Method and system for achieving high availability of service under high-load scene in distributed system
View Patent ↗Provided are a method and system for achieving high availability of service under a high-load scene in a distributed system. The method includes constructing a node selection model at a master node of a distributed cluster; constructing a request selection model in each slave node; wherein the request selection model includes weights of designated requests and trade-off parameters set for requests each having a weight greater than a set value; when the distributed system enters the high-load scene, reading, by the master node, the node selection model, and sequentially selecting the served slave nodes according to a time slice round robin policy; and reading, by each slave node, the request selection model, and sequentially returning data of each request according to the trade-off parameters of each request.
1. A method for achieving high availability of service under a high-load scene in a distributed system, comprising the following steps:
constructing a node selection model at a master node of a distributed cluster; wherein the node selection model is used for the master node to select, from the node selection model, served slave nodes and the sequence at which the slave nodes are served;
constructing a request selection model in the slave node; wherein the request selection model comprises weights of designated requests and trade-off parameters set for requests each having a weight greater than a set value; the request selection model is used for each slave node to select, from the request selection model, responded requests and the sequence of the responded requests;
when the distributed system enters the high-load scene, reading, by the master node, the node selection model, and sequentially selecting the served slave nodes according to a time slice round robin policy; and
reading, by each slave node, the request selection model, and sequentially returning data of each request according to the trade-off parameters of each request.
2. The method according to claim 1 , wherein the step of constructing a node selection model at a master node of a distributed cluster comprises:
providing a monitor on the master node to monitor the source of each request, the data size of each request, and the total time of each request;
when a load of the system is normal, periodically and statistically analyzing the number of times of requests, the type of the requests, and the data size and average time of the requests of each slave node, and calculating the average number of times of requests, the average data size of the requests and the average total time of the requests of each hour of different slave nodes; and
sorting corresponding slave nodes according to the number of times of requests to form the node selection model.
3. The method according to claim 2 , wherein the step of sorting corresponding slave nodes according to the number of times of requests to form the node selection model further comprises:
sorting the corresponding slave nodes according to the number of times of requests, and when the number of times of requests is the same, sorting the corresponding slave nodes according to the average data size of the requests to form the node selection model.
4. The method according to claim 1 , wherein the step of constructing a request selection model in each slave node comprises:
setting weights for different requests according to service importance degrees of the different requests;
setting trade-off parameters for each request with a weight greater than or equal to a set threshold; and
sorting corresponding requests according to the sizes of the weights to form the request selection model.
5. The method according to claim 2 , wherein before the step of when the distributed system enters the high-load scene, reading, by the master node, the node selection model, and sequentially selecting the served slave nodes according to a time slice round robin policy, the method comprises:
when the distributed system enters the high-load scene, stopping, by the master node, monitoring and calculation on the requests of each slave node.
6. The method according to claim 2 , wherein in the step of reading, by the master node, the node selection model, and sequentially selecting the served slave nodes according to a time slice round robin policy, the time slice served by each salve node=(the number of times of requests/the number of total nodes)×the average total time of the requests×1/N, where N is an integer greater than 3.
7. The method according to claim 1 , wherein the trade-off parameter represents a minimum parameter satisfaction degree for normal service operation.
8. The method according to claim 1 , wherein for the same priority level, the slave nodes are sorted according to a first come first served principle.
9. The method according to claim 1 , wherein for each request sent to the master node, the weight degrees of requests of each slave node are set according to actual service, and the higher the weight is, the more important the request is to service availability.
10. The method according to claim 1 , wherein the slave nodes perform requests according to first come first served policy in a normal situation of the cluster, in a high-load cluster, requests to be sent be sorted according to a priority level of weights.
11. A system for achieving high availability of service under a high-load scene in a distributed system, comprising a master node, several slave nodes, a node selection model construction module and request selection model construction modules;
the node selection model construction module is configured to construct a node selection model at a master node of a distributed cluster; the node selection model is used for the master node to select, from the node selection model, served slave nodes and the sequence at which the slave nodes are served;
request selection model construction module is configured to construct a request selection model in each slave node; wherein the request selection model comprises weights of designated requests and trade-off parameters set for requests each having a weight greater than a set value; and each request selection model is used for each slave node to select, from the request selection model, responded requests and the sequence of the responded requests;
the master node is configured to read the node selection model when the distributed system enters the high-load scene, and sequentially select the served slave nodes according to a time slice round robin policy; and
each slave node is configured to read the request selection model, and sequentially return data of each request according to the trade-off parameters of each request.
12. The system according to claim 11 , wherein the node selection model construction module comprises a calculation processor, a node selection model generator and a monitor which is provided on the master node;
the monitor is configured to monitor the source of each request, the data size of each request, and the total time of each request;
the calculation processor is configured to when a load of the system is normal, periodically and statistically analyze the number of times of requests, the type of the requests, and the data size and average time of the requests of each slave node, and calculate the average number of times of requests, the average data size of the requests and the average total time of the requests of each hour of different slave nodes; and
the node selection model generator is configured to sort corresponding slave nodes according to the number of times of requests to form the node selection model.
13. The system according to claim 12 , wherein the node selection model generator is configured to sort the corresponding slave nodes according to the number of times of requests, and when the number of times of requests is the same, sort the corresponding slave nodes according to the average data size of the requests to form the node selection model.
14. The system according to claim 11 , wherein each request selection model construction module comprises a weight setter, a trade-off parameter setter, and a request selection model generator;
the weight setter is configured to set weights for different requests according to service importance degrees of the different requests;
the trade-off parameter setter is configured to set trade-off parameters for each request with a weight greater than or equal to a set threshold; and
the request selection model generator is configured to sort corresponding requests according to the sizes of the weights to form the request selection model.
15. The system according to claim 11 , wherein wherein for each request sent to the master node, the weight degrees of requests of each slave node are set according to actual service, and the higher the weight is, the more important the request is to service availability.
16. The system according to claim 11 , wherein for the same priority level, the slave nodes are sorted according to a first come first served principle.
17. The system according to claim 11 , wherein the trade-off parameter represents a minimum parameter satisfaction degree for normal service operation.
18. The system according to claim 11 , wherein the master node is further configured to, when the distributed system enters the high-load scene, stop monitoring and calculating on the requests of each slave node.
19. The system according to claim 11 , wherein the time slice served by each salve node=(the number of times of requests/the number of total nodes)×the average total time of the requests×1/N, where N is an integer greater than 3.
20. The system according to claim 11 , wherein for each request sent to the master node, the weight degrees of requests of each slave node are set according to actual service, and the higher the weight is, the more important the request is to service availability.