-
Notifications
You must be signed in to change notification settings - Fork 113
/
IndexUtil.java
413 lines (361 loc) · 17.5 KB
/
IndexUtil.java
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
/*
* Copyright OpenSearch Contributors
* SPDX-License-Identifier: Apache-2.0
*/
package org.opensearch.knn.index.util;
import com.google.common.collect.ImmutableMap;
import com.google.common.collect.Maps;
import org.apache.commons.lang.StringUtils;
import org.opensearch.Version;
import org.opensearch.cluster.metadata.IndexMetadata;
import org.opensearch.cluster.metadata.MappingMetadata;
import org.opensearch.common.ValidationException;
import org.opensearch.knn.common.KNNConstants;
import org.opensearch.knn.index.engine.KNNMethodContext;
import org.opensearch.knn.index.KNNSettings;
import org.opensearch.knn.index.engine.MethodComponentContext;
import org.opensearch.knn.index.SpaceType;
import org.opensearch.knn.index.VectorDataType;
import org.opensearch.knn.index.mapper.KNNVectorFieldMapper;
import org.opensearch.knn.index.query.request.MethodParameter;
import org.opensearch.knn.index.engine.KNNEngine;
import org.opensearch.knn.indices.ModelDao;
import org.opensearch.knn.indices.ModelMetadata;
import org.opensearch.knn.indices.ModelUtil;
import org.opensearch.knn.jni.JNIService;
import java.io.File;
import java.util.Collections;
import java.util.HashMap;
import java.util.Locale;
import java.util.Map;
import java.util.Set;
import static org.opensearch.knn.common.KNNConstants.BYTES_PER_KILOBYTES;
import static org.opensearch.knn.common.KNNConstants.HNSW_ALGO_EF_SEARCH;
import static org.opensearch.knn.common.KNNConstants.SPACE_TYPE;
import static org.opensearch.knn.common.KNNConstants.VECTOR_DATA_TYPE_FIELD;
import static org.opensearch.knn.index.query.parser.RescoreParser.RESCORE_PARAMETER;
public class IndexUtil {
public static final String MODEL_NODE_ASSIGNMENT_KEY = KNNConstants.MODEL_NODE_ASSIGNMENT;
public static final String MODEL_METHOD_COMPONENT_CONTEXT_KEY = KNNConstants.MODEL_METHOD_COMPONENT_CONTEXT;
private static final Version MINIMAL_SUPPORTED_VERSION_FOR_IGNORE_UNMAPPED = Version.V_2_11_0;
private static final Version MINIMAL_SUPPORTED_VERSION_FOR_MODEL_NODE_ASSIGNMENT = Version.V_2_12_0;
private static final Version MINIMAL_SUPPORTED_VERSION_FOR_MODEL_METHOD_COMPONENT_CONTEXT = Version.V_2_13_0;
private static final Version MINIMAL_SUPPORTED_VERSION_FOR_RADIAL_SEARCH = Version.V_2_14_0;
private static final Version MINIMAL_SUPPORTED_VERSION_FOR_METHOD_PARAMETERS = Version.V_2_16_0;
private static final Version MINIMAL_SUPPORTED_VERSION_FOR_MODEL_VECTOR_DATA_TYPE = Version.V_2_16_0;
private static final Version MINIMAL_RESCORE_FEATURE = Version.V_2_17_0;
// public so neural search can access it
public static final Map<String, Version> minimalRequiredVersionMap = initializeMinimalRequiredVersionMap();
public static final Set<VectorDataType> VECTOR_DATA_TYPES_NOT_SUPPORTING_ENCODERS = Set.of(VectorDataType.BINARY, VectorDataType.BYTE);
/**
* Determines the size of a file on disk in kilobytes
*
* @param filePath path to the file
* @return file size in kilobytes
*/
public static int getFileSizeInKB(String filePath) {
if (filePath == null || filePath.isEmpty()) {
return 0;
}
File file = new File(filePath);
if (!file.exists() || !file.isFile()) {
return 0;
}
return Math.toIntExact((file.length() / BYTES_PER_KILOBYTES) + 1L); // Add one so that integer division rounds up
}
/**
* Validate that a field is a k-NN vector field and has the expected dimension
*
* @param indexMetadata metadata for index to validate
* @param field field name to validate
* @param expectedDimension expected dimension of the field. If this value is negative, dimension will not be
* checked
* @param modelDao used to look up dimension if field uses a model for initialization. Can be null if
* expectedDimension is negative
* @return ValidationException exception produced by field validation
*/
@SuppressWarnings("unchecked")
public static ValidationException validateKnnField(
IndexMetadata indexMetadata,
String field,
int expectedDimension,
ModelDao modelDao,
VectorDataType trainRequestVectorDataType,
KNNMethodContext trainRequestKnnMethodContext
) {
// Index metadata should not be null
if (indexMetadata == null) {
