CN115562674A - A load-aware software configuration parameter adjustment method - Google Patents

Abstract

The invention discloses a load-aware software configuration parameter adjustment method, aiming to solve the problem that the current software configuration parameters cannot be automatically adjusted to deal with load-sensitive configuration faults. The technical solution is: build a configuration parameter adjustment system composed of a configuration-related branch interaction analysis module and a dynamic monitoring module; the configuration-related branch interaction analysis module locates the configuration-related branch interaction code segment in the software source code based on the static program analysis method, and realizes code insertion ; When the software is running, the dynamic monitoring module automatically extracts the constraint information that the configuration parameters need to meet under the influence of the relevant load, and automatically adjusts the value of the configuration parameters according to the configuration constraint information when the software load changes, so as to avoid load-sensitive configuration failures. The invention can effectively obtain load-sensitive configuration constraint information, help users automatically adjust configuration parameter values to meet changing load requirements, and avoid load-sensitive configuration failures.

Description

A load-aware software configuration parameter adjustment method

technical field

The invention relates to the fields of configuration constraint extraction and configuration failure prevention in large-scale software, and in particular relates to a load-aware software configuration parameter adjustment method.

Background technique

With the continuous progress of society, software systems have been widely used in various fields, playing a pivotal role in modern society and playing an important role. As an important part of software, configuration plays an important role in improving the diversity, customizability and adaptability of software functions. Software users can use configuration to select different implementations of libraries, policies, rules, etc. to customize different functions, and effectively control the resource usage of the software to adapt to different environments and workloads and meet different needs of users. At present, in order to adapt to changes in the load environment and application requirements, large-scale software systems are developing towards high configurability, thereby improving the reliability and availability of software services.

However, with the continuous growth of software scale and the increasingly complex operating environment, software users experience the convenience of configuration while also experiencing the impact of corresponding configuration failures, resulting in frequent failures of software system services, which has gradually caused the industry widespread attention. Among them, load-sensitive configuration faults, that is, configuration faults caused by legitimate configuration values of software under specific loads, have accounted for a large proportion of software configuration faults. Zuoning Yin et al. found in a survey on the status quo of configuration failures of four commonly used open source basic software and one commercial basic software that the values of relevant configuration parameters in 46.3% to 61.9% of cases all meet the corresponding legal value range constraints, but It can still cause configuration failures in the software system. Tianyin Xu et al. also found through research that among three open source software and one commercial software, up to 53.3% of configuration failures are caused by improper configuration parameters related to the runtime environment and load.

At present, the existing configuration parameter adjustment technology focuses on configuration-oriented performance tuning, based on "Understanding and auto-adjusting performance-sensitive configurations (understanding and automatically adjusting performance-sensitive configuration parameters)" published by Shu Wang et al. at ASPLOS2018 and Miguel Velez et al. published "White-box analysis over machine learning: Modeling performance of configurable systems (research on configurable software performance model construction based on white-box analysis and machine learning)" published in ICSE2021, mainly by searching or constructing description configuration parameter values The method of relationship model with software performance, quickly finds the combination of configuration parameters that make the software performance optimal in the huge value space of software configuration parameters, so as to adjust the corresponding configuration parameters to improve software performance. However, the above-mentioned methods all require the legal value range of the configuration parameter to be provided in advance, and the existing methods for obtaining the value range of the configuration parameter mainly include two types of work. The first type of work is represented by "Do Not Blame Users for Misconfigurations" published by Tianyin Xu et al. Users set the value of configuration items within the legal range to prevent configuration failures caused by human factors. This kind of work is mainly implemented based on static analysis, and does not consider the impact of software load on configuration constraints, so it may cause load-sensitive configuration failures in the process of software configuration parameter adjustment. The second type of work, represented by "Testing Configuration Changes in Context to Prevent Production Failures" published by Xudong Sun et al. Detect incorrect configuration modifications early to avoid configuration failures in production environments. However, such work can only simulate a simple running context at present, and cannot truly simulate the diverse loads and complex contexts faced by a software system in a production environment, and the ability to prevent load-sensitive configuration failures is also limited.

To sum up, how to automatically adjust load-sensitive software configuration parameters to help users effectively deal with load-sensitive configuration failures and improve software reliability is a hot topic being discussed by those skilled in the art.

Contents of the invention

The technical problem to be solved by the present invention is to provide a load-aware software configuration parameter adjustment method, which solves the problem that the existing software configuration parameter adjustment method does not consider the influence of load on configuration parameter constraints, which leads to the inability to automatically adjust software configuration parameters and cannot cope with Load-sensitive configuration failures lead to problems with low software reliability.

The present invention performs code insertion by configuring relevant branch interaction code segments (i.e., configuring relevant program variables (referred to as configuration variables) and other program variables in branch statements to interact with other program variables to affect the behavior of the software when it is running) into the software source code, In this way, the related state of the program can be effectively monitored when the software is running, the load-sensitive configuration constraints can be extracted, and the configuration values can be adjusted according to the load-sensitive configuration constraints to prevent load-sensitive configuration failures and restore the normal operation of the software in time.

In order to solve the above-mentioned technical problems, the technical scheme of the present invention is: first build the configuration parameter adjustment system that is made of configuration-related branch interaction analysis module and dynamic monitoring module; Then, configuration-related branch interaction analysis module reads in software source code and <configuration parameter name , configuration variable name>binary pair file, based on the static program analysis method to locate the configuration-related branch interaction code segment in the software source code, to implement code instrumentation; when the software is running, the dynamic monitoring module continuously reads the instrumentation code at runtime The content of the output file automatically extracts the constraint information that the configuration parameters need to meet under the influence of the relevant load. When the software load changes, the value of the configuration parameter is automatically adjusted according to the configuration constraint information, thereby avoiding load-sensitive configuration failures.

The present invention comprises the following steps:

The first step is to build a configuration parameter adjustment system, which is composed of a configuration-related branch interaction analysis module and a dynamic monitoring module. The configuration-related branch interaction analysis module reads the software source code F and the configuration information file (stored with <configuration parameter name, configuration variable name> binary pair list) from the file system, identifies and locates the configuration-related branch interaction in the software source code F, and performs Code insertion, output the software source code F' after insertion to the file system; then the user compiles and links F' to obtain executable software F"; the dynamic monitoring module monitors the source code of F'' in the file system when it is running Output, extract and output the configuration constraint information under the influence of the current load, and automatically adjust the configuration parameter value according to the relevant configuration constraint information when the F" load changes and the current configuration value does not meet the configuration constraint information.

