CN117785457A - Resource management methods, devices, equipment and storage media - Google Patents

Abstract

The invention discloses a resource management method, device, equipment and storage medium. The method includes: when the host has deployed offline services, if an instruction to deploy online services is received, the resource allocation for offline services is reduced and the resource allocation for online services is reduced. Each microservice is configured with excess resources in the current time period; the predicted request frequency of each microservice in the next preset time period is predicted based on the pre-trained prediction model, and each microservice is configured in the next preset time period based on the predicted request frequency. Resources in the time period are pre-allocated; the real-time tail delay of each microservice is collected at preset intervals, and the tail delay target generated by each microservice in the offline analysis phase is obtained; the microservice and the tail delay target are adjusted based on the real-time tail delay and tail delay target. Resource allocation for offline business. The invention can effectively cope with the scenario of mixed deployment of multiple online and multiple offline services on the host, ensure end-to-end response delay of online services, and improve resource utilization at the same time.

Description

Resource management method, device, equipment and storage medium

Technical field

The present application relates to the field of cloud computing technology, and in particular to a resource management method, device, equipment and storage medium.

Background technique

Cloud computing is a model that provides computing resources and services over the network and can support a wide range of applications such as online applications and batch processing calls. In order to ensure the service quality of applications, resource management systems in cloud computing often over-allocate resources to applications, but this can also lead to low overall resource utilization in the cloud computing system. As a commonly used online application architecture, microservices are usually composed of multiple microservice components, and there are complex calling relationships between microservice components.

In recent years, microservice architecture has developed rapidly and is widely used in cloud computing technology. Compared with the traditional monolithic architecture, which runs all service components in one application, the microservice system decouples the application into multiple components to make it easier to manage, maintain and update. Due to the lightweight and loose coupling nature of microservices, under increasing load, Resource Manager can locate an overloaded individual microservice and scale it independently rather than scaling the entire application. .

With the widespread application of microservice architecture, microservice resource management faces new challenges. Although microservice architecture is flexible, it is necessary to handle service requests from hundreds of microservices while ensuring service level agreements. A complex dependency graph is formed between these components, making it very difficult to implement fine-grained resource management to improve resource utilization and ensure end-to-end response latency. In addition, microservice containers usually run on the same physical machine as batch applications, which may cause performance imbalance between containers of the same microservice, especially when the workload is heavy, resulting in resource interference. Most importantly, for the scenario of multiple online services and multiple offline services co-located on the host, reasonable resource allocation is particularly important, and the existing resource scheduling schemes are either resource allocation for online services alone or resource allocation for offline services alone. There is still a lack of a resource scheduling scheme that can effectively coordinate the co-location of online and offline services.

Contents of the invention

In view of this, this application provides a resource management method, device, equipment and storage medium to solve the problem of unreasonable resource scheduling when online services and offline services are mixed.

In order to solve the above technical problems, a technical solution adopted by this application is to provide a resource management method, which includes: when the host has deployed offline services, if an instruction to deploy online services is received, then reduce the resource allocation of offline services, And allocate excess resources for each microservice of the online business in the current time period; predict the predicted request frequency of each microservice in the next preset time period based on the pre-trained prediction model, and configure each microservice based on the predicted request frequency. Pre-allocate resources in the next preset time period; collect the real-time tail delay of each microservice at preset intervals, and obtain the tail delay target generated by each microservice in the offline analysis phase; based on the real-time tail delay and tail delay target Adjust resource allocation for microservices and offline businesses.

As a further improvement of this application, the predicted request frequency of each microservice in the next preset time period is predicted based on the pre-trained prediction model, and the resources of each microservice in the next preset time period are predicted based on the predicted request frequency. Allocation, including: after deploying the online business, obtaining the pre-built tail delay model and resource utilization model corresponding to each microservice with request frequency as the independent variable. The tail delay model is based on multiple preset requests for each microservice. The tail delay under frequency is constructed, and the resource utilization model is constructed based on the resource utilization of each microservice under multiple preset request frequencies; the current request frequency of each microservice in the current time period is obtained; the resource utilization model of each microservice is The current time period and current request frequency are input to the pre-trained time series prediction model, and the predicted request frequency for the next preset time period of each microservice is predicted; based on the current request frequency, predicted request frequency and each microservice's The tail delay model and resource utilization model pre-allocate resources for the next preset time period for each microservice.

As a further improvement of the present application, resources for the next preset time period are pre-allocated to each microservice based on the current request frequency, the predicted request frequency, and the tail delay model and resource utilization model of each microservice, including: calculating a first request frequency reference value according to a first preset rule using the current request frequency, the pre-acquired number of copies of the microservice, and a first preset error coefficient; comparing the predicted request frequency with the first request frequency reference value; if the predicted request frequency is greater than or equal to the first request frequency reference value, expanding the resources occupied by the microservice; if the predicted request frequency is less than the first request frequency reference value, shrinking the resources occupied by the microservice.

