diff --git pom.xml pom.xml index 21ce5cbff1..50f416e61f 100644 --- pom.xml +++ pom.xml @@ -106,7 +106,7 @@ 2.4 2.4 3.1.0 - 2.20.1 + 2.21.0 2.4 2.8 2.9 diff --git ql/src/test/org/apache/hadoop/hive/ql/TestTxnExIm.java ql/src/test/org/apache/hadoop/hive/ql/TestTxnExIm.java index 0e53697be2..6daac1b789 100644 --- ql/src/test/org/apache/hadoop/hive/ql/TestTxnExIm.java +++ ql/src/test/org/apache/hadoop/hive/ql/TestTxnExIm.java @@ -19,6 +19,7 @@ import org.apache.hadoop.hive.conf.HiveConf; import org.apache.hadoop.hive.metastore.conf.MetastoreConf; +import org.apache.hadoop.hive.metastore.tools.HiveMetaTool; import org.junit.Assert; import org.junit.Ignore; import org.junit.Test; @@ -535,4 +536,40 @@ public void testMMExportAborted() throws Exception { TestTxnCommands2.stringifyValues(data), rs); } + @Test + public void testUpgrade() throws Exception { + int[][] data = {{1,2}, {3, 4}, {5, 6}}; + int[][] dataPart = {{1, 2, 10}, {3, 4, 11}, {5, 6, 12}}; + runStatementOnDriver("drop table if exists TAcid"); + runStatementOnDriver("drop table if exists TAcidPart"); + runStatementOnDriver("drop table if exists TFlat"); + runStatementOnDriver("drop table if exists TFlatText"); + runStatementOnDriver("create table TAcid (a int, b int) stored as orc"); + runStatementOnDriver("create table TAcidPart (a int, b int) partitioned by (p int) stored" + + " as orc"); + runStatementOnDriver("create table TFlat (a int, b int) stored as orc tblproperties('transactional'='false')"); + runStatementOnDriver("create table TFlatText (a int, b int) stored as textfile tblproperties('transactional'='false')"); + + + //this needs major compaction + runStatementOnDriver("insert into TAcid" + TestTxnCommands2.makeValuesClause(data)); + runStatementOnDriver("update TAcid set a = 1 where b = 2"); + + //this table needs to be converted to Acid + runStatementOnDriver("insert into TFlat" + TestTxnCommands2.makeValuesClause(data)); + + //this table needs to be converted to MM + runStatementOnDriver("insert into TFlatText" + TestTxnCommands2.makeValuesClause(data)); + + //p=10 needs major compaction + runStatementOnDriver("insert into TAcidPart" + TestTxnCommands2.makeValuesClause(dataPart)); + runStatementOnDriver("update TAcidPart set a = 1 where b = 2 and p = 10"); + + //todo: add partitioned table that needs conversion to MM/Acid + + //todo: rename files case + String[] args = new String[1]; + args[0] = new String("-prepareAcidUpgrade"); + HiveMetaTool.main(args); + } } diff --git standalone-metastore/pom.xml standalone-metastore/pom.xml index c340fe2d24..10b1bfadf3 100644 --- standalone-metastore/pom.xml +++ standalone-metastore/pom.xml @@ -80,6 +80,7 @@ 0.9.3 2.8.2 1.10.19 + 1.4.3 2.5.0 1.3.0 2.6.0-SNAPSHOT @@ -92,6 +93,21 @@ + + org.apache.orc + orc-core + ${orc.version} + + + org.apache.hadoop + hadoop-common + + + org.apache.hive + hive-storage-api + + + com.fasterxml.jackson.core jackson-databind diff --git standalone-metastore/src/main/java/org/apache/hadoop/hive/metastore/tools/HiveMetaTool.java standalone-metastore/src/main/java/org/apache/hadoop/hive/metastore/tools/HiveMetaTool.java index f4eacd5fb9..02e8c694e8 100644 --- standalone-metastore/src/main/java/org/apache/hadoop/hive/metastore/tools/HiveMetaTool.java +++ standalone-metastore/src/main/java/org/apache/hadoop/hive/metastore/tools/HiveMetaTool.java @@ -18,8 +18,13 @@ package org.apache.hadoop.hive.metastore.tools; +import java.io.FileWriter; +import java.io.IOException; +import java.io.PrintWriter; import java.net.URI; +import java.util.ArrayList; import java.util.Collection; +import java.util.Collections; import java.util.Iterator; import java.util.List; import