add one minute chart + refactor rpc

- add one minute charts
- update disk io to use bytes
- update hub and agent connection interfaces / handlers to be more
flexible
- change agent cache to use cache time instead of session id
- refactor collection of metrics which require deltas to track
separately per cache time

agent/agent.go

@@ -12,33 +12,36 @@ import (
 	"path/filepath"
 	"strings"
 	"sync"
-	"time"
 
 	"github.com/gliderlabs/ssh"
 	"github.com/henrygd/beszel"
+	"github.com/henrygd/beszel/agent/deltatracker"
 	"github.com/henrygd/beszel/internal/entities/system"
 	"github.com/shirou/gopsutil/v4/host"
 	gossh "golang.org/x/crypto/ssh"
 )
 
 type Agent struct {
-	sync.Mutex                                   // Used to lock agent while collecting data
-	debug             bool                       // true if LOG_LEVEL is set to debug
-	zfs               bool                       // true if system has arcstats
-	memCalc           string                     // Memory calculation formula
-	fsNames           []string                   // List of filesystem device names being monitored
-	fsStats           map[string]*system.FsStats // Keeps track of disk stats for each filesystem
-	netInterfaces     map[string]struct{}        // Stores all valid network interfaces
-	netIoStats        system.NetIoStats          // Keeps track of bandwidth usage
-	dockerManager     *dockerManager             // Manages Docker API requests
-	sensorConfig      *SensorConfig              // Sensors config
-	systemInfo        system.Info                // Host system info
-	gpuManager        *GPUManager                // Manages GPU data
-	cache             *SessionCache              // Cache for system stats based on primary session ID
-	connectionManager *ConnectionManager         // Channel to signal connection events
-	server            *ssh.Server                // SSH server
-	dataDir           string                     // Directory for persisting data
-	keys              []gossh.PublicKey          // SSH public keys
+	sync.Mutex                                                                      // Used to lock agent while collecting data
+	debug                     bool                                                  // true if LOG_LEVEL is set to debug
+	zfs                       bool                                                  // true if system has arcstats
+	memCalc                   string                                                // Memory calculation formula
+	fsNames                   []string                                              // List of filesystem device names being monitored
+	fsStats                   map[string]*system.FsStats                            // Keeps track of disk stats for each filesystem
+	diskPrev                  map[uint16]map[string]prevDisk                        // Previous disk I/O counters per cache interval
+	netInterfaces             map[string]struct{}                                   // Stores all valid network interfaces
+	netIoStats                map[uint16]system.NetIoStats                          // Keeps track of bandwidth usage per cache interval
+	netInterfaceDeltaTrackers map[uint16]*deltatracker.DeltaTracker[string, uint64] // Per-cache-time NIC delta trackers
+	dockerManager             *dockerManager                                        // Manages Docker API requests
+	sensorConfig              *SensorConfig                                         // Sensors config
+	systemInfo                system.Info                                           // Host system info
+	gpuManager                *GPUManager                                           // Manages GPU data
+	cache                     *systemDataCache                                      // Cache for system stats based on cache time
+	connectionManager         *ConnectionManager                                    // Channel to signal connection events
+	handlerRegistry           *HandlerRegistry                                      // Registry for routing incoming messages
+	server                    *ssh.Server                                           // SSH server
+	dataDir                   string                                                // Directory for persisting data
+	keys                      []gossh.PublicKey                                     // SSH public keys
 }
 
 // NewAgent creates a new agent with the given data directory for persisting data.
@@ -46,9 +49,15 @@ type Agent struct {
 func NewAgent(dataDir ...string) (agent *Agent, err error) {
 	agent = &Agent{
 		fsStats: make(map[string]*system.FsStats),
-		cache:   NewSessionCache(69 * time.Second),
+		cache:   NewSystemDataCache(),
 	}
 
+	// Initialize disk I/O previous counters storage
+	agent.diskPrev = make(map[uint16]map[string]prevDisk)
+	// Initialize per-cache-time network tracking structures
+	agent.netIoStats = make(map[uint16]system.NetIoStats)
+	agent.netInterfaceDeltaTrackers = make(map[uint16]*deltatracker.DeltaTracker[string, uint64])
+
 	agent.dataDir, err = getDataDir(dataDir...)
 	if err != nil {
 		slog.Warn("Data directory not found")
@@ -79,6 +88,9 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 	// initialize connection manager
 	agent.connectionManager = newConnectionManager(agent)
 
+	// initialize handler registry
+	agent.handlerRegistry = NewHandlerRegistry()
+
 	// initialize disk info
 	agent.initializeDiskInfo()
 
@@ -97,7 +109,7 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 
 	// if debugging, print stats
 	if agent.debug {
-		slog.Debug("Stats", "data", agent.gatherStats(""))
+		slog.Debug("Stats", "data", agent.gatherStats(0))
 	}
 
 	return agent, nil
@@ -112,24 +124,24 @@ func GetEnv(key string) (value string, exists bool) {
 	return os.LookupEnv(key)
 }
 
-func (a *Agent) gatherStats(sessionID string) *system.CombinedData {
+func (a *Agent) gatherStats(cacheTimeMs uint16) *system.CombinedData {
 	a.Lock()
 	defer a.Unlock()
 
-	data, isCached := a.cache.Get(sessionID)
+	data, isCached := a.cache.Get(cacheTimeMs)
 	if isCached {
-		slog.Debug("Cached data", "session", sessionID)
+		slog.Debug("Cached data", "cacheTimeMs", cacheTimeMs)
 		return data
 	}
 
 	*data = system.CombinedData{
-		Stats: a.getSystemStats(),
+		Stats: a.getSystemStats(cacheTimeMs),
 		Info:  a.systemInfo,
 	}
-	slog.Debug("System data", "data", data)
+	// slog.Info("System data", "data", data, "cacheTimeMs", cacheTimeMs)
 
 	if a.dockerManager != nil {
-		if containerStats, err := a.dockerManager.getDockerStats(); err == nil {
+		if containerStats, err := a.dockerManager.getDockerStats(cacheTimeMs); err == nil {
 			data.Containers = containerStats
 			slog.Debug("Containers", "data", data.Containers)
 		} else {
@@ -145,7 +157,7 @@ func (a *Agent) gatherStats(sessionID string) *system.CombinedData {
 	}
 	slog.Debug("Extra FS", "data", data.Stats.ExtraFs)
 
-	a.cache.Set(sessionID, data)
+	a.cache.Set(data, cacheTimeMs)
 	return data
 }
 

agent/agent_cache.go

@@ -1,37 +1,55 @@
 package agent
 
 import (
+	"sync"
 	"time"
 
 	"github.com/henrygd/beszel/internal/entities/system"
 )
 
-// Not thread safe since we only access from gatherStats which is already locked
-type SessionCache struct {
-	data           *system.CombinedData
-	lastUpdate     time.Time
-	primarySession string
-	leaseTime      time.Duration
+type systemDataCache struct {
+	sync.RWMutex
+	cache map[uint16]*cacheNode
 }
 
-func NewSessionCache(leaseTime time.Duration) *SessionCache {
-	return &SessionCache{
-		leaseTime: leaseTime,
-		data:      &system.CombinedData{},
+type cacheNode struct {
+	data       *system.CombinedData
+	lastUpdate time.Time
+}
+
+// NewSystemDataCache creates a cache keyed by the polling interval in milliseconds.
+func NewSystemDataCache() *systemDataCache {
+	return &systemDataCache{
+		cache: make(map[uint16]*cacheNode),
 	}
 }
 
-func (c *SessionCache) Get(sessionID string) (stats *system.CombinedData, isCached bool) {
-	if sessionID != c.primarySession && time.Since(c.lastUpdate) < c.leaseTime {
-		return c.data, true
+// Get returns cached combined data when the entry is still considered fresh.
+func (c *systemDataCache) Get(cacheTimeMs uint16) (stats *system.CombinedData, isCached bool) {
+	c.RLock()
+	defer c.RUnlock()
+
+	node, ok := c.cache[cacheTimeMs]
+	if !ok {
+		return &system.CombinedData{}, false
 	}
-	return c.data, false
+	// allowedSkew := time.Second
+	// isFresh := time.Since(node.lastUpdate) < time.Duration(cacheTimeMs)*time.Millisecond-allowedSkew
+	// allow a 50% skew of the cache time
+	isFresh := time.Since(node.lastUpdate) < time.Duration(cacheTimeMs/2)*time.Millisecond
+	return node.data, isFresh
 }
 
-func (c *SessionCache) Set(sessionID string, data *system.CombinedData) {
-	if data != nil {
-		*c.data = *data
+// Set stores the latest combined data snapshot for the given interval.
+func (c *systemDataCache) Set(data *system.CombinedData, cacheTimeMs uint16) {
+	c.Lock()
+	defer c.Unlock()
+
+	node, ok := c.cache[cacheTimeMs]
+	if !ok {
+		node = &cacheNode{}
+		c.cache[cacheTimeMs] = node
 	}
-	c.primarySession = sessionID
-	c.lastUpdate = time.Now()
+	node.data = data
+	node.lastUpdate = time.Now()
 }

agent/agent_cache_test.go

@@ -8,82 +8,239 @@ import (
 	"testing/synctest"
 	"time"
 
+	"github.com/henrygd/beszel/internal/entities/container"
 	"github.com/henrygd/beszel/internal/entities/system"
-
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
 
-func TestSessionCache_GetSet(t *testing.T) {
-	synctest.Test(t, func(t *testing.T) {
-		cache := NewSessionCache(69 * time.Second)
+func createTestCacheData() *system.CombinedData {
+	return &system.CombinedData{
+		Stats: system.Stats{
+			Cpu:       50.5,
+			Mem:       8192,
+			DiskTotal: 100000,
+		},
+		Info: system.Info{
+			Hostname: "test-host",
+		},
+		Containers: []*container.Stats{
+			{
+				Name: "test-container",
+				Cpu:  25.0,
+			},
+		},
+	}
+}
 
-		testData := &system.CombinedData{
-			Info: system.Info{
-				Hostname: "test-host",
-				Cores:    4,
-			},
+func TestNewSystemDataCache(t *testing.T) {
+	cache := NewSystemDataCache()
+	require.NotNil(t, cache)
+	assert.NotNil(t, cache.cache)
+	assert.Empty(t, cache.cache)
+}
+
+func TestCacheGetSet(t *testing.T) {
+	cache := NewSystemDataCache()
+	data := createTestCacheData()
+
+	// Test setting data
+	cache.Set(data, 1000) // 1 second cache
+
+	// Test getting fresh data
+	retrieved, isCached := cache.Get(1000)
+	assert.True(t, isCached)
+	assert.Equal(t, data, retrieved)
+
+	// Test getting non-existent cache key
+	_, isCached = cache.Get(2000)
+	assert.False(t, isCached)
+}
+
+func TestCacheFreshness(t *testing.T) {
+	cache := NewSystemDataCache()
+	data := createTestCacheData()
+
+	testCases := []struct {
+		name        string
+		cacheTimeMs uint16
+		sleepMs     time.Duration
+		expectFresh bool
+	}{
+		{
+			name:        "fresh data - well within cache time",
+			cacheTimeMs: 1000, // 1 second
+			sleepMs:     100,  // 100ms
+			expectFresh: true,
+		},
+		{
+			name:        "fresh data - at 50% of cache time boundary",
+			cacheTimeMs: 1000, // 1 second, 50% = 500ms
+			sleepMs:     499,  // just under 500ms
+			expectFresh: true,
+		},
+		{
+			name:        "stale data - exactly at 50% cache time",
+			cacheTimeMs: 1000, // 1 second, 50% = 500ms
+			sleepMs:     500,  // exactly 500ms
+			expectFresh: false,
+		},
+		{
+			name:        "stale data - well beyond cache time",
+			cacheTimeMs: 1000, // 1 second
+			sleepMs:     800,  // 800ms
+			expectFresh: false,
+		},
+		{
+			name:        "short cache time",
+			cacheTimeMs: 200, // 200ms, 50% = 100ms
+			sleepMs:     150, // 150ms > 100ms
+			expectFresh: false,
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				// Set data
+				cache.Set(data, tc.cacheTimeMs)
+
+				// Wait for the specified duration
+				if tc.sleepMs > 0 {
+					time.Sleep(tc.sleepMs * time.Millisecond)
+				}
+
+				// Check freshness
+				_, isCached := cache.Get(tc.cacheTimeMs)
+				assert.Equal(t, tc.expectFresh, isCached)
+			})
+		})
+	}
+}
+
+func TestCacheMultipleIntervals(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		cache := NewSystemDataCache()
+		data1 := createTestCacheData()
+		data2 := &system.CombinedData{
 			Stats: system.Stats{
-				Cpu:     50.0,
-				MemPct:  30.0,
-				DiskPct: 40.0,
+				Cpu: 75.0,
+				Mem: 16384,
 			},
+			Info: system.Info{
+				Hostname: "test-host-2",
+			},
+			Containers: []*container.Stats{},
 		}
 
-		// Test initial state - should not be cached
-		data, isCached := cache.Get("session1")
-		assert.False(t, isCached, "Expected no cached data initially")
-		assert.NotNil(t, data, "Expected data to be initialized")
-		// Set data for session1
-		cache.Set("session1", testData)
+		// Set data for different intervals
+		cache.Set(data1, 500)  // 500ms cache
+		cache.Set(data2, 1000) // 1000ms cache
 
-		time.Sleep(15 * time.Second)
+		// Both should be fresh immediately
+		retrieved1, isCached1 := cache.Get(500)
+		assert.True(t, isCached1)
+		assert.Equal(t, data1, retrieved1)
 
-		// Get data for a different session - should be cached
-		data, isCached = cache.Get("session2")
-		assert.True(t, isCached, "Expected data to be cached for non-primary session")
-		require.NotNil(t, data, "Expected cached data to be returned")
-		assert.Equal(t, "test-host", data.Info.Hostname, "Hostname should match test data")
-		assert.Equal(t, 4, data.Info.Cores, "Cores should match test data")
-		assert.Equal(t, 50.0, data.Stats.Cpu, "CPU should match test data")
-		assert.Equal(t, 30.0, data.Stats.MemPct, "Memory percentage should match test data")
-		assert.Equal(t, 40.0, data.Stats.DiskPct, "Disk percentage should match test data")
+		retrieved2, isCached2 := cache.Get(1000)
+		assert.True(t, isCached2)
+		assert.Equal(t, data2, retrieved2)
 
-		time.Sleep(10 * time.Second)
+		// Wait 300ms - 500ms cache should be stale (250ms threshold), 1000ms should still be fresh (500ms threshold)
+		time.Sleep(300 * time.Millisecond)
 
-		// Get data for the primary session - should not be cached
-		data, isCached = cache.Get("session1")
-		assert.False(t, isCached, "Expected data not to be cached for primary session")
-		require.NotNil(t, data, "Expected data to be returned even if not cached")
-		assert.Equal(t, "test-host", data.Info.Hostname, "Hostname should match test data")
-		// if not cached, agent will update the data
-		cache.Set("session1", testData)
+		_, isCached1 = cache.Get(500)
+		assert.False(t, isCached1)
 
-		time.Sleep(45 * time.Second)
+		_, isCached2 = cache.Get(1000)
+		assert.True(t, isCached2)
 
-		// Get data for a different session - should still be cached
-		_, isCached = cache.Get("session2")
-		assert.True(t, isCached, "Expected data to be cached for non-primary session")
-
-		// Wait for the lease to expire
-		time.Sleep(30 * time.Second)
-
-		// Get data for session2 - should not be cached
-		_, isCached = cache.Get("session2")
-		assert.False(t, isCached, "Expected data not to be cached after lease expiration")
+		// Wait another 300ms (total 600ms) - now 1000ms cache should also be stale
+		time.Sleep(300 * time.Millisecond)
+		_, isCached2 = cache.Get(1000)
+		assert.False(t, isCached2)
 	})
 }
 
-func TestSessionCache_NilData(t *testing.T) {
-	// Create a new SessionCache
-	cache := NewSessionCache(30 * time.Second)
+func TestCacheOverwrite(t *testing.T) {
+	cache := NewSystemDataCache()
+	data1 := createTestCacheData()
+	data2 := &system.CombinedData{
+		Stats: system.Stats{
+			Cpu: 90.0,
+			Mem: 32768,
+		},
+		Info: system.Info{
+			Hostname: "updated-host",
+		},
+		Containers: []*container.Stats{},
+	}
 
-	// Test setting nil data (should not panic)
-	assert.NotPanics(t, func() {
-		cache.Set("session1", nil)
-	}, "Setting nil data should not panic")
+	// Set initial data
+	cache.Set(data1, 1000)
+	retrieved, isCached := cache.Get(1000)
+	assert.True(t, isCached)
+	assert.Equal(t, data1, retrieved)
 
-	// Get data - should not be nil even though we set nil
-	data, _ := cache.Get("session2")
-	assert.NotNil(t, data, "Expected data to not be nil after setting nil data")
+	// Overwrite with new data
+	cache.Set(data2, 1000)
+	retrieved, isCached = cache.Get(1000)
+	assert.True(t, isCached)
+	assert.Equal(t, data2, retrieved)
+	assert.NotEqual(t, data1, retrieved)
+}
+
+func TestCacheMiss(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		cache := NewSystemDataCache()
+
+		// Test getting from empty cache
+		_, isCached := cache.Get(1000)
+		assert.False(t, isCached)
+
+		// Set data for one interval
+		data := createTestCacheData()
+		cache.Set(data, 1000)
+
+		// Test getting different interval
+		_, isCached = cache.Get(2000)
+		assert.False(t, isCached)
+
+		// Test getting after data has expired
+		time.Sleep(600 * time.Millisecond) // 600ms > 500ms (50% of 1000ms)
+		_, isCached = cache.Get(1000)
+		assert.False(t, isCached)
+	})
+}
+
+func TestCacheZeroInterval(t *testing.T) {
+	cache := NewSystemDataCache()
+	data := createTestCacheData()
+
+	// Set with zero interval - should allow immediate cache
+	cache.Set(data, 0)
+
+	// With 0 interval, 50% is 0, so it should never be considered fresh
+	// (time.Since(lastUpdate) >= 0, which is not < 0)
+	_, isCached := cache.Get(0)
+	assert.False(t, isCached)
+}
+
+func TestCacheLargeInterval(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		cache := NewSystemDataCache()
+		data := createTestCacheData()
+
+		// Test with maximum uint16 value
+		cache.Set(data, 65535) // ~65 seconds
+
+		// Should be fresh immediately
+		_, isCached := cache.Get(65535)
+		assert.True(t, isCached)
+
+		// Should still be fresh after a short time
+		time.Sleep(100 * time.Millisecond)
+		_, isCached = cache.Get(65535)
+		assert.True(t, isCached)
+	})
 }

agent/client.go

@@ -15,6 +15,7 @@ import (
 
 	"github.com/henrygd/beszel"
 	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/entities/system"
 
 	"github.com/fxamacker/cbor/v2"
 	"github.com/lxzan/gws"
@@ -156,11 +157,15 @@ func (client *WebSocketClient) OnMessage(conn *gws.Conn, message *gws.Message) {
 		return
 	}
 
