[ML] Allow a certain number of ill-formatted rows when delimited form…

…at is specified (#55735)

While it is good to not be lenient when attempting to guess the file format, it is frustrating to users when they KNOW it is CSV but there are a few ill-formatted rows in the file (via some entry error, etc.).

This commit allows for up to 10% of sample rows to be considered "bad". These rows are effectively ignored while guessing the format.

This percentage of "allows bad rows" is only applied when the user has specified delimited formatting options. As the structure finder needs some guidance on what a "bad row" actually means.

related to #38890

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/filestructurefinder/DelimitedFileStructureFinder.java

@@ -5,6 +5,7 @@
  */
 package org.elasticsearch.xpack.ml.filestructurefinder;
 
+import org.apache.logging.log4j.message.ParameterizedMessage;
 import org.elasticsearch.common.collect.Tuple;
 import org.elasticsearch.xpack.core.ml.filestructurefinder.FieldStats;
 import org.elasticsearch.xpack.core.ml.filestructurefinder.FileStructure;
@@ -34,7 +35,6 @@ public class DelimitedFileStructureFinder implements FileStructureFinder {
  private static final String REGEX_NEEDS_ESCAPE_PATTERN = "([\\\\|()\\[\\]{}^$.+*?])";
  private static final int MAX_LEVENSHTEIN_COMPARISONS = 100;
  private static final int LONG_FIELD_THRESHOLD = 100;
-
  private final List<String> sampleMessages;
  private final FileStructure structure;
 
@@ -80,6 +80,11 @@ static DelimitedFileStructureFinder makeDelimitedFileStructureFinder(List<String
  for (int index = isHeaderInFile ? 1 : 0; index < rows.size(); ++index) {
  List<String> row = rows.get(index);
  int lineNumber = lineNumbers.get(index);
+ // Indicates an illformatted row. We allow a certain number of these
+ if (row.size() != columnNames.length) {
+ prevMessageEndLineNumber = lineNumber;
+ continue;
+ }
  Map<String, String> sampleRecord = new LinkedHashMap<>();
  Util.filterListToMap(sampleRecord, columnNames,
  trimFields ? row.stream().map(field -> (field == null) ? null : field.trim()).collect(Collectors.toList()) : row);
@@ -488,7 +493,7 @@ static boolean lineHasUnescapedQuote(String line, CsvPreference csvPreference) {
  }
 
  static boolean canCreateFromSample(List<String> explanation, String sample, int minFieldsPerRow, CsvPreference csvPreference,
- String formatName) {
+ String formatName, double allowedFractionOfBadLines) {
 
  // Logstash's CSV parser won't tolerate fields where just part of the
  // value is quoted, whereas SuperCSV will, hence this extra check
@@ -501,11 +506,13 @@ static boolean canCreateFromSample(List<String> explanation, String sample, int
  }
  }
 
+ int numberOfLinesInSample = sampleLines.length;
  try (CsvListReader csvReader = new CsvListReader(new StringReader(sample), csvPreference)) {
 
  int fieldsInFirstRow = -1;
  int fieldsInLastRow = -1;
 
+ List<Integer> illFormattedRows = new ArrayList<>();
  int numberOfRows = 0;
  try {
  List<String> row;
@@ -529,11 +536,27 @@ static boolean canCreateFromSample(List<String> explanation, String sample, int
  --fieldsInThisRow;
  }
 
- if (fieldsInLastRow != fieldsInFirstRow) {
- explanation.add("Not " + formatName + " because row [" + (numberOfRows - 1) +
- "] has a different number of fields to the first row: [" + fieldsInFirstRow + "] and [" +
- fieldsInLastRow + "]");
- return false;
+ // TODO: might be good one day to gather a distribution of the most common field counts
+ // But, this would require iterating (or at least sampling) all the lines.
+ if (fieldsInThisRow != fieldsInFirstRow) {
+ illFormattedRows.add(numberOfRows - 1);
+ // This calculation is complicated by the possibility of multi-lined CSV columns
+ // `getLineNumber` is a current count of lines, regardless of row count, so
+ // this formula is just an approximation, but gets more accurate the further
+ // through the sample you are.
+ double totalNumberOfRows = (numberOfRows + numberOfLinesInSample - csvReader.getLineNumber());
+ // We should only allow a certain percentage of ill formatted rows
+ // as it may have and down stream effects
+ if (illFormattedRows.size() > Math.ceil(allowedFractionOfBadLines * totalNumberOfRows)) {
+ explanation.add(new ParameterizedMessage(
+ "Not {} because {} or more rows did not have the same number of fields as the first row ({}). Bad rows {}",
+ formatName,
+ illFormattedRows.size(),
+ fieldsInFirstRow,
+ illFormattedRows).getFormattedMessage());
+ return false;
+ }
+ continue;
  }
 
