A system for batch computing cloud service job accountability

Abstract

The application discloses a kind of batch computing cloud service job accountability systems, it is related to information security technical field, specifically a kind of batch computing cloud service job accountability systems, including data extraction module, for extracting batch service operating condition data from cloud service log data;Failure analysis module, for integrating failure factors with cloud service operating condition, then according to subject-predicate sentence, constructs cloud service operating event vector sample;Contact discovery and event net generation module, for receiving event net vector sample data, and event abnormal vector sample is transformed into event accountability vector sample;Accountability determination module, for receiving accountability event data and unfulfilled responsibility event data, and respectively analyzing accountability event and unfulfilled responsibility event and the correlation of failure factors;The application connects batch computing service job accountability module as a whole, so that each independent accountability sub-function can cooperate with each other, and jointly complete the accountability function of batch processing service.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information security, in particular to a system for accountability of batch computing cloud service jobs. BACKGROUND

[0002] Batch computing service is a distributed cloud service suitable for large-scale parallel batch processing jobs. Users provide computing requirements such as MapReduce and machine learning to batch computing service in the form of jobs. Each batch computing job can be described in the mode of "job-task-instance", a job consists of one or more tasks, and a task contains one or more instances. Each submitted job is divided into tasks with different computing logic, and the instances in the tasks use the same computing code to process different data.

[0003] Cloud service accountability is to measure the difference between cloud service and service quality contract, and to investigate the responsibility of cloud service provider for the failure of cloud service (such as security, credibility, etc.) to meet the service quality contract. "Design Patterns" points out that framework is a set of cooperating classes that make up a reusable design of a particular software class. In the invention, the framework refers to the cooperating classes that assist the interaction of the accountability modules.

[0004] Batch computing service has the characteristics of providing user required services on demand, and service accountability has become a key problem hindering its development. The batch computing service job accountability framework can construct the accountability modules into a complete accountability system. The current research has explored key accountability technologies for batch processing service job accountability, from service failure analysis to service event contact discovery, then constructing event net, and then determining event accountability and accountability, and finally tracing the accountability event net to get the accountability chain and the responsibility proportion of the nodes.

[0005] However, there are still semantic or formal gaps in the event vector representation of each accountability module when exploring different key technologies, and it is still not possible to directly incorporate each accountability module of the current research into the accountability mechanism. And it does not pay attention to the correlation between results and failures, including the correlation between failures and node accountability, and the correlation between failures and node accountability, so that the "accountability" and "accountability" results of the accountability determination, and the "responsibility proportion" in the accountability chain lack stronger responsibility persuasion.

[0006] In order to solve the responsibility attribution problem of batch computing service events, the present application provides a system for accountability of batch computing cloud service jobs. SUMMARY

[0007] In view of the deficiencies of the prior art, the present application provides a system for accountability of batch computing cloud service jobs, which solves the problems raised in the background art.

[0008] To achieve the above object, the present application is implemented by the following technical solutions: A system for accountability of batch computing cloud service jobs comprises the following steps:

[0009] A data extraction module is configured to extract batch service running status data from cloud service log data;

[0010] A failure analysis module is configured to integrate failure factors with cloud service running status, and construct cloud service running event vector samples according to subject-predicate-object sentence patterns;

[0011] A contact discovery and event net generation module is configured to receive event net vector sample data, and transform event abnormality vector samples into event failure responsibility vector samples;

[0012] A failure responsibility determination module is configured to receive failure responsibility event data and non-failure responsibility event data, and analyze the correlation between failure responsibility events and non-failure responsibility events and failure factors, respectively;

[0013] A failure responsibility tracing module is configured to receive failure responsibility events, and trace non-failure responsibility events and their correlations;

[0014] A rectification module is configured to receive failure responsibility chains, analyze the failure responsibility chains, generate failure responsibility reports, analyze non-failure responsibility events and their correlations, and give improvement suggestions.