-	if err := cbor.NewDecoder(message.Data).Decode(client.hubRequest); err != nil {
+	var HubRequest common.HubRequest[cbor.RawMessage]
+
+	err := cbor.Unmarshal(message.Data.Bytes(), &HubRequest)
+	if err != nil {
 		slog.Error("Error parsing message", "err", err)
 		return
 	}
-	if err := client.handleHubRequest(client.hubRequest); err != nil {
+
+	if err := client.handleHubRequest(&HubRequest, HubRequest.Id); err != nil {
 		slog.Error("Error handling message", "err", err)
 	}
 }
@@ -173,7 +178,7 @@ func (client *WebSocketClient) OnPing(conn *gws.Conn, message []byte) {
 }
 
 // handleAuthChallenge verifies the authenticity of the hub and returns the system's fingerprint.
-func (client *WebSocketClient) handleAuthChallenge(msg *common.HubRequest[cbor.RawMessage]) (err error) {
+func (client *WebSocketClient) handleAuthChallenge(msg *common.HubRequest[cbor.RawMessage], requestID *uint32) (err error) {
 	var authRequest common.FingerprintRequest
 	if err := cbor.Unmarshal(msg.Data, &authRequest); err != nil {
 		return err
@@ -196,7 +201,7 @@ func (client *WebSocketClient) handleAuthChallenge(msg *common.HubRequest[cbor.R
 		_, response.Port, _ = net.SplitHostPort(serverAddr)
 	}
 
-	return client.sendMessage(response)
+	return client.sendResponse(response, requestID)
 }
 
 // verifySignature verifies the signature of the token using the public keys.
@@ -221,25 +226,17 @@ func (client *WebSocketClient) Close() {
 	}
 }
 
-// handleHubRequest routes the request to the appropriate handler.
-// It ensures the hub is verified before processing most requests.
-func (client *WebSocketClient) handleHubRequest(msg *common.HubRequest[cbor.RawMessage]) error {
-	if !client.hubVerified && msg.Action != common.CheckFingerprint {
-		return errors.New("hub not verified")
+// handleHubRequest routes the request to the appropriate handler using the handler registry.
+func (client *WebSocketClient) handleHubRequest(msg *common.HubRequest[cbor.RawMessage], requestID *uint32) error {
+	ctx := &HandlerContext{
+		Client:       client,
+		Agent:        client.agent,
+		Request:      msg,
+		RequestID:    requestID,
+		HubVerified:  client.hubVerified,
+		SendResponse: client.sendResponse,
 	}
-	switch msg.Action {
-	case common.GetData:
-		return client.sendSystemData()
-	case common.CheckFingerprint:
-		return client.handleAuthChallenge(msg)
-	}
-	return nil
-}
-
-// sendSystemData gathers and sends current system statistics to the hub.
-func (client *WebSocketClient) sendSystemData() error {
-	sysStats := client.agent.gatherStats(client.token)
-	return client.sendMessage(sysStats)
+	return client.agent.handlerRegistry.Handle(ctx)
 }
 
 // sendMessage encodes the given data to CBOR and sends it as a binary message over the WebSocket connection to the hub.
@@ -251,6 +248,36 @@ func (client *WebSocketClient) sendMessage(data any) error {
 	return client.Conn.WriteMessage(gws.OpcodeBinary, bytes)
 }
 
+// sendResponse sends a response with optional request ID for the new protocol
+func (client *WebSocketClient) sendResponse(data any, requestID *uint32) error {
+	if requestID != nil {
+		// New format with ID - use typed fields
+		response := common.AgentResponse{
+			Id: requestID,
+		}
+
+		// Set the appropriate typed field based on data type
+		switch v := data.(type) {
+		case *system.CombinedData:
+			response.SystemData = v
+		case *common.FingerprintResponse:
+			response.Fingerprint = v
+		// case []byte:
+		// 	response.RawBytes = v
+		// case string:
+		// 	response.RawBytes = []byte(v)
+		default:
+			// For any other type, convert to error
+			response.Error = fmt.Sprintf("unsupported response type: %T", data)
+		}
+
+		return client.sendMessage(response)
+	} else {
+		// Legacy format - send data directly
+		return client.sendMessage(data)
+	}
+}
+
 // getUserAgent returns one of two User-Agent strings based on current time.
 // This is used to avoid being blocked by Cloudflare or other anti-bot measures.
 func getUserAgent() string {

agent/client_test.go

@@ -301,7 +301,7 @@ func TestWebSocketClient_HandleHubRequest(t *testing.T) {
 				Data:   cbor.RawMessage{},
 			}
 
-			err := client.handleHubRequest(hubRequest)
+			err := client.handleHubRequest(hubRequest, nil)
 
 			if tc.expectError {
 				assert.Error(t, err)

agent/cpu.go

@@ -0,0 +1,66 @@
+package agent
+
+import (
+	"math"
+	"runtime"
+
+	"github.com/shirou/gopsutil/v4/cpu"
+)
+
+var lastCpuTimes = make(map[uint16]cpu.TimesStat)
+
+// init initializes the CPU monitoring by storing the initial CPU times
+// for the default 60-second cache interval.
+func init() {
+	if times, err := cpu.Times(false); err == nil {
+		lastCpuTimes[60000] = times[0]
+	}
+}
+
+// getCpuPercent calculates the CPU usage percentage using cached previous measurements.
+// It uses the specified cache time interval to determine the time window for calculation.
+// Returns the CPU usage percentage (0-100) and any error encountered.
+func getCpuPercent(cacheTimeMs uint16) (float64, error) {
+	times, err := cpu.Times(false)
+	if err != nil || len(times) == 0 {
+		return 0, err
+	}
+	// if cacheTimeMs is not in lastCpuTimes, use 60000 as fallback lastCpuTime
+	if _, ok := lastCpuTimes[cacheTimeMs]; !ok {
+		lastCpuTimes[cacheTimeMs] = lastCpuTimes[60000]
+	}
+	delta := calculateBusy(lastCpuTimes[cacheTimeMs], times[0])
+	lastCpuTimes[cacheTimeMs] = times[0]
+	return delta, nil
+}
+
+// calculateBusy calculates the CPU busy percentage between two time points.
+// It computes the ratio of busy time to total time elapsed between t1 and t2,
+// returning a percentage clamped between 0 and 100.
+func calculateBusy(t1, t2 cpu.TimesStat) float64 {
+	t1All, t1Busy := getAllBusy(t1)
+	t2All, t2Busy := getAllBusy(t2)
+
+	if t2Busy <= t1Busy {
+		return 0
+	}
+	if t2All <= t1All {
+		return 100
+	}
+	return math.Min(100, math.Max(0, (t2Busy-t1Busy)/(t2All-t1All)*100))
+}
+
+// getAllBusy calculates the total CPU time and busy CPU time from CPU times statistics.
+// On Linux, it excludes guest and guest_nice time from the total to match kernel behavior.
+// Returns total CPU time and busy CPU time (total minus idle and I/O wait time).
+func getAllBusy(t cpu.TimesStat) (float64, float64) {
+	tot := t.Total()
+	if runtime.GOOS == "linux" {
+		tot -= t.Guest     // Linux 2.6.24+
+		tot -= t.GuestNice // Linux 3.2.0+
+	}
+
+	busy := tot - t.Idle - t.Iowait
+
+	return tot, busy
+}

agent/disk.go

@@ -189,3 +189,96 @@ func (a *Agent) initializeDiskIoStats(diskIoCounters map[string]disk.IOCountersS
 		a.fsNames = append(a.fsNames, device)
 	}
 }
+
+// Updates disk usage statistics for all monitored filesystems
+func (a *Agent) updateDiskUsage(systemStats *system.Stats) {
+	// disk usage
+	for _, stats := range a.fsStats {
+		if d, err := disk.Usage(stats.Mountpoint); err == nil {
+			stats.DiskTotal = bytesToGigabytes(d.Total)
+			stats.DiskUsed = bytesToGigabytes(d.Used)
+			if stats.Root {
+				systemStats.DiskTotal = bytesToGigabytes(d.Total)
+				systemStats.DiskUsed = bytesToGigabytes(d.Used)
+				systemStats.DiskPct = twoDecimals(d.UsedPercent)
+			}
+		} else {
+			// reset stats if error (likely unmounted)
+			slog.Error("Error getting disk stats", "name", stats.Mountpoint, "err", err)
+			stats.DiskTotal = 0
+			stats.DiskUsed = 0
+			stats.TotalRead = 0
+			stats.TotalWrite = 0
+		}
+	}
+}
+
+// Updates disk I/O statistics for all monitored filesystems
+func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
+	// disk i/o (cache-aware per interval)
+	if ioCounters, err := disk.IOCounters(a.fsNames...); err == nil {
+		// Ensure map for this interval exists
+		if _, ok := a.diskPrev[cacheTimeMs]; !ok {
+			a.diskPrev[cacheTimeMs] = make(map[string]prevDisk)
+		}
+		now := time.Now()
+		for name, d := range ioCounters {
+			stats := a.fsStats[d.Name]
+			if stats == nil {
+				// skip devices not tracked
+				continue
+			}
+
+			// Previous snapshot for this interval and device
+			prev, hasPrev := a.diskPrev[cacheTimeMs][name]
+			if !hasPrev {
+				// Seed from agent-level fsStats if present, else seed from current
+				prev = prevDisk{readBytes: stats.TotalRead, writeBytes: stats.TotalWrite, at: stats.Time}
+				if prev.at.IsZero() {
+					prev = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
+				}
+			}
+
+			msElapsed := uint64(now.Sub(prev.at).Milliseconds())
+			if msElapsed < 100 {
+				// Avoid division by zero or clock issues; update snapshot and continue
+				a.diskPrev[cacheTimeMs][name] = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
+				continue
+			}
+
+			diskIORead := (d.ReadBytes - prev.readBytes) * 1000 / msElapsed
+			diskIOWrite := (d.WriteBytes - prev.writeBytes) * 1000 / msElapsed
+			readMbPerSecond := bytesToMegabytes(float64(diskIORead))
+			writeMbPerSecond := bytesToMegabytes(float64(diskIOWrite))
+
+			// validate values
+			if readMbPerSecond > 50_000 || writeMbPerSecond > 50_000 {
+				slog.Warn("Invalid disk I/O. Resetting.", "name", d.Name, "read", readMbPerSecond, "write", writeMbPerSecond)
+				// Reset interval snapshot and seed from current
+				a.diskPrev[cacheTimeMs][name] = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
+				// also refresh agent baseline to avoid future negatives
+				a.initializeDiskIoStats(ioCounters)
+				continue
+			}
+
+			// Update per-interval snapshot
+			a.diskPrev[cacheTimeMs][name] = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
+
+			// Update global fsStats baseline for cross-interval correctness
+			stats.Time = now
+			stats.TotalRead = d.ReadBytes
+			stats.TotalWrite = d.WriteBytes
+			stats.DiskReadPs = readMbPerSecond
+			stats.DiskWritePs = writeMbPerSecond
+			stats.DiskReadBytes = diskIORead
+			stats.DiskWriteBytes = diskIOWrite
+
+			if stats.Root {
+				systemStats.DiskReadPs = stats.DiskReadPs
+				systemStats.DiskWritePs = stats.DiskWritePs
+				systemStats.DiskIO[0] = diskIORead
+				systemStats.DiskIO[1] = diskIOWrite
+			}
+		}
+	}
+}

agent/docker.go

@@ -14,17 +14,25 @@ import (
 	"sync"
 	"time"
 
+	"github.com/henrygd/beszel/agent/deltatracker"
 	"github.com/henrygd/beszel/internal/entities/container"
 
 	"github.com/blang/semver"
 )
 
+const (
+	// Docker API timeout in milliseconds
+	dockerTimeoutMs = 2100
+	// Maximum realistic network speed (5 GB/s) to detect bad deltas
+	maxNetworkSpeedBps uint64 = 5e9
+)
+
 type dockerManager struct {
 	client              *http.Client                // Client to query Docker API
 	wg                  sync.WaitGroup              // WaitGroup to wait for all goroutines to finish
 	sem                 chan struct{}               // Semaphore to limit concurrent container requests
 	containerStatsMutex sync.RWMutex                // Mutex to prevent concurrent access to containerStatsMap
-	apiContainerList    []*container.ApiInfo        // List of containers from Docker API (no pointer)
+	apiContainerList    []*container.ApiInfo        // List of containers from Docker API
 	containerStatsMap   map[string]*container.Stats // Keeps track of container stats
 	validIds            map[string]struct{}         // Map of valid container ids, used to prune invalid containers from containerStatsMap
 	goodDockerVersion   bool                        // Whether docker version is at least 25.0.0 (one-shot works correctly)
@@ -32,6 +40,17 @@ type dockerManager struct {
 	buf                 *bytes.Buffer               // Buffer to store and read response bodies
 	decoder             *json.Decoder               // Reusable JSON decoder that reads from buf
 	apiStats            *container.ApiStats         // Reusable API stats object
+
+	// Cache-time-aware tracking for CPU stats (similar to cpu.go)
+	// Maps cache time intervals to container-specific CPU usage tracking
+	lastCpuContainer map[uint16]map[string]uint64    // cacheTimeMs -> containerId -> last cpu container usage
+	lastCpuSystem    map[uint16]map[string]uint64    // cacheTimeMs -> containerId -> last cpu system usage
+	lastCpuReadTime  map[uint16]map[string]time.Time // cacheTimeMs -> containerId -> last read time (Windows)
+
+	// Network delta trackers - one per cache time to avoid interference
+	// cacheTimeMs -> DeltaTracker for network bytes sent/received
+	networkSentTrackers map[uint16]*deltatracker.DeltaTracker[string, uint64]
+	networkRecvTrackers map[uint16]*deltatracker.DeltaTracker[string, uint64]
 }
 
 // userAgentRoundTripper is a custom http.RoundTripper that adds a User-Agent header to all requests
@@ -62,8 +81,8 @@ func (d *dockerManager) dequeue() {
 	}
 }
 
-// Returns stats for all running containers
-func (dm *dockerManager) getDockerStats() ([]*container.Stats, error) {
+// Returns stats for all running containers with cache-time-aware delta tracking
+func (dm *dockerManager) getDockerStats(cacheTimeMs uint16) ([]*container.Stats, error) {
 	resp, err := dm.client.Get("http://localhost/containers/json")
 	if err != nil {
 		return nil, err
@@ -87,8 +106,7 @@ func (dm *dockerManager) getDockerStats() ([]*container.Stats, error) {
 
 	var failedContainers []*container.ApiInfo
 
-	for i := range dm.apiContainerList {
-		ctr := dm.apiContainerList[i]
+	for _, ctr := range dm.apiContainerList {
 		ctr.IdShort = ctr.Id[:12]
 		dm.validIds[ctr.IdShort] = struct{}{}
 		// check if container is less than 1 minute old (possible restart)
@@ -98,9 +116,9 @@ func (dm *dockerManager) getDockerStats() ([]*container.Stats, error) {
 			dm.deleteContainerStatsSync(ctr.IdShort)
 		}
 		dm.queue()
-		go func() {
+		go func(ctr *container.ApiInfo) {
 			defer dm.dequeue()
-			err := dm.updateContainerStats(ctr)
+			err := dm.updateContainerStats(ctr, cacheTimeMs)
 			// if error, delete from map and add to failed list to retry
 			if err != nil {
 				dm.containerStatsMutex.Lock()
@@ -108,7 +126,7 @@ func (dm *dockerManager) getDockerStats() ([]*container.Stats, error) {
 				failedContainers = append(failedContainers, ctr)
 				dm.containerStatsMutex.Unlock()
 			}
-		}()
+		}(ctr)
 	}
 
 	dm.wg.Wait()
@@ -119,13 +137,12 @@ func (dm *dockerManager) getDockerStats() ([]*container.Stats, error) {
 		for i := range failedContainers {
 			ctr := failedContainers[i]
 			dm.queue()
-			go func() {
+			go func(ctr *container.ApiInfo) {
 				defer dm.dequeue()
-				err = dm.updateContainerStats(ctr)
-				if err != nil {
-					slog.Error("Error getting container stats", "err", err)
+				if err2 := dm.updateContainerStats(ctr, cacheTimeMs); err2 != nil {
+					slog.Error("Error getting container stats", "err", err2)
 				}
-			}()
+			}(ctr)
 		}
 		dm.wg.Wait()
 	}
@@ -140,18 +157,156 @@ func (dm *dockerManager) getDockerStats() ([]*container.Stats, error) {
 		}
 	}
 
+	// prepare network trackers for next interval for this cache time
+	dm.cycleNetworkDeltasForCacheTime(cacheTimeMs)
+
 	return stats, nil
 }
 
