AxeOS based firmware updates fail

[skot]

In some cases the AxeOS dashboard firmware updater fails, leaving the Bitaxe in a state where it does not boot.
Usually the failure happens during the www.bin image update.
In all of these cases (afaik) the bitaxe can be recovered with a USB firmware flash.

[tdb3]

Do users encountering the issue see a particular error response (e.g. one of the 500 reason phrases returned)? Could help see what error is encountered frequently.

Looking at the code below, it would make sense that the Bitaxe is left in a state where it doesn't boot with the web interface available. The partition is being erased optimistically with the assumption that the subsequent write/update will work. If it doesn't, there wouldn't be a www.bin to load on next boot.

ESP-Miner/main/http_server/http_server.c

Lines 422 to 457 in 11107a3

	esp_err_t POST_WWW_update(httpd_req_t * req)
	{
	char buf[1000];
	int remaining = req->content_len;
	const esp_partition_t * www_partition =
	esp_partition_find_first(ESP_PARTITION_TYPE_DATA, ESP_PARTITION_SUBTYPE_DATA_SPIFFS, "www");
	if (www_partition == NULL) {
	httpd_resp_send_err(req, HTTPD_500_INTERNAL_SERVER_ERROR, "WWW partition not found");
	return ESP_FAIL;
	}
	// Erase the entire www partition before writing
	ESP_ERROR_CHECK(esp_partition_erase_range(www_partition, 0, www_partition->size));
	while (remaining > 0) {
	int recv_len = httpd_req_recv(req, buf, MIN(remaining, sizeof(buf)));
	if (recv_len == HTTPD_SOCK_ERR_TIMEOUT) {
	continue;
	} else if (recv_len <= 0) {
	httpd_resp_send_err(req, HTTPD_500_INTERNAL_SERVER_ERROR, "Protocol Error");
	return ESP_FAIL;
	}
	if (esp_partition_write(www_partition, www_partition->size - remaining, (const void *) buf, recv_len) != ESP_OK) {
	httpd_resp_send_err(req, HTTPD_500_INTERNAL_SERVER_ERROR, "Write Error");
	return ESP_FAIL;
	}
	remaining -= recv_len;
	}
	httpd_resp_sendstr(req, "WWW update complete\n");
	return ESP_OK;
	}

Not sure if we have enough space for this, but some devices handle this by having two partitions and flipping between the two. Updates are written to the partition not currently being used. On next boot, the updated partition is tried. If loading is unsuccessful, execution falls back / fails safe to the partition that still functions, and the update can be tried again.

One alternative, if space is a luxury, would be to have three partitions, two that are very minimal in size to support bare-bones recovery, and the last to contain the bulk of the app.

POST_OTA_update() appears to do something similar with OTA updates (juggling more than one OTA partition).

ESP-Miner/main/http_server/http_server.c

Lines 462 to 506 in 11107a3

	esp_err_t POST_OTA_update(httpd_req_t * req)
	{
	char buf[1000];
	esp_ota_handle_t ota_handle;
	int remaining = req->content_len;
	const esp_partition_t * ota_partition = esp_ota_get_next_update_partition(NULL);
	ESP_ERROR_CHECK(esp_ota_begin(ota_partition, OTA_SIZE_UNKNOWN, &ota_handle));
	while (remaining > 0) {
	int recv_len = httpd_req_recv(req, buf, MIN(remaining, sizeof(buf)));
	// Timeout Error: Just retry
	if (recv_len == HTTPD_SOCK_ERR_TIMEOUT) {
	continue;
	// Serious Error: Abort OTA
	} else if (recv_len <= 0) {
	httpd_resp_send_err(req, HTTPD_500_INTERNAL_SERVER_ERROR, "Protocol Error");
	return ESP_FAIL;
	}
	// Successful Upload: Flash firmware chunk
	if (esp_ota_write(ota_handle, (const void *) buf, recv_len) != ESP_OK) {
	esp_ota_abort(ota_handle);
	httpd_resp_send_err(req, HTTPD_500_INTERNAL_SERVER_ERROR, "Flash Error");
	return ESP_FAIL;
	}
	remaining -= recv_len;
	}
	// Validate and switch to new OTA image and reboot
	if (esp_ota_end(ota_handle) != ESP_OK || esp_ota_set_boot_partition(ota_partition) != ESP_OK) {
	httpd_resp_send_err(req, HTTPD_500_INTERNAL_SERVER_ERROR, "Validation / Activation Error");
	return ESP_FAIL;
	}
	httpd_resp_sendstr(req, "Firmware update complete, rebooting now!\n");
	ESP_LOGI(TAG, "Restarting System because of Firmware update complete");
	vTaskDelay(1000 / portTICK_PERIOD_MS);
	esp_restart();
	return ESP_OK;
	}

[WantClue]

With recovery page this is been closed as fixed.