Database parameter tuning method based on parameter type customization optimization

Abstract

The invention discloses a database parameter adjusting and optimizing method based on parameter type customization optimization, and belongs to the field of database configuration optimizing.The method comprises the steps that a database parameter space is divided into a continuous type parameter subspace and a classified type parameter subspace according to parameter types, and the continuous type parameter subspace and the classified type parameter subspace are projected to a low-dimensional space; a GP model and an SMAC model are used as proxy models in the low-dimensional continuous parametric subspace and the classification parametric subspace respectively to search continuous parametric configuration and classification parametric configuration; interaction between the GP model and the SMAC model is realized through iterative optimization and a context-based communication mechanism; and mapping a parameter configuration result back to a database parameter space for performance evaluation. According to the method, the database parameter tuning performance and efficiency can be effectively improved, and a user is allowed to search database parameter configuration with higher performance in a shorter time.

Description

Technical Field

[0001] The present invention belongs to the field of database configuration tuning, and more specifically, relates to a database parameter tuning method based on parameter type customization and optimization. Background Art

[0002] Modern databases usually have hundreds or thousands of parameters that need to be set. Only when these parameters are set to appropriate values ​​can the database achieve higher throughput or lower latency. Automated database parameter tuning uses machine learning technology to automatically search for the best performance parameters in the parameter configuration space, thereby reducing the labor costs of enterprises on database administrators. Parameters in the database can be roughly divided into categorical parameters and continuous parameters. There is no obvious order relationship between the former parameters and the corresponding performance, while there is an obvious order relationship between the values ​​of the latter parameters and the level of performance. At the same time, slight changes in continuous parameters often do not cause significant changes in performance, while changes in the values ​​of categorical parameters are more likely to cause significant changes in performance. Existing tuning methods often use the same proxy model to optimize categorical and continuous parameters, resulting in poor performance. At the same time, existing methods often take a lot of time to search for relatively satisfactory parameters when searching for the optimal parameter configuration, and the tuning efficiency is not high. Summary of the invention

[0003] In view of the above defects or improvement needs of the prior art, the present invention provides a database parameter tuning method based on parameter type customization and optimization, thereby solving the problems of poor tuning performance and low tuning efficiency of the existing database parameter tuning methods.

[0004] To achieve the above object, according to a first aspect of the present invention, a method for tuning database parameters based on parameter type customization optimization is provided, comprising:

[0005] S1, decomposing the database parameter space into a continuous parameter subspace and a categorical parameter subspace according to the parameter value type, and mapping the continuous parameter subspace and the categorical parameter subspace to a low-dimensional proxy space, respectively, to obtain a continuous low-dimensional proxy space and a categorical low-dimensional proxy space;

[0006] S2, using the GP model to search for the configuration of continuous parameters in the continuous low-dimensional proxy space, using the SMAC model to search for the configuration of categorical parameters in the categorical low-dimensional proxy space, and respectively mapping the configuration of the continuous parameters and the configuration of the categorical parameters back to the database parameter space for performance testing, and storing the test results in the tuning history until the preset number of tuning rounds is reached; wherein the GP model and the SMAC model are searched alternately, and the optimal parameter configuration searched by each model in its own space is passed to the other model as a context feature;

[0007] S3: Use the categorical parameters and continuous parameters with the best performance in the tuning history as the optimal parameter configuration for the database.

[0008] According to a second aspect of the present invention, a database parameter deployment method is provided, comprising:

[0009] Using the database parameter tuning method described in the first aspect to obtain the optimal parameter configuration of the database;

[0010] If the adjusted parameters are all dynamic parameters, the corresponding parameter values ​​are written to the database through SQL commands. If the adjusted parameters also include static parameters, the corresponding parameter values ​​are written to the database configuration file, and the database is restarted to complete the parameter deployment.

[0011] According to a third aspect of the present invention, there is provided an electronic device, comprising: a computer-readable storage medium and a processor;

[0012] The computer-readable storage medium is used to store executable instructions;

[0013] The processor is used to read the executable instructions stored in the computer-readable storage medium to execute the database parameter tuning method as described in the first aspect or the database parameter deployment method as described in the second aspect.

[0014] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the database parameter tuning method as described in the first aspect or the database parameter deployment method as described in the second aspect.