The second step is to configure the relevant branch interaction analysis module to read the software source code F and configuration information files from the file system, locate and identify the configuration-related branch interaction code segment from the software source code F, and insert the program detection at the location of the configuration-related branch interaction code segment pin, and output the software source code F' after instrumentation to the file system, the method is:

2.1 The configuration-related branch interaction analysis module reads and stores the configuration information file from the file system. The configuration information file stores a list of binary pairs of <configuration parameter name, configuration variable name>, and obtains all configuration parameter names in the software source code F from the configuration information file. and the corresponding configuration variable binary pair set CS, CS={<cn ₁ , cv ₁ >, <cn ₂ , cv ₂ >, ..., <cn _n , cv _n >, ..., <cn _N , cv _N >}, where cn _n is the nth configuration parameter name in CS, cv _n is the configuration variable name corresponding to cn _n in CS in the software source code, and the binary pair relationship between cn _n and cv _n can be obtained through Shulin Zhou et al. QRS2016 published "ConfMapper: Automated Variable Finding for Configuration Items in Source Code (ConfMapper: Automatically find configuration parameters corresponding to configuration variables from software source code)" described in the method of automatically finding configuration parameters corresponding to configuration variables from software source code, N is The total number of configuration parameters in the software source code F, 1≤n≤N;

2.2 Use the Clang front-end of the LLVM compiler framework (version 10.0.0 and above) to parse the software source code F, and generate an Abstract Syntax Tree (AST _root ) corresponding to the software source code F, and each node in the AST _root represents the software source code A structure in F, such as the entire source code (TranslationUnitDecl), function declaration (FunctionDecl), If branch statement (IfStmt), function call (CallExpr), constant string (StringLiteral), binary calculation (BinaryOperator), single variable (DeclRefExpr), structure variables (MemberExpr), etc., and different structures also use the tree structure to represent the corresponding affiliation. For example, for an If branch statement in a function declaration, the corresponding IfStmt node is located with the FunctionDecl node as the root node in the subtree of

2.3 Locate and interact with branches related to cv ₁ , cv ₂ , ..., cv _n , ..., cv _N in the AST _root of the abstract syntax tree, the specific method is as follows:

2.3.1 Use the related traversal interface provided by Clang (version 10.0.0 and above) in the AST _root of the abstract syntax tree (the traversal interface in the form of VisitNodeType, where NodeType refers to the node type in the abstract syntax tree, and the specific type is as in step 2.2 As mentioned in the traversal process, the interfaces used in the traversal process include VisitFunctionDecl, VisitStringLiteral, etc., which are used to traverse the nodes of the FunctionDecl and StringLiteral types respectively. For detailed interface information, please refer to the Clang interface information document "https://clang.llvm.org/doxygen/classclang_1_1RecursiveASTVisitor. html#details") traverse the AST _root , locate the code position using cv ₁ , cv ₂ , ..., cv _n , ..., cv _N , and add it to the configuration variable use position set CL ₁ , CL ₂ , .. ., CL _n ,..., CL _N , CL _n represents the code position using cv _n in the AST _root , CL _n ={cl _n1 , cl _n2 ,..., cl _nj ,..., cl _nJ } , where cl _nj represents the position where cv _n is used for the jth time in the AST _root , J is the total number of positions where cv _n is used in the AST _root , 1≤j≤J;

2.3.2 Initialize variable n=1;

2.3.3 Initialize variable j=1, initialize cv _n related candidate branch interaction set Cand_BI _n ={}, initialize cv _n related standby configuration variable set Alternative_CV _n ={};

2.3.4 Check whether cv _n and cl _nj meet the usage conditions in the following three scenarios by:

2.3.4.1 If cv _n is directly used in the branch condition of the branch statement br of the software source code F (If statement, While statement, Switch statement in the software source code F) (that is, it is used as a part of the branch judgment condition to control program execution control flow), then add br to the relevant candidate branch interaction set Cand_BI _n of cv _n , go to 2.3.5, otherwise go to 2.3.4.2;

2.3.4.2 If cv _n is assigned to other program variables lv, and lv is directly used in the branch condition of the branch statement br, then add br to the relevant candidate branch interaction set Cand_BI _n of cv _n , and add lv to Go to Alternative_CV _n , the set of alternate configuration variables related to cv _n , and go to 2.3.5; otherwise, go to 2.3.4.3;

2.3.4.3 If the function func has a total of I (I≥1) parameters, and cv _n is used by the function func as the i-th (1≤i≤I) parameter, and in the implementation of the function func, the i-th parameter corresponding to the function func The formal parameter param _i is directly used in the branch condition of the branch statement br (the number of formal parameters implemented by the function is consistent with the actual parameter quantity of the function call, indicating that the formal parameter param _i of the function func corresponds to the i-th parameter when the function func is called ), then add br to the relevant candidate branch interaction set Cand_BI _n of cv _n , and add param _i to the alternate configuration variable set Alternative_CV _n related to cv _n , and go to 2.3.5; otherwise, go to 2.3.5 directly;

2.3.5 Let j=j+1, if j≤J, go to 2.3.4; otherwise, get the candidate branch interaction set Cand_BI _n related to cv _n , Cand_BI _n = { _{cand_bin1} , _{cand_bin2} , .., _{cand_bink} ,. .., _{cand_binK} }, where _{cand_bink} is the kth candidate branch interaction of cv _n in the Cand_BI _n set, K is the total number of candidate branch interactions of cv _n in Cand_BI _n , 1≤k≤K, go to 2.3.6;

2.3.6 Analyze and identify the branch conditions and branch processing types of the relevant branch statements in the candidate branch interaction set Cand_BI _n related to cv _n , and obtain the five-tuple <cn _n , rcv _n , ps _nk , op _nk , loc _nk > , cn _n is the configuration parameter name of cv _n , rcv _n is the value of cv _n at this position, ps _nk is the program state value interacting with cv _n , op _nk is the corresponding operator interacting with cv _n , loc _nk is the code position of the branch statement where compare_condition is located, and compare_condition is a logical expression in the leaf node of the alternate configuration variable acv that contains cv _n or contains cv _n at the corresponding position; according to the five-tuple <cn _n , rcv _n , ps _nk , op _nk , loc _nk > information is inserted at the loc _nk position of the software source code AST _root , and the position of the inserted stake in the software source code F is recorded; the method is:

