loggerd: fixed the message queue growing infinitely when the incoming encoders less than max_waiting.

[deanlee]

Issue: offset_segment_num > s->rotate_segment will be always true if the incoming encoder types less than the max_waiting:

openpilot/selfdrive/loggerd/loggerd.cc

Lines 155 to 165 in ebd25b8

	} else if (offset_segment_num > s->rotate_segment) {
	// encoderd packet has a newer segment, this means encoderd has rolled over
	if (!re.marked_ready_to_rotate) {
	re.marked_ready_to_rotate = true;
	++s->ready_to_rotate;
	LOGD("rotate %d -> %d ready %d/%d for %s",
	s->rotate_segment.load(), offset_segment_num,
	s->ready_to_rotate.load(), s->max_waiting, name.c_str());
	}
	// queue up all the new segment messages, they go in after the rotate
	re.q.push_back(msg);

Set the max_waiting according to the incoming encoders to fix this issue, resolve #27512

[bleufa]

idk if this fixes the issue, I can still see the memory usage increasing in top

[deanlee]

it's lgtm:

==440055== LEAK SUMMARY:
==440055== definitely lost: 0 bytes in 0 blocks
==440055== indirectly lost: 0 bytes in 0 blocks

have you started encoderd before loggerd?

[bleufa]

i'm not running encoderd at all, just in offroad I can see the memory usage increasing after running for a few minutes. also see possibly lost bytes in valgrind

[bleufa]

can you see the memory usage increasing?

[deanlee]

Not yet. I can't see any place where memory leaks may occur If the encoderd is not running.