[0015] According to a fifth aspect of the present invention, a computer program product is provided, comprising a computer program or instructions, wherein the computer program or instructions are executed by a processor as the database parameter tuning method described in the first aspect or the database parameter deployment method described in the second aspect.

[0016] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0017] The method provided by the present invention divides the database parameter space into a continuous parameter subspace and a categorical parameter subspace according to the parameter type when using the machine learning technology to tune the database parameters, and projects the subspace into a low-dimensional space, and uses a GP model (Gaussian process model) and a SMAC model (sequence model-based configuration algorithm model) as proxy models in the low-dimensional continuous parameter subspace and the categorical parameter subspace to search for continuous parameter configuration and categorical parameter configuration respectively; through iterative optimization and a communication mechanism based on context, the interaction between the GP model and the SMAC model is realized; and the parameter configuration result is mapped back to the database parameter space for performance evaluation; through parameter space decomposition, and customized optimization is performed for the continuous parameter space and the discrete parameter space, the ability of the GP model to model continuous relationships and the advantage of the SMAC model in processing categorical parameters can be fully utilized, so as to obtain a better tuning effect; through iterative optimization and a communication mechanism based on context features, the two models are coordinated to jointly capture the dependency relationship between parameters, so as to search for a better parameter configuration; through the parameter dimension mapping technology, the high-dimensional space is mapped to the low-dimensional proxy space, and the model search space can be reduced while ensuring that the tuning performance is not lost, thereby improving the search efficiency. Using batch Bayesian optimization technology, a batch acquisition function is used to sample multiple configurations at a time, and parameter performance is evaluated in parallel to further improve the efficiency of the tuning method.

[0018] As a further preference, the method provided by the present invention adopts a local penalty-based acquisition function for batch sampling of parameter configurations, and the local penalizer adopted by the method will give a lower weight to the configuration close to the expected optimal point; this behavior ensures that the sampled configurations are diverse, so that comprehensive information can be collected efficiently. Specifically, batch configuration sampling involves alternating maximization steps and penalty steps. In the maximization step, the configuration that has not appeared in the tuning history is obtained by maximizing the penalized acquisition function. In the penalty step, the penalized acquisition function is updated to include the selected batch configuration information. This process is repeated until the target batch size is reached; in addition, the method provided by the present invention adopts a parallel parameter evaluation strategy, and distributes a batch of sampled parameter configurations to different cloned database instances through the client, each database instance is responsible for the performance evaluation of a parameter configuration, and all cloned DBMS instances replay the same workload during the evaluation process, using a synchronous parallel evaluation mechanism, the client waits for all workloads to be replayed and receives performance data, saves this batch of observations in the tuning history, and uses it to update the tuning model, which can further improve the tuning efficiency and shorten the tuning time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 One of the flow diagrams of the parameter type customization and optimization database parameter tuning method provided in an embodiment of the present invention;

[0020] Figure 2 The second flowchart of the database parameter tuning method based on parameter type customization and optimization provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0022] Existing methods use a unified optimizer (such as GP or DDPG) to optimize both categorical parameters and discrete parameters without distinction, resulting in suboptimal tuning performance. At the same time, the tuning process of existing methods often requires more tuning rounds and consumes more tuning time.

[0023] Based on this, the present invention provides a database parameter tuning method based on parameter customization optimization, comprising:

[0024] S1, decomposing the database parameter space into a continuous parameter subspace and a categorical parameter subspace according to the parameter value type, and mapping the continuous parameter subspace and the categorical parameter subspace to a low-dimensional proxy space, respectively, to obtain a continuous low-dimensional proxy space and a categorical low-dimensional proxy space;

[0025] S2, using the GP model to search for the configuration of continuous parameters in the continuous low-dimensional proxy space, using the SMAC model to search for the configuration of categorical parameters in the categorical low-dimensional proxy space, and respectively mapping the configuration of the continuous parameters and the configuration of the categorical parameters back to the database parameter space for performance testing, and storing the test results in the tuning history until the preset number of tuning rounds is reached; wherein the GP model and the SMAC model are searched alternately, and the optimal parameter configuration searched by each model in its own space is passed to the other model as a context feature;

[0026] S3: Use the categorical parameters and continuous parameters with the best performance in the tuning history as the optimal parameter configuration for the database.