-// Updates stats for individual container
-func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo) error {
+// initializeCpuTracking initializes CPU tracking maps for a specific cache time interval
+func (dm *dockerManager) initializeCpuTracking(cacheTimeMs uint16) {
+	// Initialize cache time maps if they don't exist
+	if dm.lastCpuContainer[cacheTimeMs] == nil {
+		dm.lastCpuContainer[cacheTimeMs] = make(map[string]uint64)
+	}
+	if dm.lastCpuSystem[cacheTimeMs] == nil {
+		dm.lastCpuSystem[cacheTimeMs] = make(map[string]uint64)
+	}
+	// Ensure the outer map exists before indexing
+	if dm.lastCpuReadTime == nil {
+		dm.lastCpuReadTime = make(map[uint16]map[string]time.Time)
+	}
+	if dm.lastCpuReadTime[cacheTimeMs] == nil {
+		dm.lastCpuReadTime[cacheTimeMs] = make(map[string]time.Time)
+	}
+}
+
+// getCpuPreviousValues returns previous CPU values for a container and cache time interval
+func (dm *dockerManager) getCpuPreviousValues(cacheTimeMs uint16, containerId string) (uint64, uint64) {
+	return dm.lastCpuContainer[cacheTimeMs][containerId], dm.lastCpuSystem[cacheTimeMs][containerId]
+}
+
+// setCpuCurrentValues stores current CPU values for a container and cache time interval
+func (dm *dockerManager) setCpuCurrentValues(cacheTimeMs uint16, containerId string, cpuContainer, cpuSystem uint64) {
+	dm.lastCpuContainer[cacheTimeMs][containerId] = cpuContainer
+	dm.lastCpuSystem[cacheTimeMs][containerId] = cpuSystem
+}
+
+// calculateMemoryUsage calculates memory usage from Docker API stats
+func calculateMemoryUsage(apiStats *container.ApiStats, isWindows bool) (uint64, error) {
+	if isWindows {
+		return apiStats.MemoryStats.PrivateWorkingSet, nil
+	}
+
+	// Check if container has valid data, otherwise may be in restart loop (#103)
+	if apiStats.MemoryStats.Usage == 0 {
+		return 0, fmt.Errorf("no memory stats available")
+	}
+
+	memCache := apiStats.MemoryStats.Stats.InactiveFile
+	if memCache == 0 {
+		memCache = apiStats.MemoryStats.Stats.Cache
+	}
+
+	return apiStats.MemoryStats.Usage - memCache, nil
+}
+
+// getNetworkTracker returns the DeltaTracker for a specific cache time, creating it if needed
+func (dm *dockerManager) getNetworkTracker(cacheTimeMs uint16, isSent bool) *deltatracker.DeltaTracker[string, uint64] {
+	var trackers map[uint16]*deltatracker.DeltaTracker[string, uint64]
+	if isSent {
+		trackers = dm.networkSentTrackers
+	} else {
+		trackers = dm.networkRecvTrackers
+	}
+
+	if trackers[cacheTimeMs] == nil {
+		trackers[cacheTimeMs] = deltatracker.NewDeltaTracker[string, uint64]()
+	}
+
+	return trackers[cacheTimeMs]
+}
+
+// cycleNetworkDeltasForCacheTime cycles the network delta trackers for a specific cache time
+func (dm *dockerManager) cycleNetworkDeltasForCacheTime(cacheTimeMs uint16) {
+	if dm.networkSentTrackers[cacheTimeMs] != nil {
+		dm.networkSentTrackers[cacheTimeMs].Cycle()
+	}
+	if dm.networkRecvTrackers[cacheTimeMs] != nil {
+		dm.networkRecvTrackers[cacheTimeMs].Cycle()
+	}
+}
+
+// calculateNetworkStats calculates network sent/receive deltas using DeltaTracker
+func (dm *dockerManager) calculateNetworkStats(ctr *container.ApiInfo, apiStats *container.ApiStats, stats *container.Stats, initialized bool, name string, cacheTimeMs uint16) (uint64, uint64) {
+	var total_sent, total_recv uint64
+	for _, v := range apiStats.Networks {
+		total_sent += v.TxBytes
+		total_recv += v.RxBytes
+	}
+
+	// Get the DeltaTracker for this specific cache time
+	sentTracker := dm.getNetworkTracker(cacheTimeMs, true)
+	recvTracker := dm.getNetworkTracker(cacheTimeMs, false)
+
+	// Set current values in the cache-time-specific DeltaTracker
+	sentTracker.Set(ctr.IdShort, total_sent)
+	recvTracker.Set(ctr.IdShort, total_recv)
+
+	// Get deltas (bytes since last measurement)
+	sent_delta_raw := sentTracker.Delta(ctr.IdShort)
+	recv_delta_raw := recvTracker.Delta(ctr.IdShort)
+
+	// Calculate bytes per second independently for Tx and Rx if we have previous data
+	var sent_delta, recv_delta uint64
+	if initialized {
+		millisecondsElapsed := uint64(time.Since(stats.PrevReadTime).Milliseconds())
+		if millisecondsElapsed > 0 {
+			if sent_delta_raw > 0 {
+				sent_delta = sent_delta_raw * 1000 / millisecondsElapsed
+				if sent_delta > maxNetworkSpeedBps {
+					slog.Warn("Bad network delta", "container", name)
+					sent_delta = 0
+				}
+			}
+			if recv_delta_raw > 0 {
+				recv_delta = recv_delta_raw * 1000 / millisecondsElapsed
+				if recv_delta > maxNetworkSpeedBps {
+					slog.Warn("Bad network delta", "container", name)
+					recv_delta = 0
+				}
+			}
+		}
+	}
+
+	return sent_delta, recv_delta
+}
+
+// validateCpuPercentage checks if CPU percentage is within valid range
+func validateCpuPercentage(cpuPct float64, containerName string) error {
+	if cpuPct > 100 {
+		return fmt.Errorf("%s cpu pct greater than 100: %+v", containerName, cpuPct)
+	}
+	return nil
+}
+
+// updateContainerStatsValues updates the final stats values
+func updateContainerStatsValues(stats *container.Stats, cpuPct float64, usedMemory uint64, sent_delta, recv_delta uint64, readTime time.Time) {
+	stats.Cpu = twoDecimals(cpuPct)
+	stats.Mem = bytesToMegabytes(float64(usedMemory))
+	stats.NetworkSent = bytesToMegabytes(float64(sent_delta))
+	stats.NetworkRecv = bytesToMegabytes(float64(recv_delta))
+	stats.PrevReadTime = readTime
+}
+
+// Updates stats for individual container with cache-time-aware delta tracking
+func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeMs uint16) error {
 	name := ctr.Names[0][1:]
 
 	resp, err := dm.client.Get("http://localhost/containers/" + ctr.IdShort + "/stats?stream=0&one-shot=1")
 	if err != nil {
 		return err
 	}
-	defer resp.Body.Close()
 
 	dm.containerStatsMutex.Lock()
 	defer dm.containerStatsMutex.Unlock()
@@ -169,72 +324,58 @@ func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo) error {
 	stats.NetworkSent = 0
 	stats.NetworkRecv = 0
 
-	// docker host container stats response
-	// res := dm.getApiStats()
-	// defer dm.putApiStats(res)
-	//
-
 	res := dm.apiStats
 	res.Networks = nil
 	if err := dm.decode(resp, res); err != nil {
 		return err
 	}
 
-	// calculate cpu and memory stats
-	var usedMemory uint64
+	// Initialize CPU tracking for this cache time interval
+	dm.initializeCpuTracking(cacheTimeMs)
+
+	// Get previous CPU values
+	prevCpuContainer, prevCpuSystem := dm.getCpuPreviousValues(cacheTimeMs, ctr.IdShort)
+
+	// Calculate CPU percentage based on platform
 	var cpuPct float64
-
-	// store current cpu stats
-	prevCpuContainer, prevCpuSystem := stats.CpuContainer, stats.CpuSystem
-	stats.CpuContainer = res.CPUStats.CPUUsage.TotalUsage
-	stats.CpuSystem = res.CPUStats.SystemUsage
-
 	if dm.isWindows {
-		usedMemory = res.MemoryStats.PrivateWorkingSet
-		cpuPct = res.CalculateCpuPercentWindows(prevCpuContainer, stats.PrevReadTime)
+		prevRead := dm.lastCpuReadTime[cacheTimeMs][ctr.IdShort]
+		cpuPct = res.CalculateCpuPercentWindows(prevCpuContainer, prevRead)
 	} else {
-		// check if container has valid data, otherwise may be in restart loop (#103)
-		if res.MemoryStats.Usage == 0 {
-			return fmt.Errorf("%s - no memory stats - see https://github.com/henrygd/beszel/issues/144", name)
-		}
-		memCache := res.MemoryStats.Stats.InactiveFile
-		if memCache == 0 {
-			memCache = res.MemoryStats.Stats.Cache
-		}
-		usedMemory = res.MemoryStats.Usage - memCache
-
 		cpuPct = res.CalculateCpuPercentLinux(prevCpuContainer, prevCpuSystem)
 	}
 
-	if cpuPct > 100 {
-		return fmt.Errorf("%s cpu pct greater than 100: %+v", name, cpuPct)
+	// Calculate memory usage
+	usedMemory, err := calculateMemoryUsage(res, dm.isWindows)
+	if err != nil {
+		return fmt.Errorf("%s - %w - see https://github.com/henrygd/beszel/issues/144", name, err)
 	}
 
-	// network
+	// Store current CPU stats for next calculation
+	currentCpuContainer := res.CPUStats.CPUUsage.TotalUsage
+	currentCpuSystem := res.CPUStats.SystemUsage
+	dm.setCpuCurrentValues(cacheTimeMs, ctr.IdShort, currentCpuContainer, currentCpuSystem)
+
+	// Validate CPU percentage
+	if err := validateCpuPercentage(cpuPct, name); err != nil {
+		return err
+	}
+
+	// Calculate network stats using DeltaTracker
+	sent_delta, recv_delta := dm.calculateNetworkStats(ctr, res, stats, initialized, name, cacheTimeMs)
+
+	// Store current network values for legacy compatibility
 	var total_sent, total_recv uint64
 	for _, v := range res.Networks {
 		total_sent += v.TxBytes
 		total_recv += v.RxBytes
 	}
-	var sent_delta, recv_delta uint64
-	millisecondsElapsed := uint64(time.Since(stats.PrevReadTime).Milliseconds())
-	if initialized && millisecondsElapsed > 0 {
-		// get bytes per second
-		sent_delta = (total_sent - stats.PrevNet.Sent) * 1000 / millisecondsElapsed
-		recv_delta = (total_recv - stats.PrevNet.Recv) * 1000 / millisecondsElapsed
-		// check for unrealistic network values (> 5GB/s)
-		if sent_delta > 5e9 || recv_delta > 5e9 {
-			slog.Warn("Bad network delta", "container", name)
-			sent_delta, recv_delta = 0, 0
-		}
-	}
 	stats.PrevNet.Sent, stats.PrevNet.Recv = total_sent, total_recv
 
-	stats.Cpu = twoDecimals(cpuPct)
-	stats.Mem = bytesToMegabytes(float64(usedMemory))
-	stats.NetworkSent = bytesToMegabytes(float64(sent_delta))
-	stats.NetworkRecv = bytesToMegabytes(float64(recv_delta))
-	stats.PrevReadTime = res.Read
+	// Update final stats values
+	updateContainerStatsValues(stats, cpuPct, usedMemory, sent_delta, recv_delta, res.Read)
+	// store per-cache-time read time for Windows CPU percent calc
+	dm.lastCpuReadTime[cacheTimeMs][ctr.IdShort] = res.Read
 
 	return nil
 }
@@ -244,6 +385,15 @@ func (dm *dockerManager) deleteContainerStatsSync(id string) {
 	dm.containerStatsMutex.Lock()
 	defer dm.containerStatsMutex.Unlock()
 	delete(dm.containerStatsMap, id)
+	for ct := range dm.lastCpuContainer {
+		delete(dm.lastCpuContainer[ct], id)
+	}
+	for ct := range dm.lastCpuSystem {
+		delete(dm.lastCpuSystem[ct], id)
+	}
+	for ct := range dm.lastCpuReadTime {
+		delete(dm.lastCpuReadTime[ct], id)
+	}
 }
 
 // Creates a new http client for Docker or Podman API
@@ -283,7 +433,7 @@ func newDockerManager(a *Agent) *dockerManager {
 	}
 
 	// configurable timeout
-	timeout := time.Millisecond * 2100
+	timeout := time.Millisecond * time.Duration(dockerTimeoutMs)
 	if t, set := GetEnv("DOCKER_TIMEOUT"); set {
 		timeout, err = time.ParseDuration(t)
 		if err != nil {
@@ -308,6 +458,13 @@ func newDockerManager(a *Agent) *dockerManager {
 		sem:               make(chan struct{}, 5),
 		apiContainerList:  []*container.ApiInfo{},
 		apiStats:          &container.ApiStats{},
+
+		// Initialize cache-time-aware tracking structures
+		lastCpuContainer:    make(map[uint16]map[string]uint64),
+		lastCpuSystem:       make(map[uint16]map[string]uint64),
+		lastCpuReadTime:     make(map[uint16]map[string]time.Time),
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
 	}
 
