1.用户视角
1.1 案例描述
用户写入root.sg1.d1.s1, root.sg1.d2.s1, root.sg1.d3.s1, root.sg2.d1.s1, root.sg2.d2.s1五条序列,假设在给定的hash函数下root.sg1.d1映射为0, root.sg1.d2映射为1, root.sg1.d3映射为0,root.sg2.d1映射为0,root.sg2.d2映射为1,给定逻辑存储组数量为2,虚拟存储组数量也为2,假设写入数据均为顺序写入。
1.2 方案一目录结构(简洁起见,不考虑时间分区)
1.3 方案二目录结构(简洁起见,不考虑时间分区)
注意:由于刷盘并行度由物理存储组层控制,对于sg1来说,其刷盘并行度为2,这导致了两个虚拟存储组的数据刷在两个不同的文件中;对于sg2来说,其刷盘并行度为1,这导致两个虚拟存储组的数据刷在同一个文件中。
2.开发者视角
2.1 整体架构
2.2 方案一架构
整体架构中N = 逻辑存储组数量,M = 虚拟存储组数量,L = 1,在用户例子下,其具体结构如下:(节点中的序号是为了之后的说明使用,无实际意义)
2.3 方案一写入流程
2.3.1 写入root.sg1.d1.s1
(1) → (2) → (4) → (8)
2.3.2 写入root.sg1.d2.s1
(1) → (2) → (5) → (9)
2.3.3 写入root.sg1.d3.s1
(1) → (2) → (4) → (8)
2.3.4 写入root.sg2.d1.s1
(1) → (3) → (6) → (10)
2.3.5 写入root.sg2.d2.s1
(1) → (3) → (7) → (11)
2.4 方案二架构
整体架构中N = 1, M = 虚拟存储组数量,L = 逻辑存储组数量,在用户例子下,其具体结构如下:(节点中的序号是为了之后的说明使用,无实际意义)
2.5 方案二写入流程
2.5.1 写入root.sg1.d1.s1
(1) → (2) → (3) → (5)
2.5.2 写入root.sg1.d2.s1
(1) → (2) → (4) → (5)
2.5.3 写入root.sg1.d3.s1
(1) → (2) → (3) → (5)
2.5.4 写入root.sg2.d1.s1
(1) → (2) → (3) → (6)
2.5.5 写入root.sg2.d2.s1
(1) → (2) → (4) → (6)
3.方案一详细设计
3.1 Storage Engine
ConcurrentHashMap<PartialPath, StorageGroupProcessor> processorMap -> ConcurrentHashMap<PartialPath, VirtualStorageGroupManager> processorMap
从时间序列设备ID找到存储组的过程需要做相应修改,主要影响的函数为:
insert(InsertRowPlan)
insertTablet(InsertTabletPlan)
delete(PartialPath, long, long, long)
deleteTimeseries(PartialPath, long)
query(SingleSeriesExpression, QueryContext, QueryFileManager)
mergeLock(List<PartialPath>)
loadNewTsFileForSync(TsFileResource newTsFileResource)
loadNewTsFile(TsFileResource newTsFileResource)
deleteTsfileForSync(File deletedTsfile)
moveTsfile(File tsfileToBeMoved, File targetDir)
setPartitionVersionToMax(PartialPath storageGroup, long partitionId,long newMaxVersion)
removePartitions(PartialPath storageGroupPath, TimePartitionFilter filter)
setTTL(PartialPath storageGroup, long dataTTL)
3.2 Virtual Storage Group Manager
字段:
VirtualPartitioner partitioner; // 用于决定划分虚拟存储组的具体逻辑
StorageGroupProcessor[] virtualStorageGroupProcessor; // 虚拟存储组列表
方法:
/**
* get processor from device id
* @param partialPath device path
* @return virtual storage group processor
*/
public StorageGroupProcessor getProcessor(PartialPath partialPath, StorageGroupMNode storageGroupMNode)
throws StorageGroupProcessorException, StorageEngineException;
/**
* recover
* @param storageGroupMNode logical sg mnode
*/
public void recover(StorageGroupMNode storageGroupMNode) throws StorageGroupProcessorException;
/**
* get all virtual storage group Processor
* @return all virtual storage group Processor
*/
public StorageGroupProcessor[] getAllVirutalStorageGroupProcessor();
3.3 VirtualPartitioner
public interface VirtualPartitioner {
/**
* use device id to determine storage group id
*
* @param deviceId device id
* @return virtual storage group id
*/
public int deviceToStorageGroup(PartialPath deviceId);
/**
* release resource
*/
public void clear();
/**
* get total number of virtual storage group
*
* @return total number of virtual storage group
*/
public int getPartitionCount();
}
3.4 执行逻辑
3.4.1 设备ID层操作(例如insert, query等)
(1)Storage Engine拿到device ID,根据其逻辑存储组名找到对应的VirtualStorageGroupManager
(2)使用deviceID根据VirtualStorageGroupManager.getProcessor方法拿到虚拟存储组(方案二中就是StorageGroupProcessor)
(3)将请求发给找到的虚拟存储组
3.4.2 逻辑存储组层操作(例如removePartitions等)
(1)Storage Engine拿到device ID,根据其逻辑存储组名找到对应的VirtualStorageGroupManager
(2)使用VirtualStorageGroupManager.getAllVirutalStorageGroupProcessor方法拿到所属全部虚拟存储组(方案二中就是StorageGroupProcessor)
(3)将请求转发至所有虚拟存储组