[0027] Specifically, Figure 1 As shown, the method provided by the embodiment of the present invention includes: database parameter space decomposition, multi-agent model collaborative optimization, and parameter dimension mapping;

[0028] The parameter space decomposition includes:

[0029] According to the parameter value type, the parameter space is decomposed into a continuous parameter subspace and a categorical parameter subspace, and the two subspaces are orthogonal to each other.

[0030] In the spatial decomposition, the value type and meaning of each parameter will be checked. If the value type is integer or floating point, and the meanings corresponding to the value size have an obvious order relationship, then this parameter will be added to the continuous parameter set; if the parameter type is an enumeration type, or the parameter values ​​have no order relationship, and each value only corresponds to one option of database configuration, then the parameter will be added to the categorical parameter set. Define the continuous parameter configuration space as the Cartesian product of all possible continuous parameter value sets, and similarly define the categorical parameter subspace as the Cartesian product of all categorical parameter value sets.

[0031] The multi-agent model collaborative optimization includes:

[0032] In the classification subspace, the Sequential Model-Based Configuration Algorithm Model (SMAC) is used to search for the optimal values ​​of the classified parameters, while in the continuous space, the Gaussian Process Model (GP) is used to search for the optimal values ​​of the continuous parameters. The GP and SMAC models collaborate in an iterative optimization manner to jointly explore the entire parameter space. At the same time, the optimal parameter configuration searched by each model in its respective subspace will be passed to another model as a context feature, realizing context-based model communication to capture the dependencies between different types of parameters.

[0033] In order to enable the proxy model to support context features as input to obtain the global optimal parameter configuration, the proxy model (GP model and SMAC model) needs to be modified.

[0034] The GP model is characterized by a prior mean function μ(·) and a variance function σ2(·) defined as:

[0035]

[0036] where Φ is the continuous parameter configuration θ in the adjustment history l is a set of K, where K is the covariance matrix, and its (i,j) entry is k(θ con ,θ con ′), k(θ con ,θ con ′) is the kernel function in GP, ​​which is used to measure the similarity between continuous parameter configurations. In this embodiment of the present invention, a new kernel k((θ con ,F),(θ con ′ ,F′))=k Θ (θ con ,θ con ′)+k C(F,F ′ ) as the kernel function of GP, where k Θ is the Martin kernel, used to measure the continuous parameter configuration θ con and θ con ′. And k C is a kernel function used to measure the similarity between context features. k C The Hamming kernel is specified to measure the similarity between discrete context features or the linear kernel is specified to measure the similarity between continuous context features. It can be understood that the context features encapsulated by the categorical parameter configuration are discrete context features, and the context features encapsulated by the continuous parameter configuration are continuous context features.

[0037] In order to accurately measure the similarity between different context features F input to SMAC, for the SMAC model, the configuration features (i.e., the historical configuration of the classification parameters) and the context features are combined into a single connection input for the proxy model to use. This allows the random forest to flexibly capture changes in the continuous adjustment configuration.

[0038] The history of each tuning is recorded in the file, including the parameter configuration of each tuning step (continuous parameters and categorical parameters), the performance indicators obtained for each set of parameters, and the context features used by the model in the tuning step.

[0039] The parameter dimension mapping includes:

[0040] The high-dimensional parameter space is mapped to the low-dimensional proxy space using a random mapping matrix. The above optimization algorithm will run in the low-dimensional proxy space and map the parameter configuration explored in the low-dimensional proxy space back to the high-dimensional original space to correctly obtain the performance of the parameters.

[0041] In the parameter dimension mapping process, the HeSBO algorithm is used to define the Project_Down() interface and the Project_Up() interface, which are used to map the original space into a low-dimensional proxy space and map the points in the low-dimensional proxy space back to the high-dimensional original space, respectively.

[0042] In the Project_Down() interface, we first initialize a random mapping matrix S, each column of which has only one non-zero value (-1 or 1). Each parameter configuration θ in the original space will be It can be understood that the size of S is the number of input parameters × the dimension of the proxy space to be mapped, such as Figure 1 In the Project_Down() interface shown, the size of its random mapping matrix S is 4×2.