java.util.Map; @@ -34,7 +39,25 @@ import org.apache.commons.cli.Options; import org.apache.commons.cli.ParseException; import org.apache.hadoop.conf.Configuration; +import org.apache.hadoop.fs.ContentSummary; +import org.apache.hadoop.fs.FileStatus; +import org.apache.hadoop.fs.FileSystem; +import org.apache.hadoop.fs.PathFilter; +import org.apache.hadoop.hive.metastore.HiveMetaStoreClient; +import org.apache.hadoop.hive.metastore.TableType; +import org.apache.hadoop.hive.metastore.Warehouse; +import org.apache.hadoop.hive.metastore.api.MetaException; +import org.apache.hadoop.hive.metastore.api.Partition; +import org.apache.hadoop.hive.metastore.api.StorageDescriptor; +import org.apache.hadoop.hive.metastore.api.Table; +import org.apache.hadoop.hive.metastore.api.hive_metastoreConstants; import org.apache.hadoop.hive.metastore.conf.MetastoreConf; +import org.apache.hadoop.hive.metastore.utils.StringUtils; +import org.apache.orc.OrcFile; +import org.apache.orc.Reader; +import org.apache.orc.impl.AcidStats; +import org.apache.orc.impl.OrcAcidUtils; +import org.apache.thrift.TException; import org.slf4j.Logger; import org.slf4j.LoggerFactory; import org.apache.hadoop.fs.Path; @@ -104,6 +127,12 @@ private void init() { .withDescription("Specify the key for table property to be updated. tablePropKey option " + "is valid only with updateLocation option.") .create("tablePropKey"); + Option prepareAcidUpgrade = + OptionBuilder.withArgName("find-compactions") + .hasOptionalArg() //directory to output results to + .withDescription("Generates a set Compaction commands to run to prepare for Hive 2.x" + + " to 3.0 upgrade") + .create("prepareAcidUpgrade"); cmdLineOptions.addOption(help); cmdLineOptions.addOption(listFSRoot); @@ -112,6 +141,7 @@ private void init() { cmdLineOptions.addOption(dryRun); cmdLineOptions.addOption(serdePropKey); cmdLineOptions.addOption(tablePropKey); + cmdLineOptions.addOption(prepareAcidUpgrade); } private void initObjectStore(Configuration conf) { @@ -362,7 +392,358 @@ public void updateFSRootLocation(URI oldURI, URI newURI, String serdePropKey, St printSerdePropURIUpdateSummary(updateSerdeURIretVal, serdePropKey, isDryRun); } } + private void prepareAcidUpgrade(HiveMetaTool metaTool) { + try { + prepareAcidUpgradeInternal(); + } + catch(TException|IOException ex) { + System.err.println(StringUtils.stringifyException(ex)); + printAndExit(metaTool); + } + } + private static class CompactionMetaInfo { + /** + * total number of bytes to be compacted across all compaction commands + */ + long numberOfBytes; + } + /** + * todo: make sure compaction queue is configured and has ample capacity + * todo: what to do on failure? Suppose some table/part is not readable. should it produce + * todo: should probably suppor dryRun mode where we output scripts but instead of renaming files + * we generate a renaming script. Alternatively, always generate a renaming script and have + * user execute it - this is probably a better option. If script is not empty on rerun someone + * added files to table to be made Acid. + * commands for all other tables? + * todo: how do we test this? even if we had 2.x data it won't be readable in 3.0. even w/o any + * updates, txnids in the data won't make sense in 3.0 w/o actually performing equivalent of + * metastore upgrade to init writeid table. Also, can we even create a new table and set location + * to existing files to simulate a 2.x table? + * todo: generate some instructions in compaction script to include tottal compactions to perform, + * total data volume to handle and anything else that may help users guess at how long it will + * take. Also, highlight