As a further improvement of this application, the resources occupied by the microservice are expanded, including: confirming the minimum request frequency corresponding to the tail delay target based on the tail delay target and the tail delay model; based on the minimum request frequency, the number of copies of the microservice, and the number of copies of the microservice. A preset error coefficient calculates the second request frequency reference value according to the second preset rule; compares the predicted request frequency and the second request frequency reference value; if the predicted request frequency is less than the second request frequency reference value, the microservice The resources are expanded vertically. After vertical expansion, the resource utilization utilization predicted request frequency, number of copies, resource utilization model and second preset error coefficient are calculated according to the third preset rule; if the predicted request frequency is greater than or equal to the second If the request frequency reference value is used, the resources of the microservice will be expanded horizontally and then adjusted vertically. The number of copies for horizontal expansion is calculated according to the fourth preset rule using the predicted request frequency, tail delay target and first preset error coefficient. The adjusted maximum resource utilization utilization prediction request frequency of each copy, the number of copies, the resource utilization model and the second preset error coefficient are calculated according to the third preset rule.

As a further improvement of this application, shrinking the resources occupied by microservices includes: using the predicted request frequency, tail delay target and first preset error coefficient to calculate the reference value of the number of copies according to the fourth preset rule; compare Set the reference value of the number of copies and the size of the number of copies; if the reference value of the number of copies is equal to the number of copies, vertically shrink the resources of the microservice, and the resource utilization after vertical shrinkage predicts the request frequency, number of copies, and resource utilization. The rate model and the second preset error coefficient are calculated according to the third preset rule; if the reference value of the number of copies is less than the number of copies, the resources of the microservice will be horizontally reduced and then adjusted vertically. The number of copies after the horizontal reduction will be As a reference value for the number of copies, the longitudinally adjusted maximum resource utilization prediction request frequency, number of copies, resource utilization model and second preset error coefficient of each copy are calculated according to the third preset rule.

As a further improvement of this application, the tail delay target of each microservice is generated in the offline analysis stage, which specifically includes: in the offline analysis stage, the response delay and call dependency graph of each microservice are obtained, and the call dependency graph is based on each microservice. The call link of the service is generated. The call dependency graph includes multiple nodes, each node corresponds to a microservice; the average response delay of each node in the call dependency graph is confirmed based on the response delay of each microservice; from the in-degree to Starting from the node 0, traverse all paths in the graph, and accumulate the average response delay of the nodes passed by the path to obtain the average response delay of each path; use the average response delay of each node divided by the path where the node is located The average response delay of the node is obtained to obtain the delay ratio of the node; the delay ratio is multiplied by the pre-specified tail delay to obtain the tail delay target of each node, and when a node has multiple tail delay targets, the smallest tail delay target is selected as The ultimate tail delay target.

As a further improvement of the present application, the resource allocation of microservices and offline services is adjusted according to the real-time tail delay and the tail delay target, including: if the real-time tail delay of the target microservice is greater than the tail delay target corresponding to the target microservice, the resources of the deployed offline service are reduced by half; if the real-time tail delay of the target microservice is not greater than the tail delay target corresponding to the target microservice, a new offline service is deployed or the resources of the deployed offline service are increased.

To solve the above technical problems, another technical solution adopted by the present application is: to provide a resource management device, which includes: an online business deployment module, which is used to reduce the resource allocation of offline business when the host has deployed offline business, if an instruction to deploy online business is received, and to configure excess resources for each microservice of online business in the current time period; a pre-allocation module, which is used to predict the predicted request frequency of each microservice in the next preset time period based on a pre-trained prediction model, and pre-allocate resources for each microservice in the next preset time period according to the predicted request frequency; a collection module, which is used to collect the real-time tail delay of each microservice at intervals of a preset period, and obtain the tail delay target generated by each microservice in the offline analysis stage; an adjustment module, which is used to adjust the resource allocation of microservices and offline businesses according to the real-time tail delay and the tail delay target.

In order to solve the above technical problems, another technical solution adopted by this application is to provide a computer device. The computer device includes a processor and a memory coupled to the processor. Program instructions are stored in the memory. When the program instructions are executed by the processor, the processor is caused to execute the steps of any of the above resource management methods.

In order to solve the above technical problems, another technical solution adopted by this application is to provide a storage medium that stores program instructions that can implement any of the above resource management methods.

The beneficial effects of this application are: the resource management method of this application reduces the resources of the offline business before deploying the online business when it is necessary to deploy the online business after deploying the offline business, and manages all the microservices of the online business under the Resources within a preset time period are predicted and pre-allocated to ensure the end-to-end response delay of online services. At the same time, the tail delay of all microservices is monitored, combined with the preset tail delay target of each microservice. Dynamically adjust the resources occupied by all microservices and the resources occupied by offline services, so as to ensure the end-to-end response delay of online services while improving resource utilization, ensuring that online services and offline services can be highly efficient in the case of co-location. run.