2.3.6.1 Initialize variable k=1;

2.3.6.2 Obtain the branch condition _{branch_condition} of cand_bink through the interface getCond() provided by Clang (version 10.0.0 and above), and use the branch_condition as a binary operator && (indicating AND operation) or || (indicating or operation), the leaf nodes are all logical expressions and parsed into a binary tree, and then the logic in the leaf nodes containing the configuration variable cv _n (or the alternate configuration variable acv(acv∈Alternative_CV _n ) containing cv _n at the corresponding position) The expression is recorded as compare_condition, and all other leaf nodes adjacent to compare_condition and connected by && nodes are connected by AND operation, and the connected logical expression is recorded as pre_condition;

2.3.6.3 Check whether the compare_condition only includes cv _n or the alternate configuration variable acv (acv∈Alternative_CV _n ), constants, and other configuration variables cv _m (1≤m≤N, m≠n) containing cv _n in the corresponding position, If yes, go to 2.3.6.7; otherwise, go to 2.3.6.4;

2.3.6.4 Transform compare_condition, move cv _n or the spare configuration variable acv of cv _n at the corresponding position to the left side of the compare_condition expression operator, move all other operands to the right side of the compare_condition expression operator, and Record the statement on the right side of the expression as ps _nk , record the expression operator as op _nk , record the code position of the branch statement where the compare_condition is located as loc _nk , and record the cv _n or acv that appears in the current compare_condition as rcv _n ;

_2.3.6.5 Obtain the branch handling code segment handling_block of cand_bink through the interface getThen() provided by Clang (version 10.0.0 and above), and check whether handling_block contains any of the following three situations: i) Error handling related Log statement, ii) jump statement, and the jump label name contains keywords describing errors, iii) error handling related function calls, if any of the cases are included, go to 2.3.6.6; if none of the above three cases are included , go to 2.3.6.7;

ps_nk为数值类型程序变量++fields_read，op_nk为≥，loc_nk为字符串型代码位置信息，在此例子中为server/protocol.c：1187，记录上述插桩在软件源码F中的位置；2.3.6.6 According to the quintuple <cn _n , rcv _n , ps _nk , op _nk , loc _nk > information, perform instrumentation at the loc _nk position of the software source code AST _root . The main function of the instrumentation code segment is when the software source code F is running Record four types of information: cn _n , rcv _n , ps _nk , op _nk and output them to the file system; for example, in the HttpdWeb server software, a quintuple instance of the configuration parameter "LimitRequestFields" is as follows <"LimitRequestFields", r->server->limit_req_fields, ++fields_read, ≥, server/protocol.c: 1187>, in this example cn _n is the string type configuration parameter name LimitRequestFields, rcv _n is the numerical type program variable

ps _nk is a numerical program variable ++fields _read , op _nk is ≥, loc _nk is a string type code position information, in this example it is server/protocol. Location;

2.3.6.7 Let k=k+1, if k≤K, go to 2.3.6.2; otherwise go to 2.3.7;

2.3.7 Let n=n+1, if n≤N, go to 2.3.3; otherwise go to 2.4;

2.4 The instrumentation of the software source code F is completed, and the instrumented software source code F' is output to the file system.

The third step is to manually compile and link F' to obtain the executable software F", and run the software F".

In the fourth step, the dynamic monitoring module monitors the output of F" in the file system, extracts the configuration constraint information under the influence of the current load, and automatically adjusts the configuration parameters according to the relevant configuration constraint information when the configuration value in F" does not meet the constraint conditions value, the method is:

4.1 The dynamic monitoring module monitors the output in the file system of F" for a period of time T (can be customized, default T=1 second);

4.2 Initialize variable n=1;

4.3 The dynamic monitoring module checks all monitoring information sequences related to cn _n within time T <rcv _n , ps _n1 , op _n1 , loc _n1 >, <rcv _n , ps _n2 , op _n2 , loc _n2 >, ..., <rcv _n , ps _ns , op _ns , loc _ns >, ..., <rcv _n , ps _nS , op _nS , loc _nS > whether the value of the configuration variable and the value of the program state interacting with the configuration variable meet the configuration variable interaction requirements Corresponding to the size relationship described by the operator, where S is the total number of monitoring information related to cn _n output by F" within T time, <rcv _n , ps _ns , op _ns , loc _ns > is the sth monitoring information, 1 ≤s≤S; if the S monitoring information is all satisfied, then pass the inspection; if there is unsatisfied monitoring information in the S monitoring information, use the minimum program status threshold min_threshold to record all unsatisfied monitoring information that interacts with configuration variables The minimum value of the program state value, use the maximum program state threshold max_threshold to record the maximum value of the program state value interacting with the configuration variable in all unsatisfied monitoring information, the method is:

4.3.1 Initialize variable s=1, initialize variable minimum program state threshold min_threshold=∞, maximum program state threshold max_threshold=-1;

4.3.2 Check whether the values of rcv _n and ps _ns meet the size relationship described by op _ns , the method is:

4.3.2.1 Check the operator type of op _ns , if op _ns is "≥", go to 4.3.2.2; if op _ns is ">", go to 4.3.2.4; if op _ns is "≤", go to 4.3.2.6; if op _ns is "<", go to 4.3.2.8;

4.3.2.2 If the value of rcv _n and the value of ps _ns satisfy rcv _n ≥ ps _ns , go to 4.3.3; otherwise go to 4.3.2.3;

4.3.2.3 If ps _ns >max_threshold, set max_threshold=ps _ns and go to 4.3.3; otherwise go to 4.3.3 directly;

4.3.2.4 If the value of rcv _n and the value of ps _ns satisfy rcv _n >ps _ns , go to 4.3.3; otherwise go to 4.3.2.5;

4.3.2.5 If ps _ns >max_threshold, set max_threshold=ps _ns and go to 4.3.3; otherwise go to 4.3.3 directly;

4.3.2.6 If the value of rcv _n and the value of ps _ns satisfy rcv _n ≤ ps _ns , go to 4.3.3; otherwise go to 4.3.2.7;

4.3.2.7 If ps _ns <min_threshold, set min_threshold=ps _ns and go to 4.3.3; otherwise go to 4.3.3 directly;

4.3.2.8 If the value of rcv _n and the value of ps _ns satisfy rcv _n < ps _ns , go to 4.3.3; otherwise go to 4.3.2.9;

4.3.2.9 If ps _ns < min_threshold, set min_threshold = ps _ns and go to 4.3.3; otherwise go to 4.3.3 directly;

4.3.3 Let s=s+1, if s≤S, go to 4.3.2; otherwise go to 4.4;

4.4 The dynamic monitoring module combines the monitoring information sequence <rcv _n , ps _n1 , op _n1 , loc _n1 >, <rcv _n , ps _n2 , op _n2 , loc _n2 >, ..., <rcv _n , ps _ns , op _ns , loc _ns >,..., <rcv _n , ps _nS , op _nS , loc _nS > adjust the value of the configuration variable rcv _n , the method is:

