In typical IoT, Internet of Vehicles and Operation Monitoring scenarios, there are often many different types of data collecting devices that collect one or more different physical metrics. However, for the collection devices of the same type, there are often many specific collection devices distributed in places. BigData processing system aims to collect all kinds of data, and then calculate and analyze them. For the same kind of devices, the data collected are very regular. Taking smart meters as an example, assuming that each smart meter collects three metrics of current, voltage and phase, the collected data are similar to the following table:
In typical IoT, Internet of Vehicles and Operation Monitoring scenarios, there are often many different types of data collecting devices that collect one or more different physical metrics. However, for the collection devices of the same type, there are often many specific collection devices distributed in places. BigData processing system aims to collect all kinds of data, then calculate and analyze them. For the same kind of devices, the data collected are very regular. Taking smart meters as an example, assuming that each smart meter collects three metrics of current, voltage and phase, the collected data are similar to the following table:
<figure><table>
<thead><tr>
...
...
@@ -127,25 +127,25 @@ Since time-series data is most likely to be structured data, TDengine adopts the
### One Table for One Collection Point
To utilize this time-series and other data features, TDengine requires the user to create a table for each collection point to store collected time-series data. For example, if there are over 10 millions smart meters, means 10 millions tables shall be created. For the table above, 4 tables shall be created for devices D1001, D1002, D1003, and D1004 to store the data collected. This design has several advantages:
To utilize this time-series and other data features, TDengine requires the user to create a table for each collection point to store collected time-series data. For example, if there are over 10 million smart meters, means 10 million tables shall be created. For the table above, 4 tables shall be created for devices D1001, D1002, D1003, and D1004 to store the data collected. This design has several advantages:
1. Guarantee that all data from a collection point can be saved in a continuous memory/hard disk space block by block. If queries are applied only on one point in a time range, this design will reduce the random read latency significantly, thus increase read and query speed by orders of magnitude.
2. Since the data generation process of each collection device is completely independent, means each device has its unique data source, thus writes can be carried out in a lock-free manner to greatly improve the speed.
3. Write latency can be significantly reduced too as the data points generated by the same device will arrive in time order, the new data point will be simply appended to a block.
If the data of multiple devices are written into a table in the traditional way, due to the uncontrollable network delay, the timing of the data from different devices arriving at the server cannot be guaranteed, the writing operation must be protected by locks, and the data of one device cannot be guaranteed to continuously stored together. **The method of one table for each data collection point can ensure the optimal performance of insertion and query of a single data collection point to the greatest extent.**
If the data of multiple devices are traditionally written into a table, due to the uncontrollable network delay, the timing of the data from different devices arriving at the server cannot be guaranteed, the writing operation must be protected by locks, and the data of one device cannot be guaranteed to continuously stored together. **The method of one table for each data collection point can ensure the optimal performance of insertion and query of a single data collection point to the greatest extent.**
TDengine suggests using collection point ID as the table name (like D1001 in the above table). Each point may collect one or more metrics (like the current, voltage, phase as above). Each metric has a column in the table. The data type for a column can be int, float, string and others. In addition, the first column in the table must be a timestamp. TDengine uses the time stamp as the index, and won’t build the index on any metrics stored. All data will be stored in columns.
### STable: A Collection of Data Points in the Same Type
The method of one table for each point will bring a greatly increasing number of tables, which is difficult to manage. Moreover, applications often need to take aggregation operations between collection points, thus aggregation operations will become complicated. To support aggregation over multiple tables efficiently, the STable (Super Table) concept is introduced by TDengine.
The method of one table for each point will bring a greatly increased number of tables, which is difficult to manage. Moreover, applications often need to take aggregation operations between collection points, thus aggregation operations will become complicated. To support aggregation over multiple tables efficiently, the STable (Super Table) concept is introduced by TDengine.
STable is an abstract collection for a type of data point. A STable contains a set of points (tables) that have the same schema or data structure, but with different static attributes (tags). To describe a STable (a combination of data collection points of a specific type), in addition to defining the table structure of the collected metrics, it is also necessary to define the schema of its tag. The data type of tags can be int, float, string, and there can be multiple tags, which can be added, deleted, or modified afterward. If the whole system has N different types of data collection points, N STables need to be established.