Description of drawings

FIG1 is a schematic diagram of a process of a resource management method according to an embodiment of the present invention;

Figure 2 is a schematic diagram of the functional modules of the resource management device according to the embodiment of the present invention;

Figure 3 is a schematic structural diagram of a computer device according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of the structure of a storage medium according to an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

The terms “first”, “second” and “third” in this application are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of indicated technical features. Thus, features defined as "first", "second", and "third" may explicitly or implicitly include at least one of these features. In the description of this application, "plurality" means at least two, such as two, three, etc., unless otherwise clearly and specifically limited. All directional indications (such as up, down, left, right, front, back...) in the embodiments of this application are only used to explain the relative positional relationship between components in a specific posture (as shown in the drawings). , sports conditions, etc., if the specific posture changes, the directional indication will also change accordingly. Furthermore, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes Other steps or units inherent to such processes, methods, products or devices.

Reference herein to "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearances of this phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those skilled in the art understand, both explicitly and implicitly, that the embodiments described herein may be combined with other embodiments.

Figure 1 is a schematic flowchart of a resource management method according to an embodiment of the present invention. It should be noted that, if substantially the same results are obtained, the method of the present invention is not limited to the process sequence shown in Figure 1 . As shown in Figure 1, the resource management method includes steps:

Step S101: When the host has deployed offline services and receives an instruction to deploy online services, it reduces the resource allocation for offline services and allocates excess resources for each microservice of online services in the current time period.

It should be noted that online services include but are not limited to: long running time, delay-sensitive, high requirements for stability, unstable services that will be immediately perceived and cause losses, and obvious peaks and troughs, such as high during the day and low at night, such as advertising search services. Offline services include but are not limited to services that are not delay-sensitive, can be retried, and generally have a short running time of about tens of minutes. They are generally big data computing, machine learning and other services.

Specifically, in this embodiment, when deciding to deploy online services, it is first necessary to reduce the resource allocation of all deployed offline services to ensure that sufficient resources are allocated to online services. The method of reducing the resource allocation of offline services may be to reduce the resource allocation of offline services to half of the original value. When deploying an online business, first allocate excess resources to the microservices of the online business to ensure normal deployment of the online business, and then adjust the resource allocation of the microservices later.

Step S102: Predict the predicted request frequency of each microservice in the next preset time period based on the pre-trained prediction model, and pre-allocate the resources of each microservice in the next preset time period based on the predicted request frequency.

It should be noted that since the request frequency of online services usually changes with a certain periodicity, such as one day or one week, the time series prediction model can be used to predict the request frequency of online services.

In this embodiment, it is first necessary to divide the time periods according to the change pattern of the request frequency of each microservice of the online business to obtain multiple time periods. For example, one day is divided into 24 time periods, and each time period corresponds to one hour. . Then, the time series prediction model is trained using the pre-collected historical request frequency data of the microservices. After the online business is deployed, the corresponding time series prediction model and the request frequency of the current preset time period are used to predict each microservice, and the predicted request frequency of each microservice in the next preset time period is obtained, and then based on the The predicted request frequency of the next preset time period pre-allocates the resources of each microservice in the next preset time period. After entering the next preset time period, the resources of the microservice can be directly allocated based on the pre-allocated resources. The allocation is adjusted so that each microservice can be reasonably allocated with reasonable resources to avoid the problem of over-allocation or under-allocation of resources.

Further, step S102 specifically includes:

1. After deploying the online business, obtain the pre-built tail delay model and resource utilization model corresponding to each microservice with request frequency as the independent variable. The tail delay model is based on each microservice under multiple preset request frequencies. The tail-latency model is built based on the resource utilization of each microservice at multiple preset request frequencies.

In this embodiment, it is necessary to construct a request frequency-tail delay model and a request frequency-resource utilization model for each microservice. Both models are constructed with request frequency as an independent variable. The model building process is as follows:

First, initially set the request frequency to the lowest level (such as 10 requests per second), access the application at the lowest level of request frequency, and record the tail delay of each microservice at this time (such as the 95% response time). delay) and resource utilization.

Next, gradually increase the request frequency per second, each time increasing the request frequency to a minimum level (for example, if the minimum level is 10 requests per second, the requests after the increase are 20 requests per second, 30 requests per second, and so on), and record the tail delay (such as the response delay at the 95th percentile) and resource utilization of each microservice under this request frequency.

The collected data was then organized into two groups. In one group, the request frequency is arranged from small to large, and the tail delay corresponding to each request frequency is listed. For the second group, the request frequencies are also arranged from small to large, and the corresponding resource utilization under each request frequency is listed.

The specific format is as follows:

First group:

[Request frequency 1, request frequency 2, request frequency 3, ..., request frequency n];

[Tail delay 1, tail delay 2, tail delay 3, ..., tail delay n];

Second Group:

[Request frequency 1, request frequency 2, request frequency 3, ..., request frequency n];

[Resource utilization rate 1, resource utilization rate 2, resource utilization rate 3, ..., resource utilization rate n];

Finally, based on the two sets of collated data, a request frequency-tail delay model and a request frequency-resource utilization model were established for each microservice by fitting curves.

It should be noted that the resource utilization is specifically the CPU utilization.