转4.5；若max_threshold≠-1，将rcv_n对应的配置参数取值增大为

转4.5；If min_threshold≠∞, reduce the configuration parameter value corresponding to rcv _n to

Go to 4.5; if max_threshold≠-1, increase the value of the configuration parameter corresponding to rcv _n to

turn 4.5;

4.5 Let n=n+1, if n≤N, go to 4.3; otherwise, the current adjustment cycle ends, if the user stops running F", go to step 5; otherwise, go to 4.1 to start the next cycle adjustment.

In the fifth step, the automatic adjustment of the configuration parameters in F" by the dynamic monitoring module is completed.

Compared with the prior art, the present invention can achieve the following beneficial effects:

1. By adopting the present invention, load-sensitive configuration constraint information can be obtained. Compared with the "Do Not Blame Users for Misconfigurations (Do Not Blame Users for Misconfigurations)" published by Tianyin Xu et al. in SOSP2013, which extracts configuration constraint information through static analysis methods, the configuration constraint information extracted by the present invention under dynamic scenarios More accurate; Compared with Xudong Sun et al. published in OSDI2020 "Testing Configuration Changes in Context to Prevent Production Failures (testing configuration changes in the context to prevent production failures)" work to extract dynamic configuration constraints by simulating execution, the present invention can Obtain configuration constraints information on load impact in a real production environment.

2. The present invention provides a method for adjusting software configuration to adapt to load changes, which can help users automatically adjust configuration parameter values to meet changing load requirements, avoid load-sensitive configuration failures, and have good application scenarios.

Description of drawings

Fig. 1 is a logical structure diagram of the configuration parameter adjustment system constructed in the first step of the present invention.

Fig. 2 is the overall flow chart of the present invention.

detailed description

Below in conjunction with accompanying drawing 2, take the configuration parameter "LimitRequestFields" in the Httpd Web server software (as the software source code F) as an example to illustrate the present invention.

The first step is to build a configuration parameter adjustment system, which is composed of a configuration-related branch interaction analysis module and a dynamic monitoring module. The configuration-related branch interaction analysis module reads the Httpd software source code F and the configuration information file (stored with <configuration parameter name, configuration variable name> binary pair list) from the file system, identifies and locates the configuration-related branch interaction in the Httpd software source code F, And perform code insertion, and output the Httpd software source code F' after the insertion into the file system; then the user compiles and links F' to obtain the Httpd executable software F"; when running the software F", the dynamic monitoring module monitors The output of F'' in the file system extracts and outputs the configuration constraint information under the influence of the current load, and automatically adjusts the configuration parameters according to the relevant configuration constraint information when the F" load changes and the current configuration value does not meet the configuration constraint information value.

In the second step, the configuration-related branch interaction analysis module reads the Httpd software source code F and the Httpd configuration information file from the file system, locates and identifies the configuration-related branch interaction code segment from the Httpd software source code F, and inserts the configuration-related branch interaction code segment position The stub program probe outputs the software source code F' after stub insertion to the file system, the method is:

2.1 The configuration-related branch interaction analysis module reads and stores the Httpd software configuration information file from the file system. The configuration information file stores a list of binary pairs of <configuration parameter name, configuration variable name>, and obtains all configurations in the software source code F from the configuration information file Parameter name and corresponding configuration variable binary pair set CS, CS={<cn ₁ , cv ₁ >, <cn ₂ , cv ₂ >, ..., <cn _n , cv _n >, ..., <cn _N , cv _N >}, where cn _n is the nth configuration parameter name in CS, cv _n is the configuration variable name corresponding to cn _n in CS in the software source code, the binary pair relationship between cn _n and cv _n can be obtained through Shulin Zhou et al. "ConfMapper: Automated Variable Finding for Configuration Items in SourceCode (ConfMapper: Automatically find configuration parameters corresponding to configuration variables from software source code)" published at QRS2016 establishes the method of automatically finding configuration parameters corresponding to configuration variables from software source code, N is the total number of configuration parameters in the software source code F, N=131 (including the configuration variable binary pair information of the configuration parameter "LimitRequestFields"), 1≤n≤N;

2.2 Use the Clang front end of the LLVM compiler framework (version 10.0.0 and above) to parse the Httpd software source code F, and generate an Abstract Syntax Tree (AST _root ) corresponding to F, and each node in the AST _root represents the software source code F A structure in the whole source code (TranslationUnitDecl), function declaration (FunctionDecl), If branch statement (IfStmt), function call (CallExpr), constant string (StringLiteral), binary calculation (BinaryOperator), etc., and different The structure also uses a tree structure to represent the corresponding affiliation. For example, for an If branch statement in a function declaration, the corresponding IfStmt node is located in the subtree with the FunctionDecl node as the root node;

2.3 Locate and interact with branches related to cv ₁ , cv ₂ , ..., cv _n , ..., cv _N in the AST _root of the abstract syntax tree, the specific method is as follows:

2.3.1 Use the related traversal interface provided by Clang (version 10.0.0 and above) in the AST _root of the abstract syntax tree (the traversal interface in the form of VisitNodeType, where NodeType refers to the node type in the abstract syntax tree, and the specific type is as in step 2.2 As mentioned in , the interfaces used in the traversal process include VisitFunctionDecl, VisitStringLiteral, etc., which are used to traverse the nodes of the FunctionDecl and StringLiteral types respectively) Traversing the AST _root , positioning uses cv ₁ , cv ₂ , ..., cv _n , ..., cv The code position of _N , and added to the configuration variable use position set CL ₁ , CL ₂ , ..., CL _n , ..., CL _N , CL _n represents the code position of using cv _n in the AST _root , CL _n = {cl _n1 , cl _n2 , ..., cl _nj , ..., cl _nJ }, where cl _nj represents the position where cv _n is used for the jth time in the AST _root , and J is the position where cv _n is used in the AST _root (J corresponds to each cv _n will be different, because N=131, the value of J may change every time when facing n cycles), 1≤j≤J;

