Check system stability and run only AP and OTA if unstable

code/espurna/config/general.h

@@ -57,6 +57,14 @@
 #define TERMINAL_SUPPORT         1               // Enable terminal commands
 #endif
 
+//------------------------------------------------------------------------------
+// CRASH
+//------------------------------------------------------------------------------
+
+#define CRASH_SAFE_TIME         60000           // The system is considered stable after these many millis
+#define CRASH_COUNT_MAX         5               // After this many crashes on boot
+                                                // the system is flagged as unstable
+
 //------------------------------------------------------------------------------
 // EEPROM
 //------------------------------------------------------------------------------
@@ -65,7 +73,8 @@
 #define EEPROM_RELAY_STATUS     0               // Address for the relay status (1 byte)
 #define EEPROM_ENERGY_COUNT     1               // Address for the energy counter (4 bytes)
 #define EEPROM_CUSTOM_RESET     5               // Address for the reset reason (1 byte)
-#define EEPROM_DATA_END         6               // End of custom EEPROM data block
+#define EEPROM_CRASH_COUNTER    6               // Address for the crash counter (1 byte)
+#define EEPROM_DATA_END         7               // End of custom EEPROM data block
 
 //------------------------------------------------------------------------------
 // HEARTBEAT

code/espurna/debug.ino

@@ -33,13 +33,15 @@ void debugSend(const char * format, ...) {
     #endif
 
     #if DEBUG_UDP_SUPPORT
-        udpDebug.beginPacket(DEBUG_UDP_IP, DEBUG_UDP_PORT);
-        udpDebug.write(buffer);
-        if (len > DEBUG_MESSAGE_MAX_LENGTH) {
-            udpDebug.write(" (...)\n");
+        if (systemCheck()) {
+            udpDebug.beginPacket(DEBUG_UDP_IP, DEBUG_UDP_PORT);
+            udpDebug.write(buffer);
+            if (len > DEBUG_MESSAGE_MAX_LENGTH) {
+                udpDebug.write(" (...)\n");
+            }
+            udpDebug.endPacket();
+            delay(1);
         }
-        udpDebug.endPacket();
-        delay(1);
     #endif
 
 }
@@ -64,13 +66,15 @@ void debugSend_P(PGM_P format, ...) {
     #endif
 
     #if DEBUG_UDP_SUPPORT
-        udpDebug.beginPacket(DEBUG_UDP_IP, DEBUG_UDP_PORT);
-        udpDebug.write(buffer);
-        if (len > DEBUG_MESSAGE_MAX_LENGTH) {
-            udpDebug.write(" (...)\n");
+        if (systemCheck()) {
+            udpDebug.beginPacket(DEBUG_UDP_IP, DEBUG_UDP_PORT);
+            udpDebug.write(buffer);
+            if (len > DEBUG_MESSAGE_MAX_LENGTH) {
+                udpDebug.write(" (...)\n");
+            }
+            udpDebug.endPacket();
+            delay(1);
         }
-        udpDebug.endPacket();
-        delay(1);
     #endif
 
 }

code/espurna/espurna.ino

@@ -47,6 +47,14 @@ void hardwareSetup() {
 
 void hardwareLoop() {
 
+    // System check
+    static bool checked = false;
+    if (!checked && (millis() > CRASH_SAFE_TIME)) {
+        // Check system as stable
+        systemCheck(true);
+        checked = true;
+    }
+
     // Heartbeat
     static unsigned long last_uptime = 0;
     if ((millis() - last_uptime > HEARTBEAT_INTERVAL) || (last_uptime == 0)) {
@@ -198,15 +206,29 @@ void welcome() {
 
 void setup() {
 
+    // Init EEPROM, Serial and SPIFFS
     hardwareSetup();
+
+    // Question system stability
+    systemCheck(false);
+
+    // Show welcome message and system configuration
     welcome();
 
+    // Init persistance and terminal features
     settingsSetup();
     if (getSetting("hostname").length() == 0) {
         setSetting("hostname", getIdentifier());
         saveSettings();
     }
 
+    delay(500);
+    wifiSetup();
+    otaSetup();
+
+    // Do not run the next services if system is flagged stable
+    if (!systemCheck()) return;
+
     #if WEB_SUPPORT
         webSetup();
     #endif
@@ -217,11 +239,6 @@ void setup() {
     relaySetup();
     buttonSetup();
     ledSetup();
-
-    delay(500);
-
-    wifiSetup();
-    otaSetup();
     mqttSetup();
 
     #ifdef ITEAD_SONOFF_RFBRIDGE
@@ -276,13 +293,17 @@ void setup() {
 void loop() {
 
     hardwareLoop();
+    settingsLoop();
+    wifiLoop();
+    otaLoop();
+
+    // Do not run the next services if system is flagged stable
+    if (!systemCheck()) return;
+
     buttonLoop();
     relayLoop();
     ledLoop();
-    wifiLoop();
-    otaLoop();
     mqttLoop();
-    settingsLoop();
 
     #ifdef ITEAD_SONOFF_RFBRIDGE
         rfbLoop();

code/espurna/utils.ino

@@ -122,6 +122,8 @@ void heartbeat() {
 
 }
 
+// -----------------------------------------------------------------------------
+
 void customReset(unsigned char status) {
     EEPROM.write(EEPROM_CUSTOM_RESET, status);
     EEPROM.commit();
@@ -137,6 +139,39 @@ unsigned char customReset() {
     return status;
 }
 
+// -----------------------------------------------------------------------------
+
+// Call this method on boot with start=true to increase the crash counter
+// Call it again once the system is stable to decrease the counter
+// If the counter reaches CRASH_COUNT_MAX then the system is flagged as unstable
+// setting _systemOK = false;
+//
+// An unstable system will only have serial access, WiFi in AP mode and OTA
+
+bool _systemStable = true;
+
+void systemCheck(bool stable) {
+    unsigned char value = EEPROM.read(EEPROM_CRASH_COUNTER);
+    if (stable) {
+        value = 0;
+        DEBUG_MSG_P(PSTR("[MAIN] System OK\n"));
+    } else {
+        if (++value > CRASH_COUNT_MAX) {
+            _systemStable = false;
+            value = 0;
+            DEBUG_MSG_P(PSTR("[MAIN] System UNSTABLE\n"));
+        }
+    }
+    EEPROM.write(EEPROM_CRASH_COUNTER, value);
+    EEPROM.commit();
+}
+
+bool systemCheck() {
+    return _systemStable;
+}
+
+// -----------------------------------------------------------------------------
+
 char * ltrim(char * s) {
     char *p = s;
     while ((unsigned char) *p == ' ') ++p;

code/espurna/wifi.ino

@@ -59,6 +59,9 @@ void wifiConfigure() {
     jw.setAPMode(WIFI_AP_MODE);
     jw.cleanNetworks();
 
+    // If system is flagged unstable we do not init wifi networks
+    if (!systemCheck()) return;
+
     int i;
     for (i = 0; i< WIFI_MAX_NETWORKS; i++) {
         if (getSetting("ssid" + String(i)).length() == 0) break;