2. Obtain the current request frequency of each microservice in the current time period.

3. Input the current time period and current request frequency of each microservice into the pre-trained time series prediction model, and predict the predicted request frequency of each microservice in the next preset time period.

Specifically, after obtaining the current request frequency of the microservice, the current request frequency and the current time period are input to a pre-trained time series prediction model, and the predicted request frequency of the next preset time period is predicted.

4. Pre-allocate resources for the next preset time period to each microservice based on the current request frequency, predicted request frequency, and the tail delay model and resource utilization model of each microservice.

Specifically, after obtaining the predicted request frequency for the next preset time period, resources are pre-allocated for the microservice based on the predicted request frequency and the pre-constructed tail delay model and resource utilization model.

Furthermore, resources for the next preset time period are pre-allocated to each microservice according to the current request frequency, the predicted request frequency, the tail delay model of each microservice, and the resource utilization model, including:

4.1. Calculate the first request frequency reference value according to the first preset rule using the current request frequency, the number of copies of the microservice obtained in advance and the first preset error coefficient.

Specifically, take the resource pre-allocation process of a microservice as an example: Assume that the number of copies of the microservice is n, and the average request frequency of the current copy is R _c (that is, the current request frequency). It is predicted according to the time series prediction model. The predicted request frequency of the microservice in the next preset time period is R _n .

First, the first request frequency reference value R ₁ is calculated using the current request frequency R _c , the pre-obtained number of copies of the microservice n and the first preset error coefficient α. The calculation process is as follows:

R ₁ =R _c *n*α;

The purpose of setting the first preset error coefficient α is to eliminate the influence of the model prediction error as much as possible. The value range of the first preset error coefficient α is between 0 and 1, and is usually 0.7.

4.2. Compare the relationship between the predicted request frequency and the first request frequency reference value.

4.3. If the predicted request frequency is greater than or equal to the first request frequency reference value, the resources occupied by the microservice will be expanded.

Specifically, when the predicted request frequency is greater than or equal to the first request frequency reference value, it means that the resources occupied by the microservice in the next preset time period will increase. Therefore, the resources occupied by the microservice need to be expanded.

Furthermore, when performing expansion processing, it is necessary to analyze whether the microservice should be expanded vertically or horizontally. Therefore, the resources occupied by microservices should be expanded, including:

4.3.1. Determine the minimum request frequency corresponding to the tail latency target based on the tail latency target and the tail latency model.

4.3.2. Calculate the second request frequency reference value according to the second preset rule based on the minimum request frequency, the number of copies of the microservice and the first preset error coefficient.

Specifically, first obtain the tail delay target L _k preset for each microservice, and then combine the tail delay target and the tail delay model to confirm the minimum request frequency R _k corresponding to the tail delay target.

Then, the second request frequency reference value R ₂ is calculated using the minimum request frequency R _k , the pre-obtained number of copies of the microservice n and the first preset error coefficient α. The calculation process is as follows:

R ₂ =R _k *n*α.

4.3.3. Compare the predicted request frequency and the second request frequency reference value.

4.3.4. If the predicted request frequency is less than the second request frequency reference value, vertically expand the resources of the microservice. The resource utilization after vertical expansion uses the predicted request frequency, number of copies, resource utilization model and second preset The error coefficient is calculated according to the third preset rule.

Specifically, when the predicted request frequency is less than the second request frequency reference value, vertical expansion processing is performed, and the resource utilization rate _Cd after vertical expansion is calculated as follows: according to the resource utilization model, the resource utilization rate _Cn when the request frequency is ( _Rn /n) is confirmed, and then the resource utilization rate after expansion is calculated according to the third preset rule: _Cd = _Cn *β, β is the second preset error coefficient, the purpose is to eliminate the influence of the model prediction error as much as possible, and its value range is between 1 and 2, and it can usually be 1.4. Therefore, the final maximum resource utilization rate of the microservice will be limited to _Cd .

4.3.5. If the predicted request frequency is greater than or equal to the second request frequency reference value, the resources of the microservice will be horizontally expanded and then adjusted vertically. The number of copies for horizontal expansion will be determined by using the predicted request frequency, tail delay target and first prediction. Assume that the error coefficient is calculated according to the fourth preset rule. The vertically adjusted maximum resource utilization prediction request frequency, number of copies, resource utilization model and second preset error coefficient of each copy are calculated according to the third preset rule. get.

Specifically, when performing horizontal expansion, the number of replicas after expansion, _nd , is calculated as follows: _nd = ceil( _Rn /( _Rk *α)), where ceil() means rounding up decimals, and vertically adjusting the maximum resource utilization of each replica to _Cd .

4.4. If the predicted request frequency is less than the first request frequency reference value, the resources occupied by the microservice are scaled down.

Specifically, when the predicted request frequency is less than the first request frequency reference value, it means that the resources occupied by the microservice in the next preset time period will decrease. Therefore, the resources occupied by the microservice need to be reduced.

Furthermore, the resources occupied by the microservices are scaled down, including:

4.4.1. Use the predicted request frequency, tail delay target and first preset error coefficient to calculate the reference value of the number of copies according to the fourth preset rule.