2.3.2 Initialize variable n=1;

2.3.3 Initialize variable j=1, initialize cv _n related candidate branch interaction set Cand_BI _n ={}, initialize cv _n related standby configuration variable set Alternative_CV _n ={};

2.3.4 Check whether cv _n and cl _nj meet the usage conditions in the following three scenarios by:

2.3.4.1 If cv _n is directly used in the branch condition of the branch statement br of the software source code F (If statement, While statement, Switch statement in the software source code F) (that is, it is used as a part of the branch judgment condition to control program execution control flow), then add br to the relevant candidate branch interaction set Cand_BI _n of cv _n , go to 2.3.5, otherwise go to 2.3.4.2;

2.3.4.2 If cv _n is assigned to other program variables lv, and lv is directly used in the branch condition of the branch statement br, then add br to the relevant candidate branch interaction set Cand_BI _n of cv _n , and add lv to Go to Alternative_CV _n , the set of alternate configuration variables related to cv _n , and go to 2.3.5; otherwise, go to 2.3.4.3;

2.3.4.3 If the function func has a total of I (I ≥ 1) (I corresponds to the parameters of related function calls that have appeared in the software source code, and the number of parameters for different function calls is different) parameters, and cv _n is used by the function func as the i-th (1 ≤ i≤I) parameters are used, and in the realization of the function func, the i-th formal parameter param _i corresponding to the function func is directly used in the branch condition of the branch statement br (the formal parameter amount of the function realization and the actual parameter of the function call The amount is consistent, indicating that the formal parameter param _i of the function func corresponds to the i-th parameter when the function func is called), then add br to the relevant candidate branch interaction set Cand_BI _n of cv _n , and add param _i to cv _n In the related standby configuration variable set Alternative_CV _n , go to 2.3.5; otherwise, go to 2.3.5 directly;

2.3.5 Let j=j+1, if j≤J, go to 2.3.4; otherwise, get the candidate branch interaction set Cand_BI _n related to cv _n , Cand_BI _n = { _{cand_bin1} , _{cand_bin2} , .., _{cand_bink} ,. .., _{cand_binK} }, where _{cand_bink} is the kth candidate branch interaction of cv _n in the Cand_BI _n set, K is the total number of candidate branch interactions of cv _n in Cand_BI _n , 1≤k≤K, (K corresponds to each cv _n will be different, and the value of K may change every time when n is cycled) go to 2.3.6;

2.3.6 Analyze and identify the branch conditions and branch processing types of the relevant branch statements in the candidate branch interaction set Cand_BI _n related to cv _n , the method is:

2.3.6.1 Initialize variable k=1;

2.3.6.2 Obtain the branch condition _{branch_condition} of cand_bink through the interface getCond() provided by Clang (version 10.0.0 and above), and use the branch_condition as a binary operator && (indicating AND operation) or || (indicating or operation), the leaf nodes are all logical expressions and parsed into a binary tree, and then the logic in the leaf nodes containing the configuration variable cv _n (or the alternate configuration variable acv(acv∈Alternative_CV _n ) containing cv _n at the corresponding position) The expression is recorded as compare_condition, and all other leaf nodes adjacent to compare_condition and connected by && nodes are connected by AND operation, and the connected logical expression is recorded as pre_condition;

2.3.6.3 Check whether the compare_condition only includes cv _n or the alternate configuration variable acv (acv∈Alternative_CV _n ), constants, and other configuration variables cv _m (1≤m≤N, m≠n) containing cv _n in the corresponding position, If yes, go to 2.3.6.7; otherwise, go to 2.3.6.4;

2.3.6.4 Transform compare_condition, move cv _n or the spare configuration variable acv of cv _n at the corresponding position to the left side of the compare_condition expression operator, move all other operands to the right side of the compare_condition expression operator, and Record the statement on the right side of the expression as ps _nk , record the expression operator as op _nk , record the code position of the branch statement where the compare_condition is located as loc _nk , and record the cv _n or acv that appears in the current compare_condition as rcv _n ;

_2.3.6.5 Obtain the branch handling code segment handling_block of cand_bink through the interface getThen() provided by Clang (version 10.0.0 and above), and check whether handling_block contains any of the following three situations: i) Error handling related Log statement, ii) jump statement, and the jump label name contains keywords describing errors, iii) error handling related function calls, if any of the cases are included, go to 2.3.6.6; if none of the above three cases are included , go to 2.3.6.7;

2.3.6.6 According to the quintuple <cn _n , rcv _n , ps _nk , op _nk , loc _nk > information, perform instrumentation at the loc _nk position of the software source code AST _root . The main function of the instrumentation code segment is when the software source code F is running Record cn _n (configuration parameter name), rcv _n (the value of the configuration variable at this position), ps _nk (the value of the program state interacting with the configuration variable), op _nk (the corresponding operator interacting with the configuration variable) The information is output to the file system; for the configuration parameter "LimitRequestFields" in the HttpdWeb server software, an example of its five-tuple information is as follows <"LimitRequestFields", r->server->limit_req_fields, ++fields-read, ≥, server/protocol.c:1187>;

2.3.6.7 Let k=k+1, if k≤K, go to 2.3.6.2; otherwise go to 2.3.7;

2.3.7 Let n=n+1, if n≤N, go to 2.3.3; otherwise go to 2.4;

2.4 The instrumentation of the Httpd software source code F is completed. A total of 36 program probes are inserted (including the program probes for the location of the configuration parameter "LimitRequestFields" in F), and the instrumented software source code F' is output to the file system.

The third step is to manually compile and link F' to obtain the executable software F", and run the software F".

