent.temperatures(), [])
+        with patch.object(Path, "glob", side_effect=PermissionError()):
+            self.assertEqual(self.agent.temperatures(), [])
+
+    def test_direct_hwmon_child_keeps_identity_when_renumbered(self):
+        physical = self.chip(1, "nvme", [("temp1", "Composite", 50000)])
+        direct = physical.parent.parent / "hwmon8"
+        physical.rename(direct)
+        (self.hwmon / "hwmon1").unlink()
+        (self.hwmon / "hwmon8").symlink_to(direct, target_is_directory=True)
+        before = self.agent.temperatures()[0]["id"]
+        renamed = direct.with_name("hwmon9")
+        direct.rename(renamed)
+        (self.hwmon / "hwmon8").unlink()
+        (self.hwmon / "hwmon9").symlink_to(renamed, target_is_directory=True)
+        self.assertEqual(self.agent.temperatures()[0]["id"], before)
+
+    def test_inventory_size_is_bounded(self):
+        self.chip(0, "coretemp", [("temp" + str(i), "Core " + str(i), 65000) for i in range(1, 200)])
+        self.assertEqual(len(self.agent.temperatures()), 128)
@@ -354,6 +354,40 @@ journalctl -u electric-cherry-update.service -n 100

 ## Placement and resource capacity

+### Host temperature
+
+Agent **1.10.0** reports temperature sensors exposed by supported Linux hwmon
+drivers without installing additional packages. **Hosts** shows CPU temperature
+beneath CPU usage. A host's **Overview → Inventory** shows the selected reading
+and expandable **Temperature sensors**, including per-core and storage details.
+Stale readings are marked as last reported; absent or failed readings are
+**Unavailable**, never zero or a readiness failure.
+
+**Fleet → Capacity → Temperature history** uses the existing host selector and
+Last hour / 24 hours / 7 days controls. **All hosts** displays a separate CPU line
+for each host. Select one host to add its NVMe composite and chipset lines using
+the legend. Hover, tap, or use the arrow keys on the chart to inspect a period.
+Each host page's **Capacity** button opens the view filtered to that host.
+The game filter affects launch estimates, not temperatures.
+
+The CPU reading is the hottest reported package, falling back to a physical
+AMD die/CCD and then a core reading. AMD Tctl cooling-control values are excluded
+because they may include an offset. Supported device drivers are `coretemp`,
+`k10temp`, `nvme` and `pch_*`; sensor availability depends on hardware and drivers.
+Virtual machines and older agents may have no supported readings. Missing,
+faulty and disabled sensors are skipped individually.
+
+Temperature history starts after the agent upgrade. The controller stores one
+peak per sensor per minute, limited to the CPU summary and up to sixteen NVMe
+composite/chipset inputs per host. Longer chart periods show the maximum of
+those peaks; a period without readings remains a gap. Brief gaps within an
+aggregated period may still leave a peak for that period. **Host resource history
+retention** applies to CPU, memory and temperature samples (seven days by default;
+zero keeps them indefinitely). Series colors identify hosts/sensors and do not
+indicate thermal safety. Temperature does not change placement or trigger alerts.
+
+### CPU and memory capacity
+
 ### Host inventory and eligibility

 The Inventory panel shows CPU usage with agent 1.4.2 and later. It measures
@@ -30,6 +30,14 @@ export interface CapacityHistory {
    endAt: number;
    bucketMs: number;
    points: CapacityHistoryPoint[];
+    temperatures: TemperatureHistorySeries[];
+}
+export interface TemperatureHistorySeries {
+    id: string;
+    hostId: string;
+    label: string;
+    kind: 'cpu' | 'nvme' | 'chipset';
+    points: { at: number; celsius: number | null }[];
 }
 export interface CapacityHost {
    id: string;
@@ -305,7 +305,7 @@ export const settingHelp: Partial<Record<keyof FleetSettings, string>> = {
        'Finished non-update maintenance. Update attempts remain durable to prevent duplicate automatic retries.',
    recoveryKeepDays:
        'Resolved watchdog incidents with no unfinished recovery. Dismissed incidents remain as suppression evidence.',
-    hostResourceKeepDays: 'Host CPU and memory history.',
+    hostResourceKeepDays: 'Host CPU, memory and temperature history.',
    instanceResourceKeepDays:
        'Minute resource samples. Capacity learning continues to use its independent 24-hour window.',
    metricsKeepDays: 'Historical fleet occupancy and utilization samples.',
@@ -0,0 +1,92 @@
+export const temperatureKinds = ['cpu-package', 'cpu-die', 'cpu-core', 'nvme', 'chipset'] as const;
+export type TemperatureKind = (typeof temperatureKinds)[number];
+export interface HostTemperatureSensor {
+    id: string;
+    device: string;
+    label: string;
+    kind: TemperatureKind;
+    celsius: number;
+}
+
+/** Optional telemetry must never make an otherwise valid heartbeat fail. */
+export function validateTemperatures(value: unknown): HostTemperatureSensor[] {
+    if (!Array.isArray(value)) return [];
+    const seen = new Set<string>();
+    return value.slice(0, 128).flatMap((sensor) => {
+        if (
+            !sensor ||
+            typeof sensor !== 'object' ||
+            Array.isArray(sensor) ||
+            typeof sensor.id !== 'string' ||
+            !/^[a-zA-Z0-9_-]{1,100}$/.test(sensor.id) ||
+            seen.has(sensor.id) ||
+            !['device', 'label'].every(
+                (key) =>
+                    typeof sensor[key] === 'string' &&
+                    sensor[key].length > 0 &&
+                    sensor[key].length <= 120
+            ) ||
+            !temperatureKinds.includes(sensor.kind) ||
+            typeof sensor.celsius !== 'number' ||
+            !Number.isFinite(sensor.celsius) ||
+            sensor.celsius < -100 ||
+            sensor.celsius > 200
+        )
+            return [];
+        seen.add(sensor.id);
+        return [
+            {
+                id: sensor.id,
+                device: sensor.device,
+                label: sensor.label,
+                kind: sensor.kind,
+                celsius: sensor.celsius
+            }
+        ];
+    });
+}
+
+export function cpuTemperature(
+    sensors: HostTemperatureSensor[] = []
+): HostTemperatureSensor | null {
+    // Prefer physical package/die readings. Cores are a fallback, never mixed
+    // into an average that might hide a hot package on a multi-socket host.
+    for (const kind of ['cpu-package', 'cpu-die', 'cpu-core'] as const) {
+        const candidates = sensors.filter((sensor) => sensor.kind === kind);
+        if (candidates.length)
+            return candidates.reduce((hot, sensor) =>
+                sensor.celsius > hot.celsius ? sensor : hot
+            );
+    }
+    return null;
+}
+
+export function temperatureLabel(sensor: HostTemperatureSensor): string {
+    if (sensor.kind === 'cpu-package')
+        return `CPU package ${sensor.label.replace(/^Package id /, '')}`;
+    if (sensor.kind === 'cpu-core') return `CPU ${sensor.label.toLowerCase()}`;
+    if (sensor.kind === 'cpu-die') return `CPU die · ${sensor.label}`;
+    if (sensor.kind === 'chipset') return 'Chipset';
+    return `NVMe · ${sensor.label}`;
+}
+
+export function formatTemperature(celsius: number | null | undefined): string {
+    return typeof celsius === 'number' && Number.isFinite(celsius)
+        ? `${Number(celsius.toFixed(1))} °C`
+        : 'Unavailable';
+}
+
+/** One CPU series per host, plus bounded device series for the host drill-down. */
+export function historyTemperatures(
+    sensors: HostTemperatureSensor[] = []
+): HostTemperatureSensor[] {
+    const cpu = cpuTemperature(sensors);
+    const devices = sensors
+        .filter(
+            (sensor) =>
+                sensor.kind === 'chipset' ||
+                (sensor.kind === 'nvme' && sensor.label.toLowerCase() === 'composite')
+        )
+        .slice(0, 16);
+    return [...(cpu ? [{ ...cpu, id: 'cpu', label: 'CPU temperature' }] : []), ...devices];
+}
@@ -1,4 +1,5 @@
 import type { HostLibraryCache } from './game-files.js';
+import type { HostTemperatureSensor } from './host-temperatures.js';

 export type HostMode = 'active' | 'draining' | 'disabled';
 export type HostReadiness =
@@ -149,6 +150,7 @@ export interface HostInventory {
    cpu?: string;
    cpuCount?: number;
    cpuUsagePercent?: number;
+    temperatures?: HostTemperatureSensor[];
    memoryTotal?: number;
    memoryAvailable?: number;
    storageTotal?: number;
@@ -24,6 +24,7 @@ import { migrateLeaderboardReports } from './leaderboard-reports.js';
 import { migratePlayerScoreTotal } from './player-score-total.js';
 import { migrateBank } from './bank.js';
 import { migrateBankCredentials } from './bank-credentials.js';
+import { migrateTemperatures } from '../hosts/temperatures.js';
 import type { Migration } from './migration-runner.js';