tuning options to speed this up. + * todo: can we make the script blocking so that it waits for cleaner to delete files? + * need to poll SHOW COMPACTIONS and make sure that all partitions are in "finished" state + * todo: this should accept a file of table names to exclude from non-acid to acid conversion + * todo: change script comments to a preamble instead of a footer + * + * @throws MetaException + * @throws TException + */ + private void prepareAcidUpgradeInternal() throws MetaException, TException, IOException { + Configuration conf = MetastoreConf.newMetastoreConf(); + System.out.println("Looking for Acid tables that need to be compacted"); + //todo: check if acid is enabled and bail if not + //todo: check that running on 2.x? + HiveMetaStoreClient hms = new HiveMetaStoreClient(conf);//MetaException + List databases = hms.getAllDatabases();//TException + System.out.println("Found " + databases.size() + " databases to process"); + List compactions = new ArrayList<>(); + List convertToAcid = new ArrayList<>(); + List convertToMM = new ArrayList<>(); + final String scriptLocation = ".";//todo: get this from input + final CompactionMetaInfo compactionMetaInfo = new CompactionMetaInfo(); + for(String dbName : databases) { + List tables = hms.getAllTables(dbName); + System.out.println("found " + tables.size() + " tables in " + dbName); + for(String tableName : tables) { + Table t = hms.getTable(dbName, tableName); + + //ql depends on metastore and is not accessible here... and if it was, I would not be using + //2.6 exec jar, but 3.0.... which is not what we want + List compactionCommands = getCompactionCommands(t, conf, hms, compactionMetaInfo); + compactions.addAll(compactionCommands); + processConversion(t, convertToAcid, convertToMM, hms); + /*todo: handle renaming files somewhere + * */ + } + } + makeCompactionScript(compactions, scriptLocation, compactionMetaInfo); + makeConvertTableScript(convertToAcid, convertToMM, scriptLocation); + makeRenameFileScript(scriptLocation); + } + //todo: handle exclusion list + private static void processConversion(Table t, List convertToAcid, + List convertToMM, HiveMetaStoreClient hms) throws TException { + if(isFullAcidTable(t)) { + return; + } + if(!TableType.MANAGED_TABLE.name().equalsIgnoreCase(t.getTableType())) { + return; + } + String fullTableName = Warehouse.getQualifiedName(t); + if(t.getPartitionKeysSize() <= 0) { + if(canBeMadeAcid(fullTableName, t.getSd())) { + convertToAcid.add("ALTER TABLE " + Warehouse.getQualifiedName(t) + " SET TBLPROPERTIES (" + + "'transactional'='true')"); + } + else { + convertToMM.add("ALTER TABLE " + Warehouse.getQualifiedName(t) + " SET TBLPROPERTIES (" + + "'transactional'='true', 'transactional_properties'='insert_only')"); + } + } + else { + List partNames = hms.listPartitionNames(t.getDbName(), t.getTableName(), (short)-1); + int batchSize = 10000;//todo: right size? + int numWholeBatches = partNames.size()/batchSize; + for(int i = 0; i < numWholeBatches; i++) { + List partitionList = hms.getPartitionsByNames(t.getDbName(), t.getTableName(), + partNames.subList(i * batchSize, (i + 1) * batchSize)); + for(Partition p : partitionList) { + if(!canBeMadeAcid(fullTableName, p.getSd())) { + convertToMM.add("ALTER TABLE " + Warehouse.getQualifiedName(t) + " SET TBLPROPERTIES (" + + "'transactional'='true', 'transactional_properties'='insert_only')"); + return; + } + } + } + if(numWholeBatches * batchSize < partNames.size()) { + //last partial batch + List partitionList = hms.getPartitionsByNames(t.getDbName(), t.getTableName(), + partNames.subList(numWholeBatches * batchSize, partNames.size())); + for (Partition p : partitionList) { + if (!canBeMadeAcid(fullTableName, p.getSd())) { + convertToMM.add("ALTER TABLE " + Warehouse.getQualifiedName(t) + " SET TBLPROPERTIES (" + + "'transactional'='true', 'transactional_properties'='insert_only')"); + return; + } + } + } + //if here checked all parts and they are Acid compatible - make it acid + convertToAcid.add("ALTER TABLE " + Warehouse.getQualifiedName(t) + " SET TBLPROPERTIES (" + + "'transactional'='true')"); + } + } + private static boolean canBeMadeAcid(String fullTableName, StorageDescriptor sd) { + return isAcidInputOutputFormat(fullTableName, sd) && sd.getSortColsSize() <= 0; + } + private static boolean isAcidInputOutputFormat(String fullTableName, StorageDescriptor sd) { + try { + Class inputFormatClass = sd.getInputFormat() == null ? null : + Class.forName(sd.getInputFormat()); + Class outputFormatClass = sd.getOutputFormat() == null ? null : + Class.forName(sd.getOutputFormat()); + + if (inputFormatClass != null && outputFormatClass != null && + Class.forName("org.apache.hadoop.hive.ql.io.AcidInputFormat") + .isAssignableFrom(inputFormatClass) && + Class.forName("org.apache.hadoop.hive.ql.io.AcidOutputFormat") + .isAssignableFrom(outputFormatClass)) { + return true; + } + } catch (ClassNotFoundException e) { + //if a table is using some custom I/O format and it's not in the classpath, we won't mark + //the table for Acid, but today (Hive 3.1 and earlier) OrcInput/OutputFormat is the only + //Acid format + System.err.println("Could not determine if " + fullTableName + + " can be made Acid due to: " + e.getMessage()); + return false; + } + return false; + } + /** + * currently writes to current dir (whatever that is). + * If there is nothing to compact, outputs empty file so as not to confuse the output with a + * failed run. + * todo: add some config to tell it where to put the script + */ + private static void makeCompactionScript(List commands, String scriptLocation, + CompactionMetaInfo compactionMetaInfo) throws IOException { + try(PrintWriter pw = createScript(commands, "compacts_" + System.currentTimeMillis() + + ".sql", scriptLocation)) { + //add post script + pw.println("-- Generated total of " + commands.size() + " compaction commands"); + if(compactionMetaInfo.numberOfBytes < Math.pow(2, 20)) { + //to see it working in UTs + pw.println("-- The total volume of data to be compacted is " + + String.format("%.6fMB", compactionMetaInfo.numberOfBytes/Math.pow(2, 20))); + } + else { + pw.println("-- The total volume of data to be compacted is " + + String.format("%.3fGB", compactionMetaInfo.numberOfBytes/Math.pow(2, 30))); + } + pw.println(); + pw.println( + "-- Please note that compaction may be a heavyweight and time consuming process.\n" + + "-- Submitting all of these commands will enqueue them to a scheduling queue from\n" + + "-- which they will be picked up by compactor Workers. The max number of\n" + + "-- concurrent Workers is controlled by hive.compactor.worker.threads configured\n" + + "-- for the standalone metastore process. Compaction itself is a Map-Reduce job\n" + + "-- which is submitted to the YARN queue identified by hive.compactor.job.queue\n" + + "-- property if defined or 'default' if not defined. It's advisable to set the\n" + + "-- capacity of this queue appropriately"); + } + } + private static void makeConvertTableScript(List alterTableAcid, List alterTableMm, + String scriptLocation) throws IOException { + try(PrintWriter pw = createScript(alterTableAcid, "convertToAcid_" + System.currentTimeMillis() + ".sql", + scriptLocation)) { + pw.println("-- These commands may be executed by Hive 1.x