Pipeline Dedup

Pipeline Deduplication

Check if a failure is already tracked in Jira before investigating or creating issues.

Prerequisites

Load pipeline-jira skill first to get the current week's parent task ($PARENT).

Inputs

Deduplication Workflow

Pass 1: Exact Build Match

Check if this exact build already has a subtask:

jira issue list --parent $PARENT \
  -q "summary ~ '$JOB #$BUILD'" --plain --no-headers

If found → Return EXACT_MATCH: <JIRA-KEY>

Pass 2: Similar Failure Check (Same Job + Stream + Arch)

Check for open subtasks with same job, stream, and architecture:

# For build-arch (include architecture in search)
jira issue list --parent $PARENT -s~Closed \
  -q "summary ~ '$JOB' AND summary ~ '$STREAM' AND summary ~ '$ARCH'" \
  --plain --no-headers

# For build / build-node-image (no architecture)
jira issue list --parent $PARENT -s~Closed \
  -q "summary ~ '$JOB' AND summary ~ '$STREAM'" \
  --plain --no-headers

If found → Return RELATED_ISSUE: <JIRA-KEY>

Pass 3: Semantic Analysis

For failures not caught by Pass 1 or 2, fetch all open subtasks and analyze semantically:

# Fetch all open subtasks with summaries
jira issue list --parent $PARENT -s~Closed --plain --no-headers

Provide the failure details and fetched subtask list to analyze:

Failure to check:

• Job: $JOB
• Build: $BUILD
• Stream: $STREAM
• Arch: $ARCH (if applicable)
• Root cause hint: (if known from triage)

Analysis prompt:

Review these open Jira subtasks and determine if any track the same root cause as this failure. Consider:

• Same stream/arch combination
• Same error patterns (cloud upload, chronyd, registry auth, etc.)
• Related downstream jobs (build failure caused by build-arch failure)

Return the matching Jira key if found, or "NO_MATCH" if this appears to be a new distinct issue.

If LLM identifies a match → Return SEMANTIC_MATCH: <JIRA-KEY>

Output Format

Return one of:

Comment Template for Duplicate Occurrences

When a RELATED_ISSUE or SEMANTIC_MATCH is found, add a comment to the existing issue instead of creating a duplicate.

Note: For EXACT_MATCH, no comment is needed - the build is already tracked in the issue summary.

jira issue comment add <EXISTING-KEY> $'Additional occurrence detected:
- **Build:** [#<build>](<jenkins-url>)
- **Timestamp:** <timestamp>

Same failure pattern - consolidating under this issue.' --no-input

Batch Mode (for pipeline-monitor)

When checking multiple failures, optimize by fetching subtasks once:

# Fetch all subtasks once
jira issue list --parent $PARENT --plain --no-headers > /tmp/subtasks.txt

Then perform Pass 1 and Pass 2 locally by parsing the cached file. Only run Pass 3 (semantic analysis) for failures that pass both checks.