 // Append only: never renumber entries or change an applied migration to add an
@@ -118,7 +119,8 @@ export const migrations: readonly Migration[] = [
    },
    { version: 20, name: 'persistent player bank and transfer ledger', run: migrateBank },
    { version: 21, name: 'instance-owned bank credentials', run: migrateBankCredentials },
-    { version: 22, name: 'remembered browser sessions', run: migrateRememberedSessions }
+    { version: 22, name: 'remembered browser sessions', run: migrateRememberedSessions },
+    { version: 23, name: 'host temperature history', run: migrateTemperatures }
 ];

 export const databaseVersion = migrations.length;
@@ -1,5 +1,6 @@
 import { db } from '../db/index.js';
 import type { CapacityHistory, CapacityQuery, CapacityRange } from '../../capacity.js';
+import { readTemperatureHistory } from './temperatures.js';

 const windows: Record<CapacityRange, { duration: number; bucketMs: number }> = {
    '1h': { duration: 3_600_000, bucketMs: 60_000 },
@@ -15,10 +16,14 @@ export function readCapacityHistory(query: CapacityQuery, now = Date.now()): Cap
    const count = duration / bucketMs;
    const hosts = handle
        .prepare(
-            `SELECT id,created_at FROM ec_hosts
+            `SELECT id,name,created_at FROM ec_hosts
        WHERE deleted_at IS NULL ${query.hostId ? 'AND id=?' : ''} ORDER BY id`
        )
-        .all(...(query.hostId ? [query.hostId] : [])) as { id: string; created_at: string }[];
+        .all(...(query.hostId ? [query.hostId] : [])) as {
+        id: string;
+        name: string;
+        created_at: string;
+    }[];
    // The primary key bounds each host scan to the selected interval. Aggregate
    // before crossing the worker boundary; do not return raw heartbeat rows.
    const samples = handle.prepare(`SELECT
@@ -69,5 +74,12 @@ export function readCapacityHistory(query: CapacityQuery, now = Date.now()): Cap
            point.memoryBytesAvailable = null;
        }
    }
-    return { range: query.range, startAt, endAt: now, bucketMs, points };
+    return {
+        range: query.range,
+        startAt,
+        endAt: now,
+        bucketMs,
+        points,
+        temperatures: readTemperatureHistory(query, hosts, startAt, now, bucketMs)
+    };
 }
@@ -23,6 +23,8 @@ import { logResultMetadata } from './log-results.js';
 import { parseHostInventory } from './request-context.js';
 import { fleetSettings } from '../settings.js';
 import { recordInstanceResources, validateInstanceResources } from './calibration.js';
+import { validateTemperatures } from '../../host-temperatures.js';
+import { recordTemperatures } from './temperatures.js';
 import { refreshHostAvailability, scheduleAvailabilityRefresh } from './availability.js';
 import { hostAgentLaneLabel, hostAgentLaneState } from '../../hosts.js';
 import type {
@@ -622,6 +624,8 @@ export function validateInventory(value: unknown): HostInventory {
        throw new Error('Host inventory must be an object.');
    const inventory = value as HostInventory;
    if (JSON.stringify(inventory).length > 256_000) throw new Error('Host inventory is too large.');
+    if (inventory.temperatures !== undefined)
+        inventory.temperatures = validateTemperatures(inventory.temperatures);
    for (const key of ['hostname', 'agentVersion', 'os', 'architecture', 'cpu', 'dockerVersion']) {
        if (
            inventory[key] !== undefined &&
@@ -924,6 +928,7 @@ export function heartbeat(hostId: string, value: unknown): Host {

