NotEnoughMemoryException while publishing "long" msg

[DorBenHarush]

Hi, I'm trying to publish long massages (Images) with micro-ros. I received this error:
terminate called after throwing an instance of 'eprosima::fastcdr::exception::NotEnoughMemoryException' what(): Not enough memory in the buffer stream Aborted (core dumped)
I checked the meaning and according to eProsima it means that my msg buffer memory is too small.
The buffer default is very small, I tried to publish a string and got the error for string longer than 2 characters.
How can I increase the buffer size?

Regards,
Dor.

[pablogs9]

Hello @DorBenHarush, please can you put here a list of command and/or code to reproduce this issue?
FastCDR is not used in micro-ROS firmware so I'm not understanding something here.
Please let us replicate this issue by providing a comprehensive guide to follow your steps.

[DorBenHarush]

Hi, @pablogs9 this is the code that generates the error. It works if the string is empty or for short strings like "Hi".
I used the normal commands for configure, build and flash (esp32).
CODE:

#include <stdio.h>
#include <unistd.h>

#include <rcl/rcl.h>
#include <rcl/error_handling.h>
#include <std_msgs/msg/string.h>

#include <rclc/rclc.h>
#include <rclc/executor.h>

#ifdef ESP_PLATFORM
#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
#endif

#define RCCHECK(fn) { rcl_ret_t temp_rc = fn; if((temp_rc != RCL_RET_OK)){printf("Failed status on line %d: %d. Aborting.\n",__LINE__,(int)temp_rc);vTaskDelete(NULL);}}
#define RCSOFTCHECK(fn) { rcl_ret_t temp_rc = fn; if((temp_rc != RCL_RET_OK)){printf("Failed status on line %d: %d. Continuing.\n",__LINE__,(int)temp_rc);}}

rcl_publisher_t publisher;
std_msgs__msg__String msg;

void timer_callback(rcl_timer_t * timer, int64_t last_call_time)
{
	UNUSED(last_call_time);
	if (timer != NULL) {
		printf("sent.\n");
		RCSOFTCHECK(rcl_publish(&publisher, &msg, NULL));
	}
}

void appMain(void * arg)
{
	rcl_allocator_t allocator = rcl_get_default_allocator();
	rclc_support_t support;

	// create init_options
	RCCHECK(rclc_support_init(&support, 0, NULL, &allocator));

	// create node
	rcl_node_t node = rcl_get_zero_initialized_node();
	RCCHECK(rclc_node_init_default(&node, "freertos_String_publisher", "", &support));

	// create publisher
	RCCHECK(rclc_publisher_init_default(
		&publisher,
		&node,
		ROSIDL_GET_MSG_TYPE_SUPPORT(std_msgs, msg, String),
		"freertos_String_publisher"));

	// create timer,
	rcl_timer_t timer = rcl_get_zero_initialized_timer();
	const unsigned int timer_timeout = 1000;
	RCCHECK(rclc_timer_init_default(
		&timer,
		&support,
		RCL_MS_TO_NS(timer_timeout),
		timer_callback));

	// create executor
	rclc_executor_t executor = rclc_executor_get_zero_initialized_executor();
	RCCHECK(rclc_executor_init(&executor, &support.context, 1, &allocator));

	unsigned int rcl_wait_timeout = 100;   // in ms
	RCCHECK(rclc_executor_set_timeout(&executor, RCL_MS_TO_NS(rcl_wait_timeout)));
	RCCHECK(rclc_executor_add_timer(&executor, &timer));

	msg.data.data = "hello";

	while(1){
		rclc_executor_spin_some(&executor, 100);
		usleep(100000);
	}

	// free resources
	RCCHECK(rcl_publisher_fini(&publisher, &node))
	RCCHECK(rcl_node_fini(&node))

  	vTaskDelete(NULL);
}

[pablogs9]

You are not initializing the string structure nor allocating memory for it.