In the fourth step, the dynamic monitoring module monitors the output of F" in the file system, extracts the configuration constraint information under the influence of the current load, and automatically adjusts the configuration parameters according to the relevant configuration constraint information when the configuration value in F" does not meet the constraint conditions value, the method is:

4.1 The dynamic monitoring module monitors the output of F" in the file system for a period of time T (T=1 second);

4.2 Initialize variable n=1;

4.3 The dynamic monitoring module checks all monitoring information sequences related to cn _n within time T <rcv _n , ps _n1 , op _n1 , loc _n1 >, <rcv _n , ps _n2 , op _n2 , loc _n2 >, ..., <rcv _n , ps _ns , op _ns , loc _ns >, ..., <rcv _n , ps _nS , op _nS , loc _nS > whether the value of the configuration variable and the value of the program state interacting with the configuration variable meet the configuration variable interaction requirements Corresponding to the size relationship described by the operator, where S is the total number of monitoring information related to cn _n output by F" within T time, <rcv _n , ps _ns , op _ns , loc _ns > is the sth monitoring information, 1≤s≤S; if the S monitoring information is satisfied, the inspection will pass; if there is unsatisfied monitoring information in the S monitoring information, use the minimum program status threshold min_threshold to record all unsatisfied monitoring information and interact with configuration variables The minimum value of the program state value, use the maximum program state threshold max_threshold to record the maximum value of the program state value interacting with configuration variables in all unsatisfied monitoring information, the method is:

4.3.1 Initialize variable s=1, initialize variable minimum program state threshold min_threshold=∞, maximum program state threshold max_threshold=-1;

4.3.2 When the value of cn _n is "LimitRequestFields", check whether the values of rcv _n and ps _ns meet the size relationship described by op _ns , the method is:

4.3.2.1 For the configuration parameter "LimitRequestFields" during the insertion process, the relevant quintuple information is as described in step 2.3.6.6, indicating that the value of op _ns is "≥", then check whether the value of rcv _n and the value of ps _ns are satisfied rcv _n ≥ ps _ns , if satisfied, go to 4.3.3; otherwise, go to 4.3.2.2;

4.3.2.2 If ps _ns >max_threshold, set max_threshold=ps _ns and go to 4.3.3; otherwise go to 4.3.3 directly;

4.3.3 Let s=s+1, if s≤S, go to 4.3.2; otherwise go to 4.4;

转4.5。4.4 The dynamic monitoring module combines the monitoring information sequence <rcv _n , ps _n1 , op _n1 , loc _n1 >, <rcv _n , ps _n2 , op _n2 , loc _n2 >, ..., <rcv _n , ps _ns , op _ns , loc _ns >, ..., <rcv _n , ps _nS , op _nS , loc _nS > the configuration variable rcv _n corresponding to the configuration parameter "LimitRequestFields" (the configuration parameter "LimitRequestFields" corresponds to the variable "r->server ->limit_req_fields") to adjust the value, because at this time max_threshold≠-1 (see 4.3.2.2, when the software load changes, the program state changes, which in turn leads to a change in the value of max_threshold), set the configuration parameter corresponding to rcv _n to value increased to

Go to 4.5.

4.5 Let n=n+1, if n≤N, go to 4.3; otherwise, the current adjustment cycle ends, if the user stops running F", go to step 5; otherwise, go to 4.1 to start the next cycle adjustment.

In the fifth step, the automatic adjustment of the configuration parameter "LimitRequestFields" in F" by the dynamic monitoring module ends.

If the Httpd Web server software is not adjusted by the second step and the fourth step of the present invention to configure parameters, then the Httpd Web server may have load-sensitive configuration failures when facing the following scenarios, and the specific scenarios and failure performances are described as follows: Initially The Httpd Web server is under low load, and the number of fields in the http request header of the user accessing the Httpd Web server is small, which is lower than the value of the configuration parameter "LimitRequestFields", the Httpd Web server can provide services normally; When the number of fields in the http request header of the server increases and exceeds the value of the configuration parameter "LimitRequestFields", the Httpd Web server cannot provide services for users and returns "400error" error code to users, resulting in configuration failures;

If the configuration parameter adjustment is carried out according to the second step and the fourth step of the present invention for the Httpd software source code, then when the field number in the http request head of the user access server increases, automatically adjust and increase the value of the configuration parameter "LimitRequestFields", Make it meet the constraints that are greater than the number of fields in the user's http request header, and realize load-aware configuration parameter adjustment, thereby preventing the above-mentioned load-sensitive configuration failures.

Claims (9)