[0043] In the Project_Up interface, based on the hash function defined in the HeSBO algorithm, it can be done by something like The high-dimensional mapping is realized in this way. However, since the range of the parameters (continuous parameters and categorical parameters) mapped back (i.e. mapped to high dimensions) is [-1, 1], it is necessary to further map them back to the original parameter range.

[0044] For continuous parameters, in order to map back to the continuous parameters θ in high-dimensional space con `∈[-1,1] maps back to the original parameters Use the following formula:

[0045]

[0046] For categorical parameters, [-1,1] is evenly divided into intervals of the corresponding number according to the number of categories of the original parameter (that is, the number of options for the value), and the final parameter value is determined according to the interval in which the mapped value is located. For example, for a categorical original parameter, if its value category is 0 or 1, that is, the number of value categories is 2, then [-1,1] is divided into [-1,0] and [0,1], corresponding to the values ​​0 and 1, respectively. If the mapped value of the categorical parameter after being mapped back to the high-dimensional space is in [-1,0], the value is 0, otherwise it is 1.

[0047] Before each proxy model tuning interval begins, the optimal value (current optimal parameter configuration) currently searched by another model is first applied to the database. Then switch to another model to continue optimizing another subspace. In each model tuning interval, the model will first obtain the function sampling parameter configuration, and map it back to the high-dimensional original space to form the real parameter value, which is applied to the database to implement parameter performance evaluation, and the parameters and their performance values ​​are used to update the model.

[0048] In the process of implementing context-based model communication, when each proxy model (GP or SMAC) finishes the current tuning interval, it can obtain the performance test results of each parameter configuration, and encapsulate the best performance parameter configuration currently searched into a context feature. The context feature records the latest tuning decision of the current model and will be passed to another tuning model. The other tuning model uses the context feature as input to train the proxy model to capture the correspondence between parameter values, context features, and parameter performance.

[0049] As a further preferred embodiment of the present invention, when each model searches for parameters in its respective subspace, batch Bayesian optimization technology is used, an acquisition function based on local penalty is adopted, multiple parameter configurations are sampled at a time, and the configurations are broadcasted to multiple test environments for parallel parameter performance testing, and performance indicators are collected to update the model.

[0050] That is, in the process of batch sampling and parallel parameter evaluation, a batch acquisition function based on local penalty is used. In the process of parameter configuration sampling, this method imposes a penalty on the configuration close to the expected optimal point, that is, multiplying it by a lower penalty factor, and iterates the maximization step and the penalty step until the expected batch number of parameter configurations are sampled.

[0051] After sampling a batch of configurations, a distributed parallel evaluation is performed. The client first clones the target database management system (DBMS) instance to provide an evaluation environment. The number of cloned DBMS instances is equal to the batch size of the sampled configurations. At the same time, a batch of sampled parameter configurations is distributed to different database instances, and each database instance is responsible for the performance evaluation of one parameter configuration. All cloned DBMS instances replay the same workload during the evaluation process. Using a synchronous parallel evaluation mechanism, the client waits for all workloads to be replayed and receives performance data, saves this batch of observations to the tuning history, and uses it to update the tuning model.

[0052] It is understandable that the process of client environment preparation and preprocessing is also included in the parameter space decomposition: using the client to interact with the user, accepting the optimization target (tuning target) and tuning load related information required by the user. At the same time, the client is used to manage the database instance, perform load replay and collect database performance indicators to pass to the tuning model. The defined client needs to complete tasks such as parameter allocation in parallel tuning mode, database instance cloning, and synchronous parallel parameter verification.

[0053] That is, the method provided by the present invention comprises:

[0054] 1. Client environment preparation and preprocessing

[0055] 1.1 The client server serves as the central controller of the tuning method and is responsible for interacting with users and managing database instances, load management tools, etc. It first reads database-related information, including database startup location, parameter configuration file location, parameter list, and parameter detailed value range.

[0056] 1.2 The client also accepts user tuning requests, including target tuning load, tuning indicators (transactions per second TPS or latency, etc.), and initializes the parameter evaluation environment and related configurations according to the user's request, and replicates the database instance according to the tuning parallelism set by the user.