Try something like:

#define STR_SIZE 20

std_msgs__msg__String msg;

msg.data.capacity = STR_SIZE;
msg.data.size = 0;
msg.data.data = (char*) malloc(STR_SIZE*sizeof(char));

char * my_string = "hello";
memcpy(msg.data.data, my_string, strlen(my_string));
msg.data.size = strlen(my_string);

You also have convenience functions like std_msgs__msg__String__init to do this initialization, please check them along with your workspace.

Please close if resolved

[DorBenHarush]

Hi @pablogs9 , I'm trying to publish an Image massage and I have some questions. when I try to publish the full image the publisher returns 1 , so I reduced the size to several times and it worked with 800 bytes. Do you know why?
secondly I couldn't publish from inside the timer callback only from main.

Regards,
Dor.

CODE:

#include <esp_event_loop.h>
#include <esp_log.h>
#include <esp_system.h>
#include <nvs_flash.h>
#include <sys/param.h>
#include <string.h>

#include <stdio.h>
#include <unistd.h>

#include <rcl/rcl.h>
#include <rcl/error_handling.h>
#include <sensor_msgs/msg/image.h>

#include <rclc/rclc.h>
#include <rclc/executor.h>

#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
#include "boards.h"

#define STRING_CAPACITY 20

static const char *TAG = "example:take_picture";

#define RCCHECK(fn) { rcl_ret_t temp_rc = fn; if((temp_rc != RCL_RET_OK)){printf("Failed status on line %d: %d. Aborting.\n",__LINE__,(int)temp_rc);vTaskDelete(NULL);}}
#define RCSOFTCHECK(fn) { rcl_ret_t temp_rc = fn; if((temp_rc != RCL_RET_OK)){printf("Failed status on line %d: %d. Continuing.\n",__LINE__,(int)temp_rc);}}

rcl_publisher_t publisher;
sensor_msgs__msg__Image* msg;
camera_fb_t *pic;


static camera_config_t camera_config = {
    .pin_pwdn = CAM_PIN_PWDN,
    .pin_reset = CAM_PIN_RESET,
    .pin_xclk = CAM_PIN_XCLK,
    .pin_sscb_sda = CAM_PIN_SIOD,
    .pin_sscb_scl = CAM_PIN_SIOC,

    .pin_d7 = CAM_PIN_D7,
    .pin_d6 = CAM_PIN_D6,
    .pin_d5 = CAM_PIN_D5,
    .pin_d4 = CAM_PIN_D4,
    .pin_d3 = CAM_PIN_D3,
    .pin_d2 = CAM_PIN_D2,
    .pin_d1 = CAM_PIN_D1,
    .pin_d0 = CAM_PIN_D0,
    .pin_vsync = CAM_PIN_VSYNC,
    .pin_href = CAM_PIN_HREF,
    .pin_pclk = CAM_PIN_PCLK,

    //XCLK 20MHz or 10MHz for OV2640 double FPS (Experimental)
    .xclk_freq_hz = 20000000,
    .ledc_timer = LEDC_TIMER_0,
    .ledc_channel = LEDC_CHANNEL_0,

    .pixel_format = PIXFORMAT_YUV422, //YUV422,GRAYSCALE,RGB565,JPEG
    .frame_size = FRAMESIZE_QQVGA,    //QQVGA-UXGA Do not use sizes above QVGA when not JPEG

    .jpeg_quality = 12, //0-63 lower number means higher quality
    .fb_count = 1       //if more than one, i2s runs in continuous mode. Use only with JPEG
};

void timer_callback(rcl_timer_t * timer, int64_t last_call_time)
{
	UNUSED(last_call_time);
	if (timer != NULL) {
		
