danny0405 commented on code in PR #20085: URL: https://github.com/apache/hudi/pull/20085#discussion_r4118407671
########## hudi-common/src/main/java/org/apache/hudi/common/util/InternalSchemaHistory.java: ########## @@ -0,0 +1,339 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.hudi.common.util; + +import org.apache.hudi.common.model.HoodieCommitMetadata; +import org.apache.hudi.common.schema.HoodieSchemaUtils; +import org.apache.hudi.common.schema.internal.InternalSchema; +import org.apache.hudi.common.schema.internal.convert.InternalSchemaConverter; +import org.apache.hudi.common.schema.internal.io.FileBasedInternalSchemaStorageManager; +import org.apache.hudi.common.schema.internal.utils.SerDeHelper; +import org.apache.hudi.common.table.HoodieTableMetaClient; +import org.apache.hudi.common.table.timeline.InstantFileNameParser; +import org.apache.hudi.exception.HoodieException; +import org.apache.hudi.exception.HoodieIOException; +import org.apache.hudi.storage.StoragePath; + +import com.github.benmanes.caffeine.cache.Cache; +import com.github.benmanes.caffeine.cache.Caffeine; +import lombok.extern.slf4j.Slf4j; + +import java.io.FileNotFoundException; +import java.io.IOException; +import java.io.Serializable; +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Collections; +import java.util.HashMap; +import java.util.LinkedHashSet; +import java.util.List; +import java.util.Map; +import java.util.Set; +import java.util.TreeMap; +import java.util.concurrent.CompletableFuture; +import java.util.concurrent.CompletionException; +import java.util.concurrent.ExecutorService; +import java.util.concurrent.Executors; +import java.util.function.Function; +import java.util.stream.Collectors; + +/** + * The schema history of a table as seen by one query, used to find the {@link InternalSchema} a data file + * was written with from the file's commit time. + * + * <p>It is loaded once where the query is planned, restricted to the query's valid commits, and shipped to + * the readers, so resolving a file's schema reads no table metadata. It resolves a version id exactly like + * {@link InternalSchemaCache#getInternalSchemaByVersionId(long, String, org.apache.hudi.storage.HoodieStorage, + * String, org.apache.hudi.common.table.timeline.TimelineLayout, org.apache.hudi.common.table.HoodieTableConfig)}: + * a valid commit resolves to the {@code latest_schema} in its own commit metadata, which is not always the history + * version in effect at its commit time (a schema change requested before the commit may complete after it). + * + * <p>It is kept as configuration entries, so it can travel inside a Hadoop configuration and a reader looks up + * only the entry it needs. + */ +@Slf4j +public final class InternalSchemaHistory implements Serializable { + + private static final long serialVersionUID = 1L; + + private static final String CONFIG_PREFIX = "hoodie.internal.schema.history."; + // comma-separated version ids of the schema history + private static final String VERSIONS_KEY = CONFIG_PREFIX + "versions"; + private static final String VERSION_KEY_PREFIX = CONFIG_PREFIX + "version."; + // comma-separated "<commit time>=<version id>" for commits whose schema is a history version, + // and "<commit time>=#<n>" for commits whose schema is kept under SCHEMA_KEY_PREFIX + n + private static final String COMMITS_KEY = CONFIG_PREFIX + "commits"; + // schemas of commits that are not history versions, each distinct schema kept once + private static final String SCHEMA_KEY_PREFIX = CONFIG_PREFIX + "schema."; + private static final char SCHEMA_REF_MARKER = '#'; + + private static final int MAX_COMMIT_READ_PARALLELISM = 16; + + // Completed commit files never change, so what each one says about its schema is kept across queries. + private static final Cache<String, CommitSchema> COMMIT_SCHEMA_CACHE = Caffeine.newBuilder() + .maximumWeight(16L * 1024 * 1024) + .weigher((String commitFile, CommitSchema commitSchema) -> commitFile.length() + Math.max(1, commitSchema.weight)) + .build(); + + private final Map<String, String> configs; + + private InternalSchemaHistory(Map<String, String> configs) { + this.configs = configs; + } + + /** + * Loads the schema history of the table restricted to the given commits. + * + * @param metaClient meta client of the table + * @param validCommits comma-separated instant file names of the commits the query may read; when empty, the + * completed commits of the meta client's timeline + */ + public static