HBase Split

Region Split请求是在Region MemStore Flush之后被触发的：

boolean shouldCompact = region.flushcache();

// We just want to check the size
boolean shouldSplit = region.checkSplit() != null;

if (shouldSplit) {
    this.server.compactSplitThread.requestSplit(region);
} else if (shouldCompact) {
    server.compactSplitThread.requestCompaction(region, getName());
}

server.getMetrics().addFlush(region.getRecentFlushInfo());

Region Flush操作完成之后，会进行checkSplit的判断，如果返回值不为null（返回值为该Region的SplitPoint），表示该Region达到了进行Split的条件，发起相应的Split请求。

checkSplit方法定义如下：

/**
 * Return the splitpoint. null indicates the region isn't splittable. If the
 * splitpoint isn't explicitly specified, it will go over the stores to find
 * the best splitpoint. Currently the criteria of best splitpoint is based
 * on the size of the store.
 */
public byte[] checkSplit() {
    // Can't split ROOT/META
    if (this.regionInfo.isMetaTable()) {
        if (shouldForceSplit()) {
            LOG.warn("Cannot split root/meta regions in HBase 0.20 and above");
        }

        return null;
    }

    if (!splitPolicy.shouldSplit()) {
        return null;
    }

    byte[] ret = splitPolicy.getSplitPoint();

    if (ret != null) {
        try {
            checkRow(ret, "calculated split");
        } catch (IOException e) {
            LOG.error("Ignoring invalid split", e);

            return null;
        }
    }

    return ret;
}

由上述代码可以看出，如果当前Region属于目录信息表（ROOT/META），则是不允许进行Split操作的，否则根据当前Region的RegionSplitPolicy实例判断是否需要进行Split，流程包含两步：

（1）该Region是否允许进行Split；

（2）该Region在允许进行Split的条件下，是否可以计算出相应的SplitPoint。

RegionSplitPolicy shouldSplit

如果没有在定义表结构时进行特殊的指定，RegionSplitPolicy默认为org.apache.hadoop.hbase.regionserver.IncreasingToUpperBoundRegionSplitPolicy的实例，配置项为hbase.regionserver.region.split.policy。

方法代码如下：

@Override
protected boolean shouldSplit() {
    if (region.shouldForceSplit()) {
        return true;
    }

    boolean foundABigStore = false;

    // Get count of regions that have the same common table as this.region
    int tableRegionsCount = getCountOfCommonTableRegions();

    // Get size to check
    long sizeToCheck = getSizeToCheck(tableRegionsCount);

    for (Store store : region.getStores().values()) {
        // If any of the stores is unable to split (eg they contain
        // reference files)
        // then don't split
        if ((!store.canSplit())) {
            return false;
        }

        // Mark if any store is big enough
        long size = store.getSize();

        if (size > sizeToCheck) {
            LOG.debug("ShouldSplit because " + store.getColumnFamilyName()
                    + " size=" + size + ", sizeToCheck=" + sizeToCheck
                    + ", regionsWithCommonTable=" + tableRegionsCount);

            foundABigStore = true;

            break;
        }
    }

    return foundABigStore;
}

执行流程：

（1）如果当前Region被请求执行ForceSplit，则直接返回true；

（2）计算当前Region中的各个Store大小的上限值；

（3）循环判断当前Region中的某一Store大小是否超过上限值，如果存在这样的Store，则提前结束循环，返回true即可。

其中，进行大小判断的Region Store必须是可Split的，即该Store中不包含Reference类型的文件，如果某一Store中出现了Reference类型的文件，则表示该Region已经被Split过，不能再进行Split，此时，直接返回false即可。

重点讲述一下Region中各个Store大小的上限值的计算方法：

（1）假设当前Region所属的表为t，计算该Region所处于的RegionServer上包含表t的Online Region数目，并将结果保存至变量tableRegionsCount中；

// Get count of regions that have the same common table as this.region
int tableRegionsCount = getCountOfCommonTableRegions();

getCountOfCommonTableRegions方法代码如下：

/**
 * @return Count of regions on this server that share the table this.region
 *         belongs to
 */
private int getCountOfCommonTableRegions() {
    RegionServerServices rss = this.region.getRegionServerServices();

