Session Agenda
Presentations
Notes
Hands On
Search
Menu
Presentations
Notes
Hands On
# Monitoring Math Getting more out of your graphs --- #Metric Aggregation --- # Timeseries Data * Metrics are stored as timeseries based on combinations of the metric name and tag combinations * Queries require an aggregation function (sum, min, max, avg) to aggregate over timeseries matching the query * Aggregation is applied at each point on the graph, bringing multiple timeseries value into one point on the graph --- # Aggregation Example * Start with a basic query “sum:system.disk.used{host:oracle-xe}” * Disk metrics are tagged by disk using the “device” tag, with an individual timeseries created on each host for every device reported <!-- .element: style="background: none; box-shadow: none; width : 100%" --> --- # Aggregation Example * Using “by {device}” break the graph down to show one stacked are per disk * Sum from previous graph is now broken down by individual disks making up the previous aggregate <!-- .element: style="background: none; box-shadow: none; width : 100%" --> --- # Functions --- # Function Types * Arithmetic * Interpolation * Timeshift * Rate * Count * Smoothing * Time Aggregation * Ranking * Algorithm --- # Arithmetic ## Absolute Value * Always displays the absolute value of the metric. ## Log2/Log10 * The occasional spike may skew the scale of a graph. Log will smooth out the spike while not eliminating them. ## Cumulative Sum / Integral * Adds each value shown on the graph cumulatively. --- # Interpolation - .fill() * Not meant to fill in empty spaces, but to align multiple series together * Helpful for aggregation and arithmetic across multiple sources * Available options include linear interpolation or inserting 0’s or the most recent value * More on this in our KBs: http://dtdg.co/kb-fill --- # Timeshift ## Helpful for comparing metrics with previously reported values --- # hour_before() <!-- .element: style="background: none; box-shadow: none; width : 100%" --> --- # day_before() <!-- .element: style="background: none; box-shadow: none; width : 100%" --> --- # week_before()  --- # month_before()  --- # Rate ## Compare metric values with their previous value --- # per_second() Compare the metric with the previous value, returning difference in value divided by the difference in time <!-- .element: style="background: none; box-shadow: none; width : 100%" --> --- # per_minute() Compare the metric with the previous value, returning difference in value divided by the difference in time <!-- .element: style="background: none; box-shadow: none; width : 100%" --> Note: 60 * the per_second() value --- # per_hour() Rate Compare the metric with the previous value, returning difference in value divided by the difference in time <!-- .element: style="background: none; box-shadow: none; width : 100%" --> Note: 3600 * the per_second() value --- # dt() <!-- .element: style="background: none; box-shadow: none; width : 100%" --> Note: Time in seconds between the current and previous points --- # diff() <!-- .element: style="background: none; box-shadow: none; width : 100%" --> Note: The difference in value between the current and previous points --- # derivative() <!-- .element: style="background: none; box-shadow: none; width : 100%" --> Note: 1st order derivative function applied to the graph: diff() / dt() --- # Count ## Keep track of the number of values reported in your graphs --- # count_nonzero() <!-- .element: style="background: none; box-shadow: none; width : 100%" --> Note: Current track of all non-zero values graphed at a given moment in time --- # count_not_null() <!-- .element: style="background: none; box-shadow: none; width : 100%" --> Note: Current track of all non-null values graphed at a given moment in time --- # Smoothing ## Even out the values on your graph --- # ewma_3()  Note: Exponentially weighted moving average spanning 3 points (previous, current, next) --- # ewma_5()  Note: Exponentially weighted moving average spanning 5 points --- # ewma_10()  Note: Exponentially weighted moving average spanning 10 points --- # ewma_20()  Note: Exponentially weighted moving average spanning 20 points --- # median_3()  Note: Median value spanning 3 points (median of the previous, current, and next point) --- # median_5()  Note: Median value spanning 5 points --- # median_7()  Note: Median value spanning 7 points --- # median_9()  Note: Median value spanning 9 points --- # Time Aggregation ## Smooth out your graphs by aggregating timeseries data --- # .rollup($method, $time) * Apply aggregation $method to multiple values over $time in seconds * Available methods: sum, min, max, avg * Allows overriding the default rollup function used for smoothing graphs over larger timeframes * Various methods available can allow preservation of spikes/dips in graphs * More on our default rollups and the rollup function’s use in our KBs: http://dtdg.co/zoomsmooth --- # Ranking ## Avoid clutter by viewing only the most relevant timeseries --- # top() Avoid clutter by viewing only the most relevant timeseries  Note: * Select the top series responsive to a given query, according to a provided ranking method * Parameters for the top function include * metric query itself * the number of series you’d like returned * a ranking function (one of 'max', 'min', 'last', 'l2norm', or 'area') * sort order (either 'desc' for descending order or 'asc' for ascending order) --- # top_offset() Avoid clutter by viewing only the most relevant timeseries  Note: Similar to top(), except with an additional offset parameter, which controls where in the ordered sequence of series the graphing starts. For example, an offset of 2 would start graphing at the number 3 ranked series, according to the chosen ranking metric. --- # Algorithm ## Analyze your metrics compared to concurrent or previous metric data --- # outliers() * Compare timeseries values with each other * Highlights individual timeseries based on ones that deviate too far from the norm at a given moment in time * More on the available DBSCAN and MAD functions used here: http://dtdg.co/outlierdetect --- # anomalies() * Compares historic metric data * Allows comparison of the current value with norms calculated across previous timeframes * Can alert on anomalous spikes/dips in metric values based on deviation from this norm
Back to the Agenda
Notes
|
Hands On Instruction