In the design of TDengine, **a table is used to represent a specific data collection point, and STable is used to represent a set of data collection points of the same type**. When creating a table for a specific data collection point, the user uses the definition of STable as a template and specifies the tag value of the specific collection point (table). Compared with the traditional relational database, the table (a data collection point) has static tags, and these tags can be added, deleted, and modified afterward. **A STable contains multiple tables with the same time-series data schema but different tag values.**
When aggregating multiple data collection points with the same data type, TDEngine will first find out the tables that meet the tag filters from the STables, and then scan the time-series data of these tables to perform aggregation operation, which can greatly reduce the data sets to be scanned, thus greatly improving the performance of aggregation calculation.
When aggregating multiple data collection points with the same data type, TDEngine will first find out the tables that meet the tag filters from the STables, then scan the time-series data of these tables to perform aggregation operation, which can greatly reduce the data sets to be scanned, thus greatly improving the performance of aggregation calculation.
## <a class="anchor" id="cluster"></a> Cluster and Primary Logic Unit
...
...
@@ -162,21 +162,21 @@ Logical structure diagram of TDengine distributed architecture as following:
A complete TDengine system runs on one or more physical nodes. Logically, it includes data node (dnode), TDEngine application driver (taosc) and application (app). There are one or more data nodes in the system, which form a cluster. The application interacts with the TDengine cluster through taosc's API. The following is a brief introduction to each logical unit.
**Physical node (pnode)**: A pnode is a computer that runs independently and has its own computing, storage and network capabilities. It can be a physical machine, virtual machine or Docker container installed with OS. The physical node is identified by its configured FQDN (Fully Qualified Domain Name). TDengine relies entirely on FQDN for network communication. If you don't know about FQDN, please read the blog post "[All about FQDN of TDengine](https://www.taosdata.com/blog/2020/09/11/1824.html)".
**Physical node (pnode)**: A pnode is a computer that runs independently and has its own computing, storage and network capabilities. It can be a physical machine, virtual machine, or Docker container installed with OS. The physical node is identified by its configured FQDN (Fully Qualified Domain Name). TDengine relies entirely on FQDN for network communication. If you don't know about FQDN, please read the blog post "[All about FQDN of TDengine](https://www.taosdata.com/blog/2020/09/11/1824.html)".
**Data node (dnode):** A dnode is a running instance of the TDengine server-side execution code taosd on a physical node. A working system must have at least one data node. A dnode contains zero to multiple logical virtual nodes (VNODE), zero or at most one logical management node (mnode). The unique identification of a dnode in the system is determined by the instance's End Point (EP). EP is a combination of FQDN (Fully Qualified Domain Name) of the physical node where the dnode is located and the network port number (Port) configured by the system. By configuring different ports, a physical node (a physical machine, virtual machine or container) can run multiple instances or have multiple data nodes.
**Virtual node (vnode)**: In order to better support data sharding, load balancing and prevent data from overheating or skewing, data nodes are virtualized into multiple virtual nodes (vnode, V2, V3, V4, etc. in the figure). Each vnode is a relatively independent work unit, which is the basic unit of time-series data storage, and has independent running threads, memory space and persistent storage path. A vnode contains a certain number of tables (data collection points). When a new table is created, the system checks whether a new vnode needs to be created. The number of vnodes that can be created on a data node depends on the hardware capacities of the physical node where the data node is located. A vnode belongs to only one DB, but a DB can have multiple vnodes. In addition to the stored time-series data, a vnode also stores the schema and tag values of the included tables. A virtual node is uniquely identified in the system by the EP of the data node and the VGroup ID to which it belongs, and is created and managed by the management node.
**Virtual node (vnode)**: To better support data sharding, load balancing and prevent data from overheating or skewing, data nodes are virtualized into multiple virtual nodes (vnode, V2, V3, V4, etc. in the figure). Each vnode is a relatively independent work unit, which is the basic unit of time-series data storage and has independent running threads, memory space and persistent storage path. A vnode contains a certain number of tables (data collection points). When a new table is created, the system checks whether a new vnode needs to be created. The number of vnodes that can be created on a data node depends on the hardware capacities of the physical node where the data node is located. A vnode belongs to only one DB, but a DB can have multiple vnodes. In addition to the stored time-series data, a vnode also stores the schema and tag values of the included tables. A virtual node is uniquely identified in the system by the EP of the data node and the VGroup ID to which it belongs and is created and managed by the management node.