 	// If using podman, return client

agent/docker_test.go

@@ -0,0 +1,875 @@
+//go:build testing
+// +build testing
+
+package agent
+
+import (
+	"encoding/json"
+	"os"
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel/agent/deltatracker"
+	"github.com/henrygd/beszel/internal/entities/container"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+var defaultCacheTimeMs = uint16(60_000)
+
+// cycleCpuDeltas cycles the CPU tracking data for a specific cache time interval
+func (dm *dockerManager) cycleCpuDeltas(cacheTimeMs uint16) {
+	// Clear the CPU tracking maps for this cache time interval
+	if dm.lastCpuContainer[cacheTimeMs] != nil {
+		clear(dm.lastCpuContainer[cacheTimeMs])
+	}
+	if dm.lastCpuSystem[cacheTimeMs] != nil {
+		clear(dm.lastCpuSystem[cacheTimeMs])
+	}
+}
+
+func TestCalculateMemoryUsage(t *testing.T) {
+	tests := []struct {
+		name        string
+		apiStats    *container.ApiStats
+		isWindows   bool
+		expected    uint64
+		expectError bool
+	}{
+		{
+			name: "Linux with valid memory stats",
+			apiStats: &container.ApiStats{
+				MemoryStats: container.MemoryStats{
+					Usage: 1048576, // 1MB
+					Stats: container.MemoryStatsStats{
+						Cache:        524288, // 512KB
+						InactiveFile: 262144, // 256KB
+					},
+				},
+			},
+			isWindows:   false,
+			expected:    786432, // 1MB - 256KB (inactive_file takes precedence) = 768KB
+			expectError: false,
+		},
+		{
+			name: "Linux with zero cache uses inactive_file",
+			apiStats: &container.ApiStats{
+				MemoryStats: container.MemoryStats{
+					Usage: 1048576, // 1MB
+					Stats: container.MemoryStatsStats{
+						Cache:        0,
+						InactiveFile: 262144, // 256KB
+					},
+				},
+			},
+			isWindows:   false,
+			expected:    786432, // 1MB - 256KB = 768KB
+			expectError: false,
+		},
+		{
+			name: "Windows with valid memory stats",
+			apiStats: &container.ApiStats{
+				MemoryStats: container.MemoryStats{
+					PrivateWorkingSet: 524288, // 512KB
+				},
+			},
+			isWindows:   true,
+			expected:    524288,
+			expectError: false,
+		},
+		{
+			name: "Linux with zero usage returns error",
+			apiStats: &container.ApiStats{
+				MemoryStats: container.MemoryStats{
+					Usage: 0,
+					Stats: container.MemoryStatsStats{
+						Cache:        0,
+						InactiveFile: 0,
+					},
+				},
+			},
+			isWindows:   false,
+			expected:    0,
+			expectError: true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result, err := calculateMemoryUsage(tt.apiStats, tt.isWindows)
+
+			if tt.expectError {
+				assert.Error(t, err)
+			} else {
+				assert.NoError(t, err)
+				assert.Equal(t, tt.expected, result)
+			}
+		})
+	}
+}
+
+func TestValidateCpuPercentage(t *testing.T) {
+	tests := []struct {
+		name          string
+		cpuPct        float64
+		containerName string
+		expectError   bool
+		expectedError string
+	}{
+		{
+			name:          "valid CPU percentage",
+			cpuPct:        50.5,
+			containerName: "test-container",
+			expectError:   false,
+		},
+		{
+			name:          "zero CPU percentage",
+			cpuPct:        0.0,
+			containerName: "test-container",
+			expectError:   false,
+		},
+		{
+			name:          "CPU percentage over 100",
+			cpuPct:        150.5,
+			containerName: "test-container",
+			expectError:   true,
+			expectedError: "test-container cpu pct greater than 100: 150.5",
+		},
+		{
+			name:          "CPU percentage exactly 100",
+			cpuPct:        100.0,
+			containerName: "test-container",
+			expectError:   false,
+		},
+		{
+			name:          "negative CPU percentage",
+			cpuPct:        -10.0,
+			containerName: "test-container",
+			expectError:   false, // Function only checks for > 100, not negative
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			err := validateCpuPercentage(tt.cpuPct, tt.containerName)
+
+			if tt.expectError {
+				assert.Error(t, err)
+				assert.Contains(t, err.Error(), tt.expectedError)
+			} else {
+				assert.NoError(t, err)
+			}
+		})
+	}
+}
+
+func TestUpdateContainerStatsValues(t *testing.T) {
+	stats := &container.Stats{
+		Name:         "test-container",
+		Cpu:          0.0,
+		Mem:          0.0,
+		NetworkSent:  0.0,
+		NetworkRecv:  0.0,
+		PrevReadTime: time.Time{},
+	}
+
+	testTime := time.Now()
+	updateContainerStatsValues(stats, 75.5, 1048576, 524288, 262144, testTime)
+
+	// Check CPU percentage (should be rounded to 2 decimals)
+	assert.Equal(t, 75.5, stats.Cpu)
+
+	// Check memory (should be converted to MB: 1048576 bytes = 1 MB)
+	assert.Equal(t, 1.0, stats.Mem)
+
+	// Check network sent (should be converted to MB: 524288 bytes = 0.5 MB)
+	assert.Equal(t, 0.5, stats.NetworkSent)
+
+	// Check network recv (should be converted to MB: 262144 bytes = 0.25 MB)
+	assert.Equal(t, 0.25, stats.NetworkRecv)
+
+	// Check read time
+	assert.Equal(t, testTime, stats.PrevReadTime)
+}
+
+func TestTwoDecimals(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    float64
+		expected float64
+	}{
+		{"round down", 1.234, 1.23},
+		{"round half up", 1.235, 1.24}, // math.Round rounds half up
+		{"no rounding needed", 1.23, 1.23},
+		{"negative number", -1.235, -1.24}, // math.Round rounds half up (more negative)
+		{"zero", 0.0, 0.0},
+		{"large number", 123.456, 123.46}, // rounds 5 up
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := twoDecimals(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestBytesToMegabytes(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    float64
+		expected float64
+	}{
+		{"1 MB", 1048576, 1.0},
+		{"512 KB", 524288, 0.5},
+		{"zero", 0, 0},
+		{"large value", 1073741824, 1024}, // 1 GB = 1024 MB
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := bytesToMegabytes(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestInitializeCpuTracking(t *testing.T) {
+	dm := &dockerManager{
+		lastCpuContainer: make(map[uint16]map[string]uint64),
+		lastCpuSystem:    make(map[uint16]map[string]uint64),
+		lastCpuReadTime:  make(map[uint16]map[string]time.Time),
+	}
+
+	cacheTimeMs := uint16(30000)
+
+	// Test initializing a new cache time
+	dm.initializeCpuTracking(cacheTimeMs)
+
+	// Check that maps were created
+	assert.NotNil(t, dm.lastCpuContainer[cacheTimeMs])
+	assert.NotNil(t, dm.lastCpuSystem[cacheTimeMs])
+	assert.NotNil(t, dm.lastCpuReadTime[cacheTimeMs])
+	assert.Empty(t, dm.lastCpuContainer[cacheTimeMs])
+	assert.Empty(t, dm.lastCpuSystem[cacheTimeMs])
+
+	// Test initializing existing cache time (should not overwrite)
+	dm.lastCpuContainer[cacheTimeMs]["test"] = 100
+	dm.lastCpuSystem[cacheTimeMs]["test"] = 200
+
+	dm.initializeCpuTracking(cacheTimeMs)
+
+	// Should still have the existing values
+	assert.Equal(t, uint64(100), dm.lastCpuContainer[cacheTimeMs]["test"])
+	assert.Equal(t, uint64(200), dm.lastCpuSystem[cacheTimeMs]["test"])
+}
+
+func TestGetCpuPreviousValues(t *testing.T) {
+	dm := &dockerManager{
+		lastCpuContainer: map[uint16]map[string]uint64{
+			30000: {"container1": 100, "container2": 200},
+		},
+		lastCpuSystem: map[uint16]map[string]uint64{
+			30000: {"container1": 150, "container2": 250},
+		},
+	}
+
+	// Test getting existing values
+	container, system := dm.getCpuPreviousValues(30000, "container1")
+	assert.Equal(t, uint64(100), container)
+	assert.Equal(t, uint64(150), system)
+
+	// Test getting non-existing container
+	container, system = dm.getCpuPreviousValues(30000, "nonexistent")
+	assert.Equal(t, uint64(0), container)
+	assert.Equal(t, uint64(0), system)
+
+	// Test getting non-existing cache time
+	container, system = dm.getCpuPreviousValues(60000, "container1")
+	assert.Equal(t, uint64(0), container)
+	assert.Equal(t, uint64(0), system)
+}
+
+func TestSetCpuCurrentValues(t *testing.T) {
+	dm := &dockerManager{
+		lastCpuContainer: make(map[uint16]map[string]uint64),
+		lastCpuSystem:    make(map[uint16]map[string]uint64),
+	}
+
+	cacheTimeMs := uint16(30000)
+	containerId := "test-container"
+
+	// Initialize the cache time maps first
+	dm.initializeCpuTracking(cacheTimeMs)
+
+	// Set values
+	dm.setCpuCurrentValues(cacheTimeMs, containerId, 500, 750)
+
+	// Check that values were set
+	assert.Equal(t, uint64(500), dm.lastCpuContainer[cacheTimeMs][containerId])
+	assert.Equal(t, uint64(750), dm.lastCpuSystem[cacheTimeMs][containerId])
+}
+
+func TestCalculateNetworkStats(t *testing.T) {
+	// Create docker manager with tracker maps
+	dm := &dockerManager{
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+	}
+
+	cacheTimeMs := uint16(30000)
+
+	// Pre-populate tracker for this cache time with initial values
+	sentTracker := deltatracker.NewDeltaTracker[string, uint64]()
+	recvTracker := deltatracker.NewDeltaTracker[string, uint64]()
+	sentTracker.Set("container1", 1000)
+	recvTracker.Set("container1", 800)
+	sentTracker.Cycle() // Move to previous
+	recvTracker.Cycle()
+
+	dm.networkSentTrackers[cacheTimeMs] = sentTracker
+	dm.networkRecvTrackers[cacheTimeMs] = recvTracker
+
+	ctr := &container.ApiInfo{
+		IdShort: "container1",
+	}
+
+	apiStats := &container.ApiStats{
+		Networks: map[string]container.NetworkStats{
+			"eth0": {TxBytes: 2000, RxBytes: 1800}, // New values
+		},
+	}
+
+	stats := &container.Stats{
+		PrevReadTime: time.Now().Add(-time.Second), // 1 second ago
+	}
+
+	// Test with initialized container
+	sent, recv := dm.calculateNetworkStats(ctr, apiStats, stats, true, "test-container", cacheTimeMs)
+
+	// Should return calculated byte rates per second
+	assert.GreaterOrEqual(t, sent, uint64(0))
+	assert.GreaterOrEqual(t, recv, uint64(0))
+
+	// Cycle and test one-direction change (Tx only) is reflected independently
+	dm.cycleNetworkDeltasForCacheTime(cacheTimeMs)
+	apiStats.Networks["eth0"] = container.NetworkStats{TxBytes: 2500, RxBytes: 1800} // +500 Tx only
+	sent, recv = dm.calculateNetworkStats(ctr, apiStats, stats, true, "test-container", cacheTimeMs)
+	assert.Greater(t, sent, uint64(0))
+	assert.Equal(t, uint64(0), recv)
+}
+
+func TestDockerManagerCreation(t *testing.T) {
+	// Test that dockerManager can be created without panicking
+	dm := &dockerManager{
+		lastCpuContainer:    make(map[uint16]map[string]uint64),
+		lastCpuSystem:       make(map[uint16]map[string]uint64),
+		lastCpuReadTime:     make(map[uint16]map[string]time.Time),
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+	}
+
+	assert.NotNil(t, dm)
+	assert.NotNil(t, dm.lastCpuContainer)
+	assert.NotNil(t, dm.lastCpuSystem)
+	assert.NotNil(t, dm.networkSentTrackers)
+	assert.NotNil(t, dm.networkRecvTrackers)
+}
+
+func TestCycleCpuDeltas(t *testing.T) {
+	dm := &dockerManager{
+		lastCpuContainer: map[uint16]map[string]uint64{
+			30000: {"container1": 100, "container2": 200},
+		},
+		lastCpuSystem: map[uint16]map[string]uint64{
+			30000: {"container1": 150, "container2": 250},
+		},
+		lastCpuReadTime: map[uint16]map[string]time.Time{
+			30000: {"container1": time.Now()},
+		},
+	}
+
+	cacheTimeMs := uint16(30000)
+
+	// Verify values exist before cycling
+	assert.Equal(t, uint64(100), dm.lastCpuContainer[cacheTimeMs]["container1"])
+	assert.Equal(t, uint64(200), dm.lastCpuContainer[cacheTimeMs]["container2"])
+
+	// Cycle the CPU deltas
+	dm.cycleCpuDeltas(cacheTimeMs)
+
+	// Verify values are cleared
+	assert.Empty(t, dm.lastCpuContainer[cacheTimeMs])
+	assert.Empty(t, dm.lastCpuSystem[cacheTimeMs])
+	// lastCpuReadTime is not affected by cycleCpuDeltas
+	assert.NotEmpty(t, dm.lastCpuReadTime[cacheTimeMs])
+}
+
+func TestCycleNetworkDeltas(t *testing.T) {
+	// Create docker manager with tracker maps
+	dm := &dockerManager{
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+	}
+
+	cacheTimeMs := uint16(30000)
+
+	// Get trackers for this cache time (creates them)
+	sentTracker := dm.getNetworkTracker(cacheTimeMs, true)
+	recvTracker := dm.getNetworkTracker(cacheTimeMs, false)
+
+	// Set some test data
+	sentTracker.Set("test", 100)
+	recvTracker.Set("test", 200)
+
+	// This should not panic
+	assert.NotPanics(t, func() {
+		dm.cycleNetworkDeltasForCacheTime(cacheTimeMs)
+	})
+
+	// Verify that cycle worked by checking deltas are now zero (no previous values)
+	assert.Equal(t, uint64(0), sentTracker.Delta("test"))
+	assert.Equal(t, uint64(0), recvTracker.Delta("test"))
+}
+
+func TestConstants(t *testing.T) {
+	// Test that constants are properly defined
+	assert.Equal(t, uint16(60000), defaultCacheTimeMs)
+	assert.Equal(t, uint64(5e9), maxNetworkSpeedBps)
+	assert.Equal(t, 2100, dockerTimeoutMs)
+}
+
+func TestDockerStatsWithMockData(t *testing.T) {
+	// Create a docker manager with initialized tracking
+	dm := &dockerManager{
+		lastCpuContainer:    make(map[uint16]map[string]uint64),
+		lastCpuSystem:       make(map[uint16]map[string]uint64),
+		lastCpuReadTime:     make(map[uint16]map[string]time.Time),
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		containerStatsMap:   make(map[string]*container.Stats),
+	}
+
+	cacheTimeMs := uint16(30000)
+
+	// Test that initializeCpuTracking works
+	dm.initializeCpuTracking(cacheTimeMs)
+	assert.NotNil(t, dm.lastCpuContainer[cacheTimeMs])
+	assert.NotNil(t, dm.lastCpuSystem[cacheTimeMs])
+
+	// Test that we can set and get CPU values
+	dm.setCpuCurrentValues(cacheTimeMs, "test-container", 1000, 2000)
+	container, system := dm.getCpuPreviousValues(cacheTimeMs, "test-container")
+	assert.Equal(t, uint64(1000), container)
+	assert.Equal(t, uint64(2000), system)
+}
+
+func TestMemoryStatsEdgeCases(t *testing.T) {
+	tests := []struct {
+		name      string
+		usage     uint64
+		cache     uint64
+		inactive  uint64
+		isWindows bool
+		expected  uint64
+		hasError  bool
+	}{
+		{"Linux normal case", 1000, 200, 0, false, 800, false},
+		{"Linux with inactive file", 1000, 0, 300, false, 700, false},
+		{"Windows normal case", 0, 0, 0, true, 500, false},
+		{"Linux zero usage error", 0, 0, 0, false, 0, true},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			apiStats := &container.ApiStats{
+				MemoryStats: container.MemoryStats{
+					Usage: tt.usage,
+					Stats: container.MemoryStatsStats{
+						Cache:        tt.cache,
+						InactiveFile: tt.inactive,
+					},
+				},
+			}
+
+			if tt.isWindows {
+				apiStats.MemoryStats.PrivateWorkingSet = tt.expected
+			}
+
+			result, err := calculateMemoryUsage(apiStats, tt.isWindows)
+
+			if tt.hasError {
+				assert.Error(t, err)
+			} else {
+				assert.NoError(t, err)
+				assert.Equal(t, tt.expected, result)
+			}
+		})
+	}
+}
+
+func TestContainerStatsInitialization(t *testing.T) {
+	stats := &container.Stats{Name: "test-container"}
+
+	// Verify initial values
+	assert.Equal(t, "test-container", stats.Name)
+	assert.Equal(t, 0.0, stats.Cpu)
+	assert.Equal(t, 0.0, stats.Mem)
+	assert.Equal(t, 0.0, stats.NetworkSent)
+	assert.Equal(t, 0.0, stats.NetworkRecv)
+	assert.Equal(t, time.Time{}, stats.PrevReadTime)
+
+	// Test updating values
+	testTime := time.Now()
+	updateContainerStatsValues(stats, 45.67, 2097152, 1048576, 524288, testTime)
+
+	assert.Equal(t, 45.67, stats.Cpu)
+	assert.Equal(t, 2.0, stats.Mem)
+	assert.Equal(t, 1.0, stats.NetworkSent)
+	assert.Equal(t, 0.5, stats.NetworkRecv)
+	assert.Equal(t, testTime, stats.PrevReadTime)
+}
+
+// Test with real Docker API test data
+func TestCalculateMemoryUsageWithRealData(t *testing.T) {
+	// Load minimal container stats from test data
+	data, err := os.ReadFile("test-data/container.json")
+	require.NoError(t, err)
+
+	var apiStats container.ApiStats
+	err = json.Unmarshal(data, &apiStats)
+	require.NoError(t, err)
+
+	// Test memory calculation with real data
+	usedMemory, err := calculateMemoryUsage(&apiStats, false)
+	require.NoError(t, err)
+
+	// From the real data: usage - inactive_file = 507400192 - 165130240 = 342269952
+	expected := uint64(507400192 - 165130240)
+	assert.Equal(t, expected, usedMemory)
+}
+
+func TestCpuPercentageCalculationWithRealData(t *testing.T) {
+	// Load minimal container stats from test data
+	data1, err := os.ReadFile("test-data/container.json")
+	require.NoError(t, err)
+
+	data2, err := os.ReadFile("test-data/container2.json")
+	require.NoError(t, err)
+
+	var apiStats1, apiStats2 container.ApiStats
+	err = json.Unmarshal(data1, &apiStats1)
+	require.NoError(t, err)
+	err = json.Unmarshal(data2, &apiStats2)
+	require.NoError(t, err)
+
+	// Calculate delta manually: 314891801000 - 312055276000 = 2836525000
+	// System delta: 1368474900000000 - 1366399830000000 = 2075070000000
+	// Expected %: (2836525000 / 2075070000000) * 100 ≈ 0.1367%
+	expectedPct := float64(2836525000) / float64(2075070000000) * 100.0
+	actualPct := apiStats2.CalculateCpuPercentLinux(apiStats1.CPUStats.CPUUsage.TotalUsage, apiStats1.CPUStats.SystemUsage)
+
+	assert.InDelta(t, expectedPct, actualPct, 0.01)
+}
+
+func TestNetworkStatsCalculationWithRealData(t *testing.T) {
+	// Create synthetic test data to avoid timing issues
+	apiStats1 := &container.ApiStats{
+		Networks: map[string]container.NetworkStats{
+			"eth0": {TxBytes: 1000000, RxBytes: 500000},
+		},
+	}
+
+	apiStats2 := &container.ApiStats{
+		Networks: map[string]container.NetworkStats{
+			"eth0": {TxBytes: 3000000, RxBytes: 1500000}, // 2MB sent, 1MB received increase
+		},
+	}
+
+	// Create docker manager with tracker maps
+	dm := &dockerManager{
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+	}
+
+	ctr := &container.ApiInfo{IdShort: "test-container"}
+	cacheTimeMs := uint16(30000) // Test with 30 second cache
+
+	// Use exact timing for deterministic results
+	exactly1000msAgo := time.Now().Add(-1000 * time.Millisecond)
+	stats := &container.Stats{
+		PrevReadTime: exactly1000msAgo,
+	}
+
+	// First call sets baseline
+	sent1, recv1 := dm.calculateNetworkStats(ctr, apiStats1, stats, true, "test", cacheTimeMs)
+	assert.Equal(t, uint64(0), sent1)
+	assert.Equal(t, uint64(0), recv1)
+
+	// Cycle to establish baseline for this cache time
+	dm.cycleNetworkDeltasForCacheTime(cacheTimeMs)
+
+	// Calculate expected results precisely
+	deltaSent := uint64(2000000)                             // 3000000 - 1000000
+	deltaRecv := uint64(1000000)                             // 1500000 - 500000
+	expectedElapsedMs := uint64(1000)                        // Exactly 1000ms
+	expectedSentRate := deltaSent * 1000 / expectedElapsedMs // Should be exactly 2000000
+	expectedRecvRate := deltaRecv * 1000 / expectedElapsedMs // Should be exactly 1000000
+
+	// Second call with changed data
+	sent2, recv2 := dm.calculateNetworkStats(ctr, apiStats2, stats, true, "test", cacheTimeMs)
+
+	// Should be exactly the expected rates (no tolerance needed)
+	assert.Equal(t, expectedSentRate, sent2)
+	assert.Equal(t, expectedRecvRate, recv2)
+
+	// Bad speed cap: set absurd delta over 1ms and expect 0 due to cap
+	dm.cycleNetworkDeltasForCacheTime(cacheTimeMs)
+	stats.PrevReadTime = time.Now().Add(-1 * time.Millisecond)
+	apiStats1.Networks["eth0"] = container.NetworkStats{TxBytes: 0, RxBytes: 0}
+	apiStats2.Networks["eth0"] = container.NetworkStats{TxBytes: 10 * 1024 * 1024 * 1024, RxBytes: 0} // 10GB delta
+	_, _ = dm.calculateNetworkStats(ctr, apiStats1, stats, true, "test", cacheTimeMs)                 // baseline
+	dm.cycleNetworkDeltasForCacheTime(cacheTimeMs)
+	sent3, recv3 := dm.calculateNetworkStats(ctr, apiStats2, stats, true, "test", cacheTimeMs)
+	assert.Equal(t, uint64(0), sent3)
+	assert.Equal(t, uint64(0), recv3)
+}
+
+func TestContainerStatsEndToEndWithRealData(t *testing.T) {
+	// Load minimal container stats
+	data, err := os.ReadFile("test-data/container.json")
+	require.NoError(t, err)
+
+	var apiStats container.ApiStats
+	err = json.Unmarshal(data, &apiStats)
+	require.NoError(t, err)
+
+	// Create a docker manager with proper initialization
+	dm := &dockerManager{
+		lastCpuContainer:    make(map[uint16]map[string]uint64),
+		lastCpuSystem:       make(map[uint16]map[string]uint64),
+		lastCpuReadTime:     make(map[uint16]map[string]time.Time),
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		containerStatsMap:   make(map[string]*container.Stats),
+	}
+
+	// Initialize CPU tracking
+	cacheTimeMs := uint16(30000)
+	dm.initializeCpuTracking(cacheTimeMs)
+
+	// Create container info
+	ctr := &container.ApiInfo{
+		IdShort: "abc123",
+	}
+
+	// Initialize container stats
+	stats := &container.Stats{Name: "jellyfin"}
+	dm.containerStatsMap[ctr.IdShort] = stats
+
+	// Test individual components that we can verify
+	usedMemory, memErr := calculateMemoryUsage(&apiStats, false)
+	assert.NoError(t, memErr)
+	assert.Greater(t, usedMemory, uint64(0))
+
+	// Test CPU percentage validation
+	cpuPct := 85.5
+	err = validateCpuPercentage(cpuPct, "jellyfin")
+	assert.NoError(t, err)
+
+	err = validateCpuPercentage(150.0, "jellyfin")
+	assert.Error(t, err)
+
+	// Test stats value updates
+	testStats := &container.Stats{}
+	testTime := time.Now()
+	updateContainerStatsValues(testStats, cpuPct, usedMemory, 1000000, 500000, testTime)
+
+	assert.Equal(t, cpuPct, testStats.Cpu)
+	assert.Equal(t, bytesToMegabytes(float64(usedMemory)), testStats.Mem)
+	assert.Equal(t, bytesToMegabytes(1000000), testStats.NetworkSent)
+	assert.Equal(t, bytesToMegabytes(500000), testStats.NetworkRecv)
+	assert.Equal(t, testTime, testStats.PrevReadTime)
+}
+
+func TestEdgeCasesWithRealData(t *testing.T) {
+	// Test with minimal container stats
+	minimalStats := &container.ApiStats{
+		CPUStats: container.CPUStats{
+			CPUUsage:    container.CPUUsage{TotalUsage: 1000},
+			SystemUsage: 50000,
+		},
+		MemoryStats: container.MemoryStats{
+			Usage: 1000000,
+			Stats: container.MemoryStatsStats{
+				Cache:        0,
+				InactiveFile: 0,
+			},
+		},
+		Networks: map[string]container.NetworkStats{
+			"eth0": {TxBytes: 1000, RxBytes: 500},
+		},
+	}
+
+	// Test memory calculation with zero cache/inactive
+	usedMemory, err := calculateMemoryUsage(minimalStats, false)
+	assert.NoError(t, err)
+	assert.Equal(t, uint64(1000000), usedMemory) // Should equal usage when no cache
+
+	// Test CPU percentage calculation
+	cpuPct := minimalStats.CalculateCpuPercentLinux(0, 0) // First run
+	assert.Equal(t, 0.0, cpuPct)
+
+	// Test with Windows data
+	minimalStats.MemoryStats.PrivateWorkingSet = 800000
+	usedMemory, err = calculateMemoryUsage(minimalStats, true)
+	assert.NoError(t, err)
+	assert.Equal(t, uint64(800000), usedMemory)
+}
+
+func TestDockerStatsWorkflow(t *testing.T) {
+	// Test the complete workflow that can be tested without HTTP calls
+	dm := &dockerManager{
+		lastCpuContainer:    make(map[uint16]map[string]uint64),
+		lastCpuSystem:       make(map[uint16]map[string]uint64),
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		containerStatsMap:   make(map[string]*container.Stats),
+	}
+
+	cacheTimeMs := uint16(30000)
+
+	// Test CPU tracking workflow
+	dm.initializeCpuTracking(cacheTimeMs)
+	assert.NotNil(t, dm.lastCpuContainer[cacheTimeMs])
+
+	// Test setting and getting CPU values
+	dm.setCpuCurrentValues(cacheTimeMs, "test-container", 1000, 50000)
+	containerVal, systemVal := dm.getCpuPreviousValues(cacheTimeMs, "test-container")
+	assert.Equal(t, uint64(1000), containerVal)
+	assert.Equal(t, uint64(50000), systemVal)
+
+	// Test network tracking workflow (multi-interface summation)
+	sentTracker := dm.getNetworkTracker(cacheTimeMs, true)
+	recvTracker := dm.getNetworkTracker(cacheTimeMs, false)
+
+	// Simulate two interfaces summed by setting combined totals
+	sentTracker.Set("test-container", 1000+2000)
+	recvTracker.Set("test-container", 500+700)
+
+	deltaSent := sentTracker.Delta("test-container")
+	deltaRecv := recvTracker.Delta("test-container")
+	assert.Equal(t, uint64(0), deltaSent) // No previous value
+	assert.Equal(t, uint64(0), deltaRecv)
+
+	// Cycle and test again
+	dm.cycleNetworkDeltasForCacheTime(cacheTimeMs)
+
+	// Increase each interface total (combined totals go up by 1500 and 800)
+	sentTracker.Set("test-container", (1000+2000)+1500)
+	recvTracker.Set("test-container", (500+700)+800)
+
+	deltaSent = sentTracker.Delta("test-container")
+	deltaRecv = recvTracker.Delta("test-container")
+	assert.Equal(t, uint64(1500), deltaSent)
+	assert.Equal(t, uint64(800), deltaRecv)
+}
+
+func TestNetworkRateCalculationFormula(t *testing.T) {
+	// Test the exact formula used in calculateNetworkStats
+	testCases := []struct {
+		name         string
+		deltaBytes   uint64
+		elapsedMs    uint64
+		expectedRate uint64
+	}{
+		{"1MB over 1 second", 1000000, 1000, 1000000},
+		{"2MB over 1 second", 2000000, 1000, 2000000},
+		{"1MB over 2 seconds", 1000000, 2000, 500000},
+		{"500KB over 500ms", 500000, 500, 1000000},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			// This is the exact formula from calculateNetworkStats
+			actualRate := tc.deltaBytes * 1000 / tc.elapsedMs
+			assert.Equal(t, tc.expectedRate, actualRate,
+				"Rate calculation should be exact: %d bytes * 1000 / %d ms = %d",
+				tc.deltaBytes, tc.elapsedMs, tc.expectedRate)
+		})
+	}
+}
+
+func TestDeltaTrackerCacheTimeIsolation(t *testing.T) {
+	// Test that different cache times have separate DeltaTracker instances
+	dm := &dockerManager{
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+	}
+
+	ctr := &container.ApiInfo{IdShort: "web-server"}
+	cacheTime1 := uint16(30000)
+	cacheTime2 := uint16(60000)
+
+	// Get trackers for different cache times (creates separate instances)
+	sentTracker1 := dm.getNetworkTracker(cacheTime1, true)
+	recvTracker1 := dm.getNetworkTracker(cacheTime1, false)
+
+	sentTracker2 := dm.getNetworkTracker(cacheTime2, true)
+	recvTracker2 := dm.getNetworkTracker(cacheTime2, false)
+
+	// Verify they are different instances
+	assert.NotSame(t, sentTracker1, sentTracker2)
+	assert.NotSame(t, recvTracker1, recvTracker2)
+
+	// Set values for cache time 1
+	sentTracker1.Set(ctr.IdShort, 1000000)
+	recvTracker1.Set(ctr.IdShort, 500000)
+
+	// Set values for cache time 2
+	sentTracker2.Set(ctr.IdShort, 2000000)
+	recvTracker2.Set(ctr.IdShort, 1000000)
+
+	// Verify they don't interfere (both should return 0 since no previous values)
+	assert.Equal(t, uint64(0), sentTracker1.Delta(ctr.IdShort))
+	assert.Equal(t, uint64(0), recvTracker1.Delta(ctr.IdShort))
+	assert.Equal(t, uint64(0), sentTracker2.Delta(ctr.IdShort))
+	assert.Equal(t, uint64(0), recvTracker2.Delta(ctr.IdShort))
+
+	// Cycle cache time 1 trackers
+	dm.cycleNetworkDeltasForCacheTime(cacheTime1)
+
+	// Set new values for cache time 1
+	sentTracker1.Set(ctr.IdShort, 3000000) // 2MB increase
+	recvTracker1.Set(ctr.IdShort, 1500000) // 1MB increase
+
+	// Cache time 1 should show deltas, cache time 2 should still be 0
+	assert.Equal(t, uint64(2000000), sentTracker1.Delta(ctr.IdShort))
+	assert.Equal(t, uint64(1000000), recvTracker1.Delta(ctr.IdShort))
+	assert.Equal(t, uint64(0), sentTracker2.Delta(ctr.IdShort)) // Unaffected
+	assert.Equal(t, uint64(0), recvTracker2.Delta(ctr.IdShort)) // Unaffected
+
+	// Cycle cache time 2 and verify it works independently
+	dm.cycleNetworkDeltasForCacheTime(cacheTime2)
+	sentTracker2.Set(ctr.IdShort, 2500000) // 0.5MB increase
+	recvTracker2.Set(ctr.IdShort, 1200000) // 0.2MB increase
+
+	assert.Equal(t, uint64(500000), sentTracker2.Delta(ctr.IdShort))
+	assert.Equal(t, uint64(200000), recvTracker2.Delta(ctr.IdShort))
+}
+
+func TestConstantsAndUtilityFunctions(t *testing.T) {
+	// Test constants are properly defined
+	assert.Equal(t, uint16(60000), defaultCacheTimeMs)
+	assert.Equal(t, uint64(5e9), maxNetworkSpeedBps)
+	assert.Equal(t, 2100, dockerTimeoutMs)
+
+	// Test utility functions
+	assert.Equal(t, 1.5, twoDecimals(1.499))
+	assert.Equal(t, 1.5, twoDecimals(1.5))
+	assert.Equal(t, 1.5, twoDecimals(1.501))
+
+	assert.Equal(t, 1.0, bytesToMegabytes(1048576)) // 1 MB
+	assert.Equal(t, 0.5, bytesToMegabytes(524288))  // 512 KB
+	assert.Equal(t, 0.0, bytesToMegabytes(0))
+}

