You can reduce the amount of data that the Java agent sends to the Thundra Web Console by enabling sampling. You can either add built-in sampling rules or provide a custom sampler and implement the sampling logic by yourself. Either option allows you to sample Trace, Metric, and/or Log data using the Sampling API.

Sampling API

Sampler

This interface decides whether or not sampling should be instrumented. You can create your own custom sampler by implementing io.thundra.agent.core.sample.Sampler, which is the interface where data (such as trace, metric, log, or other data, depending on what your sampler is designed to use) that makes sampling decisions is located.

Sampler
public interface Sampler<A> {
​
    /**
     * Checks whether or not sampling should be done.
     *
     * @param arg to be used for sampling decision
     * @return <code>true</code> if sampling should be done,
     *         <code>false</code> otherwise
     */
    boolean isSampled(A arg);
​
}

CountAwareSampler

This sampler is a count-based implementation, which allows sampling for every specified count period. The default count frequency is 100, but it can be configured using the thundra_agent_sampler_countAware_countFreq environment variable.

As an example, if the count frequency is 500, data will be sampled with every 500 Lambda invocations. Here is the configuration:

CountAwareSampler
import io.thundra.agent.core.sample.impl.CountAwareSampler;
​
...
​
// Sample for every 500 call
CountAwareSampler countAwareSampler = new CountAwareSampler(500);

TimeAwareSampler

This is a time-based implementation, which allows sampling for every specified time period. The default time frequency is 300,000 milliseconds (5 minutes), but it can be configured using the thundra_agent_sampler_timeAware_timeFreq environment variable.

As an example, if the time frequency is 60,000, data will be sampled every 60 seconds (10 minutes) within consecutive Lambda invocations. Here is the configuration:

TimeAwareSampler
import io.thundra.agent.core.sample.impl.TimeAwareSampler;
​
...
​
// Sample for every 60000 milliseconds (60 seconds = 1 minute)
TimeAwareSampler timeAwareSampler = new TimeAwareSampler(60000);

CompositeSampler

This implementation composes multiple samplers with a specified composition operation (AND or OR).

CompositeSampler
import io.thundra.agent.core.sample.impl.CountAwareSampler;
import io.thundra.agent.core.sample.impl.TimeAwareSampler;
import io.thundra.agent.core.sample.impl.CompositeSampler;
​
import static io.thundra.agent.core.sample.impl.CompositeSampler.SamplerCompositionOperator.OR;
​
...
  
CountAwareSampler countAwareSampler = new CountAwareSampler(500);
TimeAwareSampler timeAwareSampler = new TimeAwareSampler(1000);
// Sample at every 500 call or 60000 milliseconds (10 minutes), whichever comes first
CompositeSampler compositeSampler = new CompositeSampler(OR, countAwareSampler, timeAwareSampler);

Trace Sampling

Samplers can be set using the setTraceSampler static method of the io.thundra.agent.trace.TraceSupport class, which samples traces (spans).

Sampling Target

For tracing, sampling is only applied to the root span (not the other internal spans) of the invocation that represents the actual Lambda invocation. You can also choose to have all of the spans in the invocation automatically sampled if the root span is sampled.

DurationAwareThundraSpanSampler

This implementation samples Thundra spans if their duration in milliseconds meets or exceeds the duration you have set. For example, say you want to send trace data only if the duration of the Lambda invocation is longer than 1 second (1000 milliseconds). Here is the configuration:

DurationAwareThundraSpanSampler
import io.thundra.agent.trace.sample.impl.DurationAwareThundraSpanSampler;
import io.thundra.agent.trace.TraceSupport;
​
...
  
// Sample only if the duration of the invocation (root span) is longer than 1 seconds  
DurationAwareThundraSpanSampler durationAwareSampler = 
  	new DurationAwareThundraSpanSampler(1000);
​
// Set sampler to be used for tracing
TraceSupport.setSampler(durationAwareSampler);

ErrorAwareThundraSpanSampler

This implementation samples Thundra spans if they are erroneous (tagged by error). For example, say you want to send trace data only if the Lambda execution fails with an error:

ErrorAwareThundraSpanSampler
import io.thundra.agent.trace.sample.impl.ErrorAwareThundraSpanSampler;
import io.thundra.agent.trace.TraceSupport;
​
...
  
// Sample only if the invocation (root span) fails with an error  
ErrorAwareThundraSpanSampler errornAwareSampler = 
  	new ErrorAwareThundraSpanSampler();
​
// Set sampler to be used for tracing
TraceSupport.setSampler(errornAwareSampler);

Metric Sampling

Samplers can be set using the setMetricSampler static method of the io.thundra.agent.metric.MetricSupport class, which is used for sampling metrics (stats).

Default Metric Sampling

By default, metrics are sampled every 100 invocations or 5 minutes, whichever comes first. Default invocation count and time frequencies can be configured with the thundra_agent_sampler_countAware_countFreq and thundra_agent_sampler_timeAware_timeFreq environment variables. You can also set your custom sampler over MetricSampler, as described above.

Log Sampling

Samplers can be set using the setLogSampler static method of the io.thundra.agent.log.LogSupport class, which is used for sampling logs.