**Management node (mnode)**: A virtual logical unit responsible for monitoring and maintaining the running status of all data nodes and load balancing among nodes (M in figure). At the same time, the management node is also responsible for the storage and management of metadata (including users, databases, tables, static tags, etc.), so it is also called Meta Node. Multiple (up to 5) mnodes can be configured in a TDengine cluster, and they are automatically constructed into a virtual management node group (M0, M1, M2 in the figure). The master/slave mechanism is used to manage between mnodes, and the data synchronization is carried out in a strong consistent way. Any data update operation can only be done on the master. The creation of mnode cluster is completed automatically by the system without manual intervention. There is at most one mnode on each dnode, which is uniquely identified by the EP of the data node to which it belongs. Each dnode automatically obtains the EP of the dnode where all mnodes in the whole cluster are located through internal messaging interaction.
**Management node (mnode)**: A virtual logical unit responsible for monitoring and maintaining the running status of all data nodes and load balancing among nodes (M in the figure). At the same time, the management node is also responsible for the storage and management of metadata (including users, databases, tables, static tags, etc.), so it is also called Meta Node. Multiple (up to 5) mnodes can be configured in a TDengine cluster, and they are automatically constructed into a virtual management node group (M0, M1, M2 in the figure). The master/slave mechanism is used to manage between mnodes, and the data synchronization is carried out in a strongly consistent way. Any data update operation can only be done on the master. The creation of mnode cluster is completed automatically by the system without manual intervention. There is at most one mnode on each dnode, which is uniquely identified by the EP of the data node to which it belongs. Each dnode automatically obtains the EP of the dnode where all mnodes in the whole cluster are located through internal messaging interaction.
**Virtual node group (VGroup)**: Vnodes on different data nodes can form a virtual node group to ensure the high reliability of the system. The virtual node group is managed in a master/slave structure. Write operations can only be performed on the master vnode, and the system synchronizes data to the slave vnode via replication, thus ensuring that one single replica of data is copied on multiple physical nodes. The number of virtual nodes in a vgroup equals the number of data replicas. If the number of replicas of a DB is N, the system must have at least N data nodes. The number of replicas can be specified by the parameter replica when creating DB, and the default is 1. Using the multi-replica feature of TDengine, the same high data reliability can be done without the need for expensive storage devices such as disk arrays. Virtual node group is created and managed by management node, and the management node assigns a system unique ID, aka VGroup ID. If two virtual nodes has the same vnode group ID, means that they belong to the same group and the data is backed up to each other. The number of virtual nodes in a virtual node group can be dynamically changed, allowing only one, that is, no data replication. VGroup ID is never changed. Even if a virtual node group is deleted, its ID will not be reused.
**Virtual node group (VGroup)**: Vnodes on different data nodes can form a virtual node group to ensure the high reliability of the system. The virtual node group is managed in a master/slave structure. Write operations can only be performed on the master vnode, and the system synchronizes data to the slave vnode via replication, thus ensuring that one single replica of data is copied on multiple physical nodes. The number of virtual nodes in a vgroup equals the number of data replicas. If the number of replicas of a DB is N, the system must have at least N data nodes. The number of replicas can be specified by the parameter replica when creating DB, and the default is 1. Using the multi-replica feature of TDengine, the same high data reliability can be done without the need for expensive storage devices such as disk arrays. Virtual node group is created and managed by the management node, and the management node assigns a system unique ID, aka VGroup ID. If two virtual nodes have the same vnode group ID, means that they belong to the same group and the data is backed up to each other. The number of virtual nodes in a virtual node group can be dynamically changed, allowing only one, that is, no data replication. VGroup ID is never changed. Even if a virtual node group is deleted, its ID will not be reused.
**TAOSC**: TAOSC is the driver provided by TDengine to applications, which is responsible for dealing with the interface interaction between application and cluster, and provides the native interface of C/C++ language, which is embedded in JDBC, C #, Python, Go, Node.js language connection libraries. Applications interact with the whole cluster through taosc instead of directly connecting to data nodes in the cluster. This module is responsible for obtaining and caching metadata; forwarding requests for insertion, query, etc. to the correct data node; when returning the results to the application, taosc also need to be responsible for the final level of aggregation, sorting, filtering and other operations. For JDBC, C/C++/C #/Python/Go/Node.js interfaces, this module runs on the physical node where the application is located. At the same time, in order to support the fully distributed RESTful interface, taosc has a running instance on each dnode of TDengine cluster.