Specifically, the reference value n ₁ for the number of replicas is calculated in the same way as the number of replicas after expansion during capacity expansion, specifically: n ₁ =ceil(R _n /(R _k *α)).

4.4.2. Compare the reference value of the number of copies and the size of the number of copies.

4.4.3. If the reference value of the number of replicas is equal to the number of replicas, perform vertical scaling of the resources of the microservice. After the vertical scaling, the resource utilization prediction request frequency, number of replicas, resource utilization model and second preset The error coefficient is calculated according to the third preset rule.

Specifically, when the replica number reference value n ₁ is equal to the number of replicas, vertical scaling is performed, and the resource utilization rate after scaling is C _d .

4.4.4. If the reference value of the number of copies is less than the number of copies, the resources of the microservice will be scaled down horizontally and then adjusted vertically. The number of copies after the horizontal reduction is the reference value for the number of copies. The number of copies of each copy after the vertical adjustment will be The maximum resource utilization prediction request frequency, number of copies, resource utilization model and second preset error coefficient are calculated according to the third preset rule.

Specifically, when the reference value of the number of replicas n ₁ is less than the number of replicas, it is necessary to perform horizontal reduction first and then adjust it vertically. The number of replicas after horizontal reduction n _s =ceil(R _n /(R _k *α), The maximum resource utilization of each replica in the vertical direction is adjusted to C _d .

It should be noted that the above reduction and expansion processes need to be executed at specific points in time: the expansion operation occurs before the time series prediction model starts the next prediction (such as 1 minute before the prediction is performed), and the reduction operation occurs at the time series After the prediction model completes the next prediction (such as 1 minute after completing the prediction).

Step S103: Collect the real-time tail delay of each microservice at preset intervals, and obtain the tail delay target generated by each microservice in the offline analysis stage.

Specifically, in this embodiment, a link tracking tool (such as Jaeger, etc.) is used to periodically sample (for example, every 5 seconds) the response delay of each microservice and monitor changes in its tail delay.

Furthermore, the tail latency target of each microservice is generated during the offline analysis phase, including:

1. In the offline analysis phase, obtain the response delay and call dependency graph of each microservice. The call dependency graph is generated based on the call link of each microservice. The call dependency graph includes multiple nodes, and each node corresponds to a microservice. .

Specifically, the link tracking tool can analyze the call relationships through the recorded call data, and then generate a dependency graph for the application based on this information, which reveals the call relationships between application components. The nodes in the call dependency graph represent components, the arrows represent the calling relationship, and the numbers on the path represent the number of calls.

2. Confirm the average response delay of each node in the call dependency graph based on the response delay of each microservice.

3. Starting from the node with in-degree 0, traverse all paths in the graph, accumulate the average response delays of the nodes passed by the path, and obtain the average response delay of each path.

4. Use the average response delay of each node divided by the average response delay of the path where the node is located to obtain the delay ratio of the node.

5. Use the delay ratio multiplied by the pre-specified tail delay to obtain the tail delay target of each node, and when a node has multiple tail delay targets, select the smallest tail delay target as the final tail delay target.

It should be noted that the user needs to specify the end-to-end total tail delay target of a service in advance (for example, the delay at the 95th percentile should be within 30 milliseconds), and then set the tail delay of each path to the total tail delay. Target.

It should be understood that a node may appear in multiple paths. Therefore, it may correspond to multiple delay ratios, causing the node to calculate multiple tail delay targets. In this embodiment, the minimum among the multiple tail delay targets is as the final tail delay target of the node.

Step S104: Adjust the resource allocation of microservices and offline services according to the real-time tail delay and tail delay target.

Specifically, corresponding adjustment measures are made based on whether the real-time tail delay of the microservice meets its tail delay target, such as increasing the resource allocation of microservices, improving the resource allocation of offline services, deploying new offline services, and suspending the execution of offline services. wait.

Further, step S104 specifically includes:

1. If the real-time tail delay of the target microservice is greater than the tail delay target corresponding to the target microservice, reduce the deployed offline business resources by half.

2. If there is no real-time tail delay of the target microservice that is greater than the tail delay target corresponding to the target microservice, deploy a new offline business or increase the resources of the deployed offline business.

Further, in order to facilitate the allocation of resources for offline services, in this embodiment, the resources for offline services are divided into multiple levels, such as 10 levels, with 20% resource utilization being the lowest level, that is, the first level. The resource utilization rate of each gear is increased by 20% based on the previous gear. The 10th gear is the highest level, and the CPU resource utilization rate is 200%.

When it is necessary to reduce the resources of deployed offline services by half, the adjustment strategy is as follows: change the resource allocation of all deployed offline services to half of the original ones. Those that were previously in odd-numbered levels will be reduced by one level and then become themselves. Half, for example, if the resource allocation of an offline business is level 6, change it to level 3; if the resource allocation of an offline business is level 3, it will be reduced by one level to level 2, and finally to level 2. Half of itself, the first gear. Offline services that were originally at the lowest level (first level) need to be suspended. At the same time, the resource restrictions on the microservice are temporarily lifted to ensure that the microservice has sufficient resources to run.