[0015] A system for accountability of batch computing cloud service jobs comprises the following steps:

[0016] S1, batch computing service running event data extraction;

[0017] S2, failure analysis;

[0018] S3, failure event semantic conversion;

[0019] S4, event contact discovery;

[0020] S5, event contact interface;

[0021] S6, event net generation;

[0022] S7, abnormal event semantic conversion;

[0023] S8, failure responsibility determination;

[0024] S9, failure responsibility correlation analysis;

[0025] S10, failure responsibility tracing.

[0026] A system for accountability of batch computing cloud service jobs further comprises the following steps:

[0027] S1, batch computing service running event data extraction;

[0028] S11, collect cloud service log data, interpret data fields;

[0029] S12, extract batch computing service event data according to screening rules;

[0030] S2, failure analysis;

[0031] S3, failure event semantic conversion;

[0032] S31, vector construction of cloud service instance event;

[0033] S32, add failure factor dimension to event vector sample;

[0034] S4, event contact discovery;

[0035] S5, event contact interface;

[0036] S6, event network generation;

[0037] S61, instance event network generation:

[0038] S62, abnormal event network generation;

[0039] S7, abnormal event semantic conversion;

[0040] S71, construct violation vector sample;

[0041] S72, construct new event network;

[0042] S8, failure determination;

[0043] S9, failure and failure correlation analysis;

[0044] S91, extract failure vector sample;

[0045] S92, calculate the correlation degree of failure factor sequence and failure event sequence;

[0046] S93, analyze experimental results and present in the form of statistical distribution chart;

[0047] S10, failure traceability.

[0048] The present application provides a kind of batch computing cloud service job accountability system, with the following beneficial effects:

[0049] 1, the present application is the batch computing service job accountability module is connected as a whole, so that each independent accountability sub-function can cooperate with each other, and the accountability function of batch processing service is completed together.

[0050] 2, the present application unifies the expression form of each functional module interface, which is more easily extended and called by the outside world.

[0051] 3. The application establishes a responsibility mechanism integrating analysis, network construction, judgment and traceability. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the application will be described clearly and completely below. Obviously, the described embodiments are only some of the embodiments of the application, not all.

[0053] Case 1

[0054] The application provides a technical solution: a system for batch computing cloud service job accountability, comprising:

[0055] A data extraction module is configured to extract batch service running status data from cloud service log data;

[0056] A failure analysis module is configured to integrate failure factors with cloud service running status, and then construct cloud service running event vector samples according to subject-predicate-object sentence patterns;

[0057] A contact discovery and event network generation module is configured to receive event network vector sample data, and transform event abnormality vector samples into event accountability vector samples;

[0058] An accountability judgment module is configured to receive accountability event data and non-accountability event data, and analyze the correlation between accountability events and non-accountability events and failure factors, respectively;

[0059] An accountability traceability module is configured to receive accountability events, and trace non-accountability events and their correlations;

[0060] A rectification module is configured to receive accountability chains, analyze the accountability chains, generate accountability reports, analyze non-accountability events and their correlations, and give improvement suggestions.

[0061] Firstly, the batch service running status data is extracted from the cloud service log data and delivered to the "failure analysis" module after integration; secondly, the failure factor data returned by the failure analysis module is received, the failure factors are integrated with the cloud service running status, and then the cloud service running event vector sample is constructed according to the subject-predicate sentence form, and finally the event vector sample is delivered to the "contact discovery and event net generation" module; thirdly, the event net vector sample data returned by the "event contact discovery and event net generation" module is received, the event abnormal vector sample is transformed into the event failure vector sample, and then it is delivered to the failure judgment module; fourthly, the "failure event" data and "inadequate event" data returned by the "failure judgment module" are received, the correlation of the failure event and the inadequate event with the failure factors is analyzed respectively, and the failure event and the correlation are delivered to the failure tracing module, while the inadequate event and the correlation are delivered to the rectification module; finally, the failure chain is received, on the one hand, the failure chain is analyzed to generate a failure report; on the other hand, the inadequate event and the correlation are analyzed to give improvement suggestions.