  fieldsInLastRow = fieldsInThisRow;

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/filestructurefinder/DelimitedFileStructureFinderFactory.java

@@ -14,6 +14,8 @@
 
 public class DelimitedFileStructureFinderFactory implements FileStructureFinderFactory {
 
+ static final double DELIMITER_OVERRIDDEN_ALLOWED_FRACTION_OF_BAD_LINES = 0.10d;
+ static final double FORMAT_OVERRIDDEN_ALLOWED_FRACTION_OF_BAD_LINES = 0.05d;
  private final CsvPreference csvPreference;
  private final int minFieldsPerRow;
  private final boolean trimFields;
@@ -44,7 +46,7 @@ public boolean canFindFormat(FileStructure.Format format) {
  * it could have been truncated when the file was sampled.
  */
  @Override
- public boolean canCreateFromSample(List<String> explanation, String sample) {
+ public boolean canCreateFromSample(List<String> explanation, String sample, double allowedFractionOfBadLines) {
  String formatName;
  switch ((char) csvPreference.getDelimiterChar()) {
  case ',':
@@ -57,7 +59,12 @@ public boolean canCreateFromSample(List<String> explanation, String sample) {
  formatName = Character.getName(csvPreference.getDelimiterChar()).toLowerCase(Locale.ROOT) + " delimited values";
  break;
  }
- return DelimitedFileStructureFinder.canCreateFromSample(explanation, sample, minFieldsPerRow, csvPreference, formatName);
+ return DelimitedFileStructureFinder.canCreateFromSample(explanation,
+ sample,
+ minFieldsPerRow,
+ csvPreference,
+ formatName,
+ allowedFractionOfBadLines);
  }
 
  @Override

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/filestructurefinder/FileStructureFinderFactory.java

@@ -25,10 +25,12 @@ public interface FileStructureFinderFactory {
  * @param explanation List of reasons for making decisions. May contain items when passed and new reasons
  * can be appended by this method.
  * @param sample A sample from the file to be ingested.
+ * @param allowedFractionOfBadLines How many lines of the passed sample are allowed to be considered "bad".
+ * Provided as a fraction from interval [0, 1]
  * @return <code>true</code> if this factory can create an appropriate
  * file structure given the sample; otherwise <code>false</code>.
  */
- boolean canCreateFromSample(List<String> explanation, String sample);
+ boolean canCreateFromSample(List<String> explanation, String sample, double allowedFractionOfBadLines);
 
  /**
  * Create an object representing the structure of a file.

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/filestructurefinder/FileStructureFinderManager.java

@@ -11,6 +11,7 @@
 import org.elasticsearch.ElasticsearchTimeoutException;
 import org.elasticsearch.common.collect.Tuple;
 import org.elasticsearch.common.unit.TimeValue;
+import org.elasticsearch.xpack.core.ml.filestructurefinder.FileStructure;
 
 import java.io.BufferedInputStream;
 import java.io.BufferedReader;
@@ -474,14 +475,17 @@ FileStructureFinder makeBestStructureFinder(List<String> explanation, String sam
  Character quote = overrides.getQuote();
  Boolean shouldTrimFields = overrides.getShouldTrimFields();
  List<FileStructureFinderFactory> factories;
+ double allowedFractionOfBadLines = 0.0;
  if (delimiter != null) {
+ allowedFractionOfBadLines = DelimitedFileStructureFinderFactory.DELIMITER_OVERRIDDEN_ALLOWED_FRACTION_OF_BAD_LINES;
 
