Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 | 236x 6584x 6584x 6584x 6584x 6409x 2x 2x 2x 6407x 6407x 12773x 6418x 6391x 6391x 6391x 6336x 55x 55x 30x 25x 6366x 6382x 6382x 244x 244x 244x 244x 181x 181x 244x 244x 5x 5x 239x 239x 440x 440x 239x 66x 239x 239x 239x 201x 185x 5x 5x 5x 5x 5x 180x 8x 1x 1x 1x 7x 172x 162x 162x 162x 10x 10x 10x 10x 16x 201x 6425x 6425x 6351x 2x 6349x 6349x 32116x 32116x 32116x 32116x 21449x 10667x 10667x 6349x 29x 29x 29x 10x 10x 10x 29x 8x 21x 21x 6x 2x 5x 5x 14x 14x 5x 9x 7x 2x 14x 7x 14x 14x 9x 14x 9x 9x 9x 5x | import { DrainBase } from "./DrainBase.js";
import { LogCluster } from "./LogCluster.js";
import { Node } from "./Node.js";
import { MatchStrategy } from "./types.js";
import type { DrainOptions } from "./types.js";
/**
* Concrete Drain algorithm implementation.
*
* Maps 1:1 to Python `Drain` class (drain.py L278-end).
*
* Implements the fixed-depth prefix tree search, similarity-based clustering,
* template merging, and LRU eviction logic exactly as specified by the
* original algorithm (He et al., ICWS 2017) and the official Drain3
* Python implementation.
*
* Algorithm complexity: O((d + cm)n) where:
* - d = tree depth
* - c = average candidate clusters per leaf
* - m = average tokens per message
* - n = total messages
*
* @example
* ```typescript
* const drain = new Drain({ depth: 4, simTh: 0.4 });
* const { cluster, changeType } = drain.addLogMessage("user alice logged in");
* console.log(cluster.getTemplate()); // "user alice logged in"
*
* const { cluster: c2 } = drain.addLogMessage("user bob logged in");
* console.log(c2.getTemplate()); // "user <*> logged in"
* ```
*/
export class Drain extends DrainBase {
/**
* @param options - Configuration options forwarded to DrainBase.
*/
constructor(options: DrainOptions = {}) {
super(options);
}
/**
* Searches the prefix tree for the best-matching cluster.
*
* Python: Drain.tree_search(root_node, tokens, sim_th, include_params)
*
* Search strategy (identical to Python):
* 1. Look up the root key node (token_count or {tc}#{pc} with param binning)
* 2. For empty tokens, return the first (only) cluster directly
* 3. Walk down the tree for up to maxNodeDepth levels:
* - Exact token match → follow that path
* - No exact match but `<*>` wildcard exists → follow wildcard
* - Neither → return null (no candidates)
* 4. At leaf node, run fastMatch over the candidate cluster list
*
* @param rootNode - Root of the prefix tree.
* @param tokens - Tokenized log message.
* @param simTh - Similarity threshold (0.4 for training, 1.0 for inference).
* @param includeParams - Whether to count parameter tokens as matches.
* @returns The best-matching cluster or null.
*/
treeSearch(
rootNode: Node,
tokens: readonly string[],
simTh: number,
includeParams: boolean,
): LogCluster | null {
// Step 1: Locate the root key node (handles param binning)
const tokenCount = tokens.length;
const rootKey = this.getRootKey(tokens);
let curNode = rootNode.keyToChildNode.get(rootKey);
if (!curNode) return null;
// Step 2: Empty tokens → return the first cluster directly
if (tokenCount === 0) {
const firstId = curNode.clusterIds[0];
Iif (firstId === undefined) return null;
return this.idToCluster.get(firstId) ?? null;
}
// Step 3: Walk down the prefix tree
let curNodeDepth = 1;
for (const token of tokens) {
// Python: if cur_node_depth >= self.max_node_depth: break
if (curNodeDepth >= this.maxNodeDepth) break;
// Python: if cur_node_depth == token_count: break
if (curNodeDepth === tokenCount) break;
const children: Map<string, Node> = curNode.keyToChildNode;
const exactNode: Node | undefined = children.get(token);
if (exactNode) {
curNode = exactNode;
} else {
const paramNode = children.get(this.paramStr);
if (paramNode) {
curNode = paramNode;
} else {
return null;
}
}
curNodeDepth++;
}
// At this point curNode is guaranteed to be a Node (not undefined),
// because the null check above and the loop both ensure valid assignments.
// However, TypeScript can't track this through the loop body
// with the early breaks. We assert the type here.
const leafNode: Node = curNode;
// Step 4: Fast match over candidate clusters
return this.fastMatch(leafNode.clusterIds, tokens, simTh, includeParams);
}
// ============================================================
// addSeqToPrefixTree (maps to Python Drain.add_seq_to_prefix_tree, drain.py L324-L388)
// ============================================================
/**
* Inserts a cluster into the prefix tree.