later"); + } + try(PrintWriter pw = createScript(alterTableMm, "convertToMM_" + System.currentTimeMillis() + ".sql", + scriptLocation)) { + pw.println("-- These commands must be executed by Hive 3.0 or later"); + } + } + + private static PrintWriter createScript(List commands, String fileName, + String scriptLocation) throws IOException { + //todo: make sure to create the file in 'scriptLocation' dir + FileWriter fw = new FileWriter(scriptLocation + "/" + fileName); + PrintWriter pw = new PrintWriter(fw); + for(String cmd : commands) { + pw.println(cmd + ";"); + } + return pw; + } + private static void makeRenameFileScript(String scriptLocation) throws IOException { + createScript(Collections.emptyList(), "normalizeFileNames_" + + System.currentTimeMillis() + ".sh", scriptLocation); + } + /** + * @return any compaction commands to run for {@code Table t} + */ + private static List getCompactionCommands(Table t, Configuration conf, + HiveMetaStoreClient hms, CompactionMetaInfo compactionMetaInfo) + throws IOException, TException { + if(!isFullAcidTable(t)) { + return Collections.emptyList(); + } + if(t.getPartitionKeysSize() <= 0) { + //not partitioned + if(!needsCompaction(new Path(t.getSd().getLocation()), conf, compactionMetaInfo)) { + return Collections.emptyList(); + } + + List cmds = new ArrayList<>(); + cmds.add(getCompactionCommand(t, null)); + return cmds; + } + List partNames = hms.listPartitionNames(t.getDbName(), t.getTableName(), (short)-1); + int batchSize = 10000;//todo: right size? + int numWholeBatches = partNames.size()/batchSize; + List compactionCommands = new ArrayList<>(); + for(int i = 0; i < numWholeBatches; i++) { + List partitionList = hms.getPartitionsByNames(t.getDbName(), t.getTableName(), partNames.subList(i * batchSize, (i + 1) * batchSize)); + for(Partition p : partitionList) { + if(needsCompaction(new Path(p.getSd().getLocation()), conf, compactionMetaInfo)) { + compactionCommands.add(getCompactionCommand(t, p)); + } + } + } + if(numWholeBatches * batchSize < partNames.size()) { + //last partial batch + List partitionList = hms.getPartitionsByNames(t.getDbName(), t.getTableName(), partNames.subList(numWholeBatches * batchSize, partNames.size())); + for (Partition p : partitionList) { + if (needsCompaction(new Path(p.getSd().getLocation()), conf, compactionMetaInfo)) { + compactionCommands.add(getCompactionCommand(t, p)); + } + } + } + return compactionCommands; + } + /** + * + * @param location - path to a partition (or table if not partitioned) dir + */ + private static boolean needsCompaction(Path location, Configuration conf, + CompactionMetaInfo compactionMetaInfo) throws IOException { + FileSystem fs = location.getFileSystem(conf); + FileStatus[] deltas = fs.listStatus(location, new PathFilter() { + @Override + public boolean accept(Path path) { + //checking for delete_delta is only so that this functionality can be exercised by code 3.0 + //which cannot produce any deltas with mix of update/insert events + return path.getName().startsWith("delta_") || path.getName().startsWith("delete_delta_"); + } + }); + if(deltas == null || deltas.length == 0) { + //base_n cannot contain update/delete. Original files are all 'insert' and we need to compact + //only if there are update/delete events. + return false; + } + deltaLoop: for(FileStatus delta : deltas) { + if(!delta.isDirectory()) { + //should never happen - just in case + continue; + } + FileStatus[] buckets = fs.listStatus(delta.getPath(), new PathFilter() { + @Override + public boolean accept(Path path) { + //since this is inside a delta dir created by Hive 2.x or earlier it can only contain + //bucket_x or