Search in sources :

Example 1 with MobCompactPartitionPolicy

use of org.apache.hadoop.hbase.client.MobCompactPartitionPolicy in project hbase by apache.

the class PartitionedMobCompactor method select.

/**
   * Selects the compacted mob/del files.
   * Iterates the candidates to find out all the del files and small mob files.
   * @param candidates All the candidates.
   * @param allFiles Whether add all mob files into the compaction.
   * @return A compaction request.
   * @throws IOException if IO failure is encountered
   */
protected PartitionedMobCompactionRequest select(List<FileStatus> candidates, boolean allFiles) throws IOException {
    final Map<CompactionPartitionId, CompactionPartition> filesToCompact = new HashMap<>();
    final CompactionPartitionId id = new CompactionPartitionId();
    final NavigableMap<CompactionDelPartitionId, CompactionDelPartition> delFilesToCompact = new TreeMap<>();
    final CompactionDelPartitionId delId = new CompactionDelPartitionId();
    final ArrayList<CompactionDelPartition> allDelPartitions = new ArrayList<>();
    int selectedFileCount = 0;
    int irrelevantFileCount = 0;
    int totalDelFiles = 0;
    MobCompactPartitionPolicy policy = column.getMobCompactPartitionPolicy();
    Calendar calendar = Calendar.getInstance();
    Date currentDate = new Date();
    Date firstDayOfCurrentMonth = null;
    Date firstDayOfCurrentWeek = null;
    if (policy == MobCompactPartitionPolicy.MONTHLY) {
        firstDayOfCurrentMonth = MobUtils.getFirstDayOfMonth(calendar, currentDate);
        firstDayOfCurrentWeek = MobUtils.getFirstDayOfWeek(calendar, currentDate);
    } else if (policy == MobCompactPartitionPolicy.WEEKLY) {
        firstDayOfCurrentWeek = MobUtils.getFirstDayOfWeek(calendar, currentDate);
    }
    // We check if there is any del files so the logic can be optimized for the following processing
    // First step is to check if there is any delete files. If there is any delete files,
    // For each Partition, it needs to read its startKey and endKey from files.
    // If there is no delete file, there is no need to read startKey and endKey from files, this
    // is an optimization.
    boolean withDelFiles = false;
    for (FileStatus file : candidates) {
        if (!file.isFile()) {
            continue;
        }
        // group the del files and small files.
        FileStatus linkedFile = file;
        if (HFileLink.isHFileLink(file.getPath())) {
            HFileLink link = HFileLink.buildFromHFileLinkPattern(conf, file.getPath());
            linkedFile = getLinkedFileStatus(link);
            if (linkedFile == null) {
                continue;
            }
        }
        if (StoreFileInfo.isDelFile(linkedFile.getPath())) {
            withDelFiles = true;
            break;
        }
    }
    for (FileStatus file : candidates) {
        if (!file.isFile()) {
            irrelevantFileCount++;
            continue;
        }
        // group the del files and small files.
        FileStatus linkedFile = file;
        if (HFileLink.isHFileLink(file.getPath())) {
            HFileLink link = HFileLink.buildFromHFileLinkPattern(conf, file.getPath());
            linkedFile = getLinkedFileStatus(link);
            if (linkedFile == null) {
                // If the linked file cannot be found, regard it as an irrelevantFileCount file
                irrelevantFileCount++;
                continue;
            }
        }
        if (withDelFiles && StoreFileInfo.isDelFile(linkedFile.getPath())) {
            // File in the Del Partition List
            // Get delId from the file
            Reader reader = HFile.createReader(fs, linkedFile.getPath(), CacheConfig.DISABLED, conf);
            try {
                delId.setStartKey(reader.getFirstRowKey());
                delId.setEndKey(reader.getLastRowKey());
            } finally {
                reader.close();
            }
            CompactionDelPartition delPartition = delFilesToCompact.get(delId);
            if (delPartition == null) {
                CompactionDelPartitionId newDelId = new CompactionDelPartitionId(delId.getStartKey(), delId.getEndKey());
                delPartition = new CompactionDelPartition(newDelId);
                delFilesToCompact.put(newDelId, delPartition);
            }
            delPartition.addDelFile(file);
            totalDelFiles++;
        } else {
            String fileName = linkedFile.getPath().getName();
            String date = MobFileName.getDateFromName(fileName);
            boolean skipCompaction = MobUtils.fillPartitionId(id, firstDayOfCurrentMonth, firstDayOfCurrentWeek, date, policy, calendar, mergeableSize);
            if (allFiles || (!skipCompaction && (linkedFile.getLen() < id.getThreshold()))) {
                // add all files if allFiles is true,
                // otherwise add the small files to the merge pool