[0062] Case two

[0063] The application will be further described in detail in combination with example one.

[0064] The object processed by the application is the Alibaba 2018 cluster tracking data set, which is composed of 4K machines, 9K online services and 4M batch computing services, and records 8 days of static and running tracking information. The data set records online and batch computing service logs. Since online services have higher priority than batch computing services, batch computing service jobs and tasks are prone to failure, so the application focuses on the accountability of batch computing services.

[0065] A system for accountability of batch computing cloud service jobs, comprising the following steps:

[0066] S1, batch computing service running event data extraction;

[0067] S11, collect cloud service log data and interpret data fields;

[0068] The Alibaba 2018 cluster tracking data set is collected, and the online service and batch computing service traces in the data set are recorded in 6 csv files. In the application, the Alibaba cloud service log data is denoted as CSLOG, which is formally expressed as

[0069] ;

[0070] represents machine meta information and machine event information, which belongs to file;

[0071] represents resource usage information per machine, belongs to file;

[0072] represents meta information and event information of containers, belongs to file;

[0073] represents resource usage information of containers, belongs to file;

[0074] represents instance information of batch computing workloads, belongs to file;

[0075] represents task information of batch computing workloads, belongs to file;

[0076] Log data contains multiple data fields in each file, machine meta information file is recorded as

[0077] ;

[0078] represents the unique identification of the machine in the service cluster;

[0079] represents the timestamp from the start of tracking to recording this record;

[0080] represents the first level of failure domain, two levels of failure domains are provided in this version of tracking, and for any application that needs fault tolerance, their instances should be distributed in many failure domains;

[0081] represents the second level of failure domain;

[0082] represents the core number of the machine;

[0083] represents the normalized machine capacity;

[0084] represents the state of the machine at a given timestamp;

[0085] Machine resource usage information file is recorded as , and ;

[0086] represents the unique identification of the machine in the service cluster;

[0087] represents Usage rate, value range in [0, 100] interval;

[0088] represents Capacity usage rate, value range in [0, 100] interval;

[0089] represents memory bandwidth usage rate;

[0090] represents cache miss rate per thousand instructions;

[0091] represents the number of incoming network packets;

[0092] represents the number of outgoing network packets;

[0093] represents disk usage rate;

[0094] Container meta information file is recorded as , and ;

[0095] represents the unique identification of the container;

[0096] represents the machine identification of the host where the container is located;

[0097] represents the deployment container unit, and containers belonging to the same deployment unit provide the same service;

[0098] represents the timestamp from the start of tracking to recording the record;

[0099] represents the planned share limit;

[0100] represents the planned memory capacity;

[0101] represents the state of the container at a given timestamp;

[0102] Container resource usage information file is recorded as , and represents the unique identification of the container;

[0103] represents the machine identification of the host where the container is located;

[0104] represents the time stamp from the beginning of the trace to the recording of this record;

[0105] represents the usage rate, whose value ranges in the interval [0, 100];

[0106] represents the cache miss rate per thousand instructions;

[0107] represents the clock cycles required for the container to execute one instruction at the current time stamp;

[0108] represents the capacity usage rate, whose value ranges in the interval [0, 100];

[0109] represents the memory bandwidth usage rate;

[0110] represents the disk usage rate;

[0111] represents the number of incoming network packets;

[0112] represents the number of outgoing network packets;

[0113] The instance information file of the batch computing workload is denoted as , and ;

[0114] represents the unique identifier of the instance, and the instance name is unique in the job;

[0115] represents the task name to which the instance belongs, and the task name is unique in the job, and the task name indicates the directed acyclic graph information;

[0116] represents the task type, and there are 12 task types;

[0117] represents the job name of the task to which the instance belongs, and the job is composed of one or more tasks:

