Tabs is a redis-backed metrics tracker for time-based events that supports counts, sums, averages, and min/max, and task based stats sliceable by the minute, hour, day, week, month, and year.
Add this line to your application's Gemfile:
gem 'tabs'
And then execute:
$ bundle
Or install it yourself as:
$ gem install tabs
Metrics come in three flavors: counters, values, and tasks.
A counter metric simply records the number of events that occur within a given timeframe. To create a counter metric called ‘website-visits’, simply call:
Tabs.create_metric("website-visits", "counter")
Tabs will also create a counter metric automatically the first time you increment the counter.
To increment a metric counter, simply call:
Tabs.increment_counter("website-visits")
If you need to retroactively increment the counter for a specific timestamp, just pass it in.
Tabs.increment_counter("wibsite-visits", Time.now - 2.days)
To retrieve the counts for a given time period just call Tabs#get_stats
with the name of the metric, a range of times defining the period for which you want stats, and the resolution at which the data should be aggregated.
Tabs.get_stats("website-visits", 10.days.ago..Time.now, :hour)
This will return stats for the last 10 days by hour in the form of a Tabs::Metrics::Counter::Stats
object. This object is enumerable so you can iterate through the results like so:
results = Tabs.get_stats("website-visits", 10.days.ago..Time.now, :hour)
results.each { |r| puts r }
#=>
{ timestamp: 2000-01-01 00:00:00 UTC, count: 1 }
{ timestamp: 2000-01-01 01:00:00 UTC, count: 0 }
{ timestamp: 2000-01-01 02:00:00 UTC, count: 10 }
{ timestamp: 2000-01-01 03:00:00 UTC, count: 1 }
{ timestamp: 2000-01-01 04:00:00 UTC, count: 0 }
{ timestamp: 2000-01-01 05:00:00 UTC, count: 0 }
{ timestamp: 2000-01-01 06:00:00 UTC, count: 3 }
{ timestamp: 2000-01-01 07:00:00 UTC, count: 0 }
...
The results object also provides the following methods:
results.total #=> The count total for the given period
results.min #=> The min count for any timestamp in the period
results.max #=> The max count for any timestamp in the period
results.avg #=> The avg count for timestamps in the period
results.period #=> The timestamp range that was requested
results.resolution #=> The resolution requested
Timestamps for the given period in which no events occurred will be "filled in" with a count value to make visualizations easier.
The timestamps are also normalized. For example, in hour resolution, the minutes and seconds of the timestamps are set to 00:00. Likewise for the week resolution, the day is set to the first day of the week.
Lastly, you can access the overall total for a counter (for all time)
using the counter_total
method.
Tabs.counter_total("website-visits") #=> 476873
Value metrics record a value at a point in time and calculate the min, max, avg, and sum for a given time resolution. Creating a value metric is easy:
To record a value, simply call Tabs#record_value
.
Tabs.record_value("new-user-age", 32)
If you need to retroactively record a value for a specific timestamp, just pass it in.
Tabs.increment_counter("new-user-age", 19, Time.now - 2.days)
This will also create a value metric the first time, you can manually create a metric as well:
Tabs.create_metric("new-user-age", "value")
Retrieving the stats for a value metric is just like retrieving a counter metric.
Tabs.get_stats("new-user-age", 6.months.ago..Time.now, :month)
This will return a Tabs::Metrics::Value::Stats
object. Again, this
object is enumerable and encapsulates all the timestamps within the
given period.
results = Tabs.get_stats("new-user-age", 6.months.ago..Time.now, :month)
results.each { |r| puts r }
#=>
{ timestamp: 2000-01-01 00:00:00, count: 9, min: 19, max: 54, sum: 226, avg: 38 }
{ timestamp: 2000-02-01 01:00:00, count: 0, min: 0, max: 0, sum: 0, avg: 0 }
{ timestamp: 2000-03-01 02:00:00, count: 2, min: 22, max: 34, sum: 180, avg: 26 }
...
