[ML] Error during classification analytics

[pheyos]

Found in version

• 8.0.0 - starting with mini stack build 6683, where [ML] specifying missing_field_value value and using it instead of empty_string elasticsearch#53108 went in

Steps to reproduce

• Create and start a classification job based in the bank-marketing dataset
• Configuration:

{
  "description": "",
  "source": {
    "index": "bank-marketing*"
  },
  "dest": {
    "index": "dest-bm_1"
  },
  "analyzed_fields": {
    "excludes": []
  },
  "analysis": {
    "classification": {
      "dependent_variable": "y",
      "training_percent": "20"
    }
  },
  "model_memory_limit": "200mb"
}

Expected result

• The classification job finishes without errors

Actual result

• The job goes into failed state with the following error in the job messages:

Updated analytics task state to [failed] with reason [[bm_1_1583497643615] analytics process stopped unexpectedly: Fatal error: 'si_signo 11, si_code: 1, si_errno: 0, address: 0x7ff15024cc91, library: /home/robert/github/machine-learning-qa/build/deployments/8.0.0/6683/elasticsearch/modules/x-pack-ml/platform/linux-x86_64/bin/../lib/libMlMaths.so, base: 0x7ff150015000, normalized address: 0x237c91', version: 8.0.0-SNAPSHOT (build 9abec24aff83ac) ]

Additional information

• When running a classification job with different input data, e.g. Kibana sample flight data, the job runs without errors, so it's probably input data related.

[droberts195]

There was a log message capturing the failure address:

Updated analytics task state to [failed] with reason [[bm_1_1583489004749] analytics process stopped unexpectedly:
Fatal error: 'si_signo 11, si_code: 1, si_errno: 0, address: 0x7f8968223941, library: /home/robert/github/machine-learning-qa/build/deployments/8.0.0/6687/elasticsearch/modules/x-pack-ml/platform/linux-x86_64/bin/../lib/libMlMaths.so, base: 0x7f8967feb000, normalized address: 0x238941', version: 8.0.0-SNAPSHOT (build 0108a6ecf8a983) ]

Looking up the function from that shows the crash happened in _ZN2ml5maths15CDataFrameUtils19metricMicWithColumnERKNS1_12CColumnValueERKNS_4core10CDataFrameERKNS5_16CPackedBitVectorESt6vectorImSaImEE, which demangles to ml::maths::CDataFrameUtils::metricMicWithColumn(ml::maths::CDataFrameUtils::CColumnValue const&, ml::core::CDataFrame const&, ml::core::CPackedBitVector const&, std::vector<unsigned long, std::allocator<unsigned long> >).