 function recordResources(hostId: string, inventory: HostInventory, timestamp: string): void {
    recordInstanceResources(hostId, inventory, timestamp);
+    recordTemperatures(hostId, inventory, timestamp);
    database()
        .prepare(
            `INSERT OR REPLACE INTO ec_host_resource_samples
@@ -0,0 +1,96 @@
+import type Database from 'better-sqlite3';
+import { db } from '../db/index.js';
+import { historyTemperatures } from '../../host-temperatures.js';
+import type { HostInventory } from '../../hosts.js';
+import type { CapacityQuery, TemperatureHistorySeries } from '../../capacity.js';
+
+export function migrateTemperatures(handle: Database.Database): void {
+    handle.exec(`CREATE TABLE IF NOT EXISTS ec_host_temperature_samples (
+        host_id TEXT NOT NULL REFERENCES ec_hosts(id), bucket TEXT NOT NULL,
+        sensor_id TEXT NOT NULL, kind TEXT NOT NULL, label TEXT NOT NULL,
+        celsius REAL NOT NULL CHECK(celsius BETWEEN -100 AND 200),
+        PRIMARY KEY(host_id,bucket,sensor_id)
+    );
+    CREATE INDEX IF NOT EXISTS ec_host_temperatures_retention ON ec_host_temperature_samples(bucket);`);
+}
+
+export function recordTemperatures(
+    hostId: string,
+    inventory: HostInventory,
+    timestamp: string
+): void {
+    const sensors = historyTemperatures(inventory.temperatures);
+    if (!sensors.length) return;
+    const bucket = new Date(Math.floor(Date.parse(timestamp) / 60_000) * 60_000).toISOString();
+    const insert = db().prepare(`INSERT INTO ec_host_temperature_samples
+        (host_id,bucket,sensor_id,kind,label,celsius) VALUES(?,?,?,?,?,?)
+        ON CONFLICT(host_id,bucket,sensor_id) DO UPDATE SET
+        celsius=MAX(celsius,excluded.celsius),kind=excluded.kind,label=excluded.label`);
+    for (const sensor of sensors)
+        insert.run(hostId, bucket, sensor.id, sensor.kind, sensor.label, sensor.celsius);
+}
+
+/** Runs on the capacity report worker, returning at most 168 points per series. */
+export function readTemperatureHistory(
+    query: CapacityQuery,
+    hosts: { id: string; name: string; created_at: string }[],
+    startAt: number,
+    endAt: number,
+    bucketMs: number
+): TemperatureHistorySeries[] {
+    const count = Math.round((endAt - startAt) / bucketMs);
+    const statement = db().prepare(`SELECT sensor_id,kind,label,
+        MIN(?,MAX(0,CAST((CAST(strftime('%s',bucket) AS INTEGER)*1000-?) / ? AS INTEGER))) AS period,
+        MAX(celsius) AS celsius
+        FROM ec_host_temperature_samples WHERE host_id=? AND bucket>=? AND bucket<=?
+        ${query.hostId ? '' : "AND sensor_id='cpu'"}
+        GROUP BY sensor_id,period ORDER BY sensor_id,period`);
+    const result: TemperatureHistorySeries[] = [];
+    for (const host of hosts) {
+        const bySensor = new Map<string, TemperatureHistorySeries>();
+        const add = (id: string, kind: TemperatureHistorySeries['kind'], label: string) => {
+            const series = {
+                id: `${host.id}:${id}`,
+                hostId: host.id,
+                kind,
+                label,
+                points: Array.from({ length: count }, (_, i) => ({
+                    at: startAt + i * bucketMs,
+                    celsius: null as number | null
+                }))
+            };
+            bySensor.set(id, series);
+            result.push(series);
+            return series;
+        };
+        add('cpu', 'cpu', query.hostId ? 'CPU temperature' : host.name);
+        const rows = statement.all(
+            count - 1,
+            startAt,
+            bucketMs,
+            host.id,
+            new Date(
+                Math.max(
+                    Math.floor(Date.parse(host.created_at) / 60_000) * 60_000,
+                    Math.floor(startAt / 60_000) * 60_000
+                )
+            ).toISOString(),
+            new Date(endAt).toISOString()
+        ) as { sensor_id: string; kind: string; label: string; period: number; celsius: number }[];
+        for (const row of rows) {
+            let series = bySensor.get(row.sensor_id);
+            if (!series) {
+                if (!['nvme', 'chipset'].includes(row.kind) || bySensor.size >= 17) continue;
+                const kind = row.kind as 'nvme' | 'chipset';
+                const number = [...bySensor.values()].filter((s) => s.kind === kind).length + 1;
+                series = add(
+                    row.sensor_id,
+                    kind,
+                    `${kind === 'nvme' ? 'NVMe drive' : 'Chipset'} ${number}`
+                );
+            }
+            if (series.points[row.period]) series.points[row.period].celsius = row.celsius;
+        }
+    }
+    return result;
+}
@@ -556,6 +556,14 @@ function samplePolicies(s = fleetSettings()) {
            value: iso(s.hostResourceKeepDays),
            enabled: s.hostResourceKeepDays > 0
        },
+        {
+            category: 'host-temperatures',
+            label: 'Host temperature samples',
+            table: 'ec_host_temperature_samples',
+            condition: 'bucket < ?',
+            value: iso(s.hostResourceKeepDays),
+            enabled: s.hostResourceKeepDays > 0
+        },
        {
            category: 'instance-resources',
            label: 'Instance resource samples',
@@ -1,5 +1,6 @@
 <script lang="ts">
    import type { Host } from '$lib/hosts.js';
+    import { cpuTemperature, formatTemperature } from '$lib/host-temperatures.js';
    import { contextHref } from '$lib/navigation-context.js';
    import {
        hostListStatus,
@@ -68,7 +69,13 @@
                            ? '—'
                            : `${host.playerCount}${host.unreportedInstances ? '+' : ''}`}</td
                    >
-                    <td class:stale>{hostCpuUsage(host.inventory.cpuUsagePercent)}</td>
+                    <td class:stale
+                        >{hostCpuUsage(host.inventory.cpuUsagePercent)}<span class="secondary"
+                            >CPU temp: {formatTemperature(
+                                cpuTemperature(host.inventory.temperatures)?.celsius
+                            )}</span
+                        ></td
+                    >
                    <td class:stale
                        >{hostUsedSpace(
                            host.inventory.memoryTotal,
@@ -0,0 +1,79 @@
+<script lang="ts">
+    import type { Host } from '$lib/hosts.js';
+    import { cpuTemperature, formatTemperature, temperatureLabel } from '$lib/host-temperatures.js';
+    import { hostReadingsStale } from '$ui/host-overview.js';
+    import { onMount } from 'svelte';
+    let { host }: { host: Host } = $props();
+    let now = $state(Date.now());
+    const stale = $derived(hostReadingsStale(host, now));
+    const sensors = $derived(host.inventory.temperatures ?? []);
+    const cpu = $derived(cpuTemperature(sensors));
+    onMount(() => {
+        const clock = setInterval(() => (now = Date.now()), 15_000);
+        return () => clearInterval(clock);
+    });
+</script>
+
+<div class:stale>
+    {formatTemperature(cpu?.celsius)}{#if stale && cpu}
+        (last reported){/if}
+    {#if cpu}<span class="source">{temperatureLabel(cpu)}</span>{/if}
+    {#if sensors.length}
+        <details>
+            <summary>Temperature sensors</summary>
+            <ul>
+                {#each sensors as sensor (sensor.id)}
+                    <li>
+                        <span
+                            >{temperatureLabel(sensor)}<small
+                                >{sensor.device} · {sensor.label} · {sensor.id.slice(-6)}</small
+                            ></span
+                        ><span>{formatTemperature(sensor.celsius)}</span>
+                    </li>
+                {/each}
+            </ul>
+            {#if stale}<p>Last reported readings; waiting for a fresh host heartbeat.</p>{/if}
+        </details>
+    {:else}<span class="source">No supported temperature sensors reported.</span>{/if}
+</div>
+
+<style>
+    .source,
+    small {
+        display: block;
+        color: var(--text-secondary);
+        font-size: var(--text-xs);
+        margin-top: var(--space-1);
+    }
+    .stale {
+        color: var(--text-secondary);
+    }
+    details {
+        margin-top: var(--space-2);
+    }
+    summary {
+        cursor: pointer;
+        font-size: var(--text-sm);
+    }
+    summary:focus-visible {
+        outline: 2px solid var(--accent-primary);
+        outline-offset: 3px;
+    }
+    ul {
+        padding: 0;
+        list-style: none;
+        margin: var(--space-3) 0;
+        display: grid;
+        gap: var(--space-3);
+    }
+    li {
+        display: flex;
+        justify-content: space-between;
+        flex-wrap: wrap;
+        gap: var(--space-2);
+        font-variant-numeric: tabular-nums;
+    }
+    p {
+        font-size: var(--text-xs);
+    }
+</style>
@@ -0,0 +1,456 @@
+<script lang="ts">
+    import type { Snippet } from 'svelte';
+    import type { CapacityHistory, TemperatureHistorySeries } from '$lib/capacity.js';
+    import { formatTemperature } from '$lib/host-temperatures.js';
+    import { timeFormat } from '$ui/stores/time-zone.js';
+
+    let {
+        history,
+        scope,
+        controls
+    }: { history: CapacityHistory; scope: string; controls?: Snippet } = $props();
+    let visibility = $state<Record<string, boolean>>({});
+    let selectedAt = $state<number | null>(null);
+    const series = $derived(history.temperatures);
+    const visible = $derived(series.filter((s) => visibility[s.id] ?? s.kind === 'cpu'));
+    const values = $derived(
+        series.flatMap((s) => s.points.flatMap((p) => (p.celsius === null ? [] : [p.celsius])))
+    );
+    const minimum = $derived(
+        values.length
+            ? Math.floor((values.reduce((a, b) => Math.min(a, b), Infinity) - 5) / 10) * 10
+            : 0
+    );
+    const maximum = $derived(
+        values.length
+            ? Math.ceil((values.reduce((a, b) => Math.max(a, b), -Infinity) + 5) / 10) * 10
+            : 100
+    );
+    const times = $derived(history.points.map((p) => p.at));
+    const measured = $derived(visible.some((s) => s.points.some((p) => p.celsius !== null)));
+    const selectedIndex = $derived.by(() => {
+        if (selectedAt !== null)
+            return times.reduce(
+                (best, at, i) =>
+                    Math.abs(at - selectedAt!) < Math.abs(times[best] - selectedAt!) ? i : best,
+                0
+            );
+        for (let i = times.length - 1; i >= 0; i--)
+            if (visible.some((s) => s.points[i]?.celsius != null)) return i;
+        return Math.max(0, times.length - 1);
+    });
+    const selectedTime = $derived(times[selectedIndex]);
+    const readout = $derived(
+        visible
+            .map((s) => `${s.label}: ${formatTemperature(s.points[selectedIndex]?.celsius)}`)
+            .join('. ')
+    );
+    const ticks = [0, 0.25, 0.5, 0.75, 1];
+    function color(s: TemperatureHistorySeries) {
+        return `var(--chart-series-${(series.indexOf(s) % 6) + 1})`;
+    }
+    function dash(s: TemperatureHistorySeries) {
+        return ['', '7 4', '2 4'][Math.floor(series.indexOf(s) / 6) % 3];
+    }
+    function x(at: number) {
+        return (
+            8 +
+            984 *
+                Math.max(
+                    0,
+                    Math.min(
+                        1,
+                        (at - history.startAt) / Math.max(1, history.endAt - history.startAt)
+                    )
+                )
+        );
+    }
+    function y(celsius: number) {
+        return 230 - (220 * (celsius - minimum)) / (maximum - minimum);
+    }
+    function segments(s: TemperatureHistorySeries) {
+        const result: { x: number; y: number }[][] = [];
+        let segment: { x: number; y: number }[] = [];
+        for (const point of s.points) {
+            if (point.celsius === null) {
+                if (segment.length) result.push(segment);
+                segment = [];
+            } else segment.push({ x: x(point.at), y: y(point.celsius) });
+        }
+        if (segment.length) result.push(segment);
+        return result;
+    }
+    function inspect(event: PointerEvent) {
+        if (!times.length || (event.type === 'pointermove' && event.pointerType === 'touch'))
+            return;
+        const rect = (event.currentTarget as HTMLElement).getBoundingClientRect();
+        selectedAt =
+            history.startAt +
+            Math.max(
+                0,
+                Math.min(1, (((event.clientX - rect.left) / rect.width) * 1000 - 8) / 984)
+            ) *
+                (history.endAt - history.startAt);
+    }
+    function move(event: KeyboardEvent) {
+        if (!times.length || !['ArrowLeft', 'ArrowRight', 'Home', 'End'].includes(event.key))
+            return;
+        event.preventDefault();
+        const next =
+            event.key === 'Home'
+                ? 0
+                : event.key === 'End'
+                  ? times.length - 1
+                  : Math.max(
+                        0,
+                        Math.min(
+                            times.length - 1,
+                            selectedIndex + (event.key === 'ArrowLeft' ? -1 : 1)
+                        )
+                    );
+        selectedAt = times[next];
+    }
+</script>
+
+<section class="panel" id="temperatures" aria-labelledby="temperature-history-title">
+    <div class="heading">
+        <div>
+            <h2 id="temperature-history-title">Temperature history</h2>
+            <p>{scope} · Peak temperature per period</p>
+        </div>
+        {@render controls?.()}
+    </div>
+    <div class="legend" role="group" aria-label="Temperature series">
+        {#each series as sensor (sensor.id)}
+            <button
+                type="button"
+                aria-pressed={visibility[sensor.id] ?? sensor.kind === 'cpu'}
+                onclick={() =>
+                    (visibility[sensor.id] = !(visibility[sensor.id] ?? sensor.kind === 'cpu'))}
+            >
+                <svg viewBox="0 0 24 8" aria-hidden="true"
+                    ><line
+                        x1="0"
+                        x2="24"
+                        y1="4"
+                        y2="4"
+                        stroke={color(sensor)}
+                        stroke-dasharray={dash(sensor)}
+                    /></svg
+                >
+                {sensor.label}
+            </button>
+        {/each}
+    </div>
+    <div class="axis-title">Temperature (°C)</div>
+    <div class="chart-grid">
+        <div class="y-axis" aria-hidden="true">
+            {#each ticks as tick}<span style:top={`${(1 - tick) * 100}%`}
+                    >{minimum + (maximum - minimum) * tick}</span
+                >{/each}
+        </div>
+        <div
+            class="plot"
+            role="slider"
+            tabindex="0"
+            aria-label="Temperature history. Use left and right arrows to inspect a period."
+            aria-valuemin="0"
+            aria-valuemax={Math.max(0, times.length - 1)}
+            aria-valuenow={selectedIndex}
+            aria-valuetext={selectedTime === undefined
+                ? 'No temperature history'
+                : `${$timeFormat.period(selectedTime)}. ${readout || 'No sensors selected'}`}
+            onpointermove={inspect}
+            onpointerdown={inspect}
+            onkeydown={move}
+        >
+            <svg viewBox="0 0 1000 240" preserveAspectRatio="none" aria-hidden="true">
+                {#each ticks as tick}<line
+                        class="grid-line"
+                        x1="0"
+                        x2="1000"
+                        y1={y(minimum + (maximum - minimum) * tick)}
+                        y2={y(minimum + (maximum - minimum) * tick)}
+                    />{/each}
+                {#each visible as sensor (sensor.id)}
+                    <g data-series={sensor.id}>
+                        {#each segments(sensor) as segment}
+                            <path
+                                class="trend"
+                                stroke={color(sensor)}
+                                stroke-dasharray={dash(sensor)}
+                                d={segment
+                                    .map((point, i) => `${i ? 'L' : 'M'}${point.x},${point.y}`)
+                                    .join(' ')}
+                            />
+                            {#if segment.length === 1}<circle
+                                    cx={segment[0].x}
+                                    cy={segment[0].y}
+                                    r="3"
+                                    fill={color(sensor)}
+                                />{/if}
+                        {/each}
+                        {#if sensor.points[selectedIndex]?.celsius != null}<circle
+                                class="selected-point"
+                                cx={x(selectedTime)}
+                                cy={y(sensor.points[selectedIndex].celsius!)}
+                                r="4"
+                                fill={color(sensor)}
+                            />{/if}
+                    </g>
+                {/each}
+                {#if selectedTime !== undefined && measured}<line
+                        class="cursor"
+                        x1={x(selectedTime)}
+                        x2={x(selectedTime)}
+                        y1="5"
+                        y2="235"
+                    />{/if}
+            </svg>
+            {#if !measured}<div class="empty-chart">
+                    <strong
+                        >{visible.length || !series.length
+                            ? 'No temperature readings in this period'
+                            : 'Select a sensor to show history'}</strong
+                    >
+                    {#if visible.length || !series.length}<span
+                            >Readings appear when an agent reports supported temperature sensors.</span
+                        >{/if}
+                </div>{/if}
+        </div>
+        <div class="x-axis" aria-hidden="true">
+            <span>{$timeFormat.period(history.startAt)}</span><span
+                >{$timeFormat.period(history.endAt)}</span
+            >
+        </div>
+    </div>
+    {#if selectedTime !== undefined && visible.length}
+        <div class="readout" aria-label="Selected temperature period">
+            <span>{$timeFormat.period(selectedTime)}</span>
+            <dl>
+                {#each visible as sensor (sensor.id)}<div>
+                        <dt>{sensor.label}</dt>
+                        <dd>{formatTemperature(sensor.points[selectedIndex]?.celsius)}</dd>
+                    </div>{/each}
+            </dl>
+        </div>
+    {/if}
+    <details>
+        <summary>About this chart</summary>
+        <p>
+            Each CPU line represents one host: its hottest reported package or physical die, falling
+            back to its hottest core. Select a host to compare its CPU, NVMe drives and chipset.
+            Per-minute peaks are retained; longer periods show their highest reading. Gaps mean no
+            measurement was received in that period. Temperatures include all workloads and are
+            unaffected by the game filter. Colors identify series; they do not indicate safe or
+            unsafe temperatures.
+        </p>
+    </details>
+</section>
+
+<style>
+    .panel {
+        background: var(--surface);
+        border: 1px solid var(--border-subtle);
+        border-radius: var(--radius-lg);
+        padding: var(--space-5);
+        min-width: 0;
+    }
+    h2 {
+        margin: 0;
+        font-size: var(--text-lg);
+    }
+    .heading {
+        display: flex;
+        align-items: flex-start;
+        justify-content: space-between;
+        gap: var(--space-4);
+        flex-wrap: wrap;
+    }
+    .heading p {
+        color: var(--text-secondary);
+        font-size: var(--text-sm);
+        margin: var(--space-2) 0 0;
+    }
+    .legend {
+        display: flex;
+        flex-wrap: wrap;
+        gap: var(--space-2) var(--space-5);
+        margin: var(--space-4) 0;
+    }
+    .legend button {
+        display: flex;
+        align-items: center;
+        gap: var(--space-2);
+        background: none;
+        border: 0;
+        padding: var(--space-2) 0;
+        color: var(--text-primary);
+        font: inherit;
+        font-size: var(--text-sm);
+        cursor: pointer;
+        overflow-wrap: anywhere;
+        max-width: 100%;
+    }
+    .legend button[aria-pressed='false'] {
+        color: var(--text-secondary);
+    }
+    .legend button[aria-pressed='false'] svg {
+        opacity: 0.35;
+    }
+    .legend svg {
+        width: 24px;
+        height: 8px;
+        flex-shrink: 0;
+    }
+    .legend line {
+        stroke-width: 3;
+    }
+    button:focus-visible,
+    .plot:focus-visible,
+    summary:focus-visible {
+        outline: 2px solid var(--accent-primary);
+        outline-offset: 3px;
+    }
+    .axis-title,
+    .x-axis,
+    .y-axis {
+        color: var(--text-tertiary);
+        font-size: var(--text-xs);
+    }
+    .chart-grid {
+        display: grid;
+        grid-template-columns: 42px minmax(0, 1fr);
+        gap: 10px;
+        margin-top: 14px;
+    }
+    .y-axis {
+        position: relative;
+        height: 220px;
+        margin-top: 10px;
+    }
+    .y-axis span {
+        position: absolute;
+        right: 0;
+        transform: translateY(-50%);
+    }
+    .plot {
+        position: relative;
+        height: 240px;
+        cursor: crosshair;
+        touch-action: pan-y;
+    }
+    .plot > svg {
+        display: block;
+        width: 100%;
+        height: 100%;
+        overflow: visible;
+    }
+    .plot path,
+    .plot line {
+        vector-effect: non-scaling-stroke;
+    }
+    .grid-line {
+        stroke: var(--border-subtle);
+        stroke-width: 1;
+    }
+    .trend {
+        fill: none;
+        stroke-width: 2;
+        stroke-linejoin: round;
+        stroke-linecap: round;
+    }
+    .selected-point {
+        stroke: var(--surface);
+        stroke-width: 2;
+    }
+    .cursor {
+        stroke: var(--text-tertiary);
+        stroke-width: 1;
+        stroke-dasharray: 4 5;
+    }
+    .x-axis {
+        grid-column: 2;
+        display: flex;
+        justify-content: space-between;
+        gap: var(--space-4);
+    }
+    .x-axis span:last-child {
+        text-align: right;
+    }
+    .empty-chart {
+        position: absolute;
+        inset: 0;
+        display: flex;
+        align-items: center;
+        justify-content: center;
+        flex-direction: column;
+        gap: var(--space-2);
+        padding: var(--space-4);
+        text-align: center;
+        pointer-events: none;
+    }
+    .empty-chart strong {
+        font-size: var(--text-sm);
+        font-weight: var(--weight-medium);
+    }
+    .empty-chart span {
+        font-size: var(--text-xs);
+        color: var(--text-secondary);
+    }
+    .readout {
+        display: flex;
+        flex-wrap: wrap;
+        gap: var(--space-4);
+        font-size: var(--text-xs);
+        margin: var(--space-4) 0;
+    }
+    .readout > span,
+    dt {
+        color: var(--text-secondary);
+    }
+    dl {
+        display: flex;
+        flex-wrap: wrap;
+        gap: var(--space-3) var(--space-5);
+        margin: 0;
+    }
+    dl > div {
+        display: flex;
+        flex-wrap: wrap;
+        min-width: 0;
+        gap: var(--space-2);
+    }
+    dt {
+        overflow-wrap: anywhere;
+    }
+    dd {
+        margin: 0;
+        font-variant-numeric: tabular-nums;
+    }
+    summary {
+        cursor: pointer;
+        font-size: var(--text-xs);
+        color: var(--text-secondary);
+        padding: var(--space-2) 0;
+    }
+    details p {
+        color: var(--text-secondary);
+        font-size: var(--text-xs);
+        line-height: 1.65;
+    }
+    @media (max-width: 640px) {
+        .panel {
+            padding: var(--space-4);
+        }
+        .chart-grid {
+            grid-template-columns: 32px minmax(0, 1fr);
+            gap: 8px;
+        }
+    }
+    @media (pointer: coarse) {
+        .legend button {
+            min-height: 44px;
+        }
+    }
+</style>
@@ -78,6 +78,12 @@