[0057] 2. Search space optimization

[0058] 2.1 According to the order relationship between parameter type and parameter value and performance, determine whether the parameter is a continuous parameter or a categorical parameter, and the continuous parameters constitute a continuous parameter subspace, while the categorical parameters constitute a categorical parameter subspace.

[0059] 2.2 Using the random mapping matrix, the high-dimensional parameter space is mapped to the low-dimensional proxy space. The above optimization algorithm will run in the low-dimensional proxy space, and the parameter configuration explored in the low-dimensional proxy space will be mapped back to the high-dimensional original space, so as to correctly obtain the performance of the parameters.

[0060] 2.3 Initialize the GP-based Bayesian optimization model in the low-dimensional continuous parameter subspace, and initialize the SMAC-based Bayesian optimization model in the low-dimensional classification parameter subspace, including defining the corresponding search space and setting the parameters of the GP and SMAC models.

[0061] 3. Communication mechanism based on contextual features

[0062] The GP model in the low-dimensional continuous space and the SMAC model in the low-dimensional classification space operate through iterative optimization (see step 4 for the specific process), that is, GP and SMAC alternately search the low-dimensional continuous space and classification space, each model runs a fixed number of rounds (tuning interval), and when the tuning interval of a model ends, the optimal parameter value is passed as the context feature to the next model as input.

[0063] 4. Collaborative optimization process

[0064] 4.1 In two low-dimensional proxy spaces, 10 configurations are sampled using Latin hypercube respectively, and the performance of these 10 parameter configurations is tested through load replay, and the observation results (parameter configuration, context features, and corresponding parameter performance) are stored in the tuning history.

[0065] 4.2 Two proxy models GP and SMAC are trained using initial observation data sampled from continuous space and discrete space, respectively.

[0066] 4.3 The optimal classification parameters of the SMAC model in the initialization phase Mapping back to the original parameter range via the Project_Up interface And apply it to the database for performance testing. Encapsulated into context feature F s , which are context features from the SMAC model.

[0067] 4.4 Start the tuning interval of the GP proxy model. Including: 1) First use the local penalty-based acquisition function to sample a batch of parameter configurations in the classification space. 2) Use the Project_Up interface to map all configurations back to the original parameter range in the original space. 3) Execute load replay on different test instances in parallel to collect the performance corresponding to this batch of parameter configurations. 4) Store the observation results of this round of tests (parameter configuration, parameter performance, contextual features) in the tuning history. 5) Use the data in the tuning history to update the proxy model GP. 6) Repeat steps 1)-5) until the tuning interval round is reached, and return the optimal parameter configuration of the continuous subspace. And the corresponding parameter configuration mapped back to the original space

[0068] 4.5 Apply the searched optimal parameter configuration to the database and Encapsulated into context feature F g .

[0069] 4.6 Start the tuning interval of the SMAC proxy model. The SMAC tuning interval is consistent with the steps in 4-4, but uses SMAC as a proxy model to search the classification parameter subspace and uses the context features from GP to update the model.

[0070] 4.7 Repeat the iterative tuning intervals of GP and SMAC until the preset tuning resources (the preset time or the preset number of tuning rounds) are exhausted.

[0071] It is understandable that the above process takes the SMAC model as an example to start the first search. The first search can also be started from the GP model, such as Figure 2 shown.

[0072] 5. The collection of categorical parameters and continuous parameters corresponding to the best round of parameter performance in the tuning history will be returned in the end As the final parameter configuration.

[0073] An embodiment of the present invention provides a database parameter deployment method, including:

[0074] Using the database parameter tuning method described in any of the above embodiments to obtain the optimal parameter configuration of the database;

[0075] If the adjusted parameters are all dynamic parameters, the corresponding parameter values ​​are written to the database through SQL commands. If the adjusted parameters also include static parameters, the corresponding parameter values ​​are written to the database configuration file, and the database is restarted to complete the parameter deployment.

[0076] An embodiment of the present invention provides an electronic device, including: a computer-readable storage medium and a processor;

[0077] The computer-readable storage medium is used to store executable instructions;

[0078] The processor is used to read the executable instructions stored in the computer-readable storage medium, and execute the database parameter tuning method described in any of the above embodiments or the database parameter deployment method described in any of the above embodiments.

[0079] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the database parameter tuning method described in any of the above embodiments or the database parameter deployment method described in any of the above embodiments.

