public static Partition fromMetastoreApiPartition(
org.apache.hadoop.hive.metastore.api.Partition partition) {
StorageDescriptor storageDescriptor = partition.getSd();
if (storageDescriptor == null) {
throw new PrestoException(
HIVE_INVALID_METADATA, "Partition does not contain a storage descriptor: " + partition);
}
Partition.Builder partitionBuilder =
Partition.builder()
.setDatabaseName(partition.getDbName())
.setTableName(partition.getTableName())
.setValues(partition.getValues())
.setColumns(
storageDescriptor
.getCols()
.stream()
.map(MetastoreUtil::fromMetastoreApiFieldSchema)
.collect(toList()))
.setParameters(partition.getParameters());
fromMetastoreApiStorageDescriptor(
storageDescriptor,
partitionBuilder.getStorageBuilder(),
format("%s.%s", partition.getTableName(), partition.getValues()));
return partitionBuilder.build();
}
public static Table fromMetastoreApiTable(org.apache.hadoop.hive.metastore.api.Table table) {
StorageDescriptor storageDescriptor = table.getSd();
if (storageDescriptor == null) {
throw new PrestoException(HIVE_INVALID_METADATA, "Table is missing storage descriptor");
}
Table.Builder tableBuilder =
Table.builder()
.setDatabaseName(table.getDbName())
.setTableName(table.getTableName())
.setOwner(nullToEmpty(table.getOwner()))
.setTableType(table.getTableType())
.setDataColumns(
storageDescriptor
.getCols()
.stream()
.map(MetastoreUtil::fromMetastoreApiFieldSchema)
.collect(toList()))
.setPartitionColumns(
table
.getPartitionKeys()
.stream()
.map(MetastoreUtil::fromMetastoreApiFieldSchema)
.collect(toList()))
.setParameters(
table.getParameters() == null ? ImmutableMap.of() : table.getParameters())
.setViewOriginalText(Optional.ofNullable(emptyToNull(table.getViewOriginalText())))
.setViewExpandedText(Optional.ofNullable(emptyToNull(table.getViewExpandedText())));
fromMetastoreApiStorageDescriptor(
storageDescriptor, tableBuilder.getStorageBuilder(), table.getTableName());
return tableBuilder.build();
}
/**
* This code block iterates over indexes on the table and populates the indexToKeys map for all
* the indexes that satisfy the rewrite criteria.
*
* @param indexTables
* @return
* @throws SemanticException
*/
Map<Index, Set<String>> getIndexToKeysMap(List<Index> indexTables) throws SemanticException {
Index index = null;
Hive hiveInstance = hiveDb;
Map<Index, Set<String>> indexToKeysMap = new LinkedHashMap<Index, Set<String>>();
for (int idxCtr = 0; idxCtr < indexTables.size(); idxCtr++) {
final Set<String> indexKeyNames = new LinkedHashSet<String>();
index = indexTables.get(idxCtr);
// Getting index key columns
StorageDescriptor sd = index.getSd();
List<FieldSchema> idxColList = sd.getCols();
for (FieldSchema fieldSchema : idxColList) {
indexKeyNames.add(fieldSchema.getName());
}
assert indexKeyNames.size() == 1;
// Check that the index schema is as expected. This code block should
// catch problems of this rewrite breaking when the AggregateIndexHandler
// index is changed.
List<String> idxTblColNames = new ArrayList<String>();
try {
Table idxTbl = hiveInstance.getTable(index.getDbName(), index.getIndexTableName());
for (FieldSchema idxTblCol : idxTbl.getCols()) {
idxTblColNames.add(idxTblCol.getName());
}
} catch (HiveException e) {
LOG.error(
"Got exception while locating index table, "
+ "skipping "
+ getName()
+ " optimization");
LOG.error(org.apache.hadoop.util.StringUtils.stringifyException(e));
throw new SemanticException(e.getMessage(), e);
}
assert (idxTblColNames.contains(IDX_BUCKET_COL));
assert (idxTblColNames.contains(IDX_OFFSETS_ARRAY_COL));
// we add all index tables which can be used for rewrite
// and defer the decision of using a particular index for later
// this is to allow choosing a index if a better mechanism is
// designed later to chose a better rewrite
indexToKeysMap.put(index, indexKeyNames);
}
return indexToKeysMap;
}
/**
* Run a compactor job.
*
* @param conf Hive configuration file
* @param jobName name to run this job with
* @param t metastore table
* @param sd metastore storage descriptor
* @param txns list of valid transactions
* @param isMajor is this a major compaction?