    /* Metrics: unknown uses a pattern as well as color. */
    --chart-active: var(--accent-primary);
+    --chart-series-1: var(--accent-primary);
+    --chart-series-2: #b99af7;
+    --chart-series-3: #69c6ba;
+    --chart-series-4: #dfb76c;
+    --chart-series-5: #df91b6;
+    --chart-series-6: #bbc7db;
    --chart-idle: #2b3442;
    --chart-unknown-base: #605b64;
    --chart-unknown-stripe: #a7a0ad;
@@ -17,6 +17,7 @@
    import Select from '$ui/components/primitives/Select.svelte';
    import CapacityHosts from '$ui/components/capacity/CapacityHosts.svelte';
    import CapacityChart from '$ui/components/capacity/CapacityChart.svelte';
+    import TemperatureChart from '$ui/components/capacity/TemperatureChart.svelte';
    import RebalancePanel from '$ui/components/capacity/RebalancePanel.svelte';
    import Freshness from '$ui/components/Freshness.svelte';
    import { plural } from '$ui/plural.js';
@@ -319,8 +320,7 @@
                : 'Waiting for host reports.'} Refreshes every 30 seconds while visible.
        </p>
    </details>
-    <CapacityChart history={report.history} scope={scopeName}>
-        {#snippet controls()}<nav class="ranges" aria-label="History range">
+    {#snippet historyRanges()}<nav class="ranges" aria-label="History range">
            {#each ranges as range}<a
                    href={scopeUrl({ range: range.value })}
                    aria-current={report.query.range === range.value ? 'page' : undefined}
@@ -328,7 +328,12 @@
                    data-sveltekit-noscroll>{range.label}</a
                >{/each}
        </nav>{/snippet}
-    </CapacityChart>
+    <CapacityChart history={report.history} scope={scopeName} controls={historyRanges} />
+    {#key report.query.hostId}<TemperatureChart
+            history={report.history}
+            scope={scopeName}
+            controls={historyRanges}
+        />{/key}
 </div>