*
* Python: Drain.add_seq_to_prefix_tree(root_node, cluster)
*
* This is the most complex method in the implementation. The logic for
* handling `maxChildren` and `parametrizeNumericTokens` follows the
* exact branching structure of the Python code.
*
* Insertion strategy:
* 1. First level: token count node
* 2. For each token in the template:
* a. Already at max depth or last token → append to leaf node's clusterIds
* b. Token exists in current node → follow existing path
* c. Token doesn't exist:
* - Contains digits + parametrizeNumericTokens → use `<*>` node
* - No digits → decide based on maxChildren and `<*>` presence
*
* @param rootNode - Root of the prefix tree.
* @param cluster - The cluster to insert.
*/
addSeqToPrefixTree(rootNode: Node, cluster: LogCluster): void {
const tokenCount = cluster.logTemplateTokens.length;
const rootKey = this.getRootKey(cluster.logTemplateTokens);
// Level 1: Root key node (token_count or {tc}#{pc})
let firstLayerNode = rootNode.keyToChildNode.get(rootKey);
if (!firstLayerNode) {
firstLayerNode = new Node();
rootNode.keyToChildNode.set(rootKey, firstLayerNode);
}
let curNode = firstLayerNode;
// Empty log message
if (tokenCount === 0) {
curNode.clusterIds = [cluster.clusterId];
return;
}
let currentDepth = 1;
for (let i = 0; i < tokenCount; i++) {
// tokenCount === logTemplateTokens.length by construction;
// array access by bounded index is always defined.
const token = cluster.logTemplateTokens[i]!;
// At maximum depth or last token → add to leaf node's clusterIds
// Python: if current_depth >= self.max_node_depth or current_depth >= token_count
if (currentDepth >= this.maxNodeDepth || currentDepth >= tokenCount) {
// Filter out clusters that have been evicted, then append the new one
const newClusterIds = curNode.clusterIds.filter((cid) =>
this.idToCluster.has(cid),
);
newClusterIds.push(cluster.clusterId);
curNode.clusterIds = newClusterIds;
break;
}
// Token doesn't exist at this level
if (!curNode.keyToChildNode.has(token)) {
if (this.parametrizeNumericTokens && Drain.hasNumbers(token)) {
// Numeric token → route through <*> wildcard node
let paramNode = curNode.keyToChildNode.get(this.paramStr);
Eif (!paramNode) {
paramNode = new Node();
curNode.keyToChildNode.set(this.paramStr, paramNode);
}
curNode = paramNode;
} else {
// Non-numeric token
if (curNode.keyToChildNode.has(this.paramStr)) {
// <*> node already exists
if (curNode.keyToChildNode.size < this.maxChildren) {
const newNode = new Node();
curNode.keyToChildNode.set(token, newNode);
curNode = newNode;
} else {
curNode = curNode.keyToChildNode.get(this.paramStr)!;
}
} else {
// <*> node does not exist
if (curNode.keyToChildNode.size + 1 < this.maxChildren) {
const newNode = new Node();
curNode.keyToChildNode.set(token, newNode);
curNode = newNode;
} else Eif (curNode.keyToChildNode.size + 1 === this.maxChildren) {
const newNode = new Node();
curNode.keyToChildNode.set(this.paramStr, newNode);
curNode = newNode;
}
// size + 1 > maxChildren is unreachable: the `=== maxChildren`
// branch above creates the <*> node before size can exceed maxChildren.
// Once <*> exists, the outer `has(this.paramStr)` branch handles routing.
}
}
} else {
// Token already exists → follow the existing path
curNode = curNode.keyToChildNode.get(token)!;
}
currentDepth++;
}
}
// ============================================================
// getSeqDistance (maps to Python Drain.get_seq_distance, drain.py L391-L413)
// ============================================================
/**
* Calculates the similarity score and parameter count between two
* token sequences.
*
* Python: Drain.get_seq_distance(seq1, seq2, include_params)
*
* Formula:
* ```
* similarity = (simTokens + (paramCount if includeParams else 0)) / len(seq1)
* ```
*
* Where:
* - seq1 is the template sequence (may contain `<*>` placeholders)
* - seq2 is the log message sequence
* - Tokens matching `<*>` in seq1 are skipped (counted as params)
* - Matching non-parameter tokens increment simTokens
*
* @param seq1 - Template tokens (may contain paramStr placeholders).
* @param seq2 - Log message tokens to compare against.
* @param includeParams - Whether to count parameter tokens toward similarity.
* @returns Similarity score [0.0, 1.0] and parameter count.
*/
getSeqDistance(
seq1: readonly string[],
seq2: readonly string[],
includeParams: boolean,
): { similarity: number; paramCount: number } {
const result = this.similarityChain.compute(
seq1,
seq2,
this.paramStr,
includeParams,
);
return { similarity: result.similarity, paramCount: result.paramCount };
}
// ============================================================
// createTemplate (maps to Python Drain.create_template, drain.py L415-L425)
// ============================================================
/**
* Creates a merged template from two token sequences.