throw new IllegalArgumentException("IndexMetadata should not be null");
}
ValidationException exception = new ValidationException();
// Check the mapping
MappingMetadata mappingMetadata = indexMetadata.mapping();
if (mappingMetadata == null) {
exception.addValidationError("Invalid index. Index does not contain a mapping");
return exception;
}
// The mapping output *should* look like this:
// "{properties={field={type=knn_vector, dimension=8}}}"
Map<String, Object> properties = (Map<String, Object>) mappingMetadata.getSourceAsMap().get("properties");
if (properties == null) {
exception.addValidationError("Properties in map does not exists. This is unexpected");
return exception;
}
// Check field path is valid
if (StringUtils.isEmpty(field)) {
exception.addValidationError(String.format(Locale.ROOT, "Field path is empty."));
return exception;
}
Object fieldMapping = getFieldMapping(properties, field);
// Check field existence
if (fieldMapping == null) {
exception.addValidationError(String.format("Field \"%s\" does not exist.", field));
return exception;
}
// Check if field is a map. If not, that is a problem
if (!(fieldMapping instanceof Map)) {
exception.addValidationError(String.format("Field info for \"%s\" is not a map.", field));
return exception;
}
Map<String, Object> fieldMap = (Map<String, Object>) fieldMapping;
// Check fields type is knn_vector
Object type = fieldMap.get("type");
if (!(type instanceof String) || !KNNVectorFieldMapper.CONTENT_TYPE.equals(type)) {
exception.addValidationError(String.format("Field \"%s\" is not of type %s.", field, KNNVectorFieldMapper.CONTENT_TYPE));
return exception;
}
if (trainRequestVectorDataType != null) {
VectorDataType trainIndexDataType = getVectorDataTypeFromFieldMapping(fieldMap);
if (trainIndexDataType != trainRequestVectorDataType) {
exception.addValidationError(
String.format(
Locale.ROOT,
"Field \"%s\" has data type %s, which is different from data type used in the training request: %s",
field,
trainIndexDataType.getValue(),
trainRequestVectorDataType.getValue()
)
);
return exception;
}
// Block binary and byte vector data type for any encoder
if (trainRequestKnnMethodContext != null) {
MethodComponentContext methodComponentContext = trainRequestKnnMethodContext.getMethodComponentContext();
Map<String, Object> parameters = methodComponentContext.getParameters();
if (parameters != null && parameters.containsKey(KNNConstants.METHOD_ENCODER_PARAMETER)) {
MethodComponentContext encoder = (MethodComponentContext) parameters.get(KNNConstants.METHOD_ENCODER_PARAMETER);
if (encoder != null && VECTOR_DATA_TYPES_NOT_SUPPORTING_ENCODERS.contains(trainRequestVectorDataType)) {
exception.addValidationError(
String.format(
Locale.ROOT,
"encoder is not supported for vector data type [%s]",
trainRequestVectorDataType.getValue()
)
);
return exception;
}
}
}
}
// Return if dimension does not need to be checked
if (expectedDimension < 0) {
return null;
}
// Check that the dimension of the method passed in matches that of the model
Object dimension = fieldMap.get(KNNConstants.DIMENSION);
// If dimension is null, the training index/field could use a model. In this case, we need to get the model id
// for the index and then fetch its dimension from the models metadata
if (dimension == null) {
String modelId = (String) fieldMap.get(KNNConstants.MODEL_ID);
if (modelId == null) {
exception.addValidationError(String.format("Field \"%s\" does not have a dimension set.", field));
return exception;
}
if (modelDao == null) {
throw new IllegalArgumentException(String.format("Field \"%s\" uses model. modelDao cannot be null.", field));
}
ModelMetadata modelMetadata = modelDao.getMetadata(modelId);
if (!ModelUtil.isModelCreated(modelMetadata)) {
exception.addValidationError(String.format("Model \"%s\" for field \"%s\" is not created.", modelId, field));
return exception;
}
dimension = modelMetadata.getDimension();
if ((Integer) dimension != expectedDimension) {
exception.addValidationError(
String.format(
"Field \"%s\" has dimension %d, which is different from " + "dimension specified in the training request: %d",
field,
dimension,
expectedDimension
)
);
return exception;
}
return null;
}
// If the dimension was found in training fields mapping, check that it equals the models proposed dimension.
