For simplicity, HistoryMetricsService will maintain 72 hours coarse-grained data
for last 72 hours, and fine-grained data for past 5 min.
For the coarse-grained data of past 72 hours, one or two sample point will be stored
for each hour.
For Counter: we will store one data point per hour.
For Meter, we will store two data points per hour, with one point which have
max mean value, the other point with min mean value.
For Histogram: we will store two data points per hour, with one point which have
max mean value, the other point with min mean value.
It is designed like this so that we are able to maintain abnormal metrics pattern,
Like a sudden rise in latency, so a sudden drop in throughput.
For fine-grained data in last 5 min, there will be 1 sample point per 15 seconds.
For simplicity, HistoryMetricsService will maintain 72 hours coarse-grained data for last 72 hours, and fine-grained data for past 5 min.
For the coarse-grained data of past 72 hours, one or two sample point will be stored for each hour.
For Counter: we will store one data point per hour. For Meter, we will store two data points per hour, with one point which have max mean value, the other point with min mean value. For Histogram: we will store two data points per hour, with one point which have max mean value, the other point with min mean value.
It is designed like this so that we are able to maintain abnormal metrics pattern, Like a sudden rise in latency, so a sudden drop in throughput.
For fine-grained data in last 5 min, there will be 1 sample point per 15 seconds.