 <style>
@@ -28,6 +28,7 @@
    import DeleteHostDialog from '$ui/components/DeleteHostDialog.svelte';
    import HostPreparation from '$ui/components/HostPreparation.svelte';
    import CapacitySamples from '$ui/components/CapacitySamples.svelte';
+    import HostTemperatures from '$ui/components/HostTemperatures.svelte';
    import ServerCard from '$ui/components/ServerCard.svelte';
    import type { ServerObservation } from '$lib/server-observation.js';
    import { syncServerTime } from '$ui/stores/servers.js';
@@ -300,6 +301,8 @@
                                : `${host.inventory.cpuUsagePercent.toFixed(1)}%`}{#if !host.connected && host.inventory.cpuUsagePercent !== undefined}
                                (last reported){/if}
                        </dd>
+                        <dt>CPU temperature</dt>
+                        <dd><HostTemperatures {host} /></dd>
                        <dt>Memory available / total</dt>
                        <dd>
                            {bytes(host.inventory.memoryAvailable)} / {bytes(
@@ -35,7 +35,7 @@ import urllib.parse
 import urllib.request
 import uuid

-VERSION = "1.9.3"
+VERSION = "1.10.0"
 LABEL = "io.electric-cherry"
 MAX_FILE = 32 * 1024 * 1024
 MANIFEST_MAX_ENTRIES = 10000
@@ -63,6 +63,7 @@ UPDATE_TIMER_ENABLED = Path("/etc/systemd/system/timers.target.wants/electric-ch
 UPDATE_JOURNAL = Path("/etc/electric-cherry/update/journal.json")
 UPDATE_STATUS = Path("/etc/electric-cherry/update-status.json")
 AGENT_HEALTH = Path("/etc/electric-cherry/agent-health.json")
+HWMON_ROOT = Path("/sys/class/hwmon")
 ROOT_MARKER = ".electric-cherry.json"


@@ -1308,6 +1309,59 @@ class Agent:
                return None
            return round(100 * (total - deltas[3] - deltas[4]) / total, 1)