The results object also provides some aggregates and other methods:
results.count #=> The total count of recorded values for the period
results.sum #=> The sum of all values for the period
results.min #=> The min value for any timestamp in the period
results.max #=> The max value for any timestamp in the period
results.avg #=> The avg value for timestamps in the period
results.period #=> The timestamp range that was requested
results.resolution #=> The resolution requested
Task metrics allow you to track the beginning and ending of a process. For example, tracking a user who downloads you mobile application and later visits your website to make a purchase.
Tabs.start_task("mobile-to-purchase", "2g4hj17787s")
The first argument is the metric key and the second is a unique token
used to identify the given process. You can use any string for the
token but it needs to be unique. Use the complete_task
method to
finish the task:
Tabs.complete_task("mobile-to-purchase", "2g4hj17787s")
If you need to retroactively start/complete a task at a specific timestamp, just pass it in.
Tabs.start_task("mobile-to-purchase", "2g4hj17787s", Time.now - 2.days)
Tabs.complete_task("mobile-to-purchase", "2g4hj17787s", Time.now - 1.days)
Retrieving stats for a task metric is just like the other types:
Tabs.get_stats("mobile-to-purchase", 6.hours.ago..Time.now, :minute)
This will return a Tabs::Metrics::Task::Stats
object:
results = Tabs.get_stats("mobile-to-purchase", 6.hours.ago..Time.now, :minute)
results.started_within_period #=> Number of items started in period
results.completed_within_period #=> Number of items completed in period
results.started_and_completed_within_period #=> Items wholly started/completed in period
results.completion_rate #=> Rate of completion in the given resolution
results.average_completion_time #=> Average time for the task to be completed
When tabs increments a counter or records a value it does so for each of the following "resolutions". You may supply any of these as the last argument to the Tabs#get_stats
method.
:minute, :hour, :day, :week, :month, :year
It automatically aggregates multiple events for the same period. For instance when you increment a counter metric, 1 will be added for each of the resolutions for the current time. Repeating the event 5 minutes later will increment a different minute slot, but the same hour, date, week, etc. When you retrieve metrics, all timestamps will be in UTC.
If the built-in resolutions above don't work you can add your own. All that's necessary is a module that conforms to the following protocol:
module SecondResolution
include Tabs::Resolutionable
extend self
PATTERN = "%Y-%m-%d-%H-%M-%S"
def name
:seconds
end
def serialize(timestamp)
timestamp.strftime(PATTERN)
end
def deserialize(str)
dt = DateTime.strptime(str, PATTERN)
self.normalize(dt)
end
def from_seconds(s)
s / 1
end
def to_seconds
1
end
def add(timestamp, num_of_seconds)
timestamp + num_of_seconds.seconds
end
def normalize(ts)
Time.utc(ts.year, ts.month, ts.day, ts.hour, ts.min, ts.sec)
end
end
A little description on each of the above methods:
name
: unique symbol used to reference registered resolution
serialize
: converts the timestamp to a string. The return value
here will be used as part of the Redis key storing values associated
with a given metric.
deserialize
: converts the string representation of a timestamp back
into an actual DateTime value.
from_seconds
: should return the number of periods in the given
number of seconds. For example, there are 60 seconds in a minute.
to_seconds
: should return the number of seconds in '1' of these time periods. For example, there are 3600 seconds in an hour.
add
: should add the number of seconds in the given resolution to the
supplied timestamp.
normalize
: should simply return the first timestamp for the period.
For example, the week resolution returns the first hour of the first day
of the week.
NOTE: If you're doing a custom resolution, you should probably look into the code a bit.
Once you have a module that conforms to the resolution protocol you need to register it with Tabs. You can do this in one of two ways:
# This call can be anywhere before you start using tabs
Tabs::Resolution.register(SecondResolution)
# or, you can use the config block described below
You can also remove any resolution (custom or built-in) by calling the unregister_resolutions
method in the config block (see config section below). Or, you can remove manually by calling:
Tabs::Resolution.unregister(:minute, :hour)
You can list all metrics using list_metrics
:
Tabs.list_metrics #=> ["website-visits", "new-user-age"]
You can check a metric's type (counter of value) by calling
metric_type
:
Tabs.metric_type("website-visits") #=> "counter"
And you can quickly check if a metric exists:
Tabs.metric_exists?("foobar") #=> false
To drop a metric, just call Tabs#drop_metric
Tabs.drop_metric!("website-visits")
This will drop all recorded values for the metric so it may not be un-done...be careful.