    // Can be null in tests
    if (rss == null) {
        return 0;
    }

    byte[] tablename = this.region.getTableDesc().getName();

    int tableRegionsCount = 0;

    try {
        List<HRegion> hri = rss.getOnlineRegions(tablename);

        tableRegionsCount = hri == null || hri.isEmpty() ? 0 : hri.size();
    } catch (IOException e) {
        LOG.debug("Failed getOnlineRegions " + Bytes.toString(tablename), e);
    }

    return tableRegionsCount;
}

首先获取该Region所处于的RegionServer实例：

RegionServerServices rss = this.region.getRegionServerServices();

然后获取该Region所对应的表的名称：

byte[] tablename = this.region.getTableDesc().getName();

最后获取表tablename在rss上的Online Region的数目：

List<HRegion> hri = rss.getOnlineRegions(tablename);

（2）根据tableRegionsCount计算上限值：

// Get size to check
long sizeToCheck = getSizeToCheck(tableRegionsCount);

getSizeToCheck方法代码如下：

/**
 * @return Region max size or
 *         <code>count of regions squared * flushsize, which ever is
 * smaller; guard against there being zero regions on this server.
 */
long getSizeToCheck(final int tableRegionsCount) {
    return tableRegionsCount == 0 ? getDesiredMaxFileSize() : Math.min(
            getDesiredMaxFileSize(), this.flushSize
                    * (tableRegionsCount * tableRegionsCount));
}

计算过程根据tableRegionsCount的值分为两种情况：

（1）tableRegionsCount值为0时（可能发生么？），直接通过方法getDesiredMaxFileSize返回结果即可（getDesiredMaxFileSize的返回值可以在创建表时指定，如果创建表时没有特殊指定，则由配置项hbase.hregion.max.filesize决定，默认值为10737418240即10G）；

（2）tableRegionsCount值不为0时，结果为getDesiredMaxFileSize（）与this.flushSize * (tableRegionsCount * tableRegionsCount)两者之间的最小值，其中flushSize在创建表时指定，如果创建表时没有特殊指定，则由配置项hbase.hregion.memstore.flush.size决定，默认值为134217728即128M。

RegionSplitPolicy getSplitPoint

进行到这一步，表示该Region是允许进行Split的，下一步应该计算该Region的SplitPoint。

方法代码如下：

/**
 * @return the key at which the region should be split, or null if it cannot
 *         be split. This will only be called if shouldSplit previously
 *         returned true.
 */
protected byte[] getSplitPoint() {
    byte[] explicitSplitPoint = this.region.getExplicitSplitPoint();
    if (explicitSplitPoint != null) {
        return explicitSplitPoint;
    }

    Map<byte[], Store> stores = region.getStores();

    byte[] splitPointFromLargestStore = null;

    long largestStoreSize = 0;

    for (Store s : stores.values()) {
        byte[] splitPoint = s.getSplitPoint();

        long storeSize = s.getSize();

        if (splitPoint != null && largestStoreSize < storeSize) {
            splitPointFromLargestStore = splitPoint;

            largestStoreSize = storeSize;
        }
    }

    return splitPointFromLargestStore;
}

执行流程如下：

（1）如果请求ForceSplit时显示指定了SplitPoint，则直接将该值返回即可；

（2）循环处理该Region的Store，分别获取该Store的大小和SplitPoint，最后Region的SplitPoint为最大的那个Store的SplitPoint。

接下来的问题是如何计算Store的SplitPoint。

Store getSplitPoint

/**
 * Determines if Store should be split
 *
 * @return byte[] if store should be split, null otherwise.
 */
public byte[] getSplitPoint() {
    this.lock.readLock().lock();

    try {
        // sanity checks
        if (this.storefiles.isEmpty()) {
            return null;
        }

        // Should already be enforced by the split policy!
        assert !this.region.getRegionInfo().isMetaRegion();

        // Not splitable if we find a reference store file present in the
        // store.
        long maxSize = 0L;