[0118] represents the instance state;

[0119] represents the start time of the instance;

[0120] represents the end time of the instance;

[0121] The machine identifier that indicates the running instance;

[0122] Indicates the number of times the instance has been retried;

[0123] This indicates the total number of times the instance attempted to run;

[0124] The average of the instances Utilization rate;

[0125] The largest instance Utilization rate;

[0126] The average of the instances Utilization rate;

[0127] The largest instance Utilization rate;

[0128] The task information file for batch computation workloads is denoted as ,and ;

[0129] Indicates the task name;

[0130] Indicates the number of instances of the task;

[0131] Indicates the task type;

[0132] Indicates the job name to which the task belongs;

[0133] Indicates the task status;

[0134] Indicates the start time of the task;

[0135] Indicates the end time of the task;

[0136] Represented as the request for each instance of the task quantity;

[0137] Represented as the request for each instance of the task capacity.

[0138] S12. Extract batch computing service event data according to the filtering rules.

[0139] According to the field meaning, the fields of the batch computing service running event are extracted to form a batch service running time data, denoted as , and .

[0140] S2, failure analysis

[0141] S3, failure event semantic conversion

[0142] S31, vector construction of cloud service instance event

[0143] According to the event vector construction method in the patent "Construction method of time-varying state space vector oriented to cloud service event contact", the batch computing service event data BCSEVENT is constructed into an event vector sample, denoted as , and ;

[0144] Event vector sample is a set formed by a 5-tuple of identification dimension , time dimension , retry dimension , state dimension , and load dimension ;

[0145] The identification dimension represents the unique identification of the instance event and its attribution, denoted as , which is a five-tuple formed by device identification , job identification , task identification , and instance identification ;

[0146] The time dimension represents the time of occurrence of the instance event, denoted as , which is a two-tuple formed by instance start time and instance end time ;

[0147] The retry dimension represents the number of instance starts, denoted as , which is a two-tuple formed by instance number of attempts and instance total number of attempts ;

[0148] The state dimension represents the state of the instance event, denoted as , which is a one-tuple formed by instance state ;

[0149] The load dimension The resource load representing instance events is denoted as Based on average CPU usage CPU maximum usage MEM average usage MEM maximum usage The resulting quadruple;

[0150] S32. Add a failure factor dimension to the event vector sample;

[0151] Event vector samples represent the operational status of batch services. To discover abnormal relationships between events, failure factors are included as one dimension in the event vector samples, forming failure event vector samples, denoted as FEVES. ;

[0152] Failure event vector samples are identified by dimension Time dimension Retry State Dimension , load dimension Failure The set formed by six tuples;

[0153] Failure dimension The factors that cause instance events to fail are denoted as... It is determined by the average satisfaction level of the CPU. CPU maximum satisfaction MEM average satisfaction MEM maximum satisfaction The resulting quadruple.

[0154] S4, Event Connection Discovery

[0155] Based on the mechanism of the "Event Correlation Discovery" function module, the failure vector samples FEVS are constructed into vector pair samples to discover the relationships between events, denoted as event correlation EVENTR. ;

[0156] Event connections are formed by failure event vector samples. The ordered binary tuples formed It is an upstream event vector sample. It is a downstream event vector sample, predicate This indicates that a connection exists;

[0157] Based on the state of the event, event connections can be divided into four categories: no connection, normal connection, unknown connection, and abnormal connection.

[0158] Denote no connection as ,and where "0" means that the two instances belong to a task without dependency, i.e. no contact;

[0159] The normal contact is denoted as , and where "1" means that the depended instance B has delivered the result to the dependency instance A, i.e. normal contact;

[0160] The unknown contact is denoted as , and .

[0161] where "2" means that the depended B is in the ready state and the dependency A is in the ready state, i.e. "ready-ready" unknown contact;

[0162] "3" means that the depended B is in the ready state and the dependency A is in the running state, i.e. "ready-running" unknown contact:

[0163] "4" means that the depended B is in the running state and the dependency A is in the ready state, i.e. "running-ready" unknown contact;

[0164] "5" means that the depended B is in the running state and the dependency A is in the running state, i.e. "running-running" unknown contact;

[0165] The abnormal contact is denoted as , and .

[0166] where "6" means that the depended B is in the ready state, and the result is not delivered to the dependency A, and the dependency A is in the finalizing state, i.e. "ready-finalizing" abnormal contact;

[0167] "7" means that the depended B is in the ready state, and the result is not delivered to the dependency A, and the dependency A is in the interrupting state, i.e. "ready-interrupting" abnormal contact;

[0168] "8" means that the depended B is in the ready state, and the result is not delivered to the dependency A, and the dependency A is in the failing state, i.e. "ready-failing" abnormal contact;

[0169] "9" means that the depended B is in the running state, and the dependency A is in the finalizing state, i.e. "running-finalizing" abnormal contact;

[0170] "10" means that the depended B is in the running state and the dependency A is in the interrupting state, i.e. "running-interrupting" abnormal contact;

[0171] "11" means that the depended B is in the running state and the dependency A is in the failing state, i.e. "running-failing" abnormal contact;

[0172] "12" indicates that the dependee B is in the terminating state and the dependency A is in the ready state, i.e., a "terminating-ready" abnormal connection;

[0173] "13" indicates that the dependee B is in the terminating state and the dependency A is in the running state, i.e., a "terminating-running" abnormal connection;

[0174] "14" indicates that the dependee B is in the terminating state and the dependency A is in the terminating state, i.e., a "terminating-terminating" abnormal connection;

[0175] "15" indicates that the dependee B is in the terminating state and the dependency A is in the interrupt state, i.e., a "terminating-interrupt" abnormal connection;

[0176] "16" indicates that the dependee B is in the terminating state and the dependency A is in the failure state, i.e., a "terminating-failure" abnormal connection;

[0177] "17" indicates that the dependee B is in the interrupt state and the dependency A is in the ready state, i.e., an "interrupt-ready" abnormal connection;

[0178] "18" indicates that the dependee B is in the interrupt state and the dependency A is in the running state, i.e., an "interrupt-running" abnormal connection;

[0179] "19" indicates that the dependee B is in the interrupt state and the dependency A is in the terminating state, i.e., an "interrupt-terminating" abnormal connection;

[0180] "20" indicates that the dependee B is in the interrupt state and the dependency A is in the interrupt state, i.e., an "interrupt-interrupt" abnormal connection;

[0181] "21" indicates that the dependee B is in the interrupt state and the dependency A is in the failure state, i.e., an "interrupt-failure" abnormal connection;

[0182] "22" indicates that the dependee B is in the failure state and the dependency A is in the ready state, i.e., a "failure-ready" abnormal connection;

[0183] "23" indicates that the dependee B is in the failure state, i.e., the result has not been delivered to the dependency A, and the dependency A is in the waiting state (waiting for the response of B), i.e., a "failure-running" abnormal connection;

[0184] "24" indicates that the dependee B is in the failure state and the dependency A is in the terminating state, i.e., a "failure-terminating" abnormal connection;

[0185] "25" indicates that the dependee B is in the failure state, i.e., the result has not been delivered to the dependency A, and the dependency A is in the interrupt state, i.e., a "failure-interrupt" abnormal connection;

[0186] "26" means that both the dependee B and the dependee A have been in the failure state, i.e. "failure-failure" abnormal contact.

[0187] S5, event contact interface

[0188] The event contact interface is responsible for normalizing the output event contact of the previous "event contact discovery" module, and also serves as an extension interface open to the outside. If the external input meets the format requirements, the next event net generation module can be used.

[0189] If the event contact i.e. meets , the data of the abnormal event contact is passed to the "event net generation" module;

[0190] S6, event net generation

[0191] S61, instance event net generation

[0192] According to the mechanism of the "event net generation" function module, the event contact EVENTR is constructed into an instance event net. The event net is an ordered pair constructed by an event set and its contact set, denoted as ENET, and ;

[0193] S62, abnormal event net generation

[0194] According to the type of the instance contact, the abnormal part of the contact is retained and constitutes an abnormal event net, denoted as , and .

[0195] The predicate indicates an abnormal contact.

[0196] S7, abnormal event semantic conversion;

[0197] S71, construction of violation vector sample

[0198] The failure vector sample FEVES of the abnormal event net and the vector sample required for the failure judgment have semantic gaps. The failure vector sample FEVES lacks the semantics of the instance number, the task start time and the task end time, which leads to the lack of the judgment condition of the "violation judgment" module. Therefore, the failure sample vector FEVES needs to be converted into a violation event vector sample according to the "vectorization method of cloud service event and service level contract data". The violation event vector sample is composed of an identification dimension , a time dimension , a retry dimension , a state dimension , a CPU load dimension , a MEM load dimension and a failure factor dimension The seven-tuple formed by the components is denoted as VEVS, and

[0199] Compared with the failure event vector sample FEVES, the violation event vector sample VEVS has a change in semantics in the time dimension, representing the time when the instance event and the task event occur, denoted as , which is formed by the instance start time , the instance end time , the task start time , and the task end time , wherein the task start time , and the task end time are the added semantics in the contact sample vector;

[0200] The violation event vector sample VEVS subdivides the load dimension into a CPU load dimension and a MEM load dimension ; wherein the CPU load dimension represents the CPU resource load condition of the instance event, denoted as , which is formed by the CPU average usage and the CPU maximum usage ; the MEM load dimension represents the MEM resource load condition of the instance event, denoted as , which is formed by the MEM average usage and the MEM maximum usage ;

[0201] The violation vector is replaced with the vector in the abnormal event net to construct a new event net, and the integrated event net has the semantics required for the failure determination, denoted as .

[0202] S8, failure determination;

[0203] The “failure determination” module receives the violation event vector sample VEVS and determines whether the event is failed and the degree of failure, and obtains the failure event, denoted as FEVENT, and

[0204] The failure event is formed by the identification dimension , the time dimension , the retry dimension , the state dimension , the CPU load dimension , the MEM load dimension , the failure factor dimension , and the label dimension ​​​The set formed by the 8-tuple of components;

[0205] Representing what type of failure the example event belongs to, and the degree of failure, denoted as , where represents the failure label, , "0" indicates normal, "1" indicates failure: represents the degree of failure, , the degree of failure is in the range of [0, 1].

[0206] S9, Failure and failure correlation analysis

[0207] Affected by failure factors, batch computing services fail to complete the specified tasks during runtime, and then the service fails; In order to verify the correlation between failure and failure, the correlation between failure factors and failure events is added in the framework, and the degree of correlation is used as the basis for failure tracing

[0208] S91, extract failure vector samples

[0209] First, extract the "failure factor" sequence: "resource satisfaction" factors, such as ① CPU average satisfaction sequence, ② CPU maximum satisfaction sequence, ③ MEM average satisfaction sequence, ④ MEM maximum satisfaction sequence; "long tail" factors, such as ⑤ long tail length sequence; second, extract the "failure event" sequence: the failure probability can represent the probability of the event sample being predicted as a certain label, so use the failure probability to represent the failure degree , prepare the failure degree sequence;

[0210] S92, calculate the correlation between failure factor sequence and failure event sequence

[0211] Calculate the correlation r between failure factor sequence ① and failure event sequence ⑥.

[0212] Calculate the correlation r between failure factor sequence ② and failure event sequence ⑥.

[0213] Calculate the correlation r between failure factor sequence ③ and failure event sequence ⑥.

[0214] Calculate the correlation r between failure factor sequence ④ and failure event sequence ⑥.