bucket_x__flush_length + return path.getName().startsWith("bucket_"); + } + }); + for(FileStatus bucket : buckets) { + if(bucket.getPath().getName().endsWith("_flush_length")) { + //streaming ingest dir - cannot have update/delete events + continue deltaLoop; + } + if(needsCompaction(bucket, fs)) { + //found delete events - this 'location' needs compacting + compactionMetaInfo.numberOfBytes += getDataSize(location, conf); + return true; + } + } + } + return false; + } + /** + * @param location - path to a partition (or table if not partitioned) dir + * @throws IOException + */ + private static long getDataSize(Path location, Configuration conf) throws IOException { + /* + * todo: Figure out the size of the partition. The + * best way is to getAcidState() and look at each file - this way it takes care of + * original files vs base and any other obsolete files. For now just brute force it, + * it's likely close enough for a rough estimate.*/ + FileSystem fs = location.getFileSystem(conf); + ContentSummary cs = fs.getContentSummary(location); + return cs.getLength(); + } + private static boolean needsCompaction(FileStatus bucket, FileSystem fs) throws IOException { + //create reader, look at footer + //no need to check side file since it can only be in a streaming ingest delta + Reader orcReader = OrcFile.createReader(bucket.getPath(),OrcFile.readerOptions(fs.getConf()) + .filesystem(fs)); + AcidStats as = OrcAcidUtils.parseAcidStats(orcReader); + if(as == null) { + //should never happen since we are reading bucket_x written by acid write + throw new IllegalStateException("AcidStats missing in " + bucket.getPath()); + } + return as.deletes > 0 || as.updates > 0; + } + private static String getCompactionCommand(Table t, Partition p) { + StringBuilder sb = new StringBuilder("ALTER TABLE ").append(Warehouse.getQualifiedName(t)); + if(t.getPartitionKeysSize() > 0) { + assert p != null : "must supply partition for partitioned table " + + Warehouse.getQualifiedName(t); + sb.append(" PARTITION("); + for (int i = 0; i < t.getPartitionKeysSize(); i++) { + //todo: should these be quoted? HiveUtils.unparseIdentifier() - if value is String should definitely quote + sb.append(t.getPartitionKeys().get(i).getName()).append('=') + .append(p.getValues().get(i)).append(","); + } + sb.setCharAt(sb.length() - 1, ')');//replace trailing ',' + } + return sb.append(" COMPACT 'major'").toString(); + } + private static boolean isFullAcidTable(Table t) { + if (t.getParametersSize() <= 0) { + //cannot be acid + return false; + } + String transacationalValue = t.getParameters() + .get(hive_metastoreConstants.TABLE_IS_TRANSACTIONAL); + if (transacationalValue != null && "true".equalsIgnoreCase(transacationalValue)) { + System.out.println("Found Acid table: " + Warehouse.getQualifiedName(t)); + return true; + } + return false; + } private static void printAndExit(HiveMetaTool metaTool) { HelpFormatter formatter = new HelpFormatter(); formatter.printHelp("metatool", metaTool.cmdLineOptions); @@ -460,7 +841,16 @@ public static void main(String[] args) { } else { metaTool.updateFSRootLocation(oldURI, newURI, serdepropKey, tablePropKey, isDryRun); } - } else { + } else if(line.hasOption("prepareAcidUpgrade")) { + String[] values = line.getOptionValues("prepareAcidUpgrade"); + String targetDir = null; + if(values != null && values.length > 0) { + if(values.length > 1) { + System.err.println("HiveMetaTool: prepareAcidUpgrade"); + } + } + metaTool.prepareAcidUpgrade(metaTool); + } else { if (line.hasOption("dryRun")) { System.err.println("HiveMetaTool: dryRun is not a valid standalone option"); } else if (line.hasOption("serdePropKey")) {