Marks the active OpenTelemetry span as ERROR when you log an error, so tail sampling keeps the trace. Logs you list in a YAML file are ignored, so expected or noisy errors don't.

In Python it's common to catch an exception, log it and return a response. The span still ends without an error status. If your tail sampler keeps traces that contain an ERROR span, it drops this one, and the trace you need to debug the failure is gone.

Version 0.1. The API may change before 1.0.

pip install otel-log-span-status

Requires Python 3.10+ and opentelemetry-api 1.20+. You also need an OpenTelemetry SDK configured in your service; this library only sets span status.

import logging

from opentelemetry import trace

logger = logging.getLogger("checkout") tracer = trace.get_tracer(name)

def charge(order): with tracer.start_as_current_span("charge"): try: gateway.charge(order) except GatewayTimeout: logger.exception("charge failed for order %s", order.id) return Response(status=502)

Before: the error is in your logs, but the charge span ends with status UNSET. To the sampler, it's a healthy trace.

After: add one call at startup and the same code marks the span ERROR:

import otel_log_span_status

otel_log_span_status.install("otel-log-span-status.yaml")

Ignoring noise

Marking every error log isn't right either. A retry that eventually succeeds, or a chatty library, would make you keep traces nobody needs. List those logs in the config file:

otel-log-span-status.yaml

ignore:

• logger: "urllib3.*" # logger name glob
• message_template: "Retrying request %s" # exact match on the unformatted message
• logger: "payments.*" # fields within one rule must all match message_regex: "card declined"

How rules match:

To skip a single call without a rule, pass extra={"span_error": False}.

This library sets the span status. Your tail sampler decides what to keep. For example, with the OpenTelemetry Collector's tail_sampling processor (in the contrib distribution):

processors: tail_sampling: decision_wait: 10s policies:

• name: keep-errors type: status_code status_code: { status_codes: [ERROR] }
• name: keep-slow type: latency latency: { threshold_ms: 2000 }
• name: sample-the-rest type: probabilistic probabilistic: { sampling_percentage: 5 }

A trace is kept if any policy matches, so errors are always kept and the rest is sampled.

Which logs: ERROR and above, including CRITICAL and logger.exception(...).

Where it attaches: install() adds a handler to the root logger, so it sees every logger that propagates to root. Loggers with propagate = False are not seen. To attach somewhere else, pass install(path, logger=some_logger).

No span: if there is no recording span, nothing happens.

Config errors: the config file is optional. With no path, every error log marks the span. If you pass a path, the file must exist and be valid, or install() raises FileNotFoundError or ConfigError.

Reloading: calling install() again replaces the handler it installed before, so you can use it to reload the config.

Safety: an error inside the handler is reported through logging's usual handleError and never breaks your logging call.

How this differs from existing OpenTelemetry logging integrations

opentelemetry-instrumentation-logging adds trace and span IDs to log records so you can find a log's trace. It does not change span status.

The SDK's LoggingHandler exports log records as OpenTelemetry logs. It does not change span status either.

Use either alongside this library.

I wrote about the problem in more detail in Error Logs and Missing Traces.