agent/gpu.go

@@ -44,6 +44,21 @@ type GPUManager struct {
 	tegrastats    bool
 	intelGpuStats bool
 	GpuDataMap    map[string]*system.GPUData
+	// lastAvgData stores the last calculated averages for each GPU
+	// Used when a collection happens before new data arrives (Count == 0)
+	lastAvgData map[string]system.GPUData
+	// Per-cache-key tracking for delta calculations
+	// cacheKey -> gpuId -> snapshot of last count/usage/power values
+	lastSnapshots map[uint16]map[string]*gpuSnapshot
+}
+
+// gpuSnapshot stores the last observed incremental values for delta tracking
+type gpuSnapshot struct {
+	count    uint32
+	usage    float64
+	power    float64
+	powerPkg float64
+	engines  map[string]float64
 }
 
 // RocmSmiJson represents the JSON structure of rocm-smi output
@@ -229,48 +244,21 @@ func (gm *GPUManager) parseAmdData(output []byte) bool {
 	return true
 }
 
-// sums and resets the current GPU utilization data since the last update
-func (gm *GPUManager) GetCurrentData() map[string]system.GPUData {
+// GetCurrentData returns GPU utilization data averaged since the last call with this cacheKey
+func (gm *GPUManager) GetCurrentData(cacheKey uint16) map[string]system.GPUData {
 	gm.Lock()
 	defer gm.Unlock()
 
-	// check for GPUs with the same name
-	nameCounts := make(map[string]int)
-	for _, gpu := range gm.GpuDataMap {
-		nameCounts[gpu.Name]++
-	}
+	gm.initializeSnapshots(cacheKey)
+	nameCounts := gm.countGPUNames()
 
-	// copy / reset the data
 	gpuData := make(map[string]system.GPUData, len(gm.GpuDataMap))
 	for id, gpu := range gm.GpuDataMap {
-		// avoid division by zero
-		count := max(gpu.Count, 1)
+		gpuAvg := gm.calculateGPUAverage(id, gpu, cacheKey)
+		gm.updateInstantaneousValues(&gpuAvg, gpu)
+		gm.storeSnapshot(id, gpu, cacheKey)
 
-		// average the data
-		gpuAvg := *gpu
-		gpuAvg.Temperature = twoDecimals(gpu.Temperature)
-		gpuAvg.Power = twoDecimals(gpu.Power / count)
-
-		// intel gpu stats doesn't provide usage, memory used, or memory total
-		if gpu.Engines != nil {
-			maxEngineUsage := 0.0
-			for name, engine := range gpu.Engines {
-				gpuAvg.Engines[name] = twoDecimals(engine / count)
-				maxEngineUsage = max(maxEngineUsage, engine/count)
-			}
-			gpuAvg.PowerPkg = twoDecimals(gpu.PowerPkg / count)
-			gpuAvg.Usage = twoDecimals(maxEngineUsage)
-		} else {
-			gpuAvg.Usage = twoDecimals(gpu.Usage / count)
-			gpuAvg.MemoryUsed = twoDecimals(gpu.MemoryUsed)
-			gpuAvg.MemoryTotal = twoDecimals(gpu.MemoryTotal)
-		}
-
-		// reset accumulators in the original gpu data for next collection
-		gpu.Usage, gpu.Power, gpu.PowerPkg, gpu.Count = gpuAvg.Usage, gpuAvg.Power, gpuAvg.PowerPkg, 1
-		gpu.Engines = gpuAvg.Engines
-
-		// append id to the name if there are multiple GPUs with the same name
+		// Append id to name if there are multiple GPUs with the same name
 		if nameCounts[gpu.Name] > 1 {
 			gpuAvg.Name = fmt.Sprintf("%s %s", gpu.Name, id)
 		}
@@ -280,6 +268,114 @@ func (gm *GPUManager) GetCurrentData() map[string]system.GPUData {
 	return gpuData
 }
 
+// initializeSnapshots ensures snapshot maps are initialized for the given cache key
+func (gm *GPUManager) initializeSnapshots(cacheKey uint16) {
+	if gm.lastAvgData == nil {
+		gm.lastAvgData = make(map[string]system.GPUData)
+	}
+	if gm.lastSnapshots == nil {
+		gm.lastSnapshots = make(map[uint16]map[string]*gpuSnapshot)
+	}
+	if gm.lastSnapshots[cacheKey] == nil {
+		gm.lastSnapshots[cacheKey] = make(map[string]*gpuSnapshot)
+	}
+}
+
+// countGPUNames returns a map of GPU names to their occurrence count
+func (gm *GPUManager) countGPUNames() map[string]int {
+	nameCounts := make(map[string]int)
+	for _, gpu := range gm.GpuDataMap {
+		nameCounts[gpu.Name]++
+	}
+	return nameCounts
+}
+
+// calculateGPUAverage computes the average GPU metrics since the last snapshot for this cache key
+func (gm *GPUManager) calculateGPUAverage(id string, gpu *system.GPUData, cacheKey uint16) system.GPUData {
+	lastSnapshot := gm.lastSnapshots[cacheKey][id]
+	currentCount := uint32(gpu.Count)
+	deltaCount := gm.calculateDeltaCount(currentCount, lastSnapshot)
+
+	// If no new data arrived, use last known average
+	if deltaCount == 0 {
+		return gm.lastAvgData[id] // zero value if not found
+	}
+
+	// Calculate new average
+	gpuAvg := *gpu
+	deltaUsage, deltaPower, deltaPowerPkg := gm.calculateDeltas(gpu, lastSnapshot)
+
+	gpuAvg.Power = twoDecimals(deltaPower / float64(deltaCount))
+
+	if gpu.Engines != nil {
+		gpuAvg.Usage = gm.calculateIntelGPUUsage(&gpuAvg, gpu, lastSnapshot, deltaCount)
+		gpuAvg.PowerPkg = twoDecimals(deltaPowerPkg / float64(deltaCount))
+	} else {
+		gpuAvg.Usage = twoDecimals(deltaUsage / float64(deltaCount))
+	}
+
+	gm.lastAvgData[id] = gpuAvg
+	return gpuAvg
+}
+
+// calculateDeltaCount returns the change in count since the last snapshot
+func (gm *GPUManager) calculateDeltaCount(currentCount uint32, lastSnapshot *gpuSnapshot) uint32 {
+	if lastSnapshot != nil {
+		return currentCount - lastSnapshot.count
+	}
+	return currentCount
+}
+
+// calculateDeltas computes the change in usage, power, and powerPkg since the last snapshot
+func (gm *GPUManager) calculateDeltas(gpu *system.GPUData, lastSnapshot *gpuSnapshot) (deltaUsage, deltaPower, deltaPowerPkg float64) {
+	if lastSnapshot != nil {
+		return gpu.Usage - lastSnapshot.usage,
+			gpu.Power - lastSnapshot.power,
+			gpu.PowerPkg - lastSnapshot.powerPkg
+	}
+	return gpu.Usage, gpu.Power, gpu.PowerPkg
+}
+
+// calculateIntelGPUUsage computes Intel GPU usage from engine metrics and returns max engine usage
+func (gm *GPUManager) calculateIntelGPUUsage(gpuAvg, gpu *system.GPUData, lastSnapshot *gpuSnapshot, deltaCount uint32) float64 {
+	maxEngineUsage := 0.0
+	for name, engine := range gpu.Engines {
+		var deltaEngine float64
+		if lastSnapshot != nil && lastSnapshot.engines != nil {
+			deltaEngine = engine - lastSnapshot.engines[name]
+		} else {
+			deltaEngine = engine
+		}
+		gpuAvg.Engines[name] = twoDecimals(deltaEngine / float64(deltaCount))
+		maxEngineUsage = max(maxEngineUsage, deltaEngine/float64(deltaCount))
+	}
+	return twoDecimals(maxEngineUsage)
+}
+
+// updateInstantaneousValues updates values that should reflect current state, not averages
+func (gm *GPUManager) updateInstantaneousValues(gpuAvg *system.GPUData, gpu *system.GPUData) {
+	gpuAvg.Temperature = twoDecimals(gpu.Temperature)
+	gpuAvg.MemoryUsed = twoDecimals(gpu.MemoryUsed)
+	gpuAvg.MemoryTotal = twoDecimals(gpu.MemoryTotal)
+}
+
+// storeSnapshot saves the current GPU state for this cache key
+func (gm *GPUManager) storeSnapshot(id string, gpu *system.GPUData, cacheKey uint16) {
+	snapshot := &gpuSnapshot{
+		count:    uint32(gpu.Count),
+		usage:    gpu.Usage,
+		power:    gpu.Power,
+		powerPkg: gpu.PowerPkg,
+	}
+	if gpu.Engines != nil {
+		snapshot.engines = make(map[string]float64, len(gpu.Engines))
+		for name, value := range gpu.Engines {
+			snapshot.engines[name] = value
+		}
+	}
+	gm.lastSnapshots[cacheKey][id] = snapshot
+}
+
 // detectGPUs checks for the presence of GPU management tools (nvidia-smi, rocm-smi, tegrastats)
 // in the system path. It sets the corresponding flags in the GPUManager struct if any of these
 // tools are found. If none of the tools are found, it returns an error indicating that no GPU

agent/gpu_test.go

@@ -332,7 +332,7 @@ func TestParseJetsonData(t *testing.T) {
 }
 
 func TestGetCurrentData(t *testing.T) {
-	t.Run("calculates averages and resets accumulators", func(t *testing.T) {
+	t.Run("calculates averages with per-cache-key delta tracking", func(t *testing.T) {
 		gm := &GPUManager{
 			GpuDataMap: map[string]*system.GPUData{
 				"0": {
@@ -365,7 +365,8 @@ func TestGetCurrentData(t *testing.T) {
 			},
 		}
 
-		result := gm.GetCurrentData()
+		cacheKey := uint16(5000)
+		result := gm.GetCurrentData(cacheKey)
 
 		// Verify name disambiguation
 		assert.Equal(t, "GPU1 0", result["0"].Name)
@@ -378,13 +379,19 @@ func TestGetCurrentData(t *testing.T) {
 		assert.InDelta(t, 30.0, result["1"].Usage, 0.01)
 		assert.InDelta(t, 60.0, result["1"].Power, 0.01)
 
-		// Verify that accumulators in the original map are reset
-		assert.EqualValues(t, float64(1), gm.GpuDataMap["0"].Count, "GPU 0 Count should be reset")
-		assert.EqualValues(t, float64(50.0), gm.GpuDataMap["0"].Usage, "GPU 0 Usage should be reset")
-		assert.Equal(t, float64(100.0), gm.GpuDataMap["0"].Power, "GPU 0 Power should be reset")
-		assert.Equal(t, float64(1), gm.GpuDataMap["1"].Count, "GPU 1 Count should be reset")
-		assert.Equal(t, float64(30), gm.GpuDataMap["1"].Usage, "GPU 1 Usage should be reset")
-		assert.Equal(t, float64(60), gm.GpuDataMap["1"].Power, "GPU 1 Power should be reset")
+		// Verify that accumulators in the original map are NOT reset (they keep growing)
+		assert.EqualValues(t, 2, gm.GpuDataMap["0"].Count, "GPU 0 Count should remain at 2")
+		assert.EqualValues(t, 100, gm.GpuDataMap["0"].Usage, "GPU 0 Usage should remain at 100")
+		assert.Equal(t, 200.0, gm.GpuDataMap["0"].Power, "GPU 0 Power should remain at 200")
+		assert.Equal(t, 1.0, gm.GpuDataMap["1"].Count, "GPU 1 Count should remain at 1")
+		assert.Equal(t, 30.0, gm.GpuDataMap["1"].Usage, "GPU 1 Usage should remain at 30")
+		assert.Equal(t, 60.0, gm.GpuDataMap["1"].Power, "GPU 1 Power should remain at 60")
+
+		// Verify snapshots were stored for this cache key
+		assert.NotNil(t, gm.lastSnapshots[cacheKey]["0"])
+		assert.Equal(t, uint32(2), gm.lastSnapshots[cacheKey]["0"].count)
+		assert.Equal(t, 100.0, gm.lastSnapshots[cacheKey]["0"].usage)
+		assert.Equal(t, 200.0, gm.lastSnapshots[cacheKey]["0"].power)
 	})
 
 	t.Run("handles zero count without panicking", func(t *testing.T) {
@@ -399,17 +406,543 @@ func TestGetCurrentData(t *testing.T) {
 			},
 		}
 
+		cacheKey := uint16(5000)
 		var result map[string]system.GPUData
 		assert.NotPanics(t, func() {
-			result = gm.GetCurrentData()
+			result = gm.GetCurrentData(cacheKey)
 		})
 
 		// Check that usage and power are 0
 		assert.Equal(t, 0.0, result["0"].Usage)
 		assert.Equal(t, 0.0, result["0"].Power)
 
