Navigating the UI

All videos in this page are running on the topology set up in the Quick Start on Storm (producing 200 data records every second.).

Clicking the icon will display useful information.

A simple first query

The default new query will get a raw record with max result count 1 - it will return when a single record is found.

Results

Since there is no projection in this query, the results are shown as a JSON. You can click the Show as Table button to switch the mode.

Example: Picking a random record from the stream

Filtering and projecting data

The Filters section allows you to pick a slice of data from the data stream.

The Output Data section allows you to retrieve a subset of fields, and optionally rename them. You can also aggregate data, or choose to see raw data records.

Example: Finding and picking out fields from events that have probability > 0.5

Stream Raw Events

A simple but useful query is a query with a filter and a Sliding Window of size 1. This query will run for the extent of your duration and stream back events that match your filters as they arrive:

Be careful when you use this query to ensure that your filter is sufficient to avoid returning TOO many results too fast. If this occurs Bullet will kill your query because of rate limiting (the default rate limit is 500 records per second).

Complex Filtering

The querybuilder allows you create nested filters (basic relational filters, ANDs and ORs).

The querybuilder is also type aware: Numeric fields only allow numeric values, String fields only allow String operation, etc.

Example: Finding and picking out the first and second events in each period that also have probability > 0.5

Count Distinct

Count Distinct will count the number of distinct elements in a field exactly up to a threshold that is established when the backend is launched (16,384 in the example below).

After this threshold the count will be approximate. As this example demonstrates, information about the precision of the count can be found in the Metadata:

Group all

Choosing the Grouped Data option with no fields will result in the metrics being applied to all the data that matches your filters (or the whole set if you have no filters).

Example: Counting, summing and averaging on the whole dataset

The metrics you apply on fields are all numeric presently. If you apply a metric on a non-numeric field, Bullet will try to type-cast your field into number and if it's not possible, the result will be null. The result will also be null if the field was not present or no data matched your filters.

Group by

You can also choose Group fields and perform metrics per group.

Example: Grouping by tuple_number

Distributions

In this example, we find distributions of the duration field. This field is generated randomly from 0 to 10,049, with a tendency to have values that are closer to 0. We should be able to see this using Bullet.

The "distribution" option allows you to pick a type of distribution:

• Quantiles lets you get various percentiles (e.g. 25th, 99th) of your numeric field
• Frequencies lets you break up the range of values of your field into intervals and get a count of how many values fell into each interval.
• Cumulative Frequencies does the same as Frequencies but each interval includes the counts of all the intervals prior to it.

Both Frequencies and Cumulative Frequencies also give you a probability of how likely a value is to fall into the interval.

You can read much more about this in the UI help by clicking the Need more help? link.

Exact

Example: Finding the various percentiles of duration

---

Example: Finding some frequency counts of duration values in an interval

Try it out youtself and see what Cumulative Frequencies does!

Approximate

Example: Approximate quantile distribution

Top K

Top K lets you get the most frequent items or the heavy hitters for the values in a set of a fields.

Exact

This example gets the Top 3 most popular type values (there are only 6 but this illustrates the idea).

Approximate

By adding duration into the fields, the number of unique values for (type, duration) is increased. However, because duration has a tendency to have low values, we will have some frequent items. The counts are now estimated.

Charting

This example shows how to get a basic chart in Bullet. The charting and pivoting modes are only enabled for queries that are not Count Distinct or Group without Group Fields. This is because these results only have a single row and it does not make sense to graph them. They are enabled for all other queries.

The charting example below shows how to get a quick chart of a Group query with 3 metrics.

Tumbling Windows

Time-Based Tumbling Windows will return results every X seconds:

This example groups-by "type" and computes a couple metrics for each 2 second window.

Additive Tumbling Windows

Additive tumbling windows will also return results every X seconds, but the results will contain all the data collected since the beginning of the query:

In this example we compute bucket'ed frequency for the "gaussian" field. As the query runs you can see the gaussian curve form.

Pivoting

If the regular chart option is insufficient for your result (for instance, you have too many groups and metrics or you want to post-aggregate your results or remove outliers etc), then there is a advanced Pivot mode available when you are in the Chart option.

The Pivot option provides a drag-and-drop interface to drag fields to breakdown and aggregate by their values. Operations such as finding standard deviations, variance, etc are available as well as easily viewing them as tables and charts.

The following example shows a Group query with multiple groups and metrics and some interactions with the Pivot table.