* @throws java.io.IOException if the job fails
*/
void run(
HiveConf conf,
String jobName,
Table t,
StorageDescriptor sd,
ValidTxnList txns,
boolean isMajor,
Worker.StatsUpdater su)
throws IOException {
JobConf job = new JobConf(conf);
job.setJobName(jobName);
job.setOutputKeyClass(NullWritable.class);
job.setOutputValueClass(NullWritable.class);
job.setJarByClass(CompactorMR.class);
LOG.debug("User jar set to " + job.getJar());
job.setMapperClass(CompactorMap.class);
job.setNumReduceTasks(0);
job.setInputFormat(CompactorInputFormat.class);
job.setOutputFormat(NullOutputFormat.class);
job.setOutputCommitter(CompactorOutputCommitter.class);
String queueName = conf.getVar(HiveConf.ConfVars.COMPACTOR_JOB_QUEUE);
if (queueName != null && queueName.length() > 0) {
job.setQueueName(queueName);
}
job.set(FINAL_LOCATION, sd.getLocation());
job.set(TMP_LOCATION, sd.getLocation() + "/" + TMPDIR + "_" + UUID.randomUUID().toString());
job.set(INPUT_FORMAT_CLASS_NAME, sd.getInputFormat());
job.set(OUTPUT_FORMAT_CLASS_NAME, sd.getOutputFormat());
job.setBoolean(IS_MAJOR, isMajor);
job.setBoolean(IS_COMPRESSED, sd.isCompressed());
job.set(TABLE_PROPS, new StringableMap(t.getParameters()).toString());
job.setInt(NUM_BUCKETS, sd.getNumBuckets());
job.set(ValidTxnList.VALID_TXNS_KEY, txns.toString());
setColumnTypes(job, sd.getCols());
// Figure out and encode what files we need to read. We do this here (rather than in
// getSplits below) because as part of this we discover our minimum and maximum transactions,
// and discovering that in getSplits is too late as we then have no way to pass it to our
// mapper.
AcidUtils.Directory dir = AcidUtils.getAcidState(new Path(sd.getLocation()), conf, txns, false);
StringableList dirsToSearch = new StringableList();
Path baseDir = null;
if (isMajor) {
// There may not be a base dir if the partition was empty before inserts or if this
// partition is just now being converted to ACID.
baseDir = dir.getBaseDirectory();
if (baseDir == null) {
List<HdfsFileStatusWithId> originalFiles = dir.getOriginalFiles();
if (!(originalFiles == null) && !(originalFiles.size() == 0)) {
// There are original format files
for (HdfsFileStatusWithId stat : originalFiles) {
Path path = stat.getFileStatus().getPath();
dirsToSearch.add(path);
LOG.debug("Adding original file " + path + " to dirs to search");
}
// Set base to the location so that the input format reads the original files.
baseDir = new Path(sd.getLocation());
}
} else {
// add our base to the list of directories to search for files in.
LOG.debug("Adding base directory " + baseDir + " to dirs to search");
dirsToSearch.add(baseDir);
}
}
List<AcidUtils.ParsedDelta> parsedDeltas = dir.getCurrentDirectories();
if (parsedDeltas == null || parsedDeltas.size() == 0) {
// Seriously, no deltas? Can't compact that.
LOG.error("No delta files found to compact in " + sd.getLocation());
return;
}
StringableList deltaDirs = new StringableList();
long minTxn = Long.MAX_VALUE;
long maxTxn = Long.MIN_VALUE;
for (AcidUtils.ParsedDelta delta : parsedDeltas) {
LOG.debug("Adding delta " + delta.getPath() + " to directories to search");
dirsToSearch.add(delta.getPath());
deltaDirs.add(delta.getPath());
minTxn = Math.min(minTxn, delta.getMinTransaction());
maxTxn = Math.max(maxTxn, delta.getMaxTransaction());
}
if (baseDir != null) job.set(BASE_DIR, baseDir.toString());
job.set(DELTA_DIRS, deltaDirs.toString());
job.set(DIRS_TO_SEARCH, dirsToSearch.toString());
job.setLong(MIN_TXN, minTxn);
job.setLong(MAX_TXN, maxTxn);
LOG.debug("Setting minimum transaction to " + minTxn);
LOG.debug("Setting maximume transaction to " + maxTxn);
RunningJob rj = JobClient.runJob(job);
LOG.info(
"Submitted "
+ (isMajor ? CompactionType.MAJOR : CompactionType.MINOR)
+ " compaction job '"
+ jobName
+ "' with jobID="
+ rj.getID()
+ " to "
+ job.getQueueName()
+ " queue. "
+ "(current delta dirs count="
+ dir.getCurrentDirectories().size()
+ ", obsolete delta dirs count="
+ dir.getObsolete());
rj.waitForCompletion();
su.gatherStats();
}