*
* Python: Drain.create_template(seq1, seq2)
*
* Comparison strategy (extended):
* - seq1[ i ] == seq2[ i ] → keep the token (it's a constant)
* - seq1[ i ] != seq2[ i ] → use strategy chain to parameterize
*
* The strategy chain enables advanced parameterization patterns:
* - Exact match: keep token as-is
* - Affix-preserving: "bytes0sent" + "bytes403sent" → "bytes<*>sent"
* - Regex-based: custom patterns for structured formats
* - Full token: replace with paramStr (Drain3 default)
*
* This is how Drain gradually generalizes templates — each time a new
* message matches a cluster, positions that differ are replaced with
* the parameter placeholder.
*
* @param seq1 - New log message tokens.
* @param seq2 - Existing template tokens.
* @returns The merged template tokens (frozen array).
*/
createTemplate(
seq1: readonly string[],
seq2: readonly string[],
): readonly string[] {
if (seq1.length !== seq2.length) {
throw new Error(
`createTemplate: sequence length mismatch (${seq1.length} vs ${seq2.length})`,
);
}
const result: string[] = [];
for (let i = 0; i < seq1.length; i++) {
const token1 = seq1[i]!;
const token2 = seq2[i]!;
Iif (this.enableMaskParamGeneralization && this.isMaskedParam(token1)) {
// Masked parameter tokens → immediately generalize to <*>
// (only when enableMaskParamGeneralization is active)
result.push(this.paramStr);
} else if (token1 === token2) {
// Exact match — keep the token
result.push(token2);
} else {
// Use strategy chain for parameterization
const paramResult = this.strategyChain.parameterize(
token1,
token2,
this.paramStr,
);
result.push(paramResult.templateToken);
}
}
return Object.freeze(result);
}
// ============================================================
// match — Inference mode (maps to Python Drain.match, drain.py L427-L472)
// ============================================================
/**
* Matches a log message against existing clusters WITHOUT modifying state.
*
* Python: Drain.match(content, full_search_strategy)
*
* This is the inference/classification mode. Key differences from training:
* - simTh is fixed at 1.0 (requires perfect match)
* - includeParams is always true (parameters count as matches)
* - No new clusters are created
* - No existing templates are modified
*
* Three search strategies:
* - "never": Tree search only. Fastest, may miss matches.
* - "fallback": Tree search first, then full search if no match.
* - "always": Full linear search over all same-length clusters.
*
* @param content - The log message to classify.
* @param fullSearchStrategy - Which search strategy to use. Default: "never".
* @returns The matching cluster, or null if no perfect match exists.
*/
match(
content: string,
fullSearchStrategy: MatchStrategy = MatchStrategy.Never,
): LogCluster | null {
const REQUIRED_SIM_TH = 1.0;
const contentTokens = this.getContentAsTokens(content);
// Python: def full_search() → return self.fast_match(all_ids, content_tokens, required_sim_th, include_params=True)
const fullSearch = (): LogCluster | null => {
const rootKey = this.getRootKey(contentTokens);
const allIds = this._getClustersIdsForRootKey(rootKey);
return this.fastMatch(allIds, contentTokens, REQUIRED_SIM_TH, true);
};
// Python: if full_search_strategy == "always": return full_search()
if (fullSearchStrategy === MatchStrategy.Always) {
return fullSearch();
}
// Python: match_cluster = self.tree_search(...)
const matchCluster = this.treeSearch(
this.rootNode,
contentTokens,
REQUIRED_SIM_TH,
true,
);
if (matchCluster !== null) return matchCluster;
// Python: if full_search_strategy == "never": return None
if (fullSearchStrategy === MatchStrategy.Never) return null;
// Python: return full_search() (fallback)
return fullSearch();
}
// ============================================================
// printTree (maps to Python Drain.print_tree/print_node, drain.py L474-end)
// ============================================================
/**
* Prints the prefix tree structure for debugging.
*
* Python: Drain.print_tree(file, max_clusters) + Drain.print_node(...)
*
* @param stream - Output stream (default: process.stdout).
* @param maxClusters - Max clusters to show per node (default: 5).
*/
printTree(stream?: NodeJS.WritableStream, maxClusters: number = 5): void {
const out = stream ?? process.stdout;
const printNode = (token: string, node: Node, depth: number): void => {
const indent = "\t".repeat(depth);
let line: string;
if (depth === 0) {
line = `<${token}>`;
} else if (depth === 1) {
// Token count layer: wrap with L= prefix if numeric
line = /^\d+$/.test(token) ? `<L=${token}>` : `<${token}>`;
} else {
line = `"${token}"`;
}
if (node.clusterIds.length > 0) {
line += ` (cluster_count=${node.clusterIds.length})`;
}
out.write(indent + line + "\n");
for (const [childToken, childNode] of node.keyToChildNode) {
printNode(childToken, childNode, depth + 1);
}
for (const cid of node.clusterIds.slice(0, maxClusters)) {
const cluster = this.idToCluster.get(cid);
Eif (cluster) {
out.write("\t".repeat(depth + 1) + cluster.toString() + "\n");
}
}
};
printNode("root", this.rootNode, 0);
}
}
|