	}
}

static esp_err_t init_camera()
{
    //initialize the camera
    esp_err_t err = esp_camera_init(&camera_config);
    if (err != ESP_OK)
    {
        ESP_LOGE(TAG, "Camera Init Failed");
        return err;
    }

    return ESP_OK;
}

void appMain(void * arg)
{
    init_camera();
    rcl_allocator_t allocator = rcl_get_default_allocator();
	rclc_support_t support;
	msg = sensor_msgs__msg__Image__create();
	// create init_options
	RCCHECK(rclc_support_init(&support, 0, NULL, &allocator));

	// create node
	rcl_node_t node = rcl_get_zero_initialized_node();
	RCCHECK(rclc_node_init_default(&node, "freertos_picture_publisher", "", &support));

	// create publisher
	RCCHECK(rclc_publisher_init_default(
		&publisher,
		&node,
		ROSIDL_GET_MSG_TYPE_SUPPORT(sensor_msgs, msg, Image),
		"freertos_picture_publisher"));

	// create timer,
	rcl_timer_t timer = rcl_get_zero_initialized_timer();
	const unsigned int timer_timeout = 1000;
	RCCHECK(rclc_timer_init_default(
		&timer,
		&support,
		RCL_MS_TO_NS(timer_timeout),
		timer_callback));

	// create executor
	rclc_executor_t executor = rclc_executor_get_zero_initialized_executor();
	RCCHECK(rclc_executor_init(&executor, &support.context, 1, &allocator));

	unsigned int rcl_wait_timeout = 1000;   // in ms
	RCCHECK(rclc_executor_set_timeout(&executor, RCL_MS_TO_NS(rcl_wait_timeout)));
	RCCHECK(rclc_executor_add_timer(&executor, &timer));

	//takes picture in yuv422 format
	pic = esp_camera_fb_get();	
	
	//set height, width, step and is_bigedian
	msg->height = pic->height;
	msg->width = pic->width;
	msg->is_bigendian = 0;
	msg->step = pic->width * 2;

	//set pixel format and frame id
	char * encoding = "yuv422";
	msg->encoding.data = (char*)heap_caps_malloc(strlen(encoding) + 1, MALLOC_CAP_8BIT); 
	msg->header.frame_id.capacity = STRING_CAPACITY;
	memset(msg->encoding.data, 0, strlen(encoding) + 1);
	memcpy(msg->encoding.data, encoding, strlen(encoding));
	msg->encoding.size = strlen(encoding) + 1;

	char* frame_id = "AI_thinker_image";
	msg->header.frame_id.data = (char*)heap_caps_malloc(strlen(frame_id) + 1, MALLOC_CAP_8BIT);
	msg->header.frame_id.capacity = STRING_CAPACITY;
	memset(msg->header.frame_id.data, 0, strlen(frame_id) + 1);
	memcpy(msg->header.frame_id.data, frame_id, strlen(frame_id));
	msg->header.frame_id.size = strlen(frame_id) + 1;


	//allocate memory for data
	msg->data.data = (uint8_t*)heap_caps_malloc(pic->len, MALLOC_CAP_8BIT);
	msg->data.capacity = pic->len;
	msg->data.size = 1000;//msg->step * pic->height ;
	printf("copying data...\n");
	memcpy(msg->data.data, pic->buf, pic->len);
	if(msg->data.data == NULL){
		printf("failed to allocate memory. restating in 5 sec.\n");
		vTaskDelay(5000 / portTICK_RATE_MS);
		esp_restart();
	}	
	//msg->header.stamp ?

	printf("height: %d \n", msg->height);
	printf("width: %d \n", msg->width);
	printf("pixel format %s \n", msg->encoding.data);
	printf("is bigendian %d \n", msg->is_bigendian);
	printf("picture step %d \n", msg->step);
	printf("frame id %s \n", msg->header.frame_id.data);
	printf("buffer length %d \n", msg->data.size);
	printf("pic->len %d \n", pic->len);

	
	while(1){
		rclc_executor_spin_some(&executor, 1000);
		usleep(100000);
		//publish picture
		printf("publisher return value: %d \n", rcl_publish(&publisher, msg, NULL));
       	vTaskDelay(1000 / portTICK_RATE_MS);
	}