**TAOSC**: TAOSC is the driver provided by TDengine to applications, which is responsible for dealing with the interface interaction between application and cluster, and provides the native interface of C/C++ language, which is embedded in JDBC, C #, Python, Go, Node.js language connection libraries. Applications interact with the whole cluster through taosc instead of directly connecting to data nodes in the cluster. This module is responsible for obtaining and caching metadata; forwarding requests for insertion, query, etc. to the correct data node; when returning the results to the application, taosc also needs to be responsible for the final level of aggregation, sorting, filtering and other operations. For JDBC, C/C++/C #/Python/Go/Node.js interfaces, this module runs on the physical node where the application is located. At the same time, to support the fully distributed RESTful interface, taosc has a running instance on each dnode of TDengine cluster.
### Node Communication
**Communication mode**: The communication among each data node of TDengine system, and among application driver and each data node is carried out through TCP/UDP. Considering an IoT scenario, the data writing packets are generally not large, so TDengine uses UDP in addition to TCP for transmission, because UDP is more efficient and is not limited by the number of connections. TDengine implements its own timeout, retransmission, confirmation and other mechanisms to ensure reliable transmission of UDP. For packets with a data volume of less than 15K, UDP is adopted for transmission, and TCP is automatically adopted for transmission of packets with a data volume of more than 15K or query operations. At the same time, TDengine will automatically compress/decompress the data, digital sign/authenticate the data according to the configuration and data packet. For data replication among data nodes, only TCP is used for data transmission.
**Communication mode**: The communication among each data node of TDengine system, and among the application driver and each data node is carried out through TCP/UDP. Considering an IoT scenario, the data writing packets are generally not large, so TDengine uses UDP in addition to TCP for transmission, because UDP is more efficient and is not limited by the number of connections. TDengine implements its own timeout, retransmission, confirmation and other mechanisms to ensure reliable transmission of UDP. For packets with a data volume of less than 15K, UDP is adopted for transmission, and TCP is automatically adopted for transmission of packets with a data volume of more than 15K or query operations. At the same time, TDengine will automatically compress/decompress the data, digital sign/authenticate the data according to the configuration and data packet. For data replication among data nodes, only TCP is used for data transmission.
**FQDN configuration:** A data node has one or more FQDNs, which can be specified in the system configuration file taos.cfg with the parameter "fqdn". If it is not specified, the system will automatically use the hostname of the computer as its FQDN. If the node is not configured with FQDN, you can directly set the configuration parameter fqdn of the node to its IP address. However, IP is not recommended because IP address is variable, and once it changes, the cluster will not work properly. The EP (End Point) of a data node consists of FQDN + Port. With FQDN, it is necessary to ensure the normal operation of DNS service, or configure hosts files on nodes and the nodes where applications are located.
...
...
@@ -184,13 +184,13 @@ A complete TDengine system runs on one or more physical nodes. Logically, it inc
**Cluster external connection**: TDengine cluster can accommodate one single, multiple or even thousands of data nodes. The application only needs to initiate a connection to any data node in the cluster. The network parameter required for connection is the End Point (FQDN plus configured port number) of a data node. When starting the application taos through CLI, the FQDN of the data node can be specified through the option-h, and the configured port number can be specified through -p. If the port is not configured, the system configuration parameter serverPort of TDengine will be adopted.
**Inter-cluster communication**: Data nodes connect with each other through TCP/UDP. When a data node starts, it will obtain the EP information of the dnode where the mnode is located, and then establish a connection with the mnode in the system to exchange information. There are three steps to obtain EP information of the mnode: 1. Check whether the mnodeEpList file exists, if it does not exist or cannot be opened normally to obtain EP information of the mnode, skip to the second step; 2: Check the system configuration file taos.cfg to obtain node configuration parameters firstEp and secondEp (the node specified by these two parameters can be a normal node without mnode, in this case, the node will try to redirect to the mnode node when connected). If these two configuration parameters do not exist or do not exist in taos.cfg, or are invalid, skip to the third step; 3: Set your own EP as a mnode EP and run it independently. After obtaining the mnode EP list, the data node initiates the connection. It will successfully join the working cluster after connected. If not successful, it will try the next item in the mnode EP list. If all attempts are made, but the connection still fails, sleep for a few seconds before trying again.
