[Bug] Task Cleanup Race Condition and Reconnection Issues

[Vladislavert]

Describe the bug

Crash on Session Cleanup (ESP-IDF Backend)

Summary

When a zenoh-pico client on ESP32 loses connection to the zenohd router (e.g., router stops), the ESP32 crashes with a spinlock assertion failure during task cleanup.

Crash Log

assert failed: spinlock_acquire spinlock.h:142 (lock->count == 0)

Backtrace: 0x40376d92:0x3fcbba80 0x4037e3b1:0x3fcbbaa0 ...

0x4037ae73: xEventGroupSetBits at /idf/components/freertos/FreeRTOS-Kernel/event_groups.c:245
0x42015fda: z_task_wrapper at zenoh-pico/src/system/espidf/system.c:64

Root Cause

In src/system/espidf/system.c, z_task_wrapper calls vTaskDelete(NULL) immediately after xEventGroupSetBits.
If _z_task_free runs immediately (e.g. on another core), it frees the Event Group while xEventGroupSetBits is still using it.
This causes a crash when xEventGroupSetBits tries to access the freed memory.

Workaround

I patched src/system/espidf/system.c to use vTaskSuspend(NULL) instead of vTaskDelete(NULL), and modified _z_task_free() to properly wait for task completion before deleting resources:

// In z_task_wrapper:
vTaskSuspend(NULL);  // Instead of vTaskDelete(NULL)

// In _z_task_free:
xEventGroupWaitBits(ptr->join_event, 1, pdFALSE, pdFALSE, pdMS_TO_TICKS(500));
if (ptr->handle != NULL) { vTaskDelete(ptr->handle); ptr->handle = NULL; }
vEventGroupDelete(ptr->join_event);  // Instead of z_free(ptr->join_event)

Related Issue

This appears related to #1033.

To reproduce

1. Flash and run the standard z_pub example on ESP32
2. Start an instance of zenohd(docker run --rm --init --net host eclipse/zenoh:latest -l udp/0.0.0.0:7447)
3. Stop zenohd
4. Start zenohd again
5. Zenoh pico will crash due to FreeRTOS attempting to acquire a spinlock on freed memory

System info

• zenoh-pico version: 1.6.2
• zenohd version: 1.6.2
• Platform: ESP32-S3, ESP-IDF v5.x
• Transport: UDP unicast