Default UDP reporter protocol limits Jaeger production trace delivery

Direct Answer

Yes. The default UDP reporter has real production limitations in segmented or zero-trust networks and should be treated as a development-only assumption. When network policies block UDP, explicitly override the agent reporter protocol to TCP or gRPC and align the collector listener to match. Confirm alignment through collector-side metrics and network-path checks, not by watching trace volume alone.

Confirmed Facts vs. Likely Explanation

Established behavior: Jaeger agents report spans to collectors over UDP by default, commonly on ports 6831 and 6832. UDP is connectionless, so the agent receives no acknowledgment; a dropped packet is invisible to the application. Jaeger also supports TCP and gRPC reporter channels, and client SDKs can bypass the agent to report directly to a collector over HTTP or gRPC.

Likely failure mode: In production, missing traces usually stem from UDP being discarded by firewalls, Kubernetes NetworkPolicies, cloud security groups, or MTU fragmentation—not from sampling. Because the reporter is fire-and-forget, these losses are silent and appear only as gaps after an incident.

Configuration Boundary and Required Steps

1. Inspect the agent reporter configuration. Identify whether the agent uses UDP compact/binary thrift or has already been switched to gRPC/TCP. The boundary is the agent's reporter channel and the collector's matching server port.
2. Override the reporter when UDP is blocked. Switch the agent to gRPC (or TCP) and ensure the collector exposes the same protocol on the target port. If you bypass the agent, configure the client SDK to report directly to the collector's HTTP or gRPC endpoint.
3. Verify the network path. Use tcpdump on the agent host to confirm packets leave the interface and to look for ICMP unreachable messages. Run nc -u -z <collector-host> 6831 to test basic UDP reachability, but treat a passing result as insufficient proof.
4. Cross-check collector-side metrics. Compare the collector's received-span counters against the agent's emitted-span counters. A widening gap confirms loss between agent and collector.

Version note: Default ports and flag names can vary across Jaeger releases and Helm charts. Confirm the exact reporter configuration keys for your deployed version before applying changes.

One Diagnostic Detail That Changes the Recommendation

Is the collector reachable over TCP or gRPC from the application pod itself? If yes, overriding the agent reporter is the lower-risk change. If the application pod cannot reach the collector directly, keep the agent but switch its reporter protocol; direct-to-collector reporting would add resource overhead and may not be feasible under the current network policy.