Files
AI_Printer/components/control_plane/src/controller_lifecycle.c

286 lines
8.7 KiB
C

#include "control_plane.h"
#include <string.h>
#include "esp_log.h"
#include "esp_timer.h"
#include "domain.h"
#include "freertos/FreeRTOS.h"
#include "freertos/semphr.h"
#include "freertos/task.h"
static const char *TAG = "controller_lifecycle";
static SemaphoreHandle_t s_lock;
static controller_lifecycle_status_t s_status = {
.state = CONTROLLER_LIFECYCLE_STATE_STOPPED,
.last_error = ESP_OK,
.start_attempt = 0,
.last_transition_ms = 0,
.last_stage = "stopped",
};
static void lifecycle_lock_init_if_needed(void) {
if (s_lock == NULL) {
s_lock = xSemaphoreCreateMutex();
}
}
static bool lifecycle_lock_take(uint32_t timeout_ms) {
lifecycle_lock_init_if_needed();
return s_lock != NULL && xSemaphoreTake(s_lock, pdMS_TO_TICKS(timeout_ms)) == pdTRUE;
}
static void lifecycle_set_state_locked(controller_lifecycle_state_t state,
esp_err_t last_error,
const char *stage) {
s_status.state = state;
s_status.last_error = last_error;
s_status.last_transition_ms = esp_timer_get_time() / 1000;
strlcpy(s_status.last_stage, stage != NULL ? stage : "unknown", sizeof(s_status.last_stage));
domain_diag_set_gauge(DOMAIN_DIAG_GAUGE_LIFECYCLE_STATE, (int32_t)state);
}
const char *controller_lifecycle_state_str(controller_lifecycle_state_t state) {
switch (state) {
case CONTROLLER_LIFECYCLE_STATE_STOPPED:
return "stopped";
case CONTROLLER_LIFECYCLE_STATE_STARTING:
return "starting";
case CONTROLLER_LIFECYCLE_STATE_RUNNING:
return "running";
case CONTROLLER_LIFECYCLE_STATE_DEGRADED:
return "degraded";
case CONTROLLER_LIFECYCLE_STATE_STOPPING:
return "stopping";
default:
return "unknown";
}
}
static esp_err_t lifecycle_start_system_runtime(void) {
return system_runtime_start();
}
static esp_err_t lifecycle_start_printer_protocol(void) {
return printer_protocol_init();
}
static esp_err_t lifecycle_start_voice(void) {
return voice_interaction_init();
}
static esp_err_t lifecycle_stop_printer_protocol(void) {
esp_err_t rc = printer_protocol_stop(domain_policy_printer_stop_timeout_ms());
if (rc == ESP_OK || rc == ESP_ERR_INVALID_STATE) {
return ESP_OK;
}
return rc;
}
static esp_err_t lifecycle_stop_voice(void) {
esp_err_t rc = voice_interaction_stop(NULL, 0);
if (rc == ESP_OK || rc == ESP_ERR_INVALID_STATE) {
return ESP_OK;
}
return rc;
}
static esp_err_t lifecycle_stop_system_runtime(void) {
return system_runtime_shutdown();
}
typedef esp_err_t (*lifecycle_step_fn_t)(void);
static esp_err_t lifecycle_run_step_with_retry(const char *stage, lifecycle_step_fn_t fn) {
uint32_t retry_count = domain_policy_lifecycle_start_retry_count();
uint32_t max_attempts = retry_count + 1;
for (uint32_t attempt = 0; attempt < max_attempts; ++attempt) {
esp_err_t rc = fn();
if (rc == ESP_OK) {
if (attempt > 0) {
ESP_LOGW(TAG, "stage %s recovered after retries, attempt=%u", stage, (unsigned)(attempt + 1));
}
return ESP_OK;
}
bool can_retry = (attempt + 1 < max_attempts) && domain_policy_is_retryable_error(rc);
ESP_LOGW(TAG,
"stage %s failed, rc=0x%x, attempt=%u/%u, retry=%d",
stage,
(unsigned)rc,
(unsigned)(attempt + 1),
(unsigned)max_attempts,
can_retry ? 1 : 0);
if (!can_retry) {
return rc;
}
domain_diag_counter_add(DOMAIN_DIAG_COUNTER_LIFECYCLE_START_RETRY, 1);
vTaskDelay(pdMS_TO_TICKS(domain_policy_lifecycle_retry_backoff_ms(attempt)));
}
return ESP_FAIL;
}
esp_err_t controller_lifecycle_start(void) {
domain_diag_counter_add(DOMAIN_DIAG_COUNTER_LIFECYCLE_START_ATTEMPT, 1);
if (!lifecycle_lock_take(1000)) {
domain_diag_counter_add(DOMAIN_DIAG_COUNTER_LIFECYCLE_START_FAILED, 1);
domain_diag_record_error("lifecycle_start", ESP_ERR_TIMEOUT, "lifecycle lock timeout");
return ESP_ERR_TIMEOUT;
}
if (s_status.state == CONTROLLER_LIFECYCLE_STATE_RUNNING) {
xSemaphoreGive(s_lock);
return ESP_OK;
}
if (s_status.state == CONTROLLER_LIFECYCLE_STATE_STARTING ||