[0080] An embodiment of the present invention provides a computer program product, including a computer program or an instruction, wherein the computer program or the instruction is executed by a processor to perform the database parameter tuning method described in any of the above embodiments or the database parameter deployment method described in any of the above embodiments.

[0081] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A database parameter tuning method based on parameter type customization and optimization, characterized in that: include: S1, decomposing the database parameter space into a continuous parameter subspace and a categorical parameter subspace according to the parameter value type, and mapping the continuous parameter subspace and the categorical parameter subspace to a low-dimensional proxy space, respectively, to obtain a continuous low-dimensional proxy space and a categorical low-dimensional proxy space; S2, using the GP model to search for the configuration of continuous parameters in the continuous low-dimensional proxy space, using the SMAC model to search for the configuration of categorical parameters in the categorical low-dimensional proxy space, and respectively mapping the configuration of the continuous parameters and the configuration of the categorical parameters back to the database parameter space for performance testing, and storing the test results in the tuning history until the preset number of tuning rounds is reached; wherein the GP model and the SMAC model are searched alternately, and the optimal parameter configuration searched by each model in its own space is passed to the other model as a context feature; S3: Use the categorical parameters and continuous parameters with the best performance in the tuning history as the optimal parameter configuration for the database.

2. The method according to claim 1, characterized in that The kernel function of the GP model is: k((θ con ,F),(θ con ′ ,F′))=k Θ (i con ,i con ′)+k C (F,F′) Among them, k Θ is the Martin kernel, used to measure the continuous parameter configuration θ con and θ con The distance between ′, k C is a Hamming kernel or a linear kernel, which is used to measure the similarity between the context features F′ and F; In the SMAC model, the historical configuration of classification parameters and context features are combined into one connection input.

3. The method according to claim 1, characterized in that In step S1, the continuous parameter subspace and the categorical parameter subspace are respectively mapped to the low-dimensional proxy space through the Project_Down() interface; In the Project_Down() interface, through Map θ to a low-dimensional proxy space, where θ is a continuous parameter in a continuous parameter subspace or a categorical parameter in a categorical parameter subspace. is a continuous parameter in a continuous low-dimensional proxy space or a categorical parameter in a categorical low-dimensional proxy space, S is a random mapping matrix, each column of which has only one non-zero value, and the non-zero value is 1 or -1; In step S2, the optimal configurations of the continuous parameters and the categorical parameters are respectively mapped back to the database parameter space through the Project_Up() interface; In the Project_Up() interface, through Will Mapped back to the database parameter space; where, for the continuous parameter θ mapped back to the database parameter space con `∈[-1,1], through θ con `Mapping back to the original parameter range yields θ con , They are θ con The maximum and minimum values ​​of the original parameter range, is θ con `Minimum value; for the classification parameter θ mapped back to the database parameter space cat `, according to θ cat `The original value category number, [-1,1] is evenly divided into the corresponding number of intervals, and according to θ cat The interval where the value of ` is located determines the corresponding original value category, thereby determining θ cat The original value of `θ cat .

4. The method according to claim 1, characterized in that In step S2, a batch of parameter configurations is sampled in a continuous low-dimensional proxy space or a categorical low-dimensional proxy space using a local penalty-based acquisition function, and the batches are respectively distributed to different clone database instances for parallel performance evaluation.

5. A database parameter deployment method, characterized in that: include: Adopting the database parameter tuning method according to any one of claims 1 to 4 to obtain the optimal parameter configuration of the database; If the adjusted parameters are all dynamic parameters, the corresponding parameter values ​​are written to the database through SQL commands. If the adjusted parameters also include static parameters, the corresponding parameter values ​​are written to the database configuration file, and the database is restarted to complete the parameter deployment.

6. An electronic device, characterized in that: include: A computer readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read the executable instructions stored in the computer-readable storage medium, and execute the database parameter tuning method according to any one of claims 1 to 4 or the database parameter deployment method according to claim 5.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the database parameter tuning method according to any one of claims 1 to 4 or the database parameter deployment method according to claim 5.

8. A computer program product comprising a computer program or instructions, characterized in that The computer program or instruction is executed by a processor as the database parameter tuning method according to any one of claims 1 to 4 or the database parameter deployment method according to claim 5.

Images