I recently had several hosts metrics plotted in the graph widget with a sum aggregation and aggregating the whole dataset. When visualized the graph was very choppy in a 90 day view. I was able to determine the cause to be that not all of the hosts  in the dataset had metrics as far back as 90 days. This becomes a problem in CSvgGraphHelper::getMetricsAggregatedData in this section here:

each ($result as $points) {
    $tick = 0;

    usort($points['data'],	
        static function (array $point_a, array $point_b): int {
            return $point_a['clock'] <=> $point_b['clock'];
        }
    );

    foreach ($points['data'] as $point) {
        if ($point['tick'] > ($tick + $approximation_tick_delta)) {
            $tick = $point['tick'];
        }
        if (array_key_exists('count', $point)) {
            $metric_points[$tick]['value'][] = $point['count'];
        }
        if (array_key_exists('value', $point)) {
            $metric_points[$tick]['value'][] = $point['value'];
        }
    }
} 

For each itemid, the $tick is initialized to 0. That means the first 'tick' value in the array during the foreach ($points['data'] as $point) { will always end up being set to $tick. This becomes a problem if one itemid has trends data for the full 90 days, and if another itemid only has trends data for less than 90 days. In other words, the final $tick array keys in the final $metric_points array for both items will be offset and cause extremely choppy and difficult to read graphs.

I think the solution is to run similar logic after the $metric_points array is finished by:

1. Running a ksort($metric_points) to sort the array by the tick (timestamps)
2. Initialize $tick to 0 again and iterate through $metric_points using the $approximation_tick_delta so that every tick key in the array is spaced apart equivalent to the $approximation_tick_delta.

Let me know if this makes sense or not.