ESP32-S3 heap decreases after Firebase Stream reconnects even with PSRAM enabled

[hasantahsindagli]

Hello everyone,

I’m investigating a progressive internal heap decrease on ESP32-S3 related to Firebase Stream reconnects, and I’d like to share exact code and log evidence to better understand whether this behavior is expected or avoidable.

🔧 Environment

MCU: ESP32-S3 N16R8 (8MB PSRAM)

Arduino core: 2.0.14

Firebase ESP Client: 4.4.14

TLS: BearSSL

PSRAM: enabled (compile-time & runtime)

All Firebase SSL RX/TX/response buffers are successfully allocated in PSRAM.

🔁 Stream Lifecycle (Code)

This is the exact reconnect sequence used in my project:

void stopFirebaseStream() {
if (Firebase.RTDB.isStream(&stream)) {
Serial.println("🔄 Stopping Firebase stream...");
Firebase.RTDB.endStream(&stream);
delay(200);
}
}

void startFirebaseStream() {
Serial.println("🔄 Starting Firebase stream...");
Firebase.RTDB.beginStream(&stream, "/HomeAutomation");
}

Heap is measured before cleanup, after cleanup, and after reconnect.

📊 Heap Measurement Code
uint32_t heapBefore = ESP.getFreeHeap();
stopFirebaseStream();
uint32_t heapAfterStop = ESP.getFreeHeap();

startFirebaseStream();
uint32_t heapAfterStart = ESP.getFreeHeap();

Serial.printf(
"Heap: before=%u KB | afterStop=%u KB | afterStart=%u KB\n",
heapBefore / 1024,
heapAfterStop / 1024,
heapAfterStart / 1024
);
🧪 Serial Log Output
🔄 Stopping Firebase stream...
Heap: before=255 KB | afterStop=247 KB | DIFF=-8 KB

🔄 Starting Firebase stream...
Heap: afterStart=248 KB

This loss is permanent. After multiple reconnects:

Reconnect #1 → -8 KB
Reconnect #2 → -16 KB
Reconnect #3 → -25 KB

After 3–4 reconnects, the device must be restarted to recover heap.

💾 PSRAM Status
PSRAM Free: 8127 KB / 8189 KB
PSRAM Usage remains stable across reconnects

This indicates the loss happens in internal heap, not PSRAM.

🧠 Observations

Same FirebaseData stream object is reused

Stream is not recreated, only ended and restarted

No dynamic allocations in user stream callback

JSON payload sizes are small

Heap loss happens even if no stream data is received

❓ Questions

Is this heap loss expected due to TLS / BearSSL reconnect behavior?

Is Firebase.RTDB.endStream() expected to fully release internal memory?

Is there a recommended stream reconnect pattern on ESP32-S3?

Would keeping the stream alive (without calling endStream) be safer?

Any clarification or best-practice guidance would be greatly appreciated.

Thanks in advance.

Best regards,

[mobizt]

That library you are mentioned is end of life and no further supports.

[hasantahsindagli]

Thanks for the clarification.

That explains the observed behavior.
I’ll proceed with a workaround on my side (persistent stream / controlled restart strategy) and will plan a migration away from this library for long-term stability.

Appreciate the confirmation.

Best regards