When it is necessary to deploy a new offline service or increase the resources of an already deployed offline service, the adjustment strategy is as follows: First, try to randomly select an offline service from the deployed offline services and increase its resource allocation by one level. If all deployed offline services are at the highest level and their resource allocation cannot be further increased, you can select a new offline service to deploy and set its resource allocation to the lowest level.

The resource management method of this embodiment reduces the resources of the offline business before deploying the online business when the online business needs to be deployed after the offline business is deployed, and all the microservices of the online business are deployed in the next preset time period. Resources are predicted and pre-allocated to ensure the end-to-end response delay of online services. At the same time, the tail delay of all microservices is monitored. Combined with the preset tail delay target of each microservice, the time occupied by all microservices is measured. The resources and resources occupied by offline services are dynamically adjusted to ensure the end-to-end response delay of online services while improving resource utilization, ensuring that online services and offline services can run efficiently even when they are co-located.

Figure 2 is a schematic diagram of the functional modules of the resource management device according to the embodiment of the present invention. As shown in FIG. 2 , the resource management device 20 includes an online service deployment module 21 , a pre-allocation module 22 , a collection module 23 and an adjustment module 24 .

The online business deployment module 21 is used to reduce the resource allocation of the offline business and configure excess resources for each microservice of the online business in the current time period if an instruction to deploy the online business is received when the host has deployed the offline business;

The pre-allocation module 22 is used to predict the predicted request frequency of each microservice in the next preset time period based on the pre-trained prediction model, and reserve the resources of each microservice in the next preset time period based on the predicted request frequency. distribute;

The collection module 23 is used to collect the real-time tail delay of each microservice at preset intervals, and obtain the tail delay target generated by each microservice in the offline analysis stage;

The adjustment module 24 is used to adjust the resource allocation of microservices and offline services according to the real-time tail delay and the tail delay target.

Optionally, the pre-allocation module 22 predicts the predicted request frequency of each microservice in the next preset time period based on the pre-trained prediction model, and allocates the resources of each microservice in the next preset time period based on the predicted request frequency. Perform pre-allocation operations, specifically including: after deploying the online business, obtain the pre-built tail delay model and resource utilization model corresponding to each microservice with request frequency as the independent variable. The tail delay model is based on the time of each microservice. The tail delay is constructed under multiple preset request frequencies, and the resource utilization model is built based on the resource utilization of each microservice under multiple preset request frequencies; obtain the current request frequency of each microservice in the current time period; The current time period and current request frequency of each microservice are input to the pre-trained time series prediction model, and the predicted request frequency of the next preset time period of each microservice is predicted; according to the current request frequency, predicted request frequency and The tail delay model and resource utilization model of each microservice pre-allocate resources for the next preset time period for each microservice.

Optionally, the pre-allocation module 22 performs the operation of pre-allocating resources for the next preset time period for each micro-service based on the current request frequency, predicted request frequency, and the tail delay model and resource utilization model of each micro-service. Specifically, It includes: using the current request frequency, the number of copies of the microservice obtained in advance and the first preset error coefficient to calculate the first request frequency reference value according to the first preset rule; comparing the predicted request frequency with the first request frequency reference value relationship; if the predicted request frequency is greater than or equal to the first request frequency reference value, the resources occupied by the microservice will be expanded; if the predicted request frequency is less than the first request frequency reference value, the resources occupied by the microservice will be reduced. Capacity processing.

Optionally, the pre-allocation module 22 performs the operation of expanding the resources occupied by the microservices, which specifically includes: confirming the minimum request frequency corresponding to the tail delay target according to the tail delay target and the tail delay model; Calculate the second request frequency reference value according to the second preset rule based on the number of copies and the first preset error coefficient; compare the predicted request frequency and the second request frequency reference value; if the predicted request frequency is less than the second request frequency reference value , then vertically expand the resources of the microservice, and the resource utilization after vertical expansion is calculated according to the predicted request frequency, number of copies, resource utilization model and second preset error coefficient according to the third preset rule; if the predicted request frequency is greater than or equal to the second request frequency reference value, the resources of the microservice will be horizontally expanded and then adjusted vertically. The number of copies for horizontal expansion will be based on the fourth preset using the predicted request frequency, tail delay target and first preset error coefficient. Calculated by the rules, the vertically adjusted maximum resource utilization prediction request frequency, number of copies, resource utilization model and second preset error coefficient of each copy are calculated according to the third preset rule.

Optionally, the pre-allocation module 22 performs the operation of shrinking the resources occupied by the microservice, which specifically includes: using the predicted request frequency, the tail delay target and the first preset error coefficient to calculate the replica according to the fourth preset rule. Quantity reference value; compare the reference value of the number of copies and the number of copies; if the reference value of the number of copies is equal to the number of copies, vertically shrink the resources of the microservice. The resource utilization after vertical shrinkage will predict the request frequency, The number of copies, the resource utilization model and the second preset error coefficient are calculated according to the third preset rule; if the reference value of the number of copies is less than the number of copies, the resources of the microservice will be scaled down horizontally and then adjusted vertically. The number of copies after storage is a reference value for the number of copies. The vertically adjusted maximum resource utilization prediction request frequency, number of copies, resource utilization model and second preset error coefficient of each copy are calculated according to the third preset rule. .