                // filter out files which are not supposed to be compacted with the
                // current policy
                id.setStartKey(MobFileName.getStartKeyFromName(fileName));
                CompactionPartition compactionPartition = filesToCompact.get(id);
                if (compactionPartition == null) {
                    CompactionPartitionId newId = new CompactionPartitionId(id.getStartKey(), id.getDate());
                    compactionPartition = new CompactionPartition(newId);
                    compactionPartition.addFile(file);
                    filesToCompact.put(newId, compactionPartition);
                    newId.updateLatestDate(date);
                } else {
                    compactionPartition.addFile(file);
                    compactionPartition.getPartitionId().updateLatestDate(date);
                }
                if (withDelFiles) {
                    // get startKey and endKey from the file and update partition
                    // TODO: is it possible to skip read of most hfiles?
                    Reader reader = HFile.createReader(fs, linkedFile.getPath(), CacheConfig.DISABLED, conf);
                    try {
                        compactionPartition.setStartKey(reader.getFirstRowKey());
                        compactionPartition.setEndKey(reader.getLastRowKey());
                    } finally {
                        reader.close();
                    }
                }
                selectedFileCount++;
            }
        }
    }
    /*
     * Merge del files so there are only non-overlapped del file lists
     */
    for (Map.Entry<CompactionDelPartitionId, CompactionDelPartition> entry : delFilesToCompact.entrySet()) {
        if (allDelPartitions.size() > 0) {
            // check if the current key range overlaps the previous one
            CompactionDelPartition prev = allDelPartitions.get(allDelPartitions.size() - 1);
            if (Bytes.compareTo(prev.getId().getEndKey(), entry.getKey().getStartKey()) >= 0) {
                // merge them together
                prev.getId().setEndKey(entry.getValue().getId().getEndKey());
                prev.addDelFileList(entry.getValue().listDelFiles());
            } else {
                allDelPartitions.add(entry.getValue());
            }
        } else {
            allDelPartitions.add(entry.getValue());
        }
    }
    PartitionedMobCompactionRequest request = new PartitionedMobCompactionRequest(filesToCompact.values(), allDelPartitions);
    if (candidates.size() == (totalDelFiles + selectedFileCount + irrelevantFileCount)) {
        // all the files are selected
        request.setCompactionType(CompactionType.ALL_FILES);
    }
    LOG.info("The compaction type is " + request.getCompactionType() + ", the request has " + totalDelFiles + " del files, " + selectedFileCount + " selected files, and " + irrelevantFileCount + " irrelevant files");
    return request;
}
Also used : CompactionDelPartitionId(org.apache.hadoop.hbase.mob.compactions.PartitionedMobCompactionRequest.CompactionDelPartitionId) HFileLink(org.apache.hadoop.hbase.io.HFileLink) CompactionPartition(org.apache.hadoop.hbase.mob.compactions.PartitionedMobCompactionRequest.CompactionPartition) FileStatus(org.apache.hadoop.fs.FileStatus) HashMap(java.util.HashMap) Calendar(java.util.Calendar) ArrayList(java.util.ArrayList) Reader(org.apache.hadoop.hbase.io.hfile.HFile.Reader) CompactionPartitionId(org.apache.hadoop.hbase.mob.compactions.PartitionedMobCompactionRequest.CompactionPartitionId) TreeMap(java.util.TreeMap) Date(java.util.Date) CompactionDelPartition(org.apache.hadoop.hbase.mob.compactions.PartitionedMobCompactionRequest.CompactionDelPartition) MobCompactPartitionPolicy(org.apache.hadoop.hbase.client.MobCompactPartitionPolicy) Map(java.util.Map) NavigableMap(java.util.NavigableMap) HashMap(java.util.HashMap) TreeMap(java.util.TreeMap)

Aggregations

ArrayList (java.util.ArrayList)1 Calendar (java.util.Calendar)1 Date (java.util.Date)1 HashMap (java.util.HashMap)1 Map (java.util.Map)1 NavigableMap (java.util.NavigableMap)1 TreeMap (java.util.TreeMap)1 FileStatus (org.apache.hadoop.fs.FileStatus)1 MobCompactPartitionPolicy (org.apache.hadoop.hbase.client.MobCompactPartitionPolicy)1 HFileLink (org.apache.hadoop.hbase.io.HFileLink)1 Reader (org.apache.hadoop.hbase.io.hfile.HFile.Reader)1 CompactionDelPartition (org.apache.hadoop.hbase.mob.compactions.PartitionedMobCompactionRequest.CompactionDelPartition)1 CompactionDelPartitionId (org.apache.hadoop.hbase.mob.compactions.PartitionedMobCompactionRequest.CompactionDelPartitionId)1 CompactionPartition (org.apache.hadoop.hbase.mob.compactions.PartitionedMobCompactionRequest.CompactionPartition)1 CompactionPartitionId (org.apache.hadoop.hbase.mob.compactions.PartitionedMobCompactionRequest.CompactionPartitionId)1