InternalSchemaHistory load(HoodieTableMetaClient metaClient, String validCommits) { + InstantFileNameParser fileNameParser = metaClient.getInstantFileNameParser(); + Set<String> commitFileNames = StringUtils.isNullOrEmpty(validCommits) + ? Collections.emptySet() + : new LinkedHashSet<>(Arrays.asList(validCommits.split(","))); + List<String> commitTimes = commitFileNames.stream().map(fileNameParser::extractTimestamp).collect(Collectors.toList()); + + String historySchemaStr = new FileBasedInternalSchemaStorageManager(metaClient).getHistorySchemaStrByGivenValidCommits(commitTimes); + TreeMap<Long, InternalSchema> history = historySchemaStr.isEmpty() ? new TreeMap<>() : SerDeHelper.parseSchemas(historySchemaStr); + + Map<String, String> configs = new HashMap<>(); + configs.put(VERSIONS_KEY, history.keySet().stream().map(String::valueOf).collect(Collectors.joining(","))); + history.forEach((versionId, schema) -> configs.put(VERSION_KEY_PREFIX + versionId, SerDeHelper.toJson(schema))); + + List<String> commitEntries = new ArrayList<>(); + Map<String, Integer> schemaRefs = new HashMap<>(); + List<String> resolvableCommitFiles = commitFileNames.stream() + .filter(commitFileName -> parseVersionId(fileNameParser.extractTimestamp(commitFileName)).isPresent()) + .collect(Collectors.toList()); + Map<String, CommitSchema> commitSchemas = readCommitSchemas(metaClient, resolvableCommitFiles); + for (String commitFileName : resolvableCommitFiles) { + CommitSchema commitSchema = commitSchemas.get(commitFileName); + if (commitSchema == null) { + continue; + } + String commitTime = fileNameParser.extractTimestamp(commitFileName); + if (commitSchema.hasLatestSchema) { + InternalSchema latestSchema = commitSchema.latestSchema; + if (latestSchema != null && latestSchema.equals(history.get(latestSchema.schemaId()))) { + commitEntries.add(commitTime + "=" + latestSchema.schemaId()); + } else { + commitEntries.add(commitTime + "=" + schemaRef(latestSchema == null ? "" : SerDeHelper.toJson(latestSchema), schemaRefs, configs)); + } + } else if (commitSchema.avroSchema != null && !history.isEmpty() && history.floorKey(parseVersionId(commitTime).get()) == null) { + commitEntries.add(commitTime + "=" + schemaRef(SerDeHelper.toJson(commitSchema.avroSchema), schemaRefs, configs)); + } + } + configs.put(COMMITS_KEY, String.join(",", commitEntries)); + return new InternalSchemaHistory(configs); + } + + /** + * Returns the reference of a commit schema that is not a history version, storing each distinct schema once. + */ + private static String schemaRef(String schemaJson, Map<String, Integer> schemaRefs, Map<String, String> configs) { + Integer ref = schemaRefs.get(schemaJson); + if (ref == null) { + ref = schemaRefs.size(); + schemaRefs.put(schemaJson, ref); + configs.put(SCHEMA_KEY_PREFIX + ref, schemaJson); + } + return SCHEMA_REF_MARKER + String.valueOf(ref); + } + + /** + * Returns the schema a file written by the given commit was written with, or an empty schema if unknown. + * The result is {@code null} when the commit metadata holds an empty internal schema. + */ + public InternalSchema getSchemaByVersionId(long versionId) { + return resolve(configs::get, versionId); + } + + /** + * Returns the history as configuration entries, to ship it inside a configuration. + */ + public Map<String, String> toConfigs() { + return Collections.unmodifiableMap(configs); + } + + /** + * Returns whether the configuration holds a history written by {@link #toConfigs()}. + * + * @param getConfig returns the value of a configuration key, or {@code null} if absent + */ + public static boolean isPresentIn(Function<String, String> getConfig) { + return getConfig.apply(VERSIONS_KEY) != null; + } + + /** + * Same as {@link #getSchemaByVersionId} on a history written by {@link #toConfigs()} into a configuration, + * reading only the entries needed for this version id. + * + * @param getConfig returns the value of a configuration key, or {@code null} if absent + */ + public static InternalSchema resolve(Function<String, String> getConfig, long versionId) { + String commitTime = String.valueOf(versionId); + for (String entry : splitList(getConfig.apply(COMMITS_KEY))) { + int separator = entry.indexOf('='); + if (entry.substring(0, separator).equals(commitTime)) { + String target = entry.substring(separator + 1); + return target.charAt(0) == SCHEMA_REF_MARKER + ? SerDeHelper.fromJson(getConfig.apply(SCHEMA_KEY_PREFIX + target.substring(1))).orElse(null) + : parseSchema(getConfig.apply(VERSION_KEY_PREFIX + target)); + } + } + Long floorVersionId = null; + for (String version : splitList(getConfig.apply(VERSIONS_KEY))) { + long id = Long.parseLong(version); + if (id <= versionId && (floorVersionId == null || id > floorVersionId)) { + floorVersionId = id; + } + } + return floorVersionId == null ? InternalSchema.getEmptyInternalSchema() : parseSchema(getConfig.apply(VERSION_KEY_PREFIX + floorVersionId)); + } + + /** + * Returns whether the key is one {@link #toConfigs()} writes. + */ + public static boolean isConfigKey(String key) { + return key.startsWith(CONFIG_PREFIX); + } + + private static InternalSchema parseSchema(String json) { + return SerDeHelper.fromJson(json).orElse(InternalSchema.getEmptyInternalSchema()); + } + + private static List<String> splitList(String value) { + return StringUtils.isNullOrEmpty(value) ? Collections.emptyList() : Arrays.asList(value.split(",")); + } + + /** + * A commit time names a schema version only in its canonical numeric form, which is how + * {@link InternalSchemaCache#getInternalSchemaByVersionId} matches it. + */ + private static Option<Long> parseVersionId(String commitTime) { + try { + long versionId = Long.parseLong(commitTime); + return String.valueOf(versionId).equals(commitTime) ? Option.of(versionId) : Option.empty(); + } catch (NumberFormatException e) { + return Option.empty(); + } + } + + /** + * Returns what each commit file says about its schema, keyed by commit file name; a commit file that no longer + * exists (archived after the timeline was loaded) is left out. Files not cached yet are read in parallel. + */ + private static Map<String, CommitSchema> readCommitSchemas(HoodieTableMetaClient metaClient, List<String> commitFileNames) { + Map<String, CommitSchema> result = new HashMap<>(); + List<StoragePath> toRead = new ArrayList<>(); + for (String commitFileName : commitFileNames) { + StoragePath commitFile = new StoragePath(metaClient.getTimelinePath(), commitFileName); + CommitSchema cached = COMMIT_SCHEMA_CACHE.getIfPresent(commitFile.toString()); + if (cached != null) { + result.put(commitFileName, cached); + } else { + toRead.add(commitFile); + } + } + if (toRead.isEmpty()) { + return result; + } + ExecutorService pool = Executors.newFixedThreadPool(Math.min(toRead.size(), MAX_COMMIT_READ_PARALLELISM), + new CustomizedThreadFactory("internal-schema-history-load", true)); + try { + List<CompletableFuture<Option<CommitSchema>>> futures = toRead.stream() + .map(commitFile -> CompletableFuture.supplyAsync(() -> readCommitSchema(metaClient, commitFile), pool)) + .collect(Collectors.toList()); + List<Option<CommitSchema>> commitSchemas = FutureUtils.allOf(futures).join(); + for (int i = 0; i < toRead.size(); i++) { + if (commitSchemas.get(i).isPresent()) { + StoragePath commitFile = toRead.get(i); + if (commitSchemas.get(i).get().cacheable) { + COMMIT_SCHEMA_CACHE.put(commitFile.toString(), commitSchemas.get(i).get()); + } + result.put(commitFile.getName(), commitSchemas.get(i).get()); + } + } + } catch (CompletionException e) { + throw e.getCause() instanceof RuntimeException ? (RuntimeException) e.getCause() : new HoodieException(e.getCause()); + } finally { + pool.shutdownNow(); + } + return result; + } + + private static Option<CommitSchema> readCommitSchema(HoodieTableMetaClient metaClient, StoragePath commitFile) { + byte[] content; + try { + content = InternalSchemaCache.readCommitFile(metaClient.getStorage(), commitFile); + } catch (FileNotFoundException e) { + log.warn("Commit file {} no longer exists, resolving its files from the schema history", commitFile); + return Option.empty(); + } catch (IOException e) { + throw new HoodieIOException("Could not read commit file " + commitFile, e); + } + HoodieCommitMetadata metadata; + try { + metadata = InternalSchemaCache.deserializeCommitMetadata(content, commitFile, metaClient.getTimelineLayout()); + } catch (Exception e) { + log.warn("Cannot parse commit file {}, resolving its files from the schema history", commitFile, e); + return Option.of(new CommitSchema(false, null, null, 0, false)); + } + return Option.of(CommitSchema.of(metadata)); + } + + /** + * The schema facts of one commit that {@link #load} needs: its {@code latest_schema}, or else the internal + * schema converted from its Avro schema. + */ + private static final class CommitSchema { + private final boolean hasLatestSchema; Review Comment: doc these variables with comments. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