To drop only a specific resolution for a metric, just call Tabs#drop_resolution_for_metric!
Tabs.drop_resolution_for_metric!("website-visits", :minute)
Even more dangerous, you can drop all metrics...be very careful.
Tabs.drop_all_metrics!
You can use the expiration features to age out old metrics that may no longer be in your operational data set. For example, you may want to keep monthly or yearly data around but the minute or day level data isn't necessary past a certain date. You can set expirations for any resolution:
Tabs.configure do |config|
config.set_expirations(minute: 1.day, day: 1.week)
end
The expiration date will start counting at the beginning of the end of the given resolution. Meaning that for a month resolution the given expiration time would start at the end of a given month. A month resolution metric recorded in January with an expiration of 2 weeks would expire after the 2nd week of February.
NOTE: You cannot expire task metrics at this time, only counter and values.
Tabs just works out of the box. However, if you want to override the default Redis connection or decimal precision, this is how:
Tabs.configure do |config|
# set it to an existing connection
config.redis = Redis.current
# pass a config hash that will be passed to Redis.new
config.redis = { :host => 'localhost', :port => 6379 }
# pass a prefix that will be used in addition to the "tabs" prefix with Redis keys
# Example: "tabs:my_app:metric_name"
config.prefix = "my_app"
# override default decimal precision (5)
# affects stat averages and task completion rate
config.decimal_precision = 2
# registers a custom resolution
config.register_resolution :second, SecondResolution
# unregisters any resolution
config.unregister_resolutions(:minute, :hour)
# sets TTL for redis keys of specific resolutions
config.set_expirations({ minute: 1.hour, hour: 1.day })
end
Many applications use a single Redis instance for a number of uses: background jobs, ephemeral data, Tabs, etc. To avoid key collisions, and to make it easier to drop all of your tabs data without affecting other parts of your system (or if more than one app shares the Redis instance) you can prefix a given 'instance'.
Setting the prefix config option will cause all of the keys that tabs stores to use this format:
tabs:#{prefix}:#{key}..."
Please note that when the library version went from v0.5.6 to v0.6.0 some of the key patterns used to store metrics in Redis were changed. If you upgrade an app to v0.6.0 the previous set of data will not be picked up by tabs. Please use v0.6.x on new applications only. However, the 'Task' metric type will only be available in v0.6.0 and above.
In version 0.8.0 and higher the get_stats method returns a more robust object instead of just an array of hashes. These stats objects are enumerable and most existing code utilizing tabs should continue to function. However, please review the docs for more information if you encounter issues when upgrading.
In version 0.8.2 and higher the storage keys for value metrics have been changed. Originally the various pieces (avg, sum, count, etc) were stored in a JSON serialized string in a single key. The intent was that this would comprise a poor-mans transaction of sorts. The downside however was a major hit on performance when doing a lot of writes or reading stats for a large date range. In v0.8.2 these component values are stored in a real Redis hash and updated atomically when a value is recorded. In future versions this will be changed to use a MULTI statement to simulate a transaction. Value data that was recorded prior to v0.8.2 will not be accessible in this or future versions so please continue to use v0.8.1 or lower if that is an issue.
WARNING: Version 1.0.0 is not compatible with previous versions of Tabs
We have made a number of changes related to how our metric keys are stored in Redis. At this point we'll be following semantic versioning and will support backwards compatability between major versions. In this release we've added a number of major features:
- Metric expiration
- Key prefixes
- Support for unregistering resolutions
- A number of small numeric fixes
- Fork it
- Create your feature branch (
git checkout -b my-new-feature
) - Commit your changes (
git commit -am 'Add some feature'
) - Push to the branch (
git push origin my-new-feature
) - Create new Pull Request