Optionally, the collection module 23 is also used to perform the operation of generating the tail delay target of each microservice in the offline analysis stage, which specifically includes: in the offline analysis stage, obtaining the response delay and call dependency graph of each microservice, calling The dependency graph is generated based on the call link of each microservice. The call dependency graph includes multiple nodes, each node corresponding to a microservice. The average response time of each node in the call dependency graph is confirmed based on the response delay of each microservice. delay; starting from the node with in-degree 0, traverse all paths in the graph, and accumulate the average response delay of the nodes the path passes through, and then obtain the average response delay of each path; use the average response time of each node The delay is divided by the average response delay of the path where the node is located to obtain the delay ratio of the node; the delay ratio is multiplied by the pre-specified tail delay to obtain the tail delay target of each node, and when a node has multiple tail delay targets, select The smallest tail delay target is used as the final tail delay target.

Optionally, the adjustment module 24 performs the operation of adjusting the resource allocation of microservices and offline services according to the real-time tail delay and the tail delay target, specifically including: if the real-time tail delay of the target microservice is greater than the tail delay target corresponding to the target microservice, Then reduce the resources of the deployed offline business by half; if there is no real-time tail delay of the target microservice that is greater than the tail delay target corresponding to the target microservice, deploy a new offline business or increase the resources of the deployed offline business.

For other details of the technical solution for implementing each module in the resource management device of the above embodiment, please refer to the description of the resource management method in the above embodiment, and will not be described again here.

It should be noted that each embodiment in this specification is described in a progressive manner. Each embodiment focuses on its differences from other embodiments. The same and similar parts between the various embodiments are referred to each other. Can. As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For relevant details, please refer to the partial description of the method embodiment.

Please refer to FIG. 3 , which is a schematic structural diagram of a computer device according to an embodiment of the present invention. As shown in FIG. 3 , the computer device 30 includes a processor 31 and a memory 32 coupled to the processor 31 . The memory 32 stores program instructions. When the program instructions are executed by the processor 31 , the processor 31 executes any of the above. Resource management method steps described in the embodiment.

The processor 31 may also be called a resource (Central Processing Unit). The processor 31 may be an integrated circuit chip with signal processing capabilities. The processor 31 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. . A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

Refer to Figure 4, which is a schematic structural diagram of a storage medium according to an embodiment of the present invention. The storage medium of the embodiment of the present invention stores program instructions 41 that can implement the above resource management method, wherein the program instructions 41 can be stored in the above storage medium in the form of a software product, including several instructions to make a computer device ( It may be a personal computer, a server, a network device, etc.) or a processor (processor) that executes all or part of the steps of the method described in each embodiment of this application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program code. Or computer equipment such as computers, servers, mobile phones, tablets, etc.

In the several embodiments provided in this application, it should be understood that the disclosed computer equipment, apparatus and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated. to another system, or some features can be ignored, or not implemented. On the other hand, the coupling or direct coupling or communication connection between each other shown or discussed may be through some interfaces, and the indirect coupling or communication connection of the devices or units may be in electrical, mechanical or other forms.

In addition, each functional unit in various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware or software functional units. The above are only embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the description and drawings of this application, or directly or indirectly applied in other related technical fields, All are similarly included in the patent protection scope of this application.

Claims (10)