if ((Integer) dimension != expectedDimension) {
exception.addValidationError(
String.format(
"Field \"%s\" has dimension %d, which is different from " + "dimension specified in the training request: %d",
field,
dimension,
expectedDimension
)
);
return exception;
}
return null;
}
/**
* Gets the load time parameters for a given engine.
*
* @param spaceType Space for this particular segment
* @param knnEngine Engine used for the native library indices being loaded in
* @param indexName Name of OpenSearch index that the segment files belong to
* @param vectorDataType Vector data type for this particular segment
* @return load parameters that will be passed to the JNI.
*/
public static Map<String, Object> getParametersAtLoading(
SpaceType spaceType,
KNNEngine knnEngine,
String indexName,
VectorDataType vectorDataType
) {
Map<String, Object> loadParameters = Maps.newHashMap(ImmutableMap.of(SPACE_TYPE, spaceType.getValue()));
// For nmslib, we need to add the dynamic ef_search parameter that needs to be passed in when the
// hnsw graphs are loaded into memory
if (KNNEngine.NMSLIB.equals(knnEngine)) {
loadParameters.put(HNSW_ALGO_EF_SEARCH, KNNSettings.getEfSearchParam(indexName));
}
loadParameters.put(VECTOR_DATA_TYPE_FIELD, vectorDataType.getValue());
return Collections.unmodifiableMap(loadParameters);
}
public static boolean isClusterOnOrAfterMinRequiredVersion(String key) {
Version minimalRequiredVersion = minimalRequiredVersionMap.get(key);
if (minimalRequiredVersion == null) {
return false;
}
return KNNClusterUtil.instance().getClusterMinVersion().onOrAfter(minimalRequiredVersion);
}
public static boolean isVersionOnOrAfterMinRequiredVersion(Version version, String key) {
Version minimalRequiredVersion = minimalRequiredVersionMap.get(key);
if (minimalRequiredVersion == null) {
return false;
}
return version.onOrAfter(minimalRequiredVersion);
}
/**
* Checks if index requires shared state
*
* @param knnEngine The knnEngine associated with the index
* @param modelId The modelId associated with the index
* @param indexAddr Address to check if loaded index requires shared state
* @return true if state can be shared; false otherwise
*/
public static boolean isSharedIndexStateRequired(KNNEngine knnEngine, String modelId, long indexAddr) {
if (StringUtils.isEmpty(modelId)) {
return false;
}
return JNIService.isSharedIndexStateRequired(indexAddr, knnEngine);
}
/**
* Tell if it is binary index or not
*
* @param knnEngine knn engine associated with an index
* @param parameters parameters associated with an index
* @return true if it is binary index
*/
public static boolean isBinaryIndex(KNNEngine knnEngine, Map<String, Object> parameters) {
return KNNEngine.FAISS == knnEngine
&& parameters.get(VECTOR_DATA_TYPE_FIELD) != null
&& parameters.get(VECTOR_DATA_TYPE_FIELD).toString().equals(VectorDataType.BINARY.getValue());
}
/**
* Update vector data type into parameters
*
* @param parameters parameters associated with an index
* @param vectorDataType vector data type
*/
public static void updateVectorDataTypeToParameters(Map<String, Object> parameters, VectorDataType vectorDataType) {
if (VectorDataType.BINARY == vectorDataType) {
parameters.put(VECTOR_DATA_TYPE_FIELD, vectorDataType.getValue());
}
if (VectorDataType.BYTE == vectorDataType) {
parameters.put(VECTOR_DATA_TYPE_FIELD, vectorDataType.getValue());
}
}
/**
* This method retrieves the field mapping by a given field path from the index metadata.