-		// Verify reset count
-		assert.EqualValues(t, 1, gm.GpuDataMap["0"].Count)
+		// Verify count remains 0
+		assert.EqualValues(t, 0, gm.GpuDataMap["0"].Count)
+	})
+
+	t.Run("uses last average when no new data arrives", func(t *testing.T) {
+		gm := &GPUManager{
+			GpuDataMap: map[string]*system.GPUData{
+				"0": {
+					Name:        "TestGPU",
+					Temperature: 55.0,
+					MemoryUsed:  1500,
+					MemoryTotal: 8000,
+					Usage:       100, // Will average to 50
+					Power:       200, // Will average to 100
+					Count:       2,
+				},
+			},
+		}
+
+		cacheKey := uint16(5000)
+
+		// First collection - should calculate averages and store them
+		result1 := gm.GetCurrentData(cacheKey)
+		assert.InDelta(t, 50.0, result1["0"].Usage, 0.01)
+		assert.InDelta(t, 100.0, result1["0"].Power, 0.01)
+		assert.EqualValues(t, 2, gm.GpuDataMap["0"].Count, "Count should remain at 2")
+
+		// Update temperature but no new usage/power data (count stays same)
+		gm.GpuDataMap["0"].Temperature = 60.0
+		gm.GpuDataMap["0"].MemoryUsed = 1600
+
+		// Second collection - should use last averages since count hasn't changed (delta = 0)
+		result2 := gm.GetCurrentData(cacheKey)
+		assert.InDelta(t, 50.0, result2["0"].Usage, 0.01, "Should use last average")
+		assert.InDelta(t, 100.0, result2["0"].Power, 0.01, "Should use last average")
+		assert.InDelta(t, 60.0, result2["0"].Temperature, 0.01, "Should use current temperature")
+		assert.InDelta(t, 1600.0, result2["0"].MemoryUsed, 0.01, "Should use current memory")
+		assert.EqualValues(t, 2, gm.GpuDataMap["0"].Count, "Count should still be 2")
+	})
+
+	t.Run("tracks separate averages per cache key", func(t *testing.T) {
+		gm := &GPUManager{
+			GpuDataMap: map[string]*system.GPUData{
+				"0": {
+					Name:        "TestGPU",
+					Temperature: 55.0,
+					MemoryUsed:  1500,
+					MemoryTotal: 8000,
+					Usage:       100, // Initial: 100 over 2 counts = 50 avg
+					Power:       200, // Initial: 200 over 2 counts = 100 avg
+					Count:       2,
+				},
+			},
+		}
+
+		cacheKey1 := uint16(5000)
+		cacheKey2 := uint16(10000)
+
+		// First check with cacheKey1 - baseline
+		result1 := gm.GetCurrentData(cacheKey1)
+		assert.InDelta(t, 50.0, result1["0"].Usage, 0.01, "CacheKey1: Initial average should be 50")
+		assert.InDelta(t, 100.0, result1["0"].Power, 0.01, "CacheKey1: Initial average should be 100")
+
+		// Simulate GPU activity - accumulate more data
+		gm.GpuDataMap["0"].Usage += 60  // Now total: 160
+		gm.GpuDataMap["0"].Power += 150 // Now total: 350
+		gm.GpuDataMap["0"].Count += 3   // Now total: 5
+
+		// Check with cacheKey1 again - should get delta since last cacheKey1 check
+		result2 := gm.GetCurrentData(cacheKey1)
+		assert.InDelta(t, 20.0, result2["0"].Usage, 0.01, "CacheKey1: Delta average should be 60/3 = 20")
+		assert.InDelta(t, 50.0, result2["0"].Power, 0.01, "CacheKey1: Delta average should be 150/3 = 50")
+
+		// Check with cacheKey2 for the first time - should get average since beginning
+		result3 := gm.GetCurrentData(cacheKey2)
+		assert.InDelta(t, 32.0, result3["0"].Usage, 0.01, "CacheKey2: Total average should be 160/5 = 32")
+		assert.InDelta(t, 70.0, result3["0"].Power, 0.01, "CacheKey2: Total average should be 350/5 = 70")
+
+		// Simulate more GPU activity
+		gm.GpuDataMap["0"].Usage += 80  // Now total: 240
+		gm.GpuDataMap["0"].Power += 160 // Now total: 510
+		gm.GpuDataMap["0"].Count += 2   // Now total: 7
+
+		// Check with cacheKey1 - should get delta since last cacheKey1 check
+		result4 := gm.GetCurrentData(cacheKey1)
+		assert.InDelta(t, 40.0, result4["0"].Usage, 0.01, "CacheKey1: New delta average should be 80/2 = 40")
+		assert.InDelta(t, 80.0, result4["0"].Power, 0.01, "CacheKey1: New delta average should be 160/2 = 80")
+
+		// Check with cacheKey2 - should get delta since last cacheKey2 check
+		result5 := gm.GetCurrentData(cacheKey2)
+		assert.InDelta(t, 40.0, result5["0"].Usage, 0.01, "CacheKey2: Delta average should be 80/2 = 40")
+		assert.InDelta(t, 80.0, result5["0"].Power, 0.01, "CacheKey2: Delta average should be 160/2 = 80")
+
+		// Verify snapshots exist for both cache keys
+		assert.NotNil(t, gm.lastSnapshots[cacheKey1])
+		assert.NotNil(t, gm.lastSnapshots[cacheKey2])
+		assert.NotNil(t, gm.lastSnapshots[cacheKey1]["0"])
+		assert.NotNil(t, gm.lastSnapshots[cacheKey2]["0"])
+	})
+}
+
+func TestCalculateDeltaCount(t *testing.T) {
+	gm := &GPUManager{}
+
+	t.Run("with no previous snapshot", func(t *testing.T) {
+		delta := gm.calculateDeltaCount(10, nil)
+		assert.Equal(t, uint32(10), delta, "Should return current count when no snapshot exists")
+	})
+
+	t.Run("with previous snapshot", func(t *testing.T) {
+		snapshot := &gpuSnapshot{count: 5}
+		delta := gm.calculateDeltaCount(15, snapshot)
+		assert.Equal(t, uint32(10), delta, "Should return difference between current and snapshot")
+	})
+
+	t.Run("with same count", func(t *testing.T) {
+		snapshot := &gpuSnapshot{count: 10}
+		delta := gm.calculateDeltaCount(10, snapshot)
+		assert.Equal(t, uint32(0), delta, "Should return zero when count hasn't changed")
+	})
+}
+
+func TestCalculateDeltas(t *testing.T) {
+	gm := &GPUManager{}
+
+	t.Run("with no previous snapshot", func(t *testing.T) {
+		gpu := &system.GPUData{
+			Usage:    100.5,
+			Power:    250.75,
+			PowerPkg: 300.25,
+		}
+		deltaUsage, deltaPower, deltaPowerPkg := gm.calculateDeltas(gpu, nil)
+		assert.Equal(t, 100.5, deltaUsage)
+		assert.Equal(t, 250.75, deltaPower)
+		assert.Equal(t, 300.25, deltaPowerPkg)
+	})
+
+	t.Run("with previous snapshot", func(t *testing.T) {
+		gpu := &system.GPUData{
+			Usage:    150.5,
+			Power:    300.75,
+			PowerPkg: 400.25,
+		}
+		snapshot := &gpuSnapshot{
+			usage:    100.5,
+			power:    250.75,
+			powerPkg: 300.25,
+		}
+		deltaUsage, deltaPower, deltaPowerPkg := gm.calculateDeltas(gpu, snapshot)
+		assert.InDelta(t, 50.0, deltaUsage, 0.01)
+		assert.InDelta(t, 50.0, deltaPower, 0.01)
+		assert.InDelta(t, 100.0, deltaPowerPkg, 0.01)
+	})
+}
+
+func TestCalculateIntelGPUUsage(t *testing.T) {
+	gm := &GPUManager{}
+
+	t.Run("with no previous snapshot", func(t *testing.T) {
+		gpuAvg := &system.GPUData{
+			Engines: make(map[string]float64),
+		}
+		gpu := &system.GPUData{
+			Engines: map[string]float64{
+				"Render/3D": 80.0,
+				"Video":     40.0,
+				"Compute":   60.0,
+			},
+		}
+		maxUsage := gm.calculateIntelGPUUsage(gpuAvg, gpu, nil, 2)
+
+		assert.Equal(t, 40.0, maxUsage, "Should return max engine usage (80/2=40)")
+		assert.Equal(t, 40.0, gpuAvg.Engines["Render/3D"])
+		assert.Equal(t, 20.0, gpuAvg.Engines["Video"])
+		assert.Equal(t, 30.0, gpuAvg.Engines["Compute"])
+	})
+
+	t.Run("with previous snapshot", func(t *testing.T) {
+		gpuAvg := &system.GPUData{
+			Engines: make(map[string]float64),
+		}
+		gpu := &system.GPUData{
+			Engines: map[string]float64{
+				"Render/3D": 180.0,
+				"Video":     100.0,
+				"Compute":   140.0,
+			},
+		}
+		snapshot := &gpuSnapshot{
+			engines: map[string]float64{
+				"Render/3D": 80.0,
+				"Video":     40.0,
+				"Compute":   60.0,
+			},
+		}
+		maxUsage := gm.calculateIntelGPUUsage(gpuAvg, gpu, snapshot, 5)
+
+		// Deltas: Render/3D=100, Video=60, Compute=80 over 5 counts
+		assert.Equal(t, 20.0, maxUsage, "Should return max engine delta (100/5=20)")
+		assert.Equal(t, 20.0, gpuAvg.Engines["Render/3D"])
+		assert.Equal(t, 12.0, gpuAvg.Engines["Video"])
+		assert.Equal(t, 16.0, gpuAvg.Engines["Compute"])
+	})
+
+	t.Run("handles missing engine in snapshot", func(t *testing.T) {
+		gpuAvg := &system.GPUData{
+			Engines: make(map[string]float64),
+		}
+		gpu := &system.GPUData{
+			Engines: map[string]float64{
+				"Render/3D": 100.0,
+				"NewEngine": 50.0,
+			},
+		}
+		snapshot := &gpuSnapshot{
+			engines: map[string]float64{
+				"Render/3D": 80.0,
+				// NewEngine doesn't exist in snapshot
+			},
+		}
+		maxUsage := gm.calculateIntelGPUUsage(gpuAvg, gpu, snapshot, 2)
+
+		assert.Equal(t, 25.0, maxUsage)
+		assert.Equal(t, 10.0, gpuAvg.Engines["Render/3D"], "Should use delta for existing engine")
+		assert.Equal(t, 25.0, gpuAvg.Engines["NewEngine"], "Should use full value for new engine")
+	})
+}
+
+func TestUpdateInstantaneousValues(t *testing.T) {
+	gm := &GPUManager{}
+
+	t.Run("updates temperature, memory used and total", func(t *testing.T) {
+		gpuAvg := &system.GPUData{
+			Temperature: 50.123,
+			MemoryUsed:  1000.456,
+			MemoryTotal: 8000.789,
+		}
+		gpu := &system.GPUData{
+			Temperature: 75.567,
+			MemoryUsed:  2500.891,
+			MemoryTotal: 8192.234,
+		}
+
+		gm.updateInstantaneousValues(gpuAvg, gpu)
+
+		assert.Equal(t, 75.57, gpuAvg.Temperature, "Should update and round temperature")
+		assert.Equal(t, 2500.89, gpuAvg.MemoryUsed, "Should update and round memory used")
+		assert.Equal(t, 8192.23, gpuAvg.MemoryTotal, "Should update and round memory total")
+	})
+}
+
+func TestStoreSnapshot(t *testing.T) {
+	gm := &GPUManager{
+		lastSnapshots: make(map[uint16]map[string]*gpuSnapshot),
+	}
+
+	t.Run("stores standard GPU snapshot", func(t *testing.T) {
+		cacheKey := uint16(5000)
+		gm.lastSnapshots[cacheKey] = make(map[string]*gpuSnapshot)
+
+		gpu := &system.GPUData{
+			Count:    10.0,
+			Usage:    150.5,
+			Power:    250.75,
+			PowerPkg: 300.25,
+		}
+
+		gm.storeSnapshot("0", gpu, cacheKey)
+
+		snapshot := gm.lastSnapshots[cacheKey]["0"]
+		assert.NotNil(t, snapshot)
+		assert.Equal(t, uint32(10), snapshot.count)
+		assert.Equal(t, 150.5, snapshot.usage)
+		assert.Equal(t, 250.75, snapshot.power)
+		assert.Equal(t, 300.25, snapshot.powerPkg)
+		assert.Nil(t, snapshot.engines, "Should not have engines for standard GPU")
+	})
+
+	t.Run("stores Intel GPU snapshot with engines", func(t *testing.T) {
+		cacheKey := uint16(10000)
+		gm.lastSnapshots[cacheKey] = make(map[string]*gpuSnapshot)
+
+		gpu := &system.GPUData{
+			Count:    5.0,
+			Usage:    100.0,
+			Power:    200.0,
+			PowerPkg: 250.0,
+			Engines: map[string]float64{
+				"Render/3D": 80.0,
+				"Video":     40.0,
+			},
+		}
+
+		gm.storeSnapshot("0", gpu, cacheKey)
+
+		snapshot := gm.lastSnapshots[cacheKey]["0"]
+		assert.NotNil(t, snapshot)
+		assert.Equal(t, uint32(5), snapshot.count)
+		assert.NotNil(t, snapshot.engines, "Should have engines for Intel GPU")
+		assert.Equal(t, 80.0, snapshot.engines["Render/3D"])
+		assert.Equal(t, 40.0, snapshot.engines["Video"])
+		assert.Len(t, snapshot.engines, 2)
+	})
+
+	t.Run("overwrites existing snapshot", func(t *testing.T) {
+		cacheKey := uint16(5000)
+		gm.lastSnapshots[cacheKey] = make(map[string]*gpuSnapshot)
+
+		// Store initial snapshot
+		gpu1 := &system.GPUData{Count: 5.0, Usage: 100.0, Power: 200.0}
+		gm.storeSnapshot("0", gpu1, cacheKey)
+
+		// Store updated snapshot
+		gpu2 := &system.GPUData{Count: 10.0, Usage: 250.0, Power: 400.0}
+		gm.storeSnapshot("0", gpu2, cacheKey)
+
+		snapshot := gm.lastSnapshots[cacheKey]["0"]
+		assert.Equal(t, uint32(10), snapshot.count, "Should overwrite previous count")
+		assert.Equal(t, 250.0, snapshot.usage, "Should overwrite previous usage")
+		assert.Equal(t, 400.0, snapshot.power, "Should overwrite previous power")
+	})
+}
+
+func TestCountGPUNames(t *testing.T) {
+	t.Run("returns empty map for no GPUs", func(t *testing.T) {
+		gm := &GPUManager{
+			GpuDataMap: make(map[string]*system.GPUData),
+		}
+		counts := gm.countGPUNames()
+		assert.Empty(t, counts)
+	})
+
+	t.Run("counts unique GPU names", func(t *testing.T) {
+		gm := &GPUManager{
+			GpuDataMap: map[string]*system.GPUData{
+				"0": {Name: "GPU A"},
+				"1": {Name: "GPU B"},
+				"2": {Name: "GPU C"},
+			},
+		}
+		counts := gm.countGPUNames()
+		assert.Equal(t, 1, counts["GPU A"])
+		assert.Equal(t, 1, counts["GPU B"])
+		assert.Equal(t, 1, counts["GPU C"])
+		assert.Len(t, counts, 3)
+	})
+
+	t.Run("counts duplicate GPU names", func(t *testing.T) {
+		gm := &GPUManager{
+			GpuDataMap: map[string]*system.GPUData{
+				"0": {Name: "RTX 4090"},
+				"1": {Name: "RTX 4090"},
+				"2": {Name: "RTX 4090"},
+				"3": {Name: "RTX 3080"},
+			},
+		}
+		counts := gm.countGPUNames()
+		assert.Equal(t, 3, counts["RTX 4090"])
+		assert.Equal(t, 1, counts["RTX 3080"])
+		assert.Len(t, counts, 2)
+	})
+}
+
+func TestInitializeSnapshots(t *testing.T) {
+	t.Run("initializes all maps from scratch", func(t *testing.T) {
+		gm := &GPUManager{}
+		cacheKey := uint16(5000)
+
+		gm.initializeSnapshots(cacheKey)
+
+		assert.NotNil(t, gm.lastAvgData)
+		assert.NotNil(t, gm.lastSnapshots)
+		assert.NotNil(t, gm.lastSnapshots[cacheKey])
+	})
+
+	t.Run("initializes only missing maps", func(t *testing.T) {
+		gm := &GPUManager{
+			lastAvgData: make(map[string]system.GPUData),
+		}
+		cacheKey := uint16(5000)
+
+		gm.initializeSnapshots(cacheKey)
+
+		assert.NotNil(t, gm.lastAvgData, "Should preserve existing lastAvgData")
+		assert.NotNil(t, gm.lastSnapshots)
+		assert.NotNil(t, gm.lastSnapshots[cacheKey])
+	})
+
+	t.Run("adds new cache key to existing snapshots", func(t *testing.T) {
+		existingKey := uint16(5000)
+		newKey := uint16(10000)
+
+		gm := &GPUManager{
+			lastSnapshots: map[uint16]map[string]*gpuSnapshot{
+				existingKey: {"0": {count: 10}},
+			},
+		}
+
+		gm.initializeSnapshots(newKey)
+
+		assert.NotNil(t, gm.lastSnapshots[existingKey], "Should preserve existing cache key")
+		assert.NotNil(t, gm.lastSnapshots[newKey], "Should add new cache key")
+		assert.NotNil(t, gm.lastSnapshots[existingKey]["0"], "Should preserve existing snapshot data")
+	})
+}
+
+func TestCalculateGPUAverage(t *testing.T) {
+	t.Run("returns zero value when deltaCount is zero", func(t *testing.T) {
+		gm := &GPUManager{
+			lastSnapshots: map[uint16]map[string]*gpuSnapshot{
+				5000: {
+					"0": {count: 10, usage: 100, power: 200},
+				},
+			},
+			lastAvgData: map[string]system.GPUData{
+				"0": {Usage: 50.0, Power: 100.0},
+			},
+		}
+
+		gpu := &system.GPUData{
+			Count: 10.0, // Same as snapshot, so delta = 0
+			Usage: 100.0,
+			Power: 200.0,
+		}
+
+		result := gm.calculateGPUAverage("0", gpu, 5000)
+
+		assert.Equal(t, 50.0, result.Usage, "Should return cached average")
+		assert.Equal(t, 100.0, result.Power, "Should return cached average")
+	})
+
+	t.Run("calculates average for standard GPU", func(t *testing.T) {
+		gm := &GPUManager{
+			lastSnapshots: map[uint16]map[string]*gpuSnapshot{
+				5000: {},
+			},
+			lastAvgData: make(map[string]system.GPUData),
+		}
+
+		gpu := &system.GPUData{
+			Name:  "Test GPU",
+			Count: 4.0,
+			Usage: 200.0, // 200 / 4 = 50
+			Power: 400.0, // 400 / 4 = 100
+		}
+
+		result := gm.calculateGPUAverage("0", gpu, 5000)
+
+		assert.Equal(t, 50.0, result.Usage)
+		assert.Equal(t, 100.0, result.Power)
+		assert.Equal(t, "Test GPU", result.Name)
+	})
+
+	t.Run("calculates average for Intel GPU with engines", func(t *testing.T) {
+		gm := &GPUManager{
+			lastSnapshots: map[uint16]map[string]*gpuSnapshot{
+				5000: {},
+			},
+			lastAvgData: make(map[string]system.GPUData),
+		}
+
+		gpu := &system.GPUData{
+			Name:     "Intel GPU",
+			Count:    5.0,
+			Power:    500.0,
+			PowerPkg: 600.0,
+			Engines: map[string]float64{
+				"Render/3D": 100.0, // 100 / 5 = 20
+				"Video":     50.0,  // 50 / 5 = 10
+			},
+		}
+
+		result := gm.calculateGPUAverage("0", gpu, 5000)
+
+		assert.Equal(t, 100.0, result.Power)
+		assert.Equal(t, 120.0, result.PowerPkg)
+		assert.Equal(t, 20.0, result.Usage, "Should use max engine usage")
+		assert.Equal(t, 20.0, result.Engines["Render/3D"])
+		assert.Equal(t, 10.0, result.Engines["Video"])
+	})
+
+	t.Run("calculates delta from previous snapshot", func(t *testing.T) {
+		gm := &GPUManager{
+			lastSnapshots: map[uint16]map[string]*gpuSnapshot{
+				5000: {
+					"0": {
+						count:    2,
+						usage:    50.0,
+						power:    100.0,
+						powerPkg: 120.0,
+					},
+				},
+			},
+			lastAvgData: make(map[string]system.GPUData),
+		}
+
+		gpu := &system.GPUData{
+			Name:     "Test GPU",
+			Count:    7.0,   // Delta = 7 - 2 = 5
+			Usage:    200.0, // Delta = 200 - 50 = 150, avg = 150/5 = 30
+			Power:    350.0, // Delta = 350 - 100 = 250, avg = 250/5 = 50
+			PowerPkg: 420.0, // Delta = 420 - 120 = 300, avg = 300/5 = 60
+		}
+
+		result := gm.calculateGPUAverage("0", gpu, 5000)
+
+		assert.Equal(t, 30.0, result.Usage)
+		assert.Equal(t, 50.0, result.Power)
+	})
+
+	t.Run("stores result in lastAvgData", func(t *testing.T) {
+		gm := &GPUManager{
+			lastSnapshots: map[uint16]map[string]*gpuSnapshot{
+				5000: {},
+			},
+			lastAvgData: make(map[string]system.GPUData),
+		}
+
+		gpu := &system.GPUData{
+			Count: 2.0,
+			Usage: 100.0,
+			Power: 200.0,
+		}
+
+		result := gm.calculateGPUAverage("0", gpu, 5000)
+
+		assert.Equal(t, result, gm.lastAvgData["0"], "Should store calculated average")
 	})
 }
 
