Distributed tracing for the PHP agent

New Relic APM's PHP agent supports distributed tracing, which lets you see the path that requests take as they travel through a distributed system. This document contains PHP-specific tips to enable distributed tracing for a PHP application or service.

Enable distributed tracing for PHP

When distributed tracing is enabled, it implicitly disables cross application tracing. The PHP agent will ignore the cross application tracing configuration value, and act as though this value is set to false.

To enable or disable distributed tracing for the PHP agent:

  1. If applicable, update your PHP agent to support distributed tracing (version 8.4 or higher).
  2. Make sure that the transaction tracer configuration option is enabled (default is true):

    newrelic.transaction_tracer.enabled = true
  3. Enable (or disable) distributed tracing for the PHP agent with the distributed_tracing_enabled configuration option:

    newrelic.distributed_tracing_enabled = true
  4. Set the transaction tracer threshold value to a suitable value. Recommendation:

    • If you want to make all transactions eligible for a distributed trace, set this value to 0 seconds.
    • If you are only interested in traces for longer running transactions, or if generating a trace for every transaction creates unacceptable application performance, set this value higher than 0 seconds.

    For example:

    newrelic.transaction_tracer.threshold = 0
  5. Be sure to restart your web server (or other PHP SAPI) so the agent will pick up the newly configured values.

Leverage automatic distributed tracing instrumentation

The PHP agent automatically will instrument a number of native PHP functions, as well as some third party HTTP clients, with distributed trace information. These include:

  • PHP function file_get_contents
  • PHP function curl_exec
  • Guzzle 4, Guzzle 5, Guzzle 6
  • Drupal's drupal_http_request function
  • Laravel's queue jobs

Set trace detail level

Distributing tracing support depends on the PHP agent's transaction tracer. When distributed tracing is enabled, the transaction tracer will behave as though newrelic.transaction_tracer.detail is set to 0. This ensures the distributed tracing user interface is not flooded with irrelevant span events.

If you require a higher level of detail than what's provided by this default configuration, use the newrelic.transaction_tracer.custom configuration setting or the newrelic_add_custom_tracer API method to add trace nodes and span events for the specific PHP function or methods you want to add to your traces.

Manually instrument applications and services

If you're using an unsupported library, or have a non-HTTP based distributed system component (such as messaging queues), you can use the PHP agent API to manually identify transactions to be included in a distributed trace. This is a two step process:

  1. Modify your service or application to create a distributed trace payload.
  2. Modify your service or application to accept distributed trace payloads created by other transactions or requests.

The following example uses a generic message/job queue. While the actual details will vary depending on what sort of system you're trying to add to a distributed trace, the core concepts are the same. Also, while we've used a job queue as an example, you can use these methods with any distributed system component.

When you create a payload, you're telling New Relic you want this request or transaction or request to participate in a distributed trace. When you accept a payload, you're linking the current request or transaction with its parent request or transaction.

For more information about using manual instrumentation to get more detail or to see connections between services, see the documentation about distributed tracing payload APIs.

Create a distributed trace payload

Somewhere in your application you'll have code that looks or acts like the following:

// create a job object
$job = new \Generic\Queue\Job;

// set the information needed to run the queue job
$job->setData('info', $info);

// save the job
$job->save();

If you want this job to appear in a distributed trace, you need to create a distributed trace payload using newrelic_create_distributed_trace_payload, and then add that payload to the job's data:

$payload = newrelic_create_distributed_trace_payload();

// create a job object
$job = new \Generic\Queue\Job;

// set the information needed to run the queue job
$job->setData('info', $info);

// add the payload data to the job data as a text json string
$job->setData('dt_payload', $payload->Text());

// save the job
$job->save();

This is step one: You've created a distributed trace payload.

Accept the distributed tracing payload

Next, somewhere in your application stack you'll have a queue runner that looks or acts like the following:

// create the job runner
$jobRunner = \Generic\Queue\Runner;

// grab jobs until there aren't any
while($job = $jobRunner->next()) {
    // run the job
    $job->run();
}

In order to accept a distributed trace payload, use the newrelic_accept_distributed_trace_payload function

// create the job runner
$jobRunner = \Generic\Queue\Runner;

while($job = $jobRunner->next()) {
    $payload = $job->getData('dt_payload');
    if($payload) {
        newrelic_accept_distributed_trace_payload($payload);
    }
    $job->run();
}
Optional: Use HTTP safe payload strings

If you need to transport the payload via a mechanism that requires HTTP safe strings, (HTTP headers, URL query strings, POST fields, etc.), the API includes distributed tracing methods and functions that will create and accept HTTP safe versions of the strings.

// create the HTTP safe payload string 
$payload = newrelic_create_distributed_trace_payload();
$httpSafePayload = $payload->httpSafe();

// ...

// accept the HTTP safe payload string
newrelic_accept_distributed_trace_payload_httpsafe($httpSafePayload);

Command line PHP programs

PHP programs run from the PHP command line are always sampled by the agent's distributed tracer. Depending on the programs you run, these processes could see an over-representation in your collection of distributed traces. In these situations, you can opt to disable command line instrumentation by using the per-directory newrelic.enabled setting in your newrelic.ini files.

For more help

Recommendations for learning more: