JP6674039B2 - Stream data distribution processing method, stream data distribution processing system, and storage medium - Google Patents

Description

  The present invention relates to a distributed processing method for stream data processing.

  Real-time information generated continuously at high rates, such as automated stock trading, advanced traffic information processing, network microburst detection, IoT (Internet of Things) sensor data, and social network analysis In view of the growing demand for instantaneous actions to be taken upon occurrence of important events, stream data processing (hereinafter, stream processing) that realizes real-time processing of high-rate data has attracted attention.

  Stream processing is a general-purpose middleware technology that can be applied to various types of data processing, so real-time data can be processed in real-time while responding to sudden changes in the business environment that would not be possible by building a system for individual projects. That can be reflected in business.

  The stream targeted for the stream processing is time-series data in which tuples, which are data with a time stamp, arrive continuously (or intermittently). When the user of the stream processing defines a monitoring rule for this stream as a query, the query registration interface of the stream processing server converts the query definition into a query graph.

  The query graph is a directed graph in which a processing unit called an operator is a node and a tuple queue between the operators is an edge. In stream processing, individual tuples constituting an input stream are processed in a data flow by passing through a query graph.

  Since the stream processing is a data flow type processing, it is possible to divide a query graph into multiple stages and perform distributed processing in a pipeline manner by using a plurality of computational resources. Also, as in the case of a chart analysis of stock prices by brand, queries that aggregate time series data of a large number of independent series can be distributed in parallel by dividing the data by series. . As described above, in the stream processing, by appropriately performing the distributed processing according to the processing of the query, it is possible to improve the throughput.

  In addition, when stream processing is applied to machine control such as monitoring and controlling manufacturing equipment in a factory based on analysis of IoT sensor data, or collision avoidance based on inter-vehicle communication of an automobile, the latency of tuples is reduced. It becomes important. In particular, distributed processing in which multiple stream processes cooperate in order to analyze sensor data from a bird's eye view of the entire manufacturing process, or to cooperate with multiple vehicles to avoid automobile collisions, not limited to individual manufacturing equipment And a reduction in latency.

  Patent Document 1 is known as a conventional technique of stream processing. Patent Literature 1 discloses a method for improving the throughput of a stream data processing system for query processing of stream data. “The technique described in this specification is based on a declarative description of a data stream processing task. And a code generation framework for creating highly optimized distributed stream processing applications.In accordance with one or more embodiments of the present invention, developers may be aware of performance impacts that are likely to be present in distributed systems. Applications can be built using fine-grained stream operators without having to use the techniques described in the present application. Avoid the inflexibility that arises.The typical code generation framework described herein Click so as to minimize the communication overhead can be assigned an application to execution units appropriate size. "Is described as.

  Patent Document 2 is known as a conventional technique of stream processing. Patent Literature 2 discloses a method for reducing latency related to query processing of stream data in a stream data processing system. “A stream data processing system constructs a single operator graph from an execution tree of a plurality of queries. An external firing operator that inputs data outside the system and an internal firing operator that generates data in a timed manner so that the execution of the streaming operation proceeds in one way from the input to the output. And the operator execution control unit repeats the process of completing the processing in the operator graph at the time according to the determined operator execution order, starting from the operator at the earliest firing time. " .

  Further, Patent Literature 2 discloses a method of executing a recursive query using a delay operation, and illustrates a state in which a delay operation, which is a delay operator for shifting the time stamp of a stream tuple to the future, is inserted. Since the time delay causes the output of the streaming operation to return to its own input with a time lag, the output of the streaming operation itself has an effect on the increase or decrease in the relation at time t1. In this way, by interposing a delay operation in the middle of a recursive query, it is possible to avoid a deadlock of a streamed operation and realize a recursive query. "

US Patent Application Publication No. 2012/0259910 JP 2014-67456 A

  However, stream processing, like a sliding window, continuously cuts out a data set within a certain range in the past from a continuous data flow, and enables continuous processing using a declarative query language. In other words, the process for the change in the time axis direction is implicitly included, and the man-hour for developing a real-time application requiring complicated control is greatly reduced.

  However, in the development of such a real-time application, due to debugging, the reproducibility of processing (that is, even if the same tuple sequence including the time stamp is processed many times, the tuple sequence of the resulting stream remains unchanged Is essential.

  In order to maintain the reproducibility of processing, distributed processing requires merging the result streams generated by multiple stream processing processes running on different computers while maintaining the time order, and propagating the processing. There is. Such a merge process involves queuing of result streams coming from a plurality of stream processing processes, and therefore, in a case where the result streams are merged in multiple stages, for example, the latency of the stream processing time and the accumulation of the inter-process communication time is reduced. An increase occurs.

  Further, in a case where the input / output relationship between a plurality of stream processing processes includes a cycle, the progress of the query processing may hang due to the interdependency between the two processes. For example, in cooperative processing such as collision avoidance of a car, the input / output circulation is necessarily included.

  Furthermore, in the conventional distributed execution technology of stream processing, the input / output relationship between queries operating on each process needs to be determined according to the input / output relationship between stream processing processes. Therefore, it is necessary for the user to design the connection relationship between the queries, that is, the query graph, while being aware of the execution position (the computer to be executed) of the process for executing each query. The restriction on the connection relationship between queries significantly deteriorates the ease of user-defined distributed queries.

  In addition, even if a part of the connection relationship between queries is changed for the query graph that has started distributed execution, the entire query graph is stopped, the query graph is redefined, and the processing of the query graph is resumed. There is a need. Accordingly, the analysis process is stopped during that time.

  In the technique of Patent Literature 1, specification of an execution position of each query can be included in the definition of the query graph. However, there is a problem in that it is necessary to define the query graph in consideration of the execution position and to stop and restart the processing accompanying the change of the query graph.

  In the technique of Patent Document 2, the processing of the query graph can be executed with low latency in a single thread of a single process on a single computer, and a recursive query among a plurality of queries can be executed. However, there is a problem in that the same effect cannot be obtained in distributed execution of a query graph across a plurality of computers because execution of a single thread is premised.

  The present invention has been made in view of the above problems, and makes it possible to simplify a query definition and to dynamically change a connection relationship between queries without stopping processing in distributed execution of stream processing. The purpose is to do.

  What is claimed is: 1. A distributed data processing method for a stream data in which a computer having a processor and a memory processes time-stamped tuples, the computer comprising: a query for processing an input tuple; and a delay time for a tuple output by the query. A first step of receiving a query definition including a recursive loop for recursively inputting the output of the variable delay operation to the query, wherein the calculator comprises: an input destination of the tuple; And a second step of receiving a configuration change instruction including a delay time, and the computer releases the recursive loop to change the output of the variable delay operation to an output destination of the configuration change instruction, A third step of changing an input destination of the tuple to an input destination of the configuration change instruction and setting a delay time of the configuration change instruction to the variable delay operation; And a fourth step