+    def temperatures(self):
+        """Best-effort, read-only samples from known millidegree-Celsius drivers."""
+        samples = []
+        try:
+            chips = sorted(HWMON_ROOT.glob("hwmon*"))[:64]
+        except OSError:
+            return samples
+        for chip in chips:
+            try:
+                device = (chip / "name").read_text().strip()
+                if device not in ("coretemp", "k10temp", "nvme") and not device.startswith("pch_"):
+                    continue
+                # hwmon numbers can change on reboot. The physical sysfs path
+                # and channel identify separate CPU sockets and NVMe devices.
+                identity = re.sub(r"/(?:hwmon/)?hwmon\d+$", "", str(chip.resolve()))
+                inputs = sorted(chip.glob("temp[0-9]*_input"))[:128]
+            except (OSError, ValueError):
+                continue
+            for source in inputs:
+                try:
+                    channel = source.name.removesuffix("_input")
+                    label_path = chip / (channel + "_label")
+                    label = label_path.read_text().strip() if label_path.exists() else channel
+                    if not label or len(label) > 120:
+                        continue
+                    fault = chip / (channel + "_fault")
+                    enabled = chip / (channel + "_enable")
+                    if (fault.exists() and int(fault.read_text()) != 0) or (enabled.exists() and int(enabled.read_text()) == 0):
+                        continue
+                    if device == "coretemp":
+                        kind = "cpu-package" if label.startswith("Package id ") else "cpu-core" if label.startswith("Core ") else None
+                    elif device == "k10temp":
+                        # Tctl can contain a fan-control offset. Never present
+                        # it as the physical die temperature or a fleet metric.
+                        kind = "cpu-die" if label == "Tdie" or re.fullmatch(r"Tccd\d+", label) else None
+                    else:
+                        kind = "nvme" if device == "nvme" else "chipset"
+                    if kind is None:
+                        continue
+                    celsius = int(source.read_text()) / 1000
+                    if not -100 <= celsius <= 200:
+                        continue
+                    sensor_id = hashlib.sha256((identity + "/" + channel).encode()).hexdigest()[:24]
+                    samples.append({"id": sensor_id, "device": device, "label": label,
+                                    "kind": kind, "celsius": round(celsius, 1)})
+                    if len(samples) >= 128:
+                        return samples
+                except (OSError, ValueError):
+                    # Read permission alone does not imply data is available
+                    # (e.g. ENODATA), and a device can disappear during a read.
+                    continue
+        return samples
+
    def instance_resources(self, observations=None):
        """One bounded Docker stats batch; unavailable samples are never zero load."""
        ids = self.docker.call("container", "ls", "--all", "--filter", "label=" + LABEL + ".managed=true",
@@ -1392,6 +1446,7 @@ class Agent:

    def inventory(self):
        cpu_usage = self.cpu_usage_percent()
+        temperatures = self.temperatures()
        usage = shutil.disk_usage(self.root)
        memory = {}
        try:
@@ -1516,9 +1571,9 @@ class Agent:
                if line.startswith("PRETTY_NAME="):
                    os_name = line.split("=", 1)[1].strip('"')
        update = self.update_inventory()
-        capabilities = ["library.remove", "plutonium.cache", "instance.cancel-create", "instance.safety-override", "instance.unhealthy-recovery", "instance.rebuild-files", "instance.autoscaling-remove", "operations.read-lane", "operations.health-lane", "rcon.status-batch", "instance.heartbeat-state", "files.manifest", "files.blobs", "logs.read-offset", "bank.settlement"] + (["host.decommission"] if self.can_uninstall() else []) + (["agent.update"] if update["enabled"] else [])
+        capabilities = ["host.temperatures", "library.remove", "plutonium.cache", "instance.cancel-create", "instance.safety-override", "instance.unhealthy-recovery", "instance.rebuild-files", "instance.autoscaling-remove", "operations.read-lane", "operations.health-lane", "rcon.status-batch", "instance.heartbeat-state", "files.manifest", "files.blobs", "logs.read-offset", "bank.settlement"] + (["host.decommission"] if self.can_uninstall() else []) + (["agent.update"] if update["enabled"] else [])
        plutonium_cache = self.plutonium_cache_inventory()
-        return {"hostname": socket.gethostname(), "agentVersion": VERSION, "protocolVersion": 1, "capabilities": capabilities, "agentUpdate": update, "agentLanes": self.lane_inventory(), "os": os_name, "osVersion": platform.release(), "architecture": platform.machine(), "cpu": platform.processor() or platform.machine(), "cpuCount": os.cpu_count() or 0, **resources, **({"cpuUsagePercent": cpu_usage} if cpu_usage is not None else {}), "memoryTotal": memory.get("MemTotal", 0), "memoryAvailable": memory.get("MemAvailable", 0), "storageTotal": usage.total, "storageAvailable": usage.free, "dockerVersion": docker_version, "dockerRunning": docker_running, "directoriesWritable": os.access(self.root, os.W_OK), "runtimeImages": sorted(set(images)), "preparedRuntimeImages": sorted(set(prepared_images)), **({"plutoniumCache": plutonium_cache} if plutonium_cache else {}), "libraries": libraries, "libraryCaches": caches, "currentOperation": self.current_operation, "portRange": self.config.get("portRange", {"min": 28960, "max": 29060})}
+        return {"hostname": socket.gethostname(), "agentVersion": VERSION, "protocolVersion": 1, "capabilities": capabilities, "agentUpdate": update, "agentLanes": self.lane_inventory(), "os": os_name, "osVersion": platform.release(), "architecture": platform.machine(), "cpu": platform.processor() or platform.machine(), "cpuCount": os.cpu_count() or 0, "temperatures": temperatures, **resources, **({"cpuUsagePercent": cpu_usage} if cpu_usage is not None else {}), "memoryTotal": memory.get("MemTotal", 0), "memoryAvailable": memory.get("MemAvailable", 0), "storageTotal": usage.total, "storageAvailable": usage.free, "dockerVersion": docker_version, "dockerRunning": docker_running, "directoriesWritable": os.access(self.root, os.W_OK), "runtimeImages": sorted(set(images)), "preparedRuntimeImages": sorted(set(prepared_images)), **({"plutoniumCache": plutonium_cache} if plutonium_cache else {}), "libraries": libraries, "libraryCaches": caches, "currentOperation": self.current_operation, "portRange": self.config.get("portRange", {"min": 28960, "max": 29060})}

    def safe_to_disrupt(self, instance_id, generation, *, allow_unhealthy=False):
        health = self.inspect(instance_id, generation)
@@ -1,8 +1,8 @@
 {
-  "version": "1.9.3",
+  "version": "1.10.0",
  "protocolVersion": 1,
  "platform": "debian-amd64",
-  "sha256": "a0baaab0f0a60e848300e63f31eeefff6293d8b749b5ca9397c36e8e75ff4a93",
+  "sha256": "1a110c92ae875ee7cd060ac7de82260b63d776d2a865e57356d629e0a99573c7",
  "updaterProtocolVersion": 2,
  "updaterSha256": "a9d7f96b87d0ec74d8a0f476dff6e3d2a7a0bc454a93a51d7b57de00c9ce23cb",
  "installerSha256": "384d5e397a51d19cb9fc9ed8793b4314353f726e217030246050859c0bc18231"
@@ -1,8 +0,0 @@
-----BEGIN PGP SIGNATURE-----
-
-iJEEABYKADkWIQSpocb6VBfQzvdIh9HF6fdSNniN+gUCas2mehsUgAAAAAAEAA5t
-YW51MiwyLjUrMS4xMiwwLDMACgkQxen3UjZ4jfpvyAD/THg0KsH9kDSCF/I2vdOV
-h2BBe9r1JWk1F0mFy2njfvAA/R9mMAMdJRkFIxMNgkIl17nQrUq4TpXQsKSSujXZ
-u5oK
-=GspF
-----END PGP SIGNATURE-----
@@ -0,0 +1,110 @@
+import { afterEach, expect, it } from 'vitest';
+import { flushSync, mount, unmount } from 'svelte';
+import TemperatureChart from '../../src/lib/ui/components/capacity/TemperatureChart.svelte';
+import type { CapacityHistory } from '../../src/lib/capacity.js';
+
+let component: ReturnType<typeof mount> | undefined;
+afterEach(async () => {
+    if (component) await unmount(component);
+    component = undefined;
+    document.body.replaceChildren();
+});
+function history(): CapacityHistory {
+    return {
+        range: '1h',
+        startAt: 1000,
+        endAt: 4000,
+        bucketMs: 1000,
+        points: [1000, 2000, 3000].map((at) => ({
+            at,
+            cpuCoresAvailable: null,
+            memoryBytesAvailable: null,
+            reportingHosts: 0,
+            expectedHosts: 1
+        })),
+        temperatures: [
+            {
+                id: 'host:cpu',
+                hostId: 'host',
+                label: 'CPU temperature',
+                kind: 'cpu',
+                points: [65, null, 67].map((celsius, i) => ({ at: 1000 + i * 1000, celsius }))
+            },
+            {
+                id: 'host:drive',
+                hostId: 'host',
+                label: 'NVMe drive 1',
+                kind: 'nvme',
+                points: [49, 50.9, 52].map((celsius, i) => ({ at: 1000 + i * 1000, celsius }))
+            }
+        ]
+    };
+}
+function show(report = history()) {
+    component = mount(TemperatureChart, {
+        target: document.body,
+        props: { history: report, scope: 'ec-2' }
+    });
+    flushSync();
+}
+
+it('defaults to CPU and toggles device lines and readouts together', () => {
+    show();
+    const buttons = document.querySelectorAll<HTMLButtonElement>('.legend button');
+    expect(buttons[0].getAttribute('aria-pressed')).toBe('true');
+    expect(buttons[1].getAttribute('aria-pressed')).toBe('false');
+    expect(document.querySelectorAll('[data-series]')).toHaveLength(1);
+    // A missing middle bucket splits the CPU line into two isolated segments.
+    expect(document.querySelectorAll('[data-series="host:cpu"] path')).toHaveLength(2);
+    buttons[1].click();
+    flushSync();
+    expect(document.querySelectorAll('[data-series]')).toHaveLength(2);
+    expect(document.querySelector('.readout')?.textContent).toContain('52 °C');
+    buttons[0].click();
+    buttons[1].click();
+    flushSync();
+    expect(document.querySelectorAll('[data-series]')).toHaveLength(0);
+    expect(document.body.textContent).toContain('Select a sensor to show history');
+    expect(document.querySelector('.readout')).toBeNull();
+});
+
+it('supports keyboard and pointer inspection without turning missing readings into zero', () => {
+    show();
+    const plot = document.querySelector<HTMLElement>('[role="slider"]')!;
+    expect(plot.getAttribute('aria-valuenow')).toBe('2');
+    plot.dispatchEvent(new KeyboardEvent('keydown', { key: 'Home', bubbles: true }));
+    flushSync();
+    expect(plot.getAttribute('aria-valuetext')).toContain('65 °C');
+    plot.dispatchEvent(new KeyboardEvent('keydown', { key: 'ArrowRight', bubbles: true }));
+    flushSync();
+    expect(plot.getAttribute('aria-valuetext')).toContain('Unavailable');
+    expect(document.querySelector('.readout')?.textContent).not.toContain('0 °C');
+    Object.defineProperty(plot, 'getBoundingClientRect', {
+        value: () => ({ left: 0, width: 1000 })
+    });
+    plot.dispatchEvent(
+        new PointerEvent('pointerdown', { clientX: 8, pointerType: 'touch', bubbles: true })
+    );
+    flushSync();
+    expect(plot.getAttribute('aria-valuenow')).toBe('0');
+    expect(document.querySelector('.readout')?.textContent).toContain('65 °C');
+});
+
+it('shows unsupported hosts as unavailable', () => {
+    const report = history();
+    report.temperatures[0].points = report.temperatures[0].points.map((point) => ({
+        ...point,
+        celsius: null
+    }));
+    show(report);
+    expect(document.body.textContent).toContain('No temperature readings in this period');
+    expect(document.body.textContent).toContain('Unavailable');
+});
+
+it('plots a genuine zero-degree reading', () => {
+    const report = history();
+    report.temperatures[0].points[2].celsius = 0;
+    show(report);
+    expect(document.querySelector('.readout')?.textContent).toContain('0 °C');
+    expect(document.querySelector('.empty-chart')).toBeNull();
+});
@@ -34,6 +34,101 @@ async function sample(id: string, at: number, cpu = 25, memory = 4 * GiB, cores
 }