[0215] Calculate the correlation r between failure factor sequence ⑤ and failure event sequence ⑥

[0216] S93, analyze the experimental results and present them in the form of statistical distribution chart.

[0217] S10, Responsibility Traceback

[0218] The "Responsibility Traceback" module receives the abnormal event net VAENET, the responsible nodes and their responsibility degrees, and converts them into a responsibility chain, denoted as BRECHAIN, and ;

[0219] The responsibility chain is an ordered pair composed of responsible nodes and edges, where N represents the set of responsible nodes, whose elements are ordered pairs of responsible events and their responsibility probabilities, represents the responsible event, represents the responsibility probability, the predicate represents that there is a responsibility relationship between the responsible events.

Claims

1. A system for batch computing cloud service job accountability, the system comprising: Comprising: a data extraction module for extracting batch service running status data from cloud service log data; a failure analysis module for integrating failure factors with cloud service running status, and then constructing cloud service running event vector samples according to subject-predicate-object sentence patterns; a contact discovery and event net generation module for receiving event net vector sample data, and transforming event anomaly vector samples into event failure vector samples; The failure judgment module is used for receiving the failure event data and the unfulfillment event data, and analyzing the correlation between the failure event and the unfulfillment event and the failure factor respectively; wherein the failure vector sample is extracted, firstly, the "failure factor" sequence is extracted: the "resource satisfaction degree" factor, such as ① CPU average satisfaction degree sequence, ② CPU maximum satisfaction degree sequence, ③ MEM average satisfaction degree sequence, ④ MEM maximum satisfaction degree sequence; the "long tail" factor, such as ⑤ long tail time length sequence; secondly, the "failure event" sequence is extracted: the failure probability can represent the probability of the event sample being predicted as a certain label, therefore, the failure probability is used to represent the failure degree , the failure degree sequence is prepared; wherein the correlation between the failure factor sequence and the failure event sequence is calculated, the correlation r between the failure factor sequence ① and the failure event sequence ⑥ is calculated; calculating the correlation degree r of failure factor sequence ② and failure event sequence ⑥; calculating the correlation degree r of failure factor sequence ③ and failure event sequence ⑥; calculating the correlation degree r of failure factor sequence ④ and failure event sequence ⑥; calculating the correlation degree r of failure factor sequence ⑤ and failure event sequence ⑥; finally analyzing the experimental results and presenting them in the form of statistical distribution graphs; a failure tracing module for receiving failure events and tracing unfulfilled events and their relevance; a rectification module for receiving failure chains, analyzing failure chains, generating failure reports, and analyzing unfulfilled events and their relevance to give improvement suggestions.

2. The system for batch computing cloud service job accountability of claim 1, wherein, Comprising the following steps: S1, batch computing service running event data extraction; S2, failure analysis; S3, failure event semantic conversion; S4, event contact discovery; S5, event contact interface; S6, event net generation; S7, anomaly event semantic conversion; S8, failure determination; S9, failure-failure correlation analysis; S10, failure tracing.

3. The system for batch computing cloud service job accountability of claim 2, wherein, Further comprising the following steps: S1, batch computing service running event data extraction; S11, collecting cloud service log data and interpreting data fields; S12, extracting batch computing service event data according to filtering rules; S2, failure analysis; S3, failure event semantic conversion; S31, vector construction of cloud service instance events; S32, adding failure factor dimensions to event vector samples; S4, event contact discovery; S5, event contact interface; S6, event net generation; S61, instance event net generation: S62, abnormal event net generation; S7, anomaly event semantic conversion; S71, constructing violation vector samples; S72, constructing new event nets; S8, failure determination; S9, failure-failure correlation analysis; S91, extracting failure vector samples; S92, calculating the correlation degree of failure factor sequence and failure event sequence; S93, analyzing the experimental results and presenting them in the form of statistical distribution graphs; S10, failure tracing.