**Inter-cluster communication**: Data nodes connect with each other through TCP/UDP. When a data node starts, it will obtain the EP information of the dnode where the mnode is located, and then establish a connection with the mnode in the system to exchange information. There are three steps to obtain EP information of the mnode: 1. Check whether the mnodeEpList file exists, if it does not exist or cannot be opened normally to obtain EP information of the mnode, skip to the second step; 2: Check the system configuration file taos.cfg to obtain node configuration parameters firstEp and secondEp (the node specified by these two parameters can be a normal node without mnode, in this case, the node will try to redirect to the mnode node when connected). If these two configuration parameters do not exist or do not exist in taos.cfg, or are invalid, skip to the third step; 3: Set your own EP as a mnode EP and run it independently. After obtaining the mnode EP list, the data node initiates the connection. It will successfully join the working cluster after connection. If not successful, it will try the next item in the mnode EP list. If all attempts are made, but the connection still fails, sleep for a few seconds before trying again.
**The choice of MNODE**: TDengine logically has a management node, but there is no separated execution code. The serverside only has a set of execution code taosd. So which data node will be the management node? This is determined automatically by the system without any manual intervention. The principle is as follows: when a data node starts, it will check its End Point and compare it with the obtained mnode EP List. If its EP exists in it, the data node shall start the mnode module and become a mnode. If your own EP is not in the mnode EP List, the mnode module will not start. During the system operation, due to load balancing, downtime and other reasons, mnode may migrate to the new dnode, while totally transparent without manual intervention. The modification of configuration parameters is the decision made by mnode itself according to resources usage.
**The choice of MNODE**: TDengine logically has a management node, but there is no separated execution code. The server-side only has a set of execution code taosd. So which data node will be the management node? This is determined automatically by the system without any manual intervention. The principle is as follows: when a data node starts, it will check its End Point and compare it with the obtained mnode EP List. If its EP exists in it, the data node shall start the mnode module and become a mnode. If your own EP is not in the mnode EP List, the mnode module will not start. During the system operation, due to load balancing, downtime and other reasons, mnode may migrate to the new dnode, while totally transparent without manual intervention. The modification of configuration parameters is the decision made by mnode itself according to resources usage.
**Add new data nodes:** After the system has a data node, it has become a working system. There are two steps to add a new node into the cluster. Step1: Connect to the existing working data node using TDengine CLI, and then add the End Point of the new data node with the command "create dnode"; Step 2: In the system configuration parameter file taos.cfg of the new data node, set the firstEp and secondEp parameters to the EP of any two data nodes in the existing cluster. Please refer to the detailed user tutorial for detailed steps. In this way, the cluster will be established step by step.
**Redirection**: No matter about dnode or taosc, the connection to the mnode shall be initiated first, but the mnode is automatically created and maintained by the system, so user does not know which dnode is running the mnode. TDengine only requires a connection to any working dnode in the system. Because any running dnode maintains the currently running mnode EP List, when receiving a connecting request from the newly started dnode or taosc, if it’s not an mnode by self, it will reply the mnode EP List back. After receiving this list, taosc or the newly started dnode will try to establish the connection again. When the mnode EP List changes, each data node quickly obtains the latest list and notifies taosc through messaging interaction among nodes.
**Redirection**: No matter about dnode or taosc, the connection to the mnode shall be initiated first, but the mnode is automatically created and maintained by the system, so the user does not know which dnode is running the mnode. TDengine only requires a connection to any working dnode in the system. Because any running dnode maintains the currently running mnode EP List, when receiving a connecting request from the newly started dnode or taosc, if it’s not a mnode by self, it will reply to the mnode EP List back. After receiving this list, taosc or the newly started dnode will try to establish the connection again. When the mnode EP List changes, each data node quickly obtains the latest list and notifies taosc through messaging interaction among nodes.
### A Typical Messaging Process
...
...
@@ -206,9 +206,9 @@ To explain the relationship between vnode, mnode, taosc and application and thei
5. After vnode inserts the data, it gives a reply to taosc, indicating that the insertion is successful. If taosc doesn't get a response from vnode for a long time, taosc will judge the node as offline. In this case, if there are multiple replicas of the inserted database, taosc will issue an insert request to the next vnode in vgroup.
