Re-process historical data after fixing version extraction

[grdumas]

Summary

After fixing version extraction in processors (issues #41-#45), all historical data in OpenSearch needs to be re-processed to populate correct test.version values.

Current State

~83% of documents in OpenSearch have incorrect test.version values:

• Fully affected tests: test.version = wrapper version or "unknown"
• Partially affected tests: test.version = wrapper version, benchmark version in config

Required Steps

1. Pre-Migration Validation

Before re-processing:

• Verify all processor fixes are merged
• Test fixes with sample data
• Validate new documents have correct versions
• Create backup of OpenSearch index

2. Migration Strategy Options

Option A: Full Re-ingest (Recommended)

• Re-process all raw result files
• Generate new documents with correct versions
• Replace existing documents

Pros: Clean, consistent results
Cons: Time-intensive, requires raw files

Option B: Partial Update

• Query existing documents
• Extract benchmark version from runs[].configuration (for partially affected)
• Update test.version field

Pros: Faster for partially affected tests
Cons: Cannot fix fully affected tests without raw data

Option C: Hybrid

• Re-ingest fully affected tests (no version in config)
• Update partially affected tests (version in config)

3. Migration Script

Create script to:

1. Identify affected documents by test.name
2. Re-process raw results OR extract from config
3. Update test.version field
4. Validate results

# Pseudo-code
affected_tests = {
    'fully': ['coremark', 'coremark_pro', 'phoronix', 'specjbb', 'streams', 'uperf'],
    'partial': ['autohpl', 'fio', 'passmark', 'speccpu2017']
}

for test_name in affected_tests['fully']:
    # Must re-ingest from raw files
    documents = opensearch.search(test_name=test_name)
    for doc in documents:
        raw_file = locate_raw_result(doc)
        new_doc = process_with_fixed_processor(raw_file)
        opensearch.update(doc.id, new_doc)

for test_name in affected_tests['partial']:
    # Can extract from existing config
    documents = opensearch.search(test_name=test_name)
    for doc in documents:
        version = extract_version_from_config(doc, test_name)
        opensearch.update(doc.id, {'test.version': version})

4. Validation

After migration:

• Query counts match pre-migration
• test.version != test.wrapper_version for affected tests
• Spot-check documents have correct benchmark versions
• Run test queries to verify version filtering works

# Validation queries
# 1. Check version distribution
GET /chronicler-runs/_search
{
  "aggs": {
    "by_test": {
      "terms": {"field": "test.name.keyword"},
      "aggs": {
        "versions": {"terms": {"field": "test.version.keyword"}}
      }
    }
  }
}

# 2. Verify no conflation (version != wrapper_version)
GET /chronicler-runs/_search
{
  "query": {
    "script": {
      "script": "doc['test.version.keyword'].value == doc['test.wrapper_version.keyword'].value"
    }
  }
}

5. Rollback Plan

If issues found:

1. Restore from backup
2. Fix processor bugs
3. Re-run migration

Dependencies

• BaseProcessor conflates wrapper version with benchmark version in test.version field #41 - BaseProcessor fix
• STREAMS processor does not extract benchmark version from CSV comments #42 - STREAMS fix
• FIO processor stores benchmark version in config instead of test.version #43 - FIO fix
• CoreMark processor does not extract benchmark version #44 - CoreMark fix
• Add benchmark version extraction for remaining affected processors #45 - Remaining processors

Estimated Impact

Assuming uniform distribution:

• Documents affected: ~83% of chronicler-runs index
• Re-processing time: Depends on data volume and strategy
• Downtime: None (can update in-place)

References

• Analysis: VERSION_CONFLATION_IMPACT.md
• Example queries in impact document