*
* @param properties Index metadata mapping properties.
* @param fieldPath The field path string that make up the path to the field mapping. e.g. "a.b.field" or "field".
* The field path is applied and checked in OpenSearch, so it is guaranteed to be valid.
*
* @return The field mapping object if found, or null if the field is not found in the index metadata.
*/
private static Object getFieldMapping(final Map<String, Object> properties, final String fieldPath) {
String[] fieldPaths = fieldPath.split("\\.");
Object currentFieldMapping = properties;
// Iterate through the field path list to retrieve the field mapping.
for (String path : fieldPaths) {
currentFieldMapping = ((Map<String, Object>) currentFieldMapping).get(path);
if (currentFieldMapping == null) {
return null;
}
if (currentFieldMapping instanceof Map<?, ?>) {
Object possibleProperties = ((Map<String, Object>) currentFieldMapping).get("properties");
if (possibleProperties instanceof Map<?, ?>) {
currentFieldMapping = possibleProperties;
}
}
}
return currentFieldMapping;
}
/**
* This method is used to get the vector data type from field mapping
* @param fieldMap field mapping
* @return vector data type
*/
private static VectorDataType getVectorDataTypeFromFieldMapping(Map<String, Object> fieldMap) {
if (fieldMap.containsKey(VECTOR_DATA_TYPE_FIELD)) {
return VectorDataType.get((String) fieldMap.get(VECTOR_DATA_TYPE_FIELD));
}
return VectorDataType.DEFAULT;
}
/**
* Initialize the minimal required version map
*
* @return minimal required version map
*/
private static Map<String, Version> initializeMinimalRequiredVersionMap() {
final Map<String, Version> versionMap = new HashMap<>() {
{
put("ignore_unmapped", MINIMAL_SUPPORTED_VERSION_FOR_IGNORE_UNMAPPED);
put(MODEL_NODE_ASSIGNMENT_KEY, MINIMAL_SUPPORTED_VERSION_FOR_MODEL_NODE_ASSIGNMENT);
put(MODEL_METHOD_COMPONENT_CONTEXT_KEY, MINIMAL_SUPPORTED_VERSION_FOR_MODEL_METHOD_COMPONENT_CONTEXT);
put(KNNConstants.RADIAL_SEARCH_KEY, MINIMAL_SUPPORTED_VERSION_FOR_RADIAL_SEARCH);
put(KNNConstants.METHOD_PARAMETER, MINIMAL_SUPPORTED_VERSION_FOR_METHOD_PARAMETERS);
put(KNNConstants.MODEL_VECTOR_DATA_TYPE_KEY, MINIMAL_SUPPORTED_VERSION_FOR_MODEL_VECTOR_DATA_TYPE);
put(RESCORE_PARAMETER, MINIMAL_RESCORE_FEATURE);
}
};
for (final MethodParameter methodParameter : MethodParameter.values()) {
if (methodParameter.getVersion() != null) {
versionMap.put(methodParameter.getName(), methodParameter.getVersion());
}
}
return Collections.unmodifiableMap(versionMap);
}
/**
* Tell if it is byte index or not
*
* @param parameters parameters associated with an index
* @return true if it is binary index
*/
public static boolean isByteIndex(Map<String, Object> parameters) {
return parameters.getOrDefault(VECTOR_DATA_TYPE_FIELD, VectorDataType.DEFAULT.getValue())
.toString()
.equals(VectorDataType.BYTE.getValue());
}
}