1. a software configuration parameter adjustment method of load perception, is characterized in that comprising the following steps: The first step is to build a configuration parameter adjustment system. The configuration parameter adjustment system is composed of a configuration-related branch interaction analysis module and a dynamic monitoring module; the configuration-related branch interaction analysis module reads in the software source code F and configuration information files from the file system. Identify, locate, and configure related branches to interact with each other, and perform code insertion, and output the software source code F' after insertion to the file system; then the user compiles and links F' to obtain executable software F"; the dynamic monitoring module monitors The output in the file system when F'' is running, extracts and outputs the configuration constraint information under the influence of the current load, and automatically adjusts according to the relevant configuration constraint information when the F" load changes and the current configuration value does not meet the configuration constraint information Configuration parameter value; The second step is to configure the relevant branch interaction analysis module to read the software source code F and configuration information files from the file system, locate and identify the configuration-related branch interaction code segment from the software source code F, and insert the program detection at the location of the configuration-related branch interaction code segment pin, and output the software source code F' after instrumentation to the file system, the method is: 2.1 The configuration-related branch interaction analysis module reads the storage configuration information file from the file system. The configuration information file stores a list of binary pairs <configuration parameter name, configuration variable name>, and obtains all configuration parameter names in the software source code F from the configuration information file. And the corresponding configuration variable binary pair set CS, CS={<cn ₁ , cv ₁ ＞, <cn ₂ , cv ₂ ＞, ..., <cn _n , cv _n >, ..., <cn _N , cv _N >}, where cn _n is the nth configuration parameter name in CS, cv _n is the configuration variable name corresponding to cn _n in the software source code in CS, N is the total number of configuration parameters in the software source code F, 1≤n≤N; 2.2 Use the Clang front end of the LLVM compiler framework to parse the software source code F, and generate the abstract syntax tree AST _root corresponding to the software source code F. Each node in the AST _root represents a structure in the software source code F, and different structures are also represented by The tree structure represents the corresponding affiliation; 2.3 Locate and interact with branches related to cv ₁ , cv ₂ , ..., cv _n , ..., cv _N in the AST _root of the abstract syntax tree, the specific method is as follows: 2.3.1 Use the relevant traversal interface provided by Clang to traverse the AST _{root in the abstract syntax tree AST root ,} locate the code position using cv ₁ , cv ₂ , ..., cv _n , ..., cv _N , and add it to the configuration In the variable use position set CL ₁ , CL ₂ , ..., CL _n , ..., CL _N , CL _n represents the code position of using cv _n in the AST _root , CL _n = {cl _n1 , cl _n2 , .. ., cl _nj ,..., cl _nJ }, where cl _nj represents the position where cv _n is used for the jth time in the AST _root , and J is the total number of positions where cv _n is used in the AST _root , 1≤j≤J; 2.3.2 Initialize variable n=1; 2.3.3 Initialize variable j=1, initialize cv _n related candidate branch interaction set Cand_BI _n ={}, initialize cv _n related standby configuration variable set Alternative_CV _n ={}; 2.3.4 Check whether cv _n and cl _nj meet the usage conditions in the following three scenarios by: 2.3.4.1 If cv _n is directly used in the branch condition of the branch statement br of the software source code F, that is, as a part of the branch judgment condition, the control flow of the program execution is controlled after operation, then br is added to the relevant candidate of cv _n In the branch interaction set Cand_BI _n , go to 2.3.5, otherwise go to 2.3.4.2; 2.3.4.2 If cv _n is assigned to other program variables lv, and lv is directly used in the branch condition of the branch statement br, then add br to the relevant candidate branch interaction set Cand_BI _n of cv _n , and add lv to Go to Alternative_CV _n , the set of alternate configuration variables related to cv _n , and go to 2.3.5; otherwise, go to 2.3.4.3; 2.3.4.3 If the function func has a total of I parameters, I≥1, and cv _n is used by the function func as the i-th parameter, 1≤i≤I, and in the implementation of the function func, the i-th formal parameter corresponding to the function func Param _i is directly used in the branch condition of the branch statement br, then add br to the relevant candidate branch interaction set Cand_BI _n of cv _n , and add param _i to the alternate configuration variable set Alternative_CV _n related to cv _n , Go to 2.3.5; otherwise go directly to 2.3.5; 2.3.5 Let j=j+1, if j≤J, go to 2.3.4; otherwise, get the candidate branch interaction set Cand_BI _n related to cv _n , Cand_BI _n = { _{cand_bin1} , _{cand_bin2} , .., _{cand_bink} ,. .., _{cand_binK} }, where _{cand_bink} is the kth candidate branch interaction of cv _n in the Cand_BI _n set, K is the total number of candidate branch interactions of cv _n in Cand_BI _n , 1≤k≤K, go to 2.3.6; 2.3.6 Analyze and identify the branch conditions and branch processing types of the relevant branch statements in the candidate branch interaction set Cand_BI _n related to cv _n , and obtain the quintuple <cn _n , rcv _n , ps _nk , op _nk , loc _nk > , rcv _n is the value of the configuration variable cv _n at this position, ps _nk is the value of the program state interacting with the configuration variable cv _n , op _nk is the corresponding operator interacting with the configuration variable cv _n , loc _nk is the branch of compare_condition Statement code position, compare_condition is a logical expression in the leaf node containing cv _n or the alternate configuration variable acv containing cv _n at the corresponding position; according to the five-tuple <cn _n , rcv _n , ps _nk , op _nk , loc _nk > The information is inserted in the loc _nk position of the AST _root of the software source code. The function of the inserted code segment is to record the four types of information cn _n , rcv _n , ps _nk , and op _nk when the software source code F is running and output them to the file system. The position of the pile in the software source code F; 2.3.7 Let n=n+1, if n≤N, go to 2.3.3; otherwise go to 2.4; 2.4 Complete the instrumentation of the software source code F, and output the instrumented software source code F' to the file system; The third step is to manually compile and link F' to obtain executable software F", and run the software F"; In the fourth step, the dynamic monitoring module monitors the output of F" in the file system, extracts the configuration constraint information under the influence of the current load, and automatically adjusts the configuration parameters according to the relevant configuration constraint information when the configuration value in F" does not meet the constraint conditions value, the method is: 4.1 The dynamic monitoring module monitors the output of F" in the file system for a period of time T; 4.2 Initialize variable n=1; 4.3 The dynamic monitoring module checks all monitoring information sequences related to cn _n within time T <rcv _n , ps _n1 , op _n1 , loc _n1 >, <rcv _n , ps _n2 , op _n2 , loc _n2 >, ..., <rcv _n , ps _ns , op _ns , loc _ns >, ..., <rcv _n , ps _nS , op _nS , loc _nS > whether the value of the configuration variable and the value of the program state interacting with the configuration variable meet the requirements of the configuration variable interaction Corresponding to the size relationship described by the operator, where S is the total number of monitoring information related to cn _n output by F" within T time, <rcv _n , ps _ns , op _ns , loc _ns > is the sth monitoring information, 1 ≤s≤S; if the S monitoring information is all satisfied, pass the inspection and go to 4.4; if there is unsatisfied monitoring information in the S monitoring information, use the minimum program status threshold min_threshold to record all unsatisfied monitoring information and configuration The minimum value of the program state value of variable interaction, use the maximum program state threshold max_threshold to record the maximum value of the program state value interacting with configuration variables in all unsatisfied monitoring information, go to 4.4; 4.4 The dynamic monitoring module combines the monitoring information sequence <rcv _n , ps _n1 , op _n1 , loc _n1 >, <rcv _n , ps _n2 , op _n2 , loc _n2 >, ..., <rcv _n , ps _ns , op _ns , loc _ns >, ..., <rcv _n , ps _nS , op _nS , loc _nS > to adjust the value of the configuration variable rcv _n , the method is: If min_threshold≠∞, reduce the configuration parameter value corresponding to rcv _n to