 describe('capacity history', () => {
+    it('keeps separate CPU series, device drill-down, real zero and missing temperature gaps', async () => {
+        await host('one');
+        await host('two');
+        await host('unsupported');
+        await host('removed');
+        const { recordTemperatures } = await import('../../src/lib/server/hosts/temperatures.js');
+        const cpu = (celsius: number) => ({
+            id: 'package',
+            kind: 'cpu-package' as const,
+            device: 'coretemp',
+            label: 'Package id 0',
+            celsius
+        });
+        const drive = {
+            id: 'drive',
+            kind: 'nvme' as const,
+            device: 'nvme',
+            label: 'Composite',
+            celsius: 50.9
+        };
+        recordTemperatures(
+            'one',
+            { temperatures: [cpu(65), drive] },
+            new Date(epoch - 55_000).toISOString()
+        );
+        recordTemperatures(
+            'one',
+            { temperatures: [cpu(75)] },
+            new Date(epoch - 30_000).toISOString()
+        );
+        recordTemperatures(
+            'one',
+            { temperatures: [cpu(0)] },
+            new Date(epoch - 180_000).toISOString()
+        );
+        recordTemperatures(
+            'two',
+            { temperatures: [cpu(60)] },
+            new Date(epoch - 40_000).toISOString()
+        );
+        recordTemperatures(
+            'removed',
+            { temperatures: [cpu(99)] },
+            new Date(epoch - 40_000).toISOString()
+        );
+        const { db } = await import('../../src/lib/server/db/index.js');
+        db()
+            .prepare('UPDATE ec_hosts SET deleted_at=? WHERE id=?')
+            .run(new Date(epoch).toISOString(), 'removed');
+        const { readCapacityHistory } =
+            await import('../../src/lib/server/hosts/capacity-history.js');
+        const fleet = readCapacityHistory(query, epoch).temperatures;
+        expect(fleet.map((s) => s.label)).toEqual(['one', 'two', 'unsupported']);
+        expect(fleet[0].points.slice(-3).map((p) => p.celsius)).toEqual([0, null, 75]);
+        expect(fleet[1].points.at(-1)?.celsius).toBe(60);
+        expect(fleet[2].points.every((p) => p.celsius === null)).toBe(true);
+        const selected = readCapacityHistory(
+            { ...query, hostId: 'one', game: 't5' },
+            epoch
+        ).temperatures;
+        expect(selected.map((s) => s.kind)).toEqual(['cpu', 'nvme']);
+        expect(selected[1].points.at(-1)?.celsius).toBe(50.9);
+        expect(
+            readCapacityHistory({ ...query, hostId: 'one', game: 't6' }, epoch).temperatures
+        ).toEqual(selected);
+        expect(readCapacityHistory({ ...query, hostId: 'missing' }, epoch).temperatures).toEqual(
+            []
+        );
+    });
+
+    it('preserves temperature peaks for long ranges and includes a host enrolled during the current minute', async () => {
+        await host('new', epoch - 20_000);
+        const { recordTemperatures } = await import('../../src/lib/server/hosts/temperatures.js');
+        recordTemperatures(
+            'new',
+            {
+                temperatures: [
+                    {
+                        id: 'pkg',
+                        device: 'coretemp',
+                        kind: 'cpu-package',
+                        label: 'Package id 0',
+                        celsius: 65
+                    }
+                ]
+            },
+            new Date(epoch - 10_000).toISOString()
+        );
+        const { readCapacityHistory } =
+            await import('../../src/lib/server/hosts/capacity-history.js');
+        expect(readCapacityHistory(query, epoch).temperatures[0].points.at(-1)?.celsius).toBe(65);
+        const long = readCapacityHistory({ ...query, range: '7d' }, epoch).temperatures[0];
+        expect(long.points).toHaveLength(168);
+        expect(long.points.at(-1)?.celsius).toBe(65);
+    });
    it('averages each host before summing fleet headroom and exposes missing observations as gaps', async () => {
        await host('one');
        await host('two');
@@ -0,0 +1,151 @@
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+import { newFixture, type Fixture } from '../helpers/fixture.js';
+import {
+    cpuTemperature,
+    historyTemperatures,
+    validateTemperatures,
+    type HostTemperatureSensor
+} from '../../src/lib/host-temperatures.js';
+
+const sensor = (changes: Partial<HostTemperatureSensor> = {}): HostTemperatureSensor => ({
+    id: 'package',
+    device: 'coretemp',
+    label: 'Package id 0',
+    kind: 'cpu-package',
+    celsius: 65,
+    ...changes
+});
+
+describe('temperature readings', () => {
+    it('keeps valid sensors while dropping malformed, duplicate and out-of-range entries', () => {
+        const good = sensor();
+        const readings = [
+            good,
+            good,
+            null,
+            'bad',
+            ...[NaN, Infinity, -Infinity, null, '65', true, -101, 201].map((celsius) => ({
+                ...good,
+                id: 'invalid',
+                celsius
+            })),
+            { ...good, id: 'x'.repeat(101) },
+            { ...good, label: '' },
+            { ...good, kind: 'cpu-control' },
+            sensor({ id: 'zero', celsius: 0 })
+        ];
+        expect(validateTemperatures(readings)).toEqual([good, sensor({ id: 'zero', celsius: 0 })]);
+        expect(validateTemperatures({})).toEqual([]);
+        expect(validateTemperatures(undefined)).toEqual([]);
+    });
+    it('chooses the hottest package and uses die/core readings only as fallbacks', () => {
+        const packages = [sensor(), sensor({ id: 'socket2', label: 'Package id 1', celsius: 72 })];
+        const core = sensor({ id: 'core', label: 'Core 0', kind: 'cpu-core', celsius: 80 });
+        const die = sensor({ id: 'die', label: 'Tdie', kind: 'cpu-die', celsius: 64 });
+        expect(cpuTemperature([...packages, core])?.celsius).toBe(72);
+        expect(cpuTemperature([die, core])?.celsius).toBe(64);
+        expect(cpuTemperature([core])?.celsius).toBe(80);
+        expect(cpuTemperature()).toBeNull();
+        const drive = sensor({
+            id: 'drive',
+            kind: 'nvme',
+            device: 'nvme',
+            label: 'Composite',
+            celsius: 50.9
+        });
+        expect(
+            historyTemperatures([
+                ...packages,
+                core,
+                drive,
+                { ...drive, id: 'internal', label: 'Sensor 1' }
+            ])
+        ).toEqual([{ ...packages[1], id: 'cpu', label: 'CPU temperature' }, drive]);
+    });
+});
+
+describe('temperature persistence', () => {
+    let fixture: Fixture;
+    beforeEach(async () => {
+        vi.resetModules();
+        fixture = await newFixture([]);
+    });
+    afterEach(async () => {
+        vi.restoreAllMocks();
+        await fixture.cleanup();
+    });
+
+    it('accepts older agents and faulty temperature payloads without losing host metrics', async () => {
+        const { createHost, heartbeat, validateInventory } =
+            await import('../../src/lib/server/hosts/index.js');
+        const host = createHost({ name: 'ec-2' });
+        expect(validateInventory({ cpuUsagePercent: 25 }).temperatures).toBeUndefined();
+        expect(
+            heartbeat(host.id, {
+                cpuUsagePercent: 25,
+                temperatures: [sensor(), { celsius: 'bad' }]
+            }).inventory
+        ).toMatchObject({ cpuUsagePercent: 25, temperatures: [sensor()] });
+        expect(
+            heartbeat(host.id, { cpuUsagePercent: 30, temperatures: [] }).inventory.temperatures
+        ).toEqual([]);
+        expect(
+            heartbeat(host.id, { cpuUsagePercent: 35, temperatures: 'bad' }).inventory.temperatures
+        ).toEqual([]);
+        const { db } = await import('../../src/lib/server/db/index.js');
+        expect(
+            db()
+                .prepare('SELECT celsius FROM ec_host_temperature_samples WHERE host_id=?')
+                .all(host.id)
+        ).toEqual([{ celsius: 65 }]);
+    });
+
+    it('retains one peak per minute and applies host history retention, including offline hosts', async () => {
+        const { createHost } = await import('../../src/lib/server/hosts/index.js');
+        const { recordTemperatures } = await import('../../src/lib/server/hosts/temperatures.js');
+        const { db } = await import('../../src/lib/server/db/index.js');
+        const { fleetSettings, saveFleetSettings } =
+            await import('../../src/lib/server/settings.js');
+        const { pruneSampleHistory } = await import('../../src/lib/server/retention.js');
+        const host = createHost({ name: 'ec-2' });
+        const minute = Math.floor((Date.now() - 3 * 86400_000) / 60_000) * 60_000;
+        for (const [i, celsius] of [65, 78, 67].entries())
+            recordTemperatures(
+                host.id,
+                { temperatures: [sensor({ celsius })] },
+                new Date(minute + i * 15_000).toISOString()
+            );
+        expect(db().prepare('SELECT celsius FROM ec_host_temperature_samples').all()).toEqual([
+            { celsius: 78 }
+        ]);
+        saveFleetSettings({ ...fleetSettings(), hostResourceKeepDays: 0 });
+        pruneSampleHistory();
+        expect(db().prepare('SELECT COUNT(*) AS n FROM ec_host_temperature_samples').get()).toEqual(
+            { n: 1 }
+        );
+        recordTemperatures(host.id, { temperatures: [sensor()] }, new Date().toISOString());
+        saveFleetSettings({ ...fleetSettings(), hostResourceKeepDays: 1 });
+        pruneSampleHistory();
+        expect(db().prepare('SELECT celsius FROM ec_host_temperature_samples').all()).toEqual([
+            { celsius: 65 }
+        ]);
+    });
+
+    it('upgrades an existing database without rewriting its resource samples', async () => {
+        const { db, closeDb } = await import('../../src/lib/server/db/index.js');
+        const { createHost } = await import('../../src/lib/server/hosts/index.js');
+        const host = createHost({ name: 'old' });
+        db()
+            .prepare(
+                'INSERT INTO ec_host_resource_samples(host_id,sampled_at,cpu_usage_percent) VALUES(?,?,?)'
+            )
+            .run(host.id, '2026-10-13T12:00:00Z', 25);
+        db().exec('DROP TABLE ec_host_temperature_samples; PRAGMA user_version=22;');
+        closeDb();
+        expect(db().prepare('SELECT * FROM ec_host_temperature_samples').all()).toEqual([]);
+        expect(
+            db().prepare('SELECT cpu_usage_percent FROM ec_host_resource_samples').all()
+        ).toEqual([{ cpu_usage_percent: 25 }]);
+        expect(db().pragma('foreign_key_check')).toEqual([]);
+    });
+});
@@ -43,6 +43,7 @@ const host = (changes: Partial<CapacityHost> = {}): CapacityHost => ({
    ...changes
 });
 const history: CapacityHistory = {
+    temperatures: [],
    range: '24h',
    startAt: 1000,
    endAt: 5000,
@@ -112,6 +113,10 @@ it('keeps current estimates and host comparison before history with scoped histo
    expect(body).not.toContain('Manage hosts');
    expect(body).toContain('Updates every 30 seconds');
    expect(body).not.toContain('>Refresh<');
+    expect(body.indexOf('id="temperature-history-title"')).toBeGreaterThan(
+        body.indexOf('id="capacity-history-title"')
+    );
+    expect(body.match(/aria-label="History range"/g)).toHaveLength(2);
 });

