The dashboard you can't stop refreshing.
Series mode gives Oliver the habit-forming part of a great metrics product: drag a time range, zoom from a month to a minute, slice by any label, and the chart is already there. Typical charts draw in under 10 milliseconds at every zoom level.
The loop that hooks people
Anyone who has lived in a good metrics dashboard knows the loop. Something looks off, so you drag across the spike. You zoom out to see if it happened last week, zoom back in to the minute it started, then split the line by region to see who it hit. Each move is a new question, and the habit only forms if every answer arrives before you have finished asking.
Most databases draw each chart from raw points, so a chart gets slower the further you zoom out. A month of data at 15-second resolution is 172,800 points per series, read again every time someone touches the time picker.
Ready at every zoom level
Series mode stores each metric and label set once, as a small number, and keeps ready-made summaries at 5 minutes, 1 hour, 4 hours and 1 day as the data arrives: count, sum, minimum, maximum, first and last values, and counter resets. A chart reads the coarsest summary that fits its time range, and the last few minutes come straight from memory.
The raw points are still there underneath. Zoom in far enough and Oliver draws every one of them, so nothing is averaged away to make the wide view fast.
Fast at every time range
We loaded the same data into a regular Oliver table and a series table, then drew the same charts from both.
A chart of the last day went from 744 milliseconds to 26. On the same data, a later build brought typical charts down to between 3.8 and 8.9 milliseconds at every range, and Oliver drew 785 charts a second from the series table against 4.6 from the regular one.
No more spinner
The chart people remember is the one that never loads. These are the slowest 1% of charts from the same runs.
The slowest chart of the last day dropped from 27.6 seconds to 0.22 seconds. That is the tile on the wall screen that used to spin, now drawing as fast as the rest.
Every label kept
Metrics get expensive when every pod, customer and endpoint becomes its own series, so teams strip labels to keep the count down. Series mode is built for the count to be high. With 450,000 distinct series loaded, a single-series chart still draws in 1.3 to 2.2 milliseconds at every time range, and Oliver does not price by the series.
For the questions that cut across thousands of series at once, series mode can also keep summaries per group. With them on, "sum by region over the last 7 days" across 225,000 series went from 196 milliseconds to 5.3, and hour-long group-bys ran 6.5 times faster.
Under a byte a point
Three days of 4,500 series, 173 million points, took 160 MB on object storage, under a byte per point. Served straight from object storage, Oliver drew 751 charts a second, against 785 from local disk. History loads at 2.3 to 2.7 million points a second.
Grafana works with it
Grafana connects to Oliver as a Prometheus data source and draws its panels from series mode, in PromQL.
Getting series mode
Series mode is available on request for Oliver deployed in your own cloud account. Tell us what you chart today and we will set it up with you.