Issue running many bumble devices

[jg-armaments]

My goal is to run multiple Android emulators in conjunction with multiple virtual peripherals. I encountered an issue where, when I create several devices simultaneously, not all of them appear to be running.

I can reproduce this with examples/run_gatt_server.py. The following steps create 8 devices. Usually 3, 4, or 5 will show up in nRF Connect's scanner. Never all 8. I see no errors when the gatt servers run.

Any advice would be greatly appreciated.

Reproduction steps:

1. Launch an Android emulator (my current AVD is AOSP API 35)
2. Install nRF Connect on the emulator and begin scanning.
3. Run the following script:

#!/usr/bin/env bash

# Launch many fake devices in background
python run_gatt_server.py device01.json android-netsim &
python run_gatt_server.py device02.json android-netsim &
python run_gatt_server.py device03.json android-netsim &
python run_gatt_server.py device04.json android-netsim &
python run_gatt_server.py device05.json android-netsim &
python run_gatt_server.py device06.json android-netsim &
python run_gatt_server.py device07.json android-netsim &
python run_gatt_server.py device08.json android-netsim &

# Terminate all launched processes on CTRL-C
trap 'kill -- -$$' EXIT

wait

The JSON files all look like this, each with different MAC addresses:

{
    "name": "Bumble",
    "address": "F0:F1:F2:F3:F4:F1",
    "advertising_interval": 1000,
    "keystore": "JsonKeyStore"
}

System details:
Mac OS Sequoia 15.1
Bumble 0.0.211
Android emulator version 35.4.9.0

[barbibulle]

This should work in theory. A couple things to try in order to troubleshoot what the issue may be:
1/ Run the same, with logging enabled (set the BUMBLE_LOGLEVEL=DEBUG env var)
2/ Run a Bumble scanner, see if you see the same devices as what the Android scanner does (python apps/scan.py android-netsim)

[barbibulle]

Also, double-check that the random MAC addrs you have assigned to your devices are indeed of the right type (there are two bits of MAC addresses that govern their type: the most significant bits should be 0b11 for a random static address for example)

[jg-armaments]

Wow, thanks for the fast response! I'll follow up on your recommendations and post any findings.

[jg-armaments]

Correction to previously provided information: according to the logs, it seems that this is Bumble version 0.0.210.

The bumble scanner always shows the same scan results as nRF Connect.

I can reliably reproduce the defect with just 3 devices (instead of 8 as originally described). The MAC addresses I'm using vary by the last digit:

F0:F1:F2:F3:F4:F1
F0:F1:F2:F3:F4:F2
F0:F1:F2:F3:F4:F3

Curiously, the results are different on each run. For example, running the script might show that only the device ending in F3 is advertising. Run again, you might see only F1 and F2 advertise. Run again, you might see only F2. On a rare occasion, all three successfully advertise.

I've attached a debug log where only F3 appeared. Everything in the log looks normal to me, but I am not deeply familiar with bumble.

3_devices_bumble.log

[zxzxwu]

The controller level emulation in this case is provided by Android Netsim / Rootcanal instead of Bumble. Here Bumble just sends HCI commands to the emulated controller, so you may need to check Android emulator launcher logs for more details.

[jg-armaments]

I found that introducing a random delay before device.start_advertising resolved the issue to my satisfaction. Maybe the root cause was advertisement collisions.

Thanks for quick responses.