APM Transactions not sampled when sample rate is 1

Describe the bug: We have micro services using Flask as framework and ElasticAPM for tracing requests. We noticed ~10% of transactions are not sampled despite not changing the default sample rate (1) for any of the services. Services send traces to APM servers running two different versions (7.14.1 and 7.16.1). With APM server 7.14.1, upsampled transactions have transaction.sampled set to false.

Environment

• Python version: 3.6, 3.7
• Framework and version: Flask 1.0.2, 1.1.2, 2.0.2, 2.1.1
• APM Server version: 7.14.1, 7.16.1
• Agent version: 5.10.1, 6.7.2

Additional context The screenshot below shows upsampled transactions associated with one HTTP request:

• transaction.sampled=false
• transaction.span_count.started=0

[basepi]

You’ll want trace_continuation_strategy=restart or trace_continuation_strategy=restart_external, depending on whether the upstream service is Elastic-monitored. (You can tell if it’s an Elastic traceparent by checking the tracestate header for an es= entry.)

Note, however, that this setting was only introduced in version 6.10.0 of the agent, and the versions you listed above are older than that.

[Alsheh]

Updating the service with trace_continuation_strategy=restart and ensuring the deployment is up to date fixes the issue. Thank you @basepi for your help in resolving this issue.