Fix deferred export by properly formatting OTLP headers for curl

[Copilot]

The deferred export mechanism in workflow instrumentation was broken because OTLP headers weren't properly formatted when forwarding telemetry data to the actual endpoint.

Root Cause

OTLP headers are provided as comma-separated key=value pairs (e.g., Authorization=Bearer token,X-Custom=value), but curl's -H @file syntax expects each header on a separate line in key: value format. The workflow instrumentation was writing the raw comma-separated string to the headers file, which curl couldn't parse correctly.

Changes

actions/instrument/workflow/main.sh - Transform headers to proper format before passing to curl:

# Before (broken):
echo "$OTEL_EXPORTER_OTLP_HEADERS" >> "$headers"
echo "$OTEL_EXPORTER_OTLP_LOGS_HEADERS" >> "$headers"

# After (fixed):
echo "$OTEL_EXPORTER_OTLP_HEADERS","$OTEL_EXPORTER_OTLP_LOGS_HEADERS" | tr ',' '\n' | grep -v '^$' | sed 's/=/: /g' >> "$headers"

The transformation:

1. Combines general (OTEL_EXPORTER_OTLP_HEADERS) and signal-specific headers (e.g., OTEL_EXPORTER_OTLP_LOGS_HEADERS)
2. Splits comma-separated headers into separate lines (tr ',' '\n')
3. Filters out empty lines (grep -v '^$')
4. Converts key=value to key: value format (sed 's/=/: /g')

This matches the exact pattern used in the job-level instrumentation for the collector configuration, ensuring consistency across the codebase.

Impact

When export_deferred_signal_file forwards deferred telemetry data via curl, it now correctly passes headers in the format curl expects, allowing the data to be successfully sent to the OTLP endpoint with proper authentication and configuration.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Deferred export is broken</issue_title>
<issue_description>Somehow the deferred export in the workflow instrumentaiton is broken</issue_description>

<agent_instructions>analayze both the code in the job level instrumentaiton and the forwarding code in the workflow instrumentation carefully, find the problem and fix it. the purprose of the deferred export is that it takes the payload, and then sends it later with proper credentials.</agent_instructions>

Comments on the Issue (you are @copilot in this section)

• Fixes Deferred export is broken #2810

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[plengauer]

@copilot that cant be it. they are already exported from the start, so unless you can show me a specific place where they are unexported, they are available. re-analyze and find the real issue

[Copilot]

@copilot that cant be it. they are already exported from the start, so unless you can show me a specific place where they are unexported, they are available. re-analyze and find the real issue

You're absolutely right - the variables were already exported. The real issue was in the header formatting.

The OTLP headers are comma-separated key=value pairs, but curl's -H @file syntax expects each header on a separate line in key: value format. The workflow code was just echoing the raw comma-separated string, which curl couldn't parse.

Fixed in commit 41df900 by applying the same header transformation used in the job-level instrumentation:

echo "$OTEL_EXPORTER_OTLP_HEADERS" | tr ',' '\n' | grep -v '^$' | sed 's/=/: /g' >> "$headers"

This splits by comma, filters empty lines, and converts = to : format.

[Copilot]

Pull request overview

This PR fixes the deferred export mechanism in workflow instrumentation by properly formatting OTLP headers for curl. The issue was that headers were provided as comma-separated key=value pairs but curl requires key: value format with each header on a separate line.

Changes:

• Transform OTLP headers from comma-separated format to curl-compatible format with proper key: value syntax
• Combine signal-specific headers with general headers before transformation
• Remove the redundant separate echo of OTEL_EXPORTER_OTLP_HEADERS