Restarting causes values in internal queues to be lost

[phijor]

Values that are put into internal input queues here

metricq-combinator/src/combinator.cpp

Lines 187 to 195 in 5d985cd

	for (MetricInputNode* input_node : input_list)
	{
	for (metricq::TimeValue tv : data)
	{
	input_node->put(tv);
	}
	Log::trace() << fmt::format("└── Put data into queue ({:p}), now has length {}",
	(void*)input_node, input_node->queue_length());
	}

but are still waiting for other values to arrive are lost when restarting the combinator.

[tilsche]

I think the idiomatic solution would be to not acknowledge message until their value is no longer needed. That requires a carefully chosen qos setting and a tracking of acknowledgements within the combinator. The latter is not possible with the current metricq/master (maybe with the async-db branch). We also should never send the same value twice in the combinator. So yeah, it's challenging.