 it('keeps fleet-wide queue scope explicit and preserves unavailable estimates', () => {
@@ -9,6 +9,7 @@ import {
    stableHostOrder
 } from '../../src/lib/ui/host-overview.js';
 import HostList from '../../src/lib/ui/components/HostList.svelte';
+import HostTemperatures from '../../src/lib/ui/components/HostTemperatures.svelte';

 const now = Date.parse('2026-10-06T12:00:00Z');
 function host(id = 'ec-1', changes: Partial<Host> = {}): Host {
@@ -43,6 +44,40 @@ function host(id = 'ec-1', changes: Partial<Host> = {}): Host {
    };
 }

+it('shows CPU temperature with stale status and keeps sensor details expandable', () => {
+    const target = host('ec-2', {
+        connected: false,
+        inventory: {
+            cpuUsagePercent: 25,
+            temperatures: [
+                {
+                    id: 'cpu',
+                    kind: 'cpu-package',
+                    device: 'coretemp',
+                    label: 'Package id 0',
+                    celsius: 65
+                }
+            ]
+        }
+    });
+    const list = render(HostList, {
+        props: {
+            hosts: [{ ...target, serverCount: 0, playerCount: 0, unreportedInstances: 0 }],
+            now
+        }
+    }).body;
+    expect(list).toContain('CPU temp: 65 °C');
+    expect(list).toContain('Resource readings stale');
+    const detail = render(HostTemperatures, { props: { host: target } }).body;
+    expect(detail).toContain('(last reported)');
+    expect(detail).toContain('Temperature sensors');
+    expect(detail).toContain('CPU package 0');
+    expect(detail).not.toMatch(/<details[^>]*open/);
+    const missing = render(HostTemperatures, { props: { host: host() } }).body;
+    expect(missing).toContain('Unavailable');
+    expect(missing).not.toContain('0 °C');
+});
+
 it('does not show an old Ready state as available when the heartbeat becomes stale', () => {
    expect(hostListStatus(host(), now).label).toBe('Ready');
    expect(hostReadingsStale(host(), now + 90_001)).toBe(true);