@@ -765,7 +1298,8 @@ func TestAccumulation(t *testing.T) {
 			}
 
 			// Verify average calculation in GetCurrentData
-			result := gm.GetCurrentData()
+			cacheKey := uint16(5000)
+			result := gm.GetCurrentData(cacheKey)
 			for id, expected := range tt.expectedValues {
 				gpu, exists := result[id]
 				assert.True(t, exists, "GPU with ID %s should exist in GetCurrentData result", id)
@@ -778,16 +1312,16 @@ func TestAccumulation(t *testing.T) {
 				assert.EqualValues(t, expected.avgPower, gpu.Power, "Average power in GetCurrentData should match")
 			}
 
-			// Verify that accumulators in the original map are reset
+			// Verify that accumulators in the original map are NOT reset (they keep growing)
 			for id, expected := range tt.expectedValues {
 				gpu, exists := gm.GpuDataMap[id]
 				assert.True(t, exists, "GPU with ID %s should still exist after GetCurrentData", id)
 				if !exists {
 					continue
 				}
-				assert.EqualValues(t, 1, gpu.Count, "Count should be reset for GPU ID %s", id)
-				assert.EqualValues(t, expected.avgUsage, gpu.Usage, "Usage should be reset for GPU ID %s", id)
-				assert.EqualValues(t, expected.avgPower, gpu.Power, "Power should be reset for GPU ID %s", id)
+				assert.EqualValues(t, expected.count, gpu.Count, "Count should remain at accumulated value for GPU ID %s", id)
+				assert.EqualValues(t, expected.usage, gpu.Usage, "Usage should remain at accumulated value for GPU ID %s", id)
+				assert.EqualValues(t, expected.power, gpu.Power, "Power should remain at accumulated value for GPU ID %s", id)
 			}
 		})
 	}

agent/handlers.go

@@ -0,0 +1,101 @@
+package agent
+
+import (
+	"errors"
+	"fmt"
+
+	"github.com/fxamacker/cbor/v2"
+	"github.com/henrygd/beszel/internal/common"
+)
+
+// HandlerContext provides context for request handlers
+type HandlerContext struct {
+	Client      *WebSocketClient
+	Agent       *Agent
+	Request     *common.HubRequest[cbor.RawMessage]
+	RequestID   *uint32
+	HubVerified bool
+	// SendResponse abstracts how a handler sends responses (WS or SSH)
+	SendResponse func(data any, requestID *uint32) error
+}
+
+// RequestHandler defines the interface for handling specific websocket request types
+type RequestHandler interface {
+	// Handle processes the request and returns an error if unsuccessful
+	Handle(hctx *HandlerContext) error
+}
+
+// Responder sends handler responses back to the hub (over WS or SSH)
+type Responder interface {
+	SendResponse(data any, requestID *uint32) error
+}
+
+// HandlerRegistry manages the mapping between actions and their handlers
+type HandlerRegistry struct {
+	handlers map[common.WebSocketAction]RequestHandler
+}
+
+// NewHandlerRegistry creates a new handler registry with default handlers
+func NewHandlerRegistry() *HandlerRegistry {
+	registry := &HandlerRegistry{
+		handlers: make(map[common.WebSocketAction]RequestHandler),
+	}
+
+	registry.Register(common.GetData, &GetDataHandler{})
+	registry.Register(common.CheckFingerprint, &CheckFingerprintHandler{})
+
+	return registry
+}
+
+// Register registers a handler for a specific action type
+func (hr *HandlerRegistry) Register(action common.WebSocketAction, handler RequestHandler) {
+	hr.handlers[action] = handler
+}
+
+// Handle routes the request to the appropriate handler
+func (hr *HandlerRegistry) Handle(hctx *HandlerContext) error {
+	handler, exists := hr.handlers[hctx.Request.Action]
+	if !exists {
+		return fmt.Errorf("unknown action: %d", hctx.Request.Action)
+	}
+
+	// Check verification requirement - default to requiring verification
+	if hctx.Request.Action != common.CheckFingerprint && !hctx.HubVerified {
+		return errors.New("hub not verified")
+	}
+
+	// Log handler execution for debugging
+	// slog.Debug("Executing handler", "action", hctx.Request.Action)
+
+	return handler.Handle(hctx)
+}
+
+// GetHandler returns the handler for a specific action
+func (hr *HandlerRegistry) GetHandler(action common.WebSocketAction) (RequestHandler, bool) {
+	handler, exists := hr.handlers[action]
+	return handler, exists
+}
+
+////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////
+
+// GetDataHandler handles system data requests
+type GetDataHandler struct{}
+
+func (h *GetDataHandler) Handle(hctx *HandlerContext) error {
+	var options common.DataRequestOptions
+	_ = cbor.Unmarshal(hctx.Request.Data, &options)
+
+	sysStats := hctx.Agent.gatherStats(options.CacheTimeMs)
+	return hctx.SendResponse(sysStats, hctx.RequestID)
+}
+
+////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////
+
+// CheckFingerprintHandler handles authentication challenges
+type CheckFingerprintHandler struct{}
+
+func (h *CheckFingerprintHandler) Handle(hctx *HandlerContext) error {
+	return hctx.Client.handleAuthChallenge(hctx.Request, hctx.RequestID)
+}

agent/handlers_test.go

@@ -0,0 +1,112 @@
+//go:build testing
+// +build testing
+
+package agent
+
+import (
+	"testing"
+
+	"github.com/fxamacker/cbor/v2"
+	"github.com/henrygd/beszel/internal/common"
+	"github.com/stretchr/testify/assert"
+)
+
+// MockHandler for testing
+type MockHandler struct {
+	requiresVerification bool
+	description          string
+	handleFunc           func(ctx *HandlerContext) error
+}
+
+func (m *MockHandler) Handle(ctx *HandlerContext) error {
+	if m.handleFunc != nil {
+		return m.handleFunc(ctx)
+	}
+	return nil
+}
+
+func (m *MockHandler) RequiresVerification() bool {
+	return m.requiresVerification
+}
+
+// TestHandlerRegistry tests the handler registry functionality
+func TestHandlerRegistry(t *testing.T) {
+	t.Run("default registration", func(t *testing.T) {
+		registry := NewHandlerRegistry()
+
+		// Check default handlers are registered
+		getDataHandler, exists := registry.GetHandler(common.GetData)
+		assert.True(t, exists)
+		assert.IsType(t, &GetDataHandler{}, getDataHandler)
+
+		fingerprintHandler, exists := registry.GetHandler(common.CheckFingerprint)
+		assert.True(t, exists)
+		assert.IsType(t, &CheckFingerprintHandler{}, fingerprintHandler)
+	})
+
+	t.Run("custom handler registration", func(t *testing.T) {
+		registry := NewHandlerRegistry()
+		mockHandler := &MockHandler{
+			requiresVerification: true,
+			description:          "Test handler",
+		}
+
+		// Register a custom handler for a mock action
+		const mockAction common.WebSocketAction = 99
+		registry.Register(mockAction, mockHandler)
+
+		// Verify registration
+		handler, exists := registry.GetHandler(mockAction)
+		assert.True(t, exists)
+		assert.Equal(t, mockHandler, handler)
+	})
+
+	t.Run("unknown action", func(t *testing.T) {
+		registry := NewHandlerRegistry()
+		ctx := &HandlerContext{
+			Request: &common.HubRequest[cbor.RawMessage]{
+				Action: common.WebSocketAction(255), // Unknown action
+			},
+			HubVerified: true,
+		}
+
+		err := registry.Handle(ctx)
+		assert.Error(t, err)
+		assert.Contains(t, err.Error(), "unknown action: 255")
+	})
+
+	t.Run("verification required", func(t *testing.T) {
+		registry := NewHandlerRegistry()
+		ctx := &HandlerContext{
+			Request: &common.HubRequest[cbor.RawMessage]{
+				Action: common.GetData, // Requires verification
+			},
+			HubVerified: false, // Not verified
+		}
+
+		err := registry.Handle(ctx)
+		assert.Error(t, err)
+		assert.Contains(t, err.Error(), "hub not verified")
+	})
+}
+
+// TestCheckFingerprintHandler tests the CheckFingerprint handler
+func TestCheckFingerprintHandler(t *testing.T) {
+	handler := &CheckFingerprintHandler{}
+
+	t.Run("handle with invalid data", func(t *testing.T) {
+		client := &WebSocketClient{}
+		ctx := &HandlerContext{
+			Client:      client,
+			HubVerified: false,
+			Request: &common.HubRequest[cbor.RawMessage]{
+				Action: common.CheckFingerprint,
+				Data:   cbor.RawMessage{}, // Empty/invalid data
+			},
+		}
+
+		// Should fail to decode the fingerprint request
+		err := handler.Handle(ctx)
+		assert.Error(t, err)
+	})
+}

agent/network.go

@@ -12,8 +12,6 @@ import (
 	psutilNet "github.com/shirou/gopsutil/v4/net"
 )
 
-var netInterfaceDeltaTracker = deltatracker.NewDeltaTracker[string, uint64]()
-
 // NicConfig controls inclusion/exclusion of network interfaces via the NICS env var
 //
 // Behavior mirrors SensorConfig's matching logic:
@@ -77,75 +75,17 @@ func isValidNic(nicName string, cfg *NicConfig) bool {
 	return cfg.isBlacklist
 }
 
-func (a *Agent) updateNetworkStats(systemStats *system.Stats) {
+func (a *Agent) updateNetworkStats(cacheTimeMs uint16, systemStats *system.Stats) {
 	// network stats
-	if len(a.netInterfaces) == 0 {
-		// if no network interfaces, initialize again
-		// this is a fix if agent started before network is online (#466)
-		// maybe refactor this in the future to not cache interface names at all so we
-		// don't miss an interface that's been added after agent started in any circumstance
-		a.initializeNetIoStats()
-	}
+	a.ensureNetInterfacesInitialized()
 
-	if systemStats.NetworkInterfaces == nil {
-		systemStats.NetworkInterfaces = make(map[string][4]uint64, 0)
-	}
+	a.ensureNetworkInterfacesMap(systemStats)
 
 	if netIO, err := psutilNet.IOCounters(true); err == nil {
-		msElapsed := uint64(time.Since(a.netIoStats.Time).Milliseconds())
-		a.netIoStats.Time = time.Now()
-		totalBytesSent := uint64(0)
-		totalBytesRecv := uint64(0)
-		netInterfaceDeltaTracker.Cycle()
-		// sum all bytes sent and received
-		for _, v := range netIO {
-			// skip if not in valid network interfaces list
-			if _, exists := a.netInterfaces[v.Name]; !exists {
-				continue
-			}
-			totalBytesSent += v.BytesSent
-			totalBytesRecv += v.BytesRecv
-
-			// track deltas for each network interface
-			var upDelta, downDelta uint64
-			upKey, downKey := fmt.Sprintf("%sup", v.Name), fmt.Sprintf("%sdown", v.Name)
-			netInterfaceDeltaTracker.Set(upKey, v.BytesSent)
-			netInterfaceDeltaTracker.Set(downKey, v.BytesRecv)
-			if msElapsed > 0 {
-				upDelta = netInterfaceDeltaTracker.Delta(upKey) * 1000 / msElapsed
-				downDelta = netInterfaceDeltaTracker.Delta(downKey) * 1000 / msElapsed
-			}
-			// add interface to systemStats
-			systemStats.NetworkInterfaces[v.Name] = [4]uint64{upDelta, downDelta, v.BytesSent, v.BytesRecv}
-		}
-
-		// add to systemStats
-		var bytesSentPerSecond, bytesRecvPerSecond uint64
-		if msElapsed > 0 {
-			bytesSentPerSecond = (totalBytesSent - a.netIoStats.BytesSent) * 1000 / msElapsed
-			bytesRecvPerSecond = (totalBytesRecv - a.netIoStats.BytesRecv) * 1000 / msElapsed
-		}
-		networkSentPs := bytesToMegabytes(float64(bytesSentPerSecond))
-		networkRecvPs := bytesToMegabytes(float64(bytesRecvPerSecond))
-		// add check for issue (#150) where sent is a massive number
-		if networkSentPs > 10_000 || networkRecvPs > 10_000 {
-			slog.Warn("Invalid net stats. Resetting.", "sent", networkSentPs, "recv", networkRecvPs)
-			for _, v := range netIO {
-				if _, exists := a.netInterfaces[v.Name]; !exists {
-					continue
-				}
-				slog.Info(v.Name, "recv", v.BytesRecv, "sent", v.BytesSent)
-			}
-			// reset network I/O stats
-			a.initializeNetIoStats()
-		} else {
-			systemStats.NetworkSent = networkSentPs
-			systemStats.NetworkRecv = networkRecvPs
-			systemStats.Bandwidth[0], systemStats.Bandwidth[1] = bytesSentPerSecond, bytesRecvPerSecond
-			// update netIoStats
-			a.netIoStats.BytesSent = totalBytesSent
-			a.netIoStats.BytesRecv = totalBytesRecv
-		}
+		nis, msElapsed := a.loadAndTickNetBaseline(cacheTimeMs)
+		totalBytesSent, totalBytesRecv := a.sumAndTrackPerNicDeltas(cacheTimeMs, msElapsed, netIO, systemStats)
+		bytesSentPerSecond, bytesRecvPerSecond := a.computeBytesPerSecond(msElapsed, totalBytesSent, totalBytesRecv, nis)
+		a.applyNetworkTotals(cacheTimeMs, netIO, systemStats, nis, totalBytesSent, totalBytesRecv, bytesSentPerSecond, bytesRecvPerSecond)
 	}
 }
 
@@ -160,13 +100,8 @@ func (a *Agent) initializeNetIoStats() {
 		nicCfg = newNicConfig(nicsEnvVal)
 	}
 
-	// reset network I/O stats
-	a.netIoStats.BytesSent = 0
-	a.netIoStats.BytesRecv = 0
-
-	// get intial network I/O stats
+	// get current network I/O stats and record valid interfaces
 	if netIO, err := psutilNet.IOCounters(true); err == nil {
-		a.netIoStats.Time = time.Now()
 		for _, v := range netIO {
 			if nicsEnvExists && !isValidNic(v.Name, nicCfg) {
 				continue
@@ -175,12 +110,116 @@ func (a *Agent) initializeNetIoStats() {
 				continue
 			}
 			slog.Info("Detected network interface", "name", v.Name, "sent", v.BytesSent, "recv", v.BytesRecv)
-			a.netIoStats.BytesSent += v.BytesSent
-			a.netIoStats.BytesRecv += v.BytesRecv
 			// store as a valid network interface
 			a.netInterfaces[v.Name] = struct{}{}
 		}
 	}
+
+	// Reset per-cache-time trackers and baselines so they will reinitialize on next use
+	a.netInterfaceDeltaTrackers = make(map[uint16]*deltatracker.DeltaTracker[string, uint64])
+	a.netIoStats = make(map[uint16]system.NetIoStats)
+}
+
+// ensureNetInterfacesInitialized re-initializes NICs if none are currently tracked
+func (a *Agent) ensureNetInterfacesInitialized() {
+	if len(a.netInterfaces) == 0 {
+		// if no network interfaces, initialize again
+		// this is a fix if agent started before network is online (#466)
+		// maybe refactor this in the future to not cache interface names at all so we
+		// don't miss an interface that's been added after agent started in any circumstance
+		a.initializeNetIoStats()
+	}
+}
+
+// ensureNetworkInterfacesMap ensures systemStats.NetworkInterfaces map exists
+func (a *Agent) ensureNetworkInterfacesMap(systemStats *system.Stats) {
+	if systemStats.NetworkInterfaces == nil {
+		systemStats.NetworkInterfaces = make(map[string][4]uint64, 0)
+	}
+}
+
+// loadAndTickNetBaseline returns the NetIoStats baseline and milliseconds elapsed, updating time
+func (a *Agent) loadAndTickNetBaseline(cacheTimeMs uint16) (netIoStat system.NetIoStats, msElapsed uint64) {
+	netIoStat = a.netIoStats[cacheTimeMs]
+	if netIoStat.Time.IsZero() {
+		netIoStat.Time = time.Now()
+		msElapsed = 0
+	} else {
+		msElapsed = uint64(time.Since(netIoStat.Time).Milliseconds())
+		netIoStat.Time = time.Now()
+	}
+	return netIoStat, msElapsed
+}
+
+// sumAndTrackPerNicDeltas accumulates totals and records per-NIC up/down deltas into systemStats
+func (a *Agent) sumAndTrackPerNicDeltas(cacheTimeMs uint16, msElapsed uint64, netIO []psutilNet.IOCountersStat, systemStats *system.Stats) (totalBytesSent, totalBytesRecv uint64) {
+	tracker := a.netInterfaceDeltaTrackers[cacheTimeMs]
+	if tracker == nil {
+		tracker = deltatracker.NewDeltaTracker[string, uint64]()
+		a.netInterfaceDeltaTrackers[cacheTimeMs] = tracker
+	}
+	tracker.Cycle()
+
+	for _, v := range netIO {
+		if _, exists := a.netInterfaces[v.Name]; !exists {
+			continue
+		}
+		totalBytesSent += v.BytesSent
+		totalBytesRecv += v.BytesRecv
+
+		var upDelta, downDelta uint64
+		upKey, downKey := fmt.Sprintf("%sup", v.Name), fmt.Sprintf("%sdown", v.Name)
+		tracker.Set(upKey, v.BytesSent)
+		tracker.Set(downKey, v.BytesRecv)
+		if msElapsed > 0 {
+			upDelta = tracker.Delta(upKey) * 1000 / msElapsed
+			downDelta = tracker.Delta(downKey) * 1000 / msElapsed
+		}
+		systemStats.NetworkInterfaces[v.Name] = [4]uint64{upDelta, downDelta, v.BytesSent, v.BytesRecv}
+	}
+
+	return totalBytesSent, totalBytesRecv
+}
+
+// computeBytesPerSecond calculates per-second totals from elapsed time and totals
+func (a *Agent) computeBytesPerSecond(msElapsed, totalBytesSent, totalBytesRecv uint64, nis system.NetIoStats) (bytesSentPerSecond, bytesRecvPerSecond uint64) {
+	if msElapsed > 0 {
+		bytesSentPerSecond = (totalBytesSent - nis.BytesSent) * 1000 / msElapsed
+		bytesRecvPerSecond = (totalBytesRecv - nis.BytesRecv) * 1000 / msElapsed
+	}
+	return bytesSentPerSecond, bytesRecvPerSecond
+}
+
+// applyNetworkTotals validates and writes computed network stats, or resets on anomaly
+func (a *Agent) applyNetworkTotals(
+	cacheTimeMs uint16,
+	netIO []psutilNet.IOCountersStat,
+	systemStats *system.Stats,
+	nis system.NetIoStats,
+	totalBytesSent, totalBytesRecv uint64,
+	bytesSentPerSecond, bytesRecvPerSecond uint64,
+) {
+	networkSentPs := bytesToMegabytes(float64(bytesSentPerSecond))
+	networkRecvPs := bytesToMegabytes(float64(bytesRecvPerSecond))
+	if networkSentPs > 10_000 || networkRecvPs > 10_000 {
+		slog.Warn("Invalid net stats. Resetting.", "sent", networkSentPs, "recv", networkRecvPs)
+		for _, v := range netIO {
+			if _, exists := a.netInterfaces[v.Name]; !exists {
+				continue
+			}
+			slog.Info(v.Name, "recv", v.BytesRecv, "sent", v.BytesSent)
+		}
+		a.initializeNetIoStats()
+		delete(a.netIoStats, cacheTimeMs)
+		delete(a.netInterfaceDeltaTrackers, cacheTimeMs)
+	}
+
+	systemStats.NetworkSent = networkSentPs
+	systemStats.NetworkRecv = networkRecvPs
+	systemStats.Bandwidth[0], systemStats.Bandwidth[1] = bytesSentPerSecond, bytesRecvPerSecond
+	nis.BytesSent = totalBytesSent
+	nis.BytesRecv = totalBytesRecv
+	a.netIoStats[cacheTimeMs] = nis
 }
 
 func (a *Agent) skipNetworkInterface(v psutilNet.IOCountersStat) bool {

agent/network_test.go

@@ -4,7 +4,11 @@ package agent
 
 import (
 	"testing"
+	"time"
 