  // If a precise delimiter is specified, we only need one structure finder
  // factory, and we'll tolerate as little as one column in the input
  factories = Collections.singletonList(new DelimitedFileStructureFinderFactory(delimiter, (quote == null) ? '"' : quote, 1,
  (shouldTrimFields == null) ? (delimiter == '|') : shouldTrimFields));
 
- } else if (quote != null || shouldTrimFields != null) {
+ } else if (quote != null || shouldTrimFields != null || FileStructure.Format.DELIMITED.equals(overrides.getFormat())) {
+ allowedFractionOfBadLines = DelimitedFileStructureFinderFactory.FORMAT_OVERRIDDEN_ALLOWED_FRACTION_OF_BAD_LINES;
 
  // The delimiter is not specified, but some other aspect of delimited files is,
  // so clone our default delimited factories altering the overridden values
@@ -499,7 +503,7 @@ FileStructureFinder makeBestStructureFinder(List<String> explanation, String sam
 
  for (FileStructureFinderFactory factory : factories) {
  timeoutChecker.check("high level format detection");
- if (factory.canCreateFromSample(explanation, sample)) {
+ if (factory.canCreateFromSample(explanation, sample, allowedFractionOfBadLines)) {
  return factory.createFromSample(explanation, sample, charsetName, hasByteOrderMarker, lineMergeSizeLimit, overrides,
  timeoutChecker);
  }

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/filestructurefinder/NdJsonFileStructureFinderFactory.java

@@ -30,7 +30,7 @@ public boolean canFindFormat(FileStructure.Format format) {
  * documents must be non-empty, to prevent lines containing "{}" from matching.
  */
  @Override
- public boolean canCreateFromSample(List<String> explanation, String sample) {
+ public boolean canCreateFromSample(List<String> explanation, String sample, double allowedFractionOfBadLines) {
 
  int completeDocCount = 0;
 

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/filestructurefinder/TextLogFileStructureFinderFactory.java

@@ -25,7 +25,7 @@ public boolean canFindFormat(FileStructure.Format format) {
  * non-blank lines.
  */
  @Override
- public boolean canCreateFromSample(List<String> explanation, String sample) {
+ public boolean canCreateFromSample(List<String> explanation, String sample, double allowedFractionOfBadLines) {
  if (sample.indexOf('\n') < 0) {
  explanation.add("Not text because sample contains no newlines");
  return false;

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/filestructurefinder/XmlFileStructureFinderFactory.java

@@ -43,7 +43,7 @@ public boolean canFindFormat(FileStructure.Format format) {
  * necessarily have to be complete (as the sample could have truncated it).
  */
  @Override
- public boolean canCreateFromSample(List<String> explanation, String sample) {
+ public boolean canCreateFromSample(List<String> explanation, String sample, double allowedFractionOfBadLines) {
 
  int completeDocCount = 0;
  String commonRootElementName = null;

x-pack/plugin/ml/src/test/java/org/elasticsearch/xpack/ml/filestructurefinder/DelimitedFileStructureFinderFactoryTests.java

@@ -16,78 +16,78 @@ public class DelimitedFileStructureFinderFactoryTests extends FileStructureTestC
 
  public void testCanCreateCsvFromSampleGivenCsv() {
 
- assertTrue(csvFactory.canCreateFromSample(explanation, CSV_SAMPLE));
+ assertTrue(csvFactory.canCreateFromSample(explanation, CSV_SAMPLE, 0.0));
  }
 
  public void testCanCreateCsvFromSampleGivenTsv() {
 
- assertFalse(csvFactory.canCreateFromSample(explanation, TSV_SAMPLE));
+ assertFalse(csvFactory.canCreateFromSample(explanation, TSV_SAMPLE, 0.0));
  }
 
  public void testCanCreateCsvFromSampleGivenSemiColonDelimited() {
 
- assertFalse(csvFactory.canCreateFromSample(explanation, SEMI_COLON_DELIMITED_SAMPLE));
+ assertFalse(csvFactory.canCreateFromSample(explanation, SEMI_COLON_DELIMITED_SAMPLE, 0.0));
  }
 
  public void testCanCreateCsvFromSampleGivenPipeDelimited() {
 
- assertFalse(csvFactory.canCreateFromSample(explanation, PIPE_DELIMITED_SAMPLE));
+ assertFalse(csvFactory.canCreateFromSample(explanation, PIPE_DELIMITED_SAMPLE, 0.0));
  }
 
  public void testCanCreateCsvFromSampleGivenText() {
 
- assertFalse(csvFactory.canCreateFromSample(explanation, TEXT_SAMPLE));
+ assertFalse(csvFactory.canCreateFromSample(explanation, TEXT_SAMPLE, 0.0));
  }
 
