tinygo-org / bluetooth

Cross-platform Bluetooth API for Go and TinyGo. Supports Linux, macOS, Windows, and bare metal using Nordic SoftDevice or HCI
https://tinygo.org
Other
761 stars 138 forks source link

dbus/decode high memory use/leak #296

Open AndyRN opened 1 month ago

AndyRN commented 1 month ago

Using pprof, the following was generated from a Golang application running on Linux (arm64). The application is quite simple, and mirrors mostly what the heartrate-monitor example portrays:

pprof

Over time, memory usage will slowly climb as the loop continues to accept connections, decode payloads, and disconnect from devices - which can be reconnected each time they advertise an update.

Possibly related issues:

AndyRN commented 4 weeks ago

After some investigation, it appears to be this block of code: https://github.com/tinygo-org/bluetooth/blob/release/gap_linux.go#L362

err := device.device.Call("org.bluez.Device1.Connect", 0).Err
if err != nil {
    return Device{}, fmt.Errorf("bluetooth: failed to connect: %w", err)
}

The call to the bluez device has the potential to hang indefinitely if there is an issue with the bluetooth device, and my particular device is proving to be quite flakey.

Really this needs to make use of the ConnectionTimeout parameters that are passed into the Connect() call, unless I'm misunderstanding where they're meant to be referenced.

Testing the change below seems to solve it for me, though unsure what wider impact this may have on the rest of the library:

ctx, ctxCancel := context.WithCancel(context.Background())
ctxTimer := time.AfterFunc(params.ConnectionTimeout.AsTimeDuration(), ctxCancel)
err := device.device.CallWithContext(ctx, "org.bluez.Device1.Connect", 0).Err
if err != nil {
    return Device{}, fmt.Errorf("bluetooth: failed to connect: %w", err)
}
ctxTimer.Stop()

Would appreciate any thoughts/opinions on this, and if anyone can see if there's a reason to not timeout the connection attempts?

deadprogram commented 4 weeks ago

@aykevl any thoughts on this?

AndyRN commented 3 weeks ago

goroutines

Related to the high memory usage, but potentially unrelated to the connection issue detailed above, there is a high number of goroutines being continuously started without being cleaned up which I believe is related to the scanning.

After some experimentation, restarting the scan callback on the adapter every hour or so prevents this steady climb of goroutines, which somewhat confirms the theory, however doesn't explain where the misbehaving code lies in the library (or any underlying dependency).