Watchdog Timeout Sizing

Watchdog Timeout Sizing

animation
actuators
Try Leave Safety factor at 1.5, then reduce Worst-case loop time (ms).
Observe The opening recommended timeout is 495 ms. Reducing Worst-case loop time (ms) lowers it.
Explain Worst-case loop time (ms) multiplied by Safety factor produces the timeout. The range control snaps the requested 325 ms to 330 ms, which gives 495 ms at a factor of 1.5.

Technical boundaries. The sizing rule assumes the entered worst-case duration is representative. The output is rounded to whole milliseconds and does not automatically check the separate actuator safety constraint described below.

Watchdog Timeout Sizing

Scenario: An IoT-controlled automated warehouse uses a robotic arm to pick items from shelves. The control loop performs these steps every cycle:

  1. Read load sensor (5 ms)
  2. Calculate pick trajectory (20 ms)
  3. Send motor commands via CAN bus (15 ms)
  4. Wait for arm position feedback (30 ms)
  5. Verify grip force sensor (5 ms)
  6. Log to SD card (10 ms)

Total expected loop time: 85 ms

Question: What watchdog timer (WDT) timeout should you configure to detect control loop hangs without triggering false positives?

Step 1: Account for worst-case execution variance

Real-world timing varies due to: - CAN bus arbitration delays (priority collisions) - SD card write delays (wear leveling, flash erase) - Interrupt handling (network stack, sensor polling)

Measured worst-case loop times: - Typical: 85 ms - 95th percentile: 120 ms - 99th percentile: 180 ms - Worst-case loop (75 ms of other work plus a 250 ms SD card write): 325 ms

Step 2: Add safety margin

Watchdog timeout = Worst-case time × Safety factor

Using 1.5× safety factor: 325 ms × 1.5 = 487.5 ms

Step 3: Validate against actuator safety constraints

If the control loop hangs, what is the maximum safe time before forcing a system reset?

  • Robot arm moving at 0.5 m/s
  • Collision hazard if control lost for >300 ms (150mm uncontrolled travel)

487.5 ms candidate watchdog timeout exceeds the 300 ms safe-stop limit by 187.5 ms!

Step 4: Redesign control architecture

The 250 ms SD card write is the bottleneck creating the safety conflict. Solution: Decouple logging from critical loop

// BEFORE: Unsafe monolithic loop
void control_loop() {
    read_sensors();        // 5ms
    calculate_path();      // 20ms
    send_motor_commands(); // 15ms
    wait_for_feedback();   // 30ms
    verify_grip();         // 5ms
    log_to_sd();          // 10-250ms <-- PROBLEM!
    feed_watchdog();
}

// AFTER: Two-tier architecture
void critical_control_loop() {  // Runs every 100ms
    read_sensors();             // 5ms
    calculate_path();           // 20ms
    send_motor_commands();      // 15ms
    wait_for_feedback();        // 30ms
    verify_grip();              // 5ms
    queue_log_event();          // <1ms (just adds to queue)
    feed_watchdog();            // WDT timeout = 150ms (100ms × 1.5)
}

void background_logger() {      // Runs in lower-priority task
    if (log_queue_not_empty()) {
        write_to_sd_card();     // 10-250ms (doesn't block control)
    }
}

Final watchdog configuration:

esp_task_wdt_config_t cfg = { .timeout_ms = 150, .idle_core_mask = 0, .trigger_panic = true };
esp_err_t wdt_result = esp_task_wdt_init(&cfg);  // ESP-IDF 5.x; call after the scheduler starts
if (wdt_result == ESP_ERR_INVALID_STATE) {
    wdt_result = esp_task_wdt_reconfigure(&cfg);  // Already initialized
}
if (wdt_result != ESP_OK) {
    // Handle configuration failure; do not claim the watchdog is active.
    return;
}
wdt_result = esp_task_wdt_add(NULL);  // Add current task to WDT
if (wdt_result != ESP_OK) {
    // Handle subscription failure before relying on the watchdog.
    return;
}

while (1) {
    critical_control_loop();     // Must complete in <150ms
    esp_task_wdt_reset();        // Feed watchdog
}

Verification: The ideal timer-only travel is 150 ms × 0.5 m/s = 75 mm (within the 150 mm target). This is not a guaranteed stopping distance: measure reset and output latency, and use an independent motion stop.

Key lesson: Watchdog timeout is a safety-critical parameter. Always: 1. Measure real worst-case timing, not theoretical 2. Account for I/O delays (SD, network, sensors) 3. Ensure timeout is shorter than actuator safety constraint 4. Decouple slow non-critical operations from control loop

Back to the chapter