    // free resources
	RCCHECK(rcl_publisher_fini(&publisher, &node))
	RCCHECK(rcl_node_fini(&node))
	heap_caps_free(msg->encoding.data);
	heap_caps_free(msg->header.frame_id.data);
	heap_caps_free(msg->data.data);
	sensor_msgs__msg__Image__destroy(msg);
	vTaskDelete(NULL);
}

[pablogs9]

Some points:

• You are not setting msg->encoding.data.capacity.
• You are setting capacity to STRING_CAPACITY but the real capacity is the memory that you have allocated, which is based in strlen().
• You are setting msg->data.size = 1000 but you are copying pic->len bytes, here you should set the capacity to the length of the memory that you are allocating and the size to the actual size of the byte array that you are copying. Also, I think that it should be checked if the size is less than the capacity.
• I guess that the msg->data.data == NULL check should be done just after the memory allocation because allocation can fail and you are setting memory before checking the correctness of the allocation.

There are some other important points to take into account when dealing with big messages. First of all, you should be aware that the default MTU for the UDP packages that the micro-ROS middleware sends is 512 Bytes. The middleware also has a "historic" which is by default 4. So, the internal middleware layer uses a buffer (from static memory) of 512x4 = 2048 Bytes. Check this for more info

Whit that said, you have to be aware that if you are using a reliable publisher (rclc_publisher_init_default()) and the message length is greater than the MTU, it will be fragmented into the "historic" slots available. In the other case, if you are using best effort publisher (rclc_publisher_init_best_effort()) the message length must be less than the MTU.

So, it is important to know the maximum size of your messages, and then you can configure the RMW in the colcon.meta file of your application. See a sample here.

If you want to tune your RMW buffers for publication it should looks like:

{
    "names": {
        "rmw_microxrcedds": {
            "cmake-args": [
                   ...
                "-DRMW_UXRCE_STREAM_HISTORY=4",
                   ...
            ]
        },
        "microxrcedds_client": {
            "cmake-args": [
                   ...
                "-DUCLIENT_UDP_TRANSPORT_MTU=512",
                   ...
            ]
        }
    }
}

Take care of the formatting of this JSON, it often fails just because of an error in a comma.

You have here two options:

• Increase the MTU, so the client will send bigger packages to the agent
• Increase the history, so the client will have more available "historic" slots to fragment you whole message into 512 chunks that will be sent to the agent

Of course, you can also increase both.

[DorBenHarush]

I changed the parameters as you suggested but it didn't work. Furthermore, I found out that my defaults were 512 and 1.
800 bytes msg works with 512 and 1 so I don't know what the problem is.

[pablogs9]

800 Bytes works with 512 Bytes of MTU and 1 RMW_UXRCE_STREAM_HISTORY? So, I have not understood the problem... Is your client publishing? Can you receive data using ros2 topic echo ...?

[DorBenHarush]

Yes, I receive the data.

[pablogs9]

Ok cool, can we close the issue or you have another doubt?

[DorBenHarush]

the problem is that I cant publish data bigger than 800 bytes even if i increase the parameters to 10 and 4096.

[pablogs9]

One more thing, when you modify the colcon.meta file are you cleaning all builds?

Usually when modifying the micro-ROS configuration you have to do:

ros2 run micro_ros_setup build_firmware.sh clean  #clean is only valid for the ESP32 port
ros2 run micro_ros_setup configure_firmware.sh [config]
ros2 run micro_ros_setup build_firmware.sh

You can check if your changes has been effective looking:
include/rmw_microxrcedds_c/config.h and
include/uxr/client/config.h

please make sure that in those files you see the values that you have set in your colcon.metas

[DorBenHarush]

I checked files and the values didnt change even after doing the steps you mentioned, so I chnaged the values inside
include/rmw_microxrcedds_c/config.h and
include/uxr/client/config.h
to make sure that values are correct and it didnt work.