stats
Buckets consist of a lower limit and an upper limit.
Basic implementation of exception stat recording that allows exceptions to be categorised and optional rollup.
A StatsReceiver that provides a linting rule to track metrics that have different underlying names, yet are exported under the same name.
A writeable Counter.
A writeable Counter. Only sums are kept of Counters. An example Counter is "number of requests served".
Represents a counter with its assorted metadata.
Should be mixed into StatsReceivers that delegate to other StatsReceivers.
A denylisting StatsReceiver.
A denylisting StatsReceiver. If the name for a metric is found to be denylisted, nothing is recorded.
Exception Stats Recorder.
Exposes the value of a function.
Exposes the value of a function. For example, one could add a gauge for a computed health metric.
Represents a gauge with its assorted metadata.
Details of a histogram's data points.
Represents a histogram with its assorted metadata.
An in-memory implementation of StatsReceiver, which is mostly used for testing.
An in-memory implementation of StatsReceiver, which is mostly used for testing.
Note that an InMemoryStatsReceiver does not conflate Seq("a", "b") and Seq("a/b")
names no matter how they look when printed.
val isr = new InMemoryStatsReceiver isr.counter("a", "b", "foo") isr.counter("a/b", "bar") isr.print(Console.out) // will print two lines "a/b/foo 0" and "a/b/bar 0" assert(isr.counters(Seq("a", "b", "foo") == 0)) // ok assert(isr.counters(Seq("a", "b", "bar") == 0)) // fail
Wraps an underlying StatsReceiver to ensure that derived counters and
stats will not start exporting metrics until incr or add is first called
on them.
Wraps an underlying StatsReceiver to ensure that derived counters and
stats will not start exporting metrics until incr or add is first called
on them.
This should be used when integrating with tools that create metrics eagerly, but you don't know whether you're going to actually use those metrics or not. One example might be if you're speaking to a remote peer that exposes many endpoints, and you eagerly create metrics for all of those endpoints, but aren't going to use all of the different methods.
We don't apply this very widely automatically--it can mess with caching, and adds an extra allocation when you construct a new counter or stat, so please be judicious when using it.
does not change the way gauges are used, since there isn't a way of modeling whether a gauge is "used" or not.
A builder class used to configure settings and metadata for metrics prior to instantiating them.
A builder class used to configure settings and metadata for metrics prior to instantiating them. Calling any of the three build methods (counter, gauge, or histogram) will cause the metric to be instantiated in the underlying StatsReceiver.
This trait represents a metric with the full set of its metadata.
Represents the units this metric is measured in.
Represents the units this metric is measured in.
Common units for metrics are: Bytes/Kilobytes/Megabytes (for payload size, data written to disk) Milliseconds (for latency, GC durations) Requests (for successes, failures, and requests) Percentage (for CPU Util, Memory Usage)
A StatsReceiver receiver proxy that translates all counter, stat, and gauge
names according to a translate function.
A no-op StatsReceiver.
A no-op StatsReceiver. Metrics are not recorded, making this receiver useful in unit tests and as defaults in situations where metrics are not strictly required.
A variation of Counter that also supports reading of the current value via the apply method.
A variation of Stat that also supports reading of the current time series via the apply method.
A RollupStatsReceiver reports stats on multiple Counter/Stat/Gauge based on the sequence of names you pass.
A RollupStatsReceiver reports stats on multiple Counter/Stat/Gauge based on the sequence of names you pass. e.g. counter("errors", "clientErrors", "java_net_ConnectException").incr() will actually increment those three counters: - "/errors" - "/errors/clientErrors" - "/errors/clientErrors/java_net_ConnectException"
Represents the "role" this service plays with respect to this metric.
Represents the "role" this service plays with respect to this metric.
Usually either Server (the service is processing a request) or Client (the server has sent a request to another service). In many cases, there is no relevant role for a metric, in which case NoRole should be used.
An append-only collection of time-series data.
An append-only collection of time-series data. Example Stats are "queue depth" or "query width in a stream of requests".
Utilities for timing synchronous execution and asynchronous execution are on the companion object (Stat.time(Stat) and Stat.timeFuture(Stat).
An interface for recording metrics.
An interface for recording metrics. Named Counters, Stats, and Gauges can be accessed through the corresponding methods of this class.
Each metric created via a stats receiver has a verbosity level attached to it. Distinguishing verbosity levels for metrics is optional and is up to a concrete implementation. Doing this, however, helps to separate debug metrics (only helpful in troubleshooting) from their operationally-required counterparts (provide a corresponding degree of visibility into a healthy process) thus potentially reducing the observability cost.
Metrics created w/o an explicitly specified Verbosity level, will use Verbosity.Default. Use VerbosityAdjustingStatsReceiver to adjust this behaviour.
A proxy StatsReceiver that delegates all calls to the self stats receiver.
Represent a verbosity level for a given metric.
A StatsReceiver that adjusts the passed Verbosity of an underlying stats receiver to
a given defaultVerbosity.
Allow StatsReceivers to provide snapshots of histogram counts.
Allow StatsReceivers to provide snapshots of histogram counts. Implementations must expose a map where keys are the name of the stat and values are the contents of the histogram.
BroadcastCounter is a helper object that create a Counter wrapper around multiple Counters (n).
BroadcastCounter is a helper object that create a Counter wrapper around multiple Counters (n). For performance reason, we have specialized cases if n == (0, 1, 2, 3 or 4)
BroadcastStat is a helper object that create a Counter wrapper around multiple Stats (n).
BroadcastStat is a helper object that create a Counter wrapper around multiple Stats (n). For performance reason, we have specialized cases if n == (0, 1, 2, 3 or 4)
BroadcastStatsReceiver is a helper object that create a StatsReceiver wrapper around multiple StatsReceivers (n).
API for deciding where request exceptions are reported in stats.
API for deciding where request exceptions are reported in stats. Typical implementations may report any cancellations or validation errors separately so success rate can from valid non cancelled requests.
Null for a no-op handler.
Stat utility methods for ease of use from java.
Helpers for working with histograms.
Helpers for working with histograms.
Java-friendly versions can be found in com.twitter.finagle.stats.JStats.
StatsReceiver utility methods for ease of use from java.
StatsReceiver utility methods for ease of use from java.
(Since version 2020-06-10) Use StatsReceiver addGauge and provideGauge methods directly
Buckets consist of a lower limit and an upper limit. In histograms, all values that fall inside these limits are counted as the same.
The lower limit is inclusive and the upper limit is exclusive
number of data points which landed in that bucket.