6. Taosc notifies APP that writing is successful.
For Step 2 and 3, when taosc starts, it does not know the End Point of mnode, so it will directly initiate a request to the externally serving End Point of the configured cluster. If the dnode that received the request does not have an mnode configured, it will inform the mnode EP list in a reply message, so that taosc will re-issue a request to obtain meta-data to the EP of another new mnode.
For Step 2 and 3, when taosc starts, it does not know the End Point of mnode, so it will directly initiate a request to the externally serving End Point of the configured cluster. If the dnode that received the request does not have a mnode configured, it will inform the mnode EP list in a reply message, so that taosc will re-issue a request to obtain meta-data to the EP of another new mnode.
For Step 4 and 5, without caching, taosc can't recognize the master in the virtual node group, so assumes that the first vnodeID is the master and send a request to it. If the requested vnode is not the master, it will reply the actual master as a new target taosc makes a request to. Once the reply of successful insertion is obtained, taosc will cache the information of master node.
For Step 4 and 5, without caching, taosc can't recognize the master in the virtual node group, so assumes that the first vnodeID is the master and send a request to it. If the requested vnode is not the master, it will reply to the actual master as a new target taosc requests to. Once the reply of successful insertion is obtained, taosc will cache the information of master node.
The above is the process of inserting data, and the processes of querying and calculating are completely consistent. Taosc encapsulates and shields all these complicated processes, and has no perception and no special treatment for applications.
...
...
@@ -220,26 +220,26 @@ Through taosc caching mechanism, mnode needs to be accessed only when a table is
The data stored by TDengine include collected time-series data, metadata related to libraries and tables, tag data, etc. These data are specifically divided into three parts:
- Time-series data: stored in vnode and composed of data, head and last files. The amount of data is large and query amount depends on the application scenario. Out-of-order writing is allowed, but delete operation is not supported for the time being, and update operation is only allowed when update parameter is set to 1. By adopting the model with one table for each collection point, the data of a given time period is continuously stored, and the writing against one single table is a simple add operation. Multiple records can be read at one time, thus ensuring the insert and query operation of a single collection point with best performance.
- Time-series data: stored in vnode and composed of data, head and last files. The amount of data is large and query amount depends on the application scenario. Out-of-order writing is allowed, but delete operation is not supported for the time being, and update operation is only allowed when update parameter is set to 1. By adopting the model with one table for each collection point, the data of a given time period is continuously stored, and the writing against one single table is a simple add operation. Multiple records can be read at one time, thus ensuring the insert and query operation of a single collection point with the best performance.
- Tag data: meta files stored in vnode support four standard operations of add, delete, modify and check. The amount of data is not large. If there are N tables, there are N records, so all can be stored in memory. If there are many tag filtering operations, queries will be very frequent and TDengine supports multi-core and multi-threaded concurrent queries. As long as the computing resources are sufficient, even in face of millions of tables, the filtering results will return in milliseconds.
- Metadata: stored in mnode, including system node, user, DB, Table Schema and other information. Four standard operations of add, delete, modify and query are supported. The amount of these data are not large and can be stored in memory, moreover the query amount is not large because of the client cache. Therefore, TDengine uses centralized storage management, however, there will be no performance bottleneck.
- Metadata: stored in mnode, including system node, user, DB, Table Schema and other information. Four standard operations of add, delete, modify and query are supported. The amount of these data are not large and can be stored in memory, moreover, the query amount is not large because of the client cache. Therefore, TDengine uses centralized storage management, however, there will be no performance bottleneck.
Compared with the typical NoSQL storage model, TDengine stores tag data and time-series data completely separately, which has two major advantages:
- Greatly reduce the redundancy of tag data storage: general NoSQL database or time-series database adopts K-V storage, in which Key includes timestamp, device ID and various tags. Each record carries these duplicates, so wasting storage space. Moreover, if the application needs to add, modify or delete tags on historical data, it has to traverse the data and rewrite again, which is extremely expensive to operate.
- Realize extremely efficient aggregation query between multiple tables: when doing aggregation query between multiple tables, it firstly finds out the tag filtered tables, and then find out the corresponding data blocks of these tables to greatly reduce the data sets to be scanned, thus greatly improving the query efficiency. Moreover, tag data is managed and maintained in a full-memory structure, and tag data queries in tens of millions can return in milliseconds.