s_status.state == CONTROLLER_LIFECYCLE_STATE_STOPPING) {
xSemaphoreGive(s_lock);
return ESP_ERR_INVALID_STATE;
}
s_status.start_attempt++;
lifecycle_set_state_locked(CONTROLLER_LIFECYCLE_STATE_STARTING, ESP_OK, "bootstrap");
xSemaphoreGive(s_lock);
bool system_started = false;
bool printer_started = false;
bool voice_started = false;
const char *failed_stage = "unknown";
failed_stage = "system_runtime";
esp_err_t rc = lifecycle_run_step_with_retry("system_runtime", lifecycle_start_system_runtime);
if (rc != ESP_OK) {
goto start_failed;
}
system_started = true;
failed_stage = "printer_protocol";
rc = lifecycle_run_step_with_retry("printer_protocol", lifecycle_start_printer_protocol);
if (rc != ESP_OK) {
goto start_failed;
}
printer_started = true;
failed_stage = "voice_interaction";
rc = lifecycle_run_step_with_retry("voice_interaction", lifecycle_start_voice);
if (rc != ESP_OK) {
goto start_failed;
}
voice_started = true;
image_generation_schedule_prewarm();
if (lifecycle_lock_take(1000)) {
lifecycle_set_state_locked(CONTROLLER_LIFECYCLE_STATE_RUNNING, ESP_OK, "running");
xSemaphoreGive(s_lock);
}
domain_diag_counter_add(DOMAIN_DIAG_COUNTER_LIFECYCLE_START_SUCCESS, 1);
ESP_LOGI(TAG, "controller lifecycle started");
return ESP_OK;
start_failed:
if (voice_started) {
(void)lifecycle_stop_voice();
}
if (printer_started) {
(void)lifecycle_stop_printer_protocol();
}
if (system_started) {
(void)lifecycle_stop_system_runtime();
}
if (lifecycle_lock_take(1000)) {
lifecycle_set_state_locked(CONTROLLER_LIFECYCLE_STATE_DEGRADED, rc, failed_stage);
xSemaphoreGive(s_lock);
}
domain_diag_counter_add(DOMAIN_DIAG_COUNTER_LIFECYCLE_START_FAILED, 1);
domain_diag_record_error("lifecycle_start", rc, failed_stage);
ESP_LOGE(TAG,
"controller lifecycle start failed at stage=%s, rc=0x%x",
failed_stage,
(unsigned)rc);
return rc;
}
esp_err_t controller_lifecycle_stop(void) {
domain_diag_counter_add(DOMAIN_DIAG_COUNTER_LIFECYCLE_STOP_ATTEMPT, 1);
if (!lifecycle_lock_take(1000)) {
domain_diag_counter_add(DOMAIN_DIAG_COUNTER_LIFECYCLE_STOP_FAILED, 1);
domain_diag_record_error("lifecycle_stop", ESP_ERR_TIMEOUT, "lifecycle lock timeout");
return ESP_ERR_TIMEOUT;
}
if (s_status.state == CONTROLLER_LIFECYCLE_STATE_STOPPED) {
xSemaphoreGive(s_lock);
return ESP_OK;
}
if (s_status.state == CONTROLLER_LIFECYCLE_STATE_STARTING ||
s_status.state == CONTROLLER_LIFECYCLE_STATE_STOPPING) {
xSemaphoreGive(s_lock);
return ESP_ERR_INVALID_STATE;
}
lifecycle_set_state_locked(CONTROLLER_LIFECYCLE_STATE_STOPPING, ESP_OK, "stopping");
xSemaphoreGive(s_lock);
esp_err_t first_err = ESP_OK;
esp_err_t rc = lifecycle_stop_voice();
if (rc != ESP_OK && first_err == ESP_OK) {
first_err = rc;
}
rc = lifecycle_stop_printer_protocol();
if (rc != ESP_OK && first_err == ESP_OK) {
first_err = rc;
}
rc = lifecycle_stop_system_runtime();
if (rc != ESP_OK && first_err == ESP_OK) {
first_err = rc;
}
if (lifecycle_lock_take(1000)) {
if (first_err == ESP_OK) {
lifecycle_set_state_locked(CONTROLLER_LIFECYCLE_STATE_STOPPED, ESP_OK, "stopped");
} else {
lifecycle_set_state_locked(CONTROLLER_LIFECYCLE_STATE_DEGRADED, first_err, "stop_failed");
}
xSemaphoreGive(s_lock);
}
if (first_err == ESP_OK) {
domain_diag_counter_add(DOMAIN_DIAG_COUNTER_LIFECYCLE_STOP_SUCCESS, 1);
ESP_LOGI(TAG, "controller lifecycle stopped");
} else {
domain_diag_counter_add(DOMAIN_DIAG_COUNTER_LIFECYCLE_STOP_FAILED, 1);
domain_diag_record_error("lifecycle_stop", first_err, "controller stop failed");
ESP_LOGE(TAG, "controller lifecycle stop failed, rc=0x%x", (unsigned)first_err);
}
return first_err;
}
void controller_lifecycle_get_status(controller_lifecycle_status_t *out_status) {
if (out_status == NULL) {
return;
}
memset(out_status, 0, sizeof(*out_status));
if (!lifecycle_lock_take(200)) {
return;
}
*out_status = s_status;
xSemaphoreGive(s_lock);
}