  // TSV - no need to check NDJSON, XML or CSV because they come earlier in the order we check formats
 
  public void testCanCreateTsvFromSampleGivenTsv() {
 
- assertTrue(tsvFactory.canCreateFromSample(explanation, TSV_SAMPLE));
+ assertTrue(tsvFactory.canCreateFromSample(explanation, TSV_SAMPLE, 0.0));
  }
 
  public void testCanCreateTsvFromSampleGivenSemiColonDelimited() {
 
- assertFalse(tsvFactory.canCreateFromSample(explanation, SEMI_COLON_DELIMITED_SAMPLE));
+ assertFalse(tsvFactory.canCreateFromSample(explanation, SEMI_COLON_DELIMITED_SAMPLE, 0.0));
  }
 
  public void testCanCreateTsvFromSampleGivenPipeDelimited() {
 
- assertFalse(tsvFactory.canCreateFromSample(explanation, PIPE_DELIMITED_SAMPLE));
+ assertFalse(tsvFactory.canCreateFromSample(explanation, PIPE_DELIMITED_SAMPLE, 0.0));
  }
 
  public void testCanCreateTsvFromSampleGivenText() {
 
- assertFalse(tsvFactory.canCreateFromSample(explanation, TEXT_SAMPLE));
+ assertFalse(tsvFactory.canCreateFromSample(explanation, TEXT_SAMPLE, 0.0));
  }
 
  // Semi-colon delimited - no need to check NDJSON, XML, CSV or TSV because they come earlier in the order we check formats
 
  public void testCanCreateSemiColonDelimitedFromSampleGivenSemiColonDelimited() {
 
- assertTrue(semiColonDelimitedfactory.canCreateFromSample(explanation, SEMI_COLON_DELIMITED_SAMPLE));
+ assertTrue(semiColonDelimitedfactory.canCreateFromSample(explanation, SEMI_COLON_DELIMITED_SAMPLE, 0.0));
  }
 
  public void testCanCreateSemiColonDelimitedFromSampleGivenPipeDelimited() {
 
- assertFalse(semiColonDelimitedfactory.canCreateFromSample(explanation, PIPE_DELIMITED_SAMPLE));
+ assertFalse(semiColonDelimitedfactory.canCreateFromSample(explanation, PIPE_DELIMITED_SAMPLE, 0.0));
  }
 
  public void testCanCreateSemiColonDelimitedFromSampleGivenText() {
 
- assertFalse(semiColonDelimitedfactory.canCreateFromSample(explanation, TEXT_SAMPLE));
+ assertFalse(semiColonDelimitedfactory.canCreateFromSample(explanation, TEXT_SAMPLE, 0.0));
  }
 
  // Pipe delimited - no need to check NDJSON, XML, CSV, TSV or semi-colon delimited
  // values because they come earlier in the order we check formats
 
  public void testCanCreatePipeDelimitedFromSampleGivenPipeDelimited() {
 
- assertTrue(pipeDelimitedFactory.canCreateFromSample(explanation, PIPE_DELIMITED_SAMPLE));
+ assertTrue(pipeDelimitedFactory.canCreateFromSample(explanation, PIPE_DELIMITED_SAMPLE, 0.0));
  }
 
  public void testCanCreatePipeDelimitedFromSampleGivenText() {
 
- assertFalse(pipeDelimitedFactory.canCreateFromSample(explanation, TEXT_SAMPLE));
+ assertFalse(pipeDelimitedFactory.canCreateFromSample(explanation, TEXT_SAMPLE, 0.0));
  }
 }