- Greatly reduce the redundancy of tag data storage: general NoSQL database or time-series database adopts K-V storage, in which Key includes a timestamp, a device ID and various tags. Each record carries these duplicates, so wasting storage space. Moreover, if the application needs to add, modify or delete tags on historical data, it has to traverse the data and rewrite again, which is extremely expensive to operate.
- Realize extremely efficient aggregation query between multiple tables: when doing aggregation query between multiple tables, it first finds out the tag filtered tables, and then find out the corresponding data blocks of these tables to greatly reduce the data sets to be scanned, thus greatly improving the query efficiency. Moreover, tag data is managed and maintained in a full-memory structure, and tag data queries in tens of millions can return in milliseconds.
### Data Sharding
For large-scale data management, to achieve scale-out, it is generally necessary to adopt the a Partitioning strategy as Sharding. TDengine implements data sharding via vnode, and time-series data partitioning via one data file for each time range.
VNode (Virtual Data Node) is responsible for providing writing, query and calculation functions for collected time-series data. To facilitate load balancing, data recovery and support heterogeneous environments, TDengine splits a data node into multiple vnodes according to its computing and storage resources. The management of these vnodes is done automatically by TDengine and completely transparent to the application.
VNode (Virtual Data Node) is responsible for providing writing, query and calculation functions for collected time-series data. To facilitate load balancing, data recovery and support heterogeneous environments, TDengine splits a data node into multiple vnodes according to its computing and storage resources. The management of these vnodes is done automatically by TDengine and is completely transparent to the application.
For a single data collection point, regardless of the amount of data, a vnode (or vnode group, if the number of replicas is greater than 1) has enough computing resource and storage resource to process (if a 16-byte record is generated per second, the original data generated in one year will be less than 0.5 G), so TDengine stores all the data of a table (a data collection point) in one vnode instead of distributing the data to two or more dnodes. Moreover, a vnode can store data from multiple data collection points (tables), and the upper limit of the tables’ quantity for a vnode is one million. By design, all tables in a vnode belong to the same DB. On a data node, unless specially configured, the number of vnodes owned by a DB will not exceed the number of system cores.
When creating a DB, the system does not allocate resources immediately. However, when creating a table, the system will check if there is an allocated vnode with free tablespace. If so, the table will be created in the vacant vnode immediately. If not, the system will create a new vnode on a dnode from the cluster according to the current workload, and then a table. If there are multiple replicas of a DB, the system does not create only one vnode, but a vgroup (virtual data node group). The system has no limit on the number of vnodes, which is just limited by the computing and storage resources of physical nodes.
The meda data of each table (including schema, tags, etc.) is also stored in vnode instead of centralized storage in mnode. In fact, this means sharding of meta data, which is convenient for efficient and parallel tag filtering operations.
The meta data of each table (including schema, tags, etc.) is also stored in vnode instead of centralized storage in mnode. In fact, this means sharding of meta data, which is convenient for efficient and parallel tag filtering operations.
### Data Partitioning
...
...
@@ -249,7 +249,7 @@ In general, **TDengine splits big data by vnode and time as two dimensions**, wh
### Load Balancing
Each dnode regularly reports its status (including hard disk space, memory size, CPU, network, number of virtual nodes, etc.) to the mnode (virtual management node) for declaring the status of the entire cluster. Based on the overall state, when an mnode finds an overloaded dnode, it will migrate one or more vnodes to other dnodes. In the process, external services keep running and the data insertion, query and calculation operations are not affected.
Each dnode regularly reports its status (including hard disk space, memory size, CPU, network, number of virtual nodes, etc.) to the mnode (virtual management node) for declaring the status of the entire cluster. Based on the overall state, when a mnode finds an overloaded dnode, it will migrate one or more vnodes to other dnodes. In the process, external services keep running and the data insertion, query and calculation operations are not affected.
If the mnode has not received the dnode status for a period of time, the dnode will be judged as offline. When offline lasts a certain period of time (the duration is determined by the configuration parameter ‘offlineThreshold’), the dnode will be forcibly removed from the cluster by mnode. If the number of replicas of vnodes on this dnode is greater than one, the system will automatically create new replicas on other dnodes to ensure the replica number. If there are other mnodes on this dnode and the number of mnodes replicas is greater than one, the system will automatically create new mnodes on other dnodes to ensure t the replica number.