+	"github.com/henrygd/beszel/agent/deltatracker"
+	"github.com/henrygd/beszel/internal/entities/system"
+	psutilNet "github.com/shirou/gopsutil/v4/net"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
@@ -257,3 +261,202 @@ func TestNewNicConfig(t *testing.T) {
 		})
 	}
 }
+func TestEnsureNetworkInterfacesMap(t *testing.T) {
+	var a Agent
+	var stats system.Stats
+
+	// Initially nil
+	assert.Nil(t, stats.NetworkInterfaces)
+	// Ensure map is created
+	a.ensureNetworkInterfacesMap(&stats)
+	assert.NotNil(t, stats.NetworkInterfaces)
+	// Idempotent
+	a.ensureNetworkInterfacesMap(&stats)
+	assert.NotNil(t, stats.NetworkInterfaces)
+}
+
+func TestLoadAndTickNetBaseline(t *testing.T) {
+	a := &Agent{netIoStats: make(map[uint16]system.NetIoStats)}
+
+	// First call initializes time and returns 0 elapsed
+	ni, elapsed := a.loadAndTickNetBaseline(100)
+	assert.Equal(t, uint64(0), elapsed)
+	assert.False(t, ni.Time.IsZero())
+
+	// Store back what loadAndTick returns to mimic updateNetworkStats behavior
+	a.netIoStats[100] = ni
+
+	time.Sleep(2 * time.Millisecond)
+
+	// Next call should produce >= 0 elapsed and update time
+	ni2, elapsed2 := a.loadAndTickNetBaseline(100)
+	assert.True(t, elapsed2 > 0)
+	assert.False(t, ni2.Time.IsZero())
+}
+
+func TestComputeBytesPerSecond(t *testing.T) {
+	a := &Agent{}
+
+	// No elapsed -> zero rate
+	bytesUp, bytesDown := a.computeBytesPerSecond(0, 2000, 3000, system.NetIoStats{BytesSent: 1000, BytesRecv: 1000})
+	assert.Equal(t, uint64(0), bytesUp)
+	assert.Equal(t, uint64(0), bytesDown)
+
+	// With elapsed -> per-second calculation
+	bytesUp, bytesDown = a.computeBytesPerSecond(500, 6000, 11000, system.NetIoStats{BytesSent: 1000, BytesRecv: 1000})
+	// (6000-1000)*1000/500 = 10000; (11000-1000)*1000/500 = 20000
+	assert.Equal(t, uint64(10000), bytesUp)
+	assert.Equal(t, uint64(20000), bytesDown)
+}
+
+func TestSumAndTrackPerNicDeltas(t *testing.T) {
+	a := &Agent{
+		netInterfaces:             map[string]struct{}{"eth0": {}, "wlan0": {}},
+		netInterfaceDeltaTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+	}
+
+	// Two samples for same cache interval to verify delta behavior
+	cache := uint16(42)
+	net1 := []psutilNet.IOCountersStat{{Name: "eth0", BytesSent: 1000, BytesRecv: 2000}}
+	stats1 := &system.Stats{}
+	a.ensureNetworkInterfacesMap(stats1)
+	tx1, rx1 := a.sumAndTrackPerNicDeltas(cache, 0, net1, stats1)
+	assert.Equal(t, uint64(1000), tx1)
+	assert.Equal(t, uint64(2000), rx1)
+
+	// Second cycle with elapsed, larger counters -> deltas computed inside
+	net2 := []psutilNet.IOCountersStat{{Name: "eth0", BytesSent: 4000, BytesRecv: 9000}}
+	stats := &system.Stats{}
+	a.ensureNetworkInterfacesMap(stats)
+	tx2, rx2 := a.sumAndTrackPerNicDeltas(cache, 1000, net2, stats)
+	assert.Equal(t, uint64(4000), tx2)
+	assert.Equal(t, uint64(9000), rx2)
+	// Up/Down deltas per second should be (4000-1000)/1s = 3000 and (9000-2000)/1s = 7000
+	ni, ok := stats.NetworkInterfaces["eth0"]
+	assert.True(t, ok)
+	assert.Equal(t, uint64(3000), ni[0])
+	assert.Equal(t, uint64(7000), ni[1])
+}
+
+func TestApplyNetworkTotals(t *testing.T) {
+	tests := []struct {
+		name                  string
+		bytesSentPerSecond    uint64
+		bytesRecvPerSecond    uint64
+		totalBytesSent        uint64
+		totalBytesRecv        uint64
+		expectReset           bool
+		expectedNetworkSent   float64
+		expectedNetworkRecv   float64
+		expectedBandwidthSent uint64
+		expectedBandwidthRecv uint64
+	}{
+		{
+			name:                  "Valid network stats - normal values",
+			bytesSentPerSecond:    1000000, // 1 MB/s
+			bytesRecvPerSecond:    2000000, // 2 MB/s
+			totalBytesSent:        10000000,
+			totalBytesRecv:        20000000,
+			expectReset:           false,
+			expectedNetworkSent:   0.95, // ~1 MB/s rounded to 2 decimals
+			expectedNetworkRecv:   1.91, // ~2 MB/s rounded to 2 decimals
+			expectedBandwidthSent: 1000000,
+			expectedBandwidthRecv: 2000000,
+		},
+		{
+			name:               "Invalid network stats - sent exceeds threshold",
+			bytesSentPerSecond: 11000000000, // ~10.5 GB/s > 10 GB/s threshold
+			bytesRecvPerSecond: 1000000,     // 1 MB/s
+			totalBytesSent:     10000000,
+			totalBytesRecv:     20000000,
+			expectReset:        true,
+		},
+		{
+			name:               "Invalid network stats - recv exceeds threshold",
+			bytesSentPerSecond: 1000000,     // 1 MB/s
+			bytesRecvPerSecond: 11000000000, // ~10.5 GB/s > 10 GB/s threshold
+			totalBytesSent:     10000000,
+			totalBytesRecv:     20000000,
+			expectReset:        true,
+		},
+		{
+			name:               "Invalid network stats - both exceed threshold",
+			bytesSentPerSecond: 12000000000, // ~11.4 GB/s
+			bytesRecvPerSecond: 13000000000, // ~12.4 GB/s
+			totalBytesSent:     10000000,
+			totalBytesRecv:     20000000,
+			expectReset:        true,
+		},
+		{
+			name:                  "Valid network stats - at threshold boundary",
+			bytesSentPerSecond:    10485750000, // ~9999.99 MB/s (rounds to 9999.99)
+			bytesRecvPerSecond:    10485750000, // ~9999.99 MB/s (rounds to 9999.99)
+			totalBytesSent:        10000000,
+			totalBytesRecv:        20000000,
+			expectReset:           false,
+			expectedNetworkSent:   9999.99,
+			expectedNetworkRecv:   9999.99,
+			expectedBandwidthSent: 10485750000,
+			expectedBandwidthRecv: 10485750000,
+		},
+		{
+			name:                  "Zero values",
+			bytesSentPerSecond:    0,
+			bytesRecvPerSecond:    0,
+			totalBytesSent:        0,
+			totalBytesRecv:        0,
+			expectReset:           false,
+			expectedNetworkSent:   0.0,
+			expectedNetworkRecv:   0.0,
+			expectedBandwidthSent: 0,
+			expectedBandwidthRecv: 0,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			// Setup agent with initialized maps
+			a := &Agent{
+				netInterfaces:             make(map[string]struct{}),
+				netIoStats:                make(map[uint16]system.NetIoStats),
+				netInterfaceDeltaTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+			}
+
+			cacheTimeMs := uint16(100)
+			netIO := []psutilNet.IOCountersStat{
+				{Name: "eth0", BytesSent: 1000, BytesRecv: 2000},
+			}
+			systemStats := &system.Stats{}
+			nis := system.NetIoStats{}
+
+			a.applyNetworkTotals(
+				cacheTimeMs,
+				netIO,
+				systemStats,
+				nis,
+				tt.totalBytesSent,
+				tt.totalBytesRecv,
+				tt.bytesSentPerSecond,
+				tt.bytesRecvPerSecond,
+			)
+
+			if tt.expectReset {
+				// Should have reset network tracking state - delta trackers should be cleared
+				// Note: initializeNetIoStats resets the maps, then applyNetworkTotals sets nis back
+				assert.Contains(t, a.netIoStats, cacheTimeMs, "cache entry should exist after reset")
+				assert.NotContains(t, a.netInterfaceDeltaTrackers, cacheTimeMs, "tracker should be cleared on reset")
+			} else {
+				// Should have applied stats
+				assert.Equal(t, tt.expectedNetworkSent, systemStats.NetworkSent)
+				assert.Equal(t, tt.expectedNetworkRecv, systemStats.NetworkRecv)
+				assert.Equal(t, tt.expectedBandwidthSent, systemStats.Bandwidth[0])
+				assert.Equal(t, tt.expectedBandwidthRecv, systemStats.Bandwidth[1])
+
+				// Should have updated NetIoStats
+				updatedNis := a.netIoStats[cacheTimeMs]
+				assert.Equal(t, tt.totalBytesSent, updatedNis.BytesSent)
+				assert.Equal(t, tt.totalBytesRecv, updatedNis.BytesRecv)
+			}
+		})
+	}
+}

agent/server.go

@@ -127,15 +127,75 @@ func (a *Agent) handleSession(s ssh.Session) {
 
 	hubVersion := a.getHubVersion(sessionID, sessionCtx)
 
-	stats := a.gatherStats(sessionID)
-
-	err := a.writeToSession(s, stats, hubVersion)
-	if err != nil {
-		slog.Error("Error encoding stats", "err", err, "stats", stats)
-		s.Exit(1)
-	} else {
-		s.Exit(0)
+	// Legacy one-shot behavior for older hubs
+	if hubVersion.LT(beszel.MinVersionAgentResponse) {
+		if err := a.handleLegacyStats(s, hubVersion); err != nil {
+			slog.Error("Error encoding stats", "err", err)
+			s.Exit(1)
+			return
+		}
 	}
+
+	var req common.HubRequest[cbor.RawMessage]
+	if err := cbor.NewDecoder(s).Decode(&req); err != nil {
+		// Fallback to legacy one-shot if the first decode fails
+		if err2 := a.handleLegacyStats(s, hubVersion); err2 != nil {
+			slog.Error("Error encoding stats (fallback)", "err", err2)
+			s.Exit(1)
+			return
+		}
+		s.Exit(0)
+		return
+	}
+	if err := a.handleSSHRequest(s, &req); err != nil {
+		slog.Error("SSH request handling failed", "err", err)
+		s.Exit(1)
+		return
+	}
+	s.Exit(0)
+}
+
+// handleSSHRequest builds a handler context and dispatches to the shared registry
+func (a *Agent) handleSSHRequest(w io.Writer, req *common.HubRequest[cbor.RawMessage]) error {
+	// SSH does not support fingerprint auth action
+	if req.Action == common.CheckFingerprint {
+		return cbor.NewEncoder(w).Encode(common.AgentResponse{Error: "unsupported action"})
+	}
+
+	// responder that writes AgentResponse to stdout
+	sshResponder := func(data any, requestID *uint32) error {
+		response := common.AgentResponse{Id: requestID}
+		switch v := data.(type) {
+		case *system.CombinedData:
+			response.SystemData = v
+		default:
+			response.Error = fmt.Sprintf("unsupported response type: %T", data)
+		}
+		return cbor.NewEncoder(w).Encode(response)
+	}
+
+	ctx := &HandlerContext{
+		Client:       nil,
+		Agent:        a,
+		Request:      req,
+		RequestID:    nil,
+		HubVerified:  true,
+		SendResponse: sshResponder,
+	}
+
+	if handler, ok := a.handlerRegistry.GetHandler(req.Action); ok {
+		if err := handler.Handle(ctx); err != nil {
+			return cbor.NewEncoder(w).Encode(common.AgentResponse{Error: err.Error()})
+		}
+		return nil
+	}
+	return cbor.NewEncoder(w).Encode(common.AgentResponse{Error: fmt.Sprintf("unknown action: %d", req.Action)})
+}
+
+// handleLegacyStats serves the legacy one-shot stats payload for older hubs
+func (a *Agent) handleLegacyStats(w io.Writer, hubVersion semver.Version) error {
+	stats := a.gatherStats(60_000)
+	return a.writeToSession(w, stats, hubVersion)
 }
 
 // writeToSession encodes and writes system statistics to the session.

agent/system.go

@@ -14,12 +14,18 @@ import (
 	"github.com/henrygd/beszel/internal/entities/system"
 
 	"github.com/shirou/gopsutil/v4/cpu"
-	"github.com/shirou/gopsutil/v4/disk"
 	"github.com/shirou/gopsutil/v4/host"
 	"github.com/shirou/gopsutil/v4/load"
 	"github.com/shirou/gopsutil/v4/mem"
 )
 
+// prevDisk stores previous per-device disk counters for a given cache interval
+type prevDisk struct {
+	readBytes  uint64
+	writeBytes uint64
+	at         time.Time
+}
+
 // Sets initial / non-changing values about the host system
 func (a *Agent) initializeSystemInfo() {
 	a.systemInfo.AgentVersion = beszel.Version
@@ -68,7 +74,7 @@ func (a *Agent) initializeSystemInfo() {
 }
 
 // Returns current info, stats about the host system
-func (a *Agent) getSystemStats() system.Stats {
+func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 	var systemStats system.Stats
 
 	// battery
@@ -77,11 +83,11 @@ func (a *Agent) getSystemStats() system.Stats {
 	}
 
 	// cpu percent
-	cpuPct, err := cpu.Percent(0, false)
-	if err != nil {
+	cpuPercent, err := getCpuPercent(cacheTimeMs)
+	if err == nil {
+		systemStats.Cpu = twoDecimals(cpuPercent)
+	} else {
 		slog.Error("Error getting cpu percent", "err", err)
-	} else if len(cpuPct) > 0 {
-		systemStats.Cpu = twoDecimals(cpuPct[0])
 	}
 
 	// load average
@@ -131,56 +137,13 @@ func (a *Agent) getSystemStats() system.Stats {
 	}
 
 	// disk usage
-	for _, stats := range a.fsStats {
-		if d, err := disk.Usage(stats.Mountpoint); err == nil {
-			stats.DiskTotal = bytesToGigabytes(d.Total)
-			stats.DiskUsed = bytesToGigabytes(d.Used)
-			if stats.Root {
-				systemStats.DiskTotal = bytesToGigabytes(d.Total)
-				systemStats.DiskUsed = bytesToGigabytes(d.Used)
-				systemStats.DiskPct = twoDecimals(d.UsedPercent)
-			}
-		} else {
-			// reset stats if error (likely unmounted)
-			slog.Error("Error getting disk stats", "name", stats.Mountpoint, "err", err)
-			stats.DiskTotal = 0
-			stats.DiskUsed = 0
-			stats.TotalRead = 0
-			stats.TotalWrite = 0
-		}
-	}
+	a.updateDiskUsage(&systemStats)
 
-	// disk i/o
-	if ioCounters, err := disk.IOCounters(a.fsNames...); err == nil {
-		for _, d := range ioCounters {
-			stats := a.fsStats[d.Name]
-			if stats == nil {
-				continue
-			}
-			secondsElapsed := time.Since(stats.Time).Seconds()
-			readPerSecond := bytesToMegabytes(float64(d.ReadBytes-stats.TotalRead) / secondsElapsed)
-			writePerSecond := bytesToMegabytes(float64(d.WriteBytes-stats.TotalWrite) / secondsElapsed)
-			// check for invalid values and reset stats if so
-			if readPerSecond < 0 || writePerSecond < 0 || readPerSecond > 50_000 || writePerSecond > 50_000 {
-				slog.Warn("Invalid disk I/O. Resetting.", "name", d.Name, "read", readPerSecond, "write", writePerSecond)
-				a.initializeDiskIoStats(ioCounters)
-				break
-			}
-			stats.Time = time.Now()
-			stats.DiskReadPs = readPerSecond
-			stats.DiskWritePs = writePerSecond
-			stats.TotalRead = d.ReadBytes
-			stats.TotalWrite = d.WriteBytes
-			// if root filesystem, update system stats
-			if stats.Root {
-				systemStats.DiskReadPs = stats.DiskReadPs
-				systemStats.DiskWritePs = stats.DiskWritePs
-			}
-		}
-	}
+	// disk i/o (cache-aware per interval)
+	a.updateDiskIo(cacheTimeMs, &systemStats)
 
-	// network stats
-	a.updateNetworkStats(&systemStats)
+	// network stats (per cache interval)
+	a.updateNetworkStats(cacheTimeMs, &systemStats)
 
 	// temperatures
 	// TODO: maybe refactor to methods on systemStats
@@ -191,7 +154,7 @@ func (a *Agent) getSystemStats() system.Stats {
 		// reset high gpu percent
 		a.systemInfo.GpuPct = 0
 		// get current GPU data
-		if gpuData := a.gpuManager.GetCurrentData(); len(gpuData) > 0 {
+		if gpuData := a.gpuManager.GetCurrentData(cacheTimeMs); len(gpuData) > 0 {
 			systemStats.GPUData = gpuData
 
 			// add temperatures

agent/test-data/container.json

@@ -0,0 +1,24 @@
+{
+	"cpu_stats": {
+		"cpu_usage": {
+			"total_usage": 312055276000
+		},
+		"system_cpu_usage": 1366399830000000
+	},
+	"memory_stats": {
+		"usage": 507400192,
+		"stats": {
+			"inactive_file": 165130240
+		}
+	},
+	"networks": {
+		"eth0": {
+			"tx_bytes": 20376558,
+			"rx_bytes": 537029455
+		},
+		"eth1": {
+			"tx_bytes": 2003766,
+			"rx_bytes": 6241
+		}
+	}
+}

agent/test-data/container2.json

@@ -0,0 +1,24 @@
+{
+	"cpu_stats": {
+		"cpu_usage": {
+			"total_usage": 314891801000
+		},
+		"system_cpu_usage": 1368474900000000
+	},
+	"memory_stats": {
+		"usage": 507400192,
+		"stats": {
+			"inactive_file": 165130240
+		}
+	},
+	"networks": {
+		"eth0": {
+			"tx_bytes": 20376558,
+			"rx_bytes": 537029455
+		},
+		"eth1": {
+			"tx_bytes": 2003766,
+			"rx_bytes": 6241
+		}
+	}
+}