        StoreFile largestSf = null;

        for (StoreFile sf : storefiles) {
            if (sf.isReference()) {
                // Should already be enforced since we return false in this
                // case
                assert false : "getSplitPoint() called on a region that can't split!";

                return null;
            }

            StoreFile.Reader r = sf.getReader();

            if (r == null) {
                LOG.warn("Storefile " + sf + " Reader is null");

                continue;
            }

            long size = r.length();

            if (size > maxSize) {
                // This is the largest one so far
                maxSize = size;

                largestSf = sf;
            }
        }

        StoreFile.Reader r = largestSf.getReader();

        if (r == null) {
            LOG.warn("Storefile " + largestSf + " Reader is null");

            return null;
        }

        // Get first, last, and mid keys. Midkey is the key that starts
        // block
        // in middle of hfile. Has column and timestamp. Need to return just
        // the row we want to split on as midkey.
        byte[] midkey = r.midkey();

        if (midkey != null) {
            KeyValue mk = KeyValue.createKeyValueFromKey(midkey, 0,
                    midkey.length);

            byte[] fk = r.getFirstKey();
            KeyValue firstKey = KeyValue.createKeyValueFromKey(fk, 0,
                    fk.length);

            byte[] lk = r.getLastKey();
            KeyValue lastKey = KeyValue.createKeyValueFromKey(lk, 0,
                    lk.length);

            // if the midkey is the same as the first or last keys, then we
            // cannot
            // (ever) split this region.
            if (this.comparator.compareRows(mk, firstKey) == 0
                    || this.comparator.compareRows(mk, lastKey) == 0) {
                if (LOG.isDebugEnabled()) {
                    LOG.debug("cannot split because midkey is the same as first or "
                            + "last row");
                }

                return null;
            }

            return mk.getRow();
        }
    } catch (IOException e) {
        LOG.warn("Failed getting store size for " + this, e);
    } finally {
        this.lock.readLock().unlock();
    }

    return null;
}

执行流程

（1）选择Store StoreFiles中的最大的那个StoreFile largestSf；

long maxSize = 0L;

StoreFile largestSf = null;

for (StoreFile sf : storefiles) {
    if (sf.isReference()) {
        // Should already be enforced since we return false in this
        // case
        assert false : "getSplitPoint() called on a region that can't split!";

        return null;
    }

    StoreFile.Reader r = sf.getReader();

    if (r == null) {
        LOG.warn("Storefile " + sf + " Reader is null");

        continue;
    }

    long size = r.length();

    if (size > maxSize) {
        // This is the largest one so far
        maxSize = size;

        largestSf = sf;
    }
}

（2）获取largestSf的MidKey、FirstKey、LastKey，如果MidKey与FirstKey相等或者MidKey与LastKey相等，则返回null（为什么？）；否则返回MidKey。

// Get first, last, and mid keys. Midkey is the key that starts
// block
// in middle of hfile. Has column and timestamp. Need to return just
// the row we want to split on as midkey.
byte[] midkey = r.midkey();

if (midkey != null) {
    KeyValue mk = KeyValue.createKeyValueFromKey(midkey, 0,
            midkey.length);

    byte[] fk = r.getFirstKey();
    KeyValue firstKey = KeyValue.createKeyValueFromKey(fk, 0,
            fk.length);

    byte[] lk = r.getLastKey();
    KeyValue lastKey = KeyValue.createKeyValueFromKey(lk, 0,
            lk.length);

    // if the midkey is the same as the first or last keys, then we
    // cannot
    // (ever) split this region.
    if (this.comparator.compareRows(mk, firstKey) == 0
            || this.comparator.compareRows(mk, lastKey) == 0) {
        if (LOG.isDebugEnabled()) {
            LOG.debug("cannot split because midkey is the same as first or "
                    + "last row");
        }

        return null;
    }

    return mk.getRow();
}

StoreFile是由多个Block组成的（这里的Block不同于HDFS的Block），每个Block的第一个RowKey会被存储到StoreFile中的特殊位置中，因此，这里的MidKey、FirstKey、LastKey指的就是StoreFile中MidBlock、FirstBlock、LastBlock各自的第一个RowKey。

Region Split是以Row作为最小切分单位的，即同一行的数据会完整的出现在某一Region中，如果MidKey与FirstKey相等或者MidKey与LastKey相等，则表示如果进行切分则会出现某Region中的RowKey是完全一样的，即该Region中仅包含一个行的数据，这种情况出现中HBase中是不合理的，因此不允许MidKey与FirstKey相等或者MidKey与LastKey相等时进行Split。

综上所述，如果某一Region满足Split的条件且可以计算出SplitPoint，则可以发起Split请求：

this.server.compactSplitThread.requestSplit(region);