1. A resource management method, characterized in that it includes: When the host has deployed offline services, if it receives an instruction to deploy online services, it will reduce the resource allocation for offline services and allocate excess resources for each microservice of online services in the current time period; Predict the predicted request frequency of each microservice in the next preset time period based on the pre-trained prediction model, and pre-allocate the resources of each microservice in the next preset time period based on the predicted request frequency; Collect the real-time tail latency of each microservice at preset intervals, and obtain the tail latency target generated by each microservice during the offline analysis phase; Adjust the resource allocation of the microservice and the offline business according to the real-time tail delay and the tail delay target. 2. The resource management method according to claim 1, characterized in that the predicted request frequency of each microservice in the next preset time period is predicted based on a pre-trained prediction model, and the resources of each microservice in the next preset time period are pre-allocated according to the predicted request frequency, comprising: After deploying the online business, obtain the pre-built tail delay model and resource utilization model corresponding to each microservice with request frequency as the independent variable. The tail delay model is based on each microservice under multiple preset request frequencies. The tail delay is constructed, and the resource utilization model is constructed based on the resource utilization of each microservice under multiple preset request frequencies; Get the current request frequency of each microservice in the current time period; Input the current time period and the current request frequency of each microservice into a pre-trained time series prediction model, and predict the predicted request frequency of the next preset time period for each microservice; Pre-allocate resources for the next preset time period to each microservice based on the current request frequency, the predicted request frequency, the tail delay model and the resource utilization model of each microservice. 3. The resource management method according to claim 2, characterized in that the tail delay model and the resource utilization model based on the current request frequency, the predicted request frequency and each microservice are: Each microservice is pre-allocated resources for the next preset time period, including: Calculate the first request frequency reference value according to the first preset rule using the current request frequency, the number of copies of the microservice obtained in advance and the first preset error coefficient; Compare the magnitude relationship between the predicted request frequency and the first request frequency reference value; If the predicted request frequency is greater than or equal to the first request frequency reference value, expand the resources occupied by the microservice; If the predicted request frequency is less than the first request frequency reference value, the resources occupied by the microservice are scaled down. 4. The resource management method according to claim 3, wherein the expansion of the resources occupied by the microservices includes: Confirm the minimum request frequency corresponding to the tail delay target according to the tail delay target and the tail delay model; Calculate a second request frequency reference value according to the minimum request frequency, the number of copies of the microservice, and the first preset error coefficient according to a second preset rule; Compare the size of the predicted request frequency and the second request frequency reference value; If the predicted request frequency is less than the second request frequency reference value, the resources of the microservice are vertically expanded, and the resource utilization after the vertical expansion is calculated according to a third preset rule using the predicted request frequency, the number of replicas, the resource utilization model and the second preset error coefficient; If the predicted request frequency is greater than or equal to the second request frequency reference value, the resources of the microservice are horizontally expanded and then vertically adjusted. The number of copies of the horizontal expansion is determined by the predicted request frequency and the total number of copies. The tail delay target and the first preset error coefficient are calculated according to the fourth preset rule. The vertically adjusted maximum resource utilization of each copy utilizes the predicted request frequency, the number of copies, the resource utilization The rate model and the second preset error coefficient are calculated according to the third preset rule. 5. The resource management method according to claim 4, characterized in that the scaling down of resources occupied by the microservices comprises: Calculate the copy number reference value according to a fourth preset rule using the prediction request frequency, the tail delay target and the first preset error coefficient; Compare the reference value of the number of copies with the size of the number of copies; If the reference value of the number of replicas is equal to the number of replicas, the resources of the microservice are vertically scaled down, and the resource utilization after the vertical scaling is determined by using the predicted request frequency, the number of replicas, and the resources. The utilization model and the second preset error coefficient are calculated according to the third preset rule; If the reference value of the number of copies is less than the number of copies, the resources of the microservice are horizontally reduced and then adjusted vertically. The number of copies after the horizontal reduction is the reference value of the number of copies, and the number of copies after the vertical adjustment is The maximum resource utilization of each copy is calculated according to a third preset rule using the predicted request frequency, the number of copies, the resource utilization model and the second preset error coefficient. 6. The resource management method according to claim 1, characterized in that the tail delay target of each microservice is generated in the offline analysis stage, specifically including: In the offline analysis stage, the response delay and call dependency graph of each microservice are obtained. The call dependency graph is generated based on the call link of each microservice. The call dependency graph includes multiple nodes, each node corresponds to one microservices; Confirm the average response delay of each node in the call dependency graph based on the response delay of each microservice; Starting from the node with in-degree 0, traverse all paths in the graph, accumulate the average response delays of the nodes passed by the path, and then get the average response delay of each path; The average response delay of each node is divided by the average response delay of the path where the node is located to obtain the delay ratio of the node; The tail delay target of each node is obtained by multiplying the delay ratio by the pre-specified tail delay, and when a node has multiple tail delay targets, the smallest tail delay target is selected as the final tail delay target. 7. The resource management method according to claim 1, wherein adjusting the resource allocation of the microservice and the offline business according to the real-time tail delay and the tail delay target includes: If there is a real-time tail delay of the target microservice that is greater than the tail delay target corresponding to the target microservice, reduce the deployed offline business resources by half; If there is no real-time tail delay of the target microservice that is greater than the tail delay target corresponding to the target microservice, then deploy a new offline service or increase the resources of the deployed offline service. 8. A resource management device, characterized in that it includes: The online business deployment module is used to reduce the resource allocation of the offline business and allocate excess resources for each microservice of the online business in the current time period if the host has deployed an offline business and receives an instruction to deploy an online business; The pre-allocation module is used to predict the predicted request frequency of each microservice in the next preset time period based on the pre-trained prediction model, and perform resource allocation for each microservice in the next preset time period based on the predicted request frequency. preallocation; The collection module is used to collect the real-time tail delay of each microservice at preset intervals, and obtain the tail delay target generated by each microservice in the offline analysis phase; An adjustment module, configured to adjust the resource allocation of the microservice and the offline business according to the real-time tail delay and the tail delay target. 9. A computer device, characterized in that the computer device includes a processor and a memory coupled to the processor, and program instructions are stored in the memory. When the program instructions are executed by the processor, The processor is caused to execute the steps of the resource management method according to any one of claims 1-7. 10. A storage medium, characterized by storing program instructions capable of implementing the resource management method according to any one of claims 1-7.