Go to 4.5; if max_threshold≠-1, then increase the value of the configuration parameter corresponding to rcv _n to

turn 4.5; 4.5 Let n=n+1, if n≤N, go to 4.3; otherwise, the current adjustment cycle ends, if the user stops running F", go to step 5; otherwise, go to 4.1 to start the next cycle adjustment; In the fifth step, the automatic adjustment of the configuration parameters in F" by the dynamic monitoring module is completed. 2. The software configuration parameter adjustment method of a kind of load perception as claimed in claim 1, it is characterized in that 2.1 step described cn _n and cv _n binary pair relation is established by automatically searching configuration parameter corresponding configuration variable method from software source code . 3. A load-aware software configuration parameter adjustment method as claimed in claim 1, characterized in that said Clang front end is version 10.0.0 and above. 4. the software configuration parameter adjustment method of a kind of load perception as claimed in claim 1, it is characterized in that the structure in the software source code F of the node representation in the described AST _root of step 2.2 comprises: whole source code TranslationUnitDecl, function declaration FunctionDecl , If branch statement IfStmt, function call CallExpr, constant string StringLiteral, binary calculation BinaryOperator, single variable DeclRefExpr, structure variable MemberExpr. 5. the software configuration parameter adjustment method of a kind of load perception as claimed in claim 1, it is characterized in that the relevant traversal interface that Clang provides in 2.3.1 step refers to the traversal interface of VisitNodeType form, and wherein NodeType refers to abstract syntax tree The node type in . 6. the software configuration parameter adjustment method of a kind of load perception as claimed in claim 1, it is characterized in that the branch statement br of software source code F described in 2.3.4.1 step refers to If statement, While statement, Switch statement in software source code F . 7. the software configuration parameter adjustment method of a kind of load perception as claimed in claim 1, it is characterized in that described in 2.3.6 step to the branch condition and the branch processing of relevant branch statement in the relevant candidate branch interaction set Cand_BI _n of cv _n The type is analyzed and identified, and the quintuple <cn _n , rcv _n , ps _nk , op _nk , loc _nk > is obtained. According to the quintuple information, the method of inserting stubs at the loc _nk position of the software source code AST _root is: 2.3.6.1 Initialize variable k=1; 2.3.6.2 Obtain the branch condition _{branch_condition} of cand_bink through the interface getCond() provided by Clang, and analyze branch_condition according to the form that non-leaf nodes are binary AND operators && or OR operators ||, and leaf nodes are logical expressions is a binary tree, and then record the logical expression in the leaf node containing the configuration variable cv _n or the alternate configuration variable acv containing cv _n in the corresponding position as compare_condition, acv∈Alternative_CV _n , and connect the others adjacent to compare_condition and using && node All leaf nodes of are connected by AND operation, and the logical expression after connection is recorded as pre_condition; 2.3.6.3 Check whether the compare_condition only includes cv _n or the spare configuration variable acv, constant, and other configuration variables cv _m containing cv _n in the corresponding position, 1≤m≤N, m≠n; if so, go to 2.3.6.7; Otherwise, go to 2.3.6.4; 2.3.6.4 Transform compare_condition, move cv _n or the spare configuration variable acv of cv _n at the corresponding position to the left side of the compare_condition expression operator, move all other operands to the right side of the compare_condition expression operator, and Record the statement on the right side of the expression as ps _nk , record the expression operator as op _nk , record the code position of the branch statement where the compare_condition is located as loc _nk , and record the cv _n or acv that appears in the current compare_condition as rcv _n ; _2.3.6.5 Obtain the branch processing code segment handling_block of cand_bink through the interface getThen() provided by Clang, and check whether the handling_block contains any of the following three situations: i) Error handling related log statements, ii) Jump statements , and the name of the jump label contains keywords describing the error, iii) error handling related function calls, if any of the cases are included, go to 2.3.6.6; if none of the above three cases are included, go to 2.3.6.7; 2.3.6.6 According to the quintuple <cn _n , rcv _n , ps _nk , op _nk , loc _nk > information, perform instrumentation at the loc _nk position of the software source code AST _root , and the function of inserting code segments is recorded when the software source code F is running The configuration parameter name cn _n , the value rcv _n of the configuration variable at this location, the value ps _nk of the program state interacting with the configuration variable, and the corresponding operator op _nk interacting with the configuration variable are output to the file system, and Record the position of the above-mentioned stub in the software source code F; 2.3.6.7 Let k=k+1, if k≤K, go to 2.3.6.2; otherwise end. 8. A load-aware software configuration parameter adjustment method as claimed in claim 1, characterized in that T=1 second in step 4.1. 9. The software configuration parameter adjustment method of a kind of load perception as claimed in claim 1, it is characterized in that all monitoring information sequences of cn _n related in the inspection time T of the dynamic monitoring module described in step 4.3<rcv _n , ps _n1 , op _n1 , loc _n1 >, <rcv _n , ps _n2 , op _n2 , loc _n2 >, ..., <rcv _n , ps _ns , op _ns , loc _ns >, ..., <rcv _n , ps _nS , op _nS , loc _nS > Whether the value of the configuration variable and the value of the program state interacting with the configuration variable meet the size relationship described by the corresponding operator of the interaction of the configuration variable is as follows: 4.3.1 Initialize variable s=1, initialize variable minimum program state threshold min_threshold=∞, maximum program state threshold max_threshold=-1; 4.3.2 Check whether the values of rcv _n and ps _ns meet the size relationship described by op _ns , the method is: 4.3.2.1 Check the operator type of op _ns , if op _ns is "≥", go to 4.3.2.2; if op _ns is ">", go to 4.3.2.4; if op _ns is "≤", go to 4.3.2.6; if op _ns is "<", go to 4.3.2.8; 4.3.2.2 If the value of rcv _n and the value of ps _ns satisfy rcv _n ≥ ps _ns , go to 4.3.3; otherwise go to 4.3.2.3; 4.3.2.3 If ps _ns > max_threshold, set max_threshold = ps _ns and go to 4.3.3; otherwise go to 4.3.3 directly; 4.3.2.4 If the value of rcv _n and the value of ps _ns satisfy rcv _n > ps _ns , go to 4.3.3; otherwise go to 4.3.2.5; 4.3.2.5 If ps _ns > max_threshold, set max_threshold = ps _ns and go to 4.3.3; otherwise go to 4.3.3 directly; 4.3.2.6 If the value of rcv _n and the value of ps _ns satisfy rcv _n ≤ ps _ns , go to 4.3.3; otherwise go to 4.3.2.7; 4.3.2.7 If ps _ns < min_threshold, set min_threshold = ps _ns and go to 4.3.3; otherwise go to 4.3.3 directly; 4.3.2.8 If the value of rcv _n and the value of ps _ns satisfy rcv _n < ps _ns , go to 4.3.3; otherwise go to 4.3.2.9; 4.3.2.9 If ps _ns < min_threshold, set min_threshold = ps _ns and go to 4.3.3; otherwise go to 4.3.3 directly; 4.3.3 Let s=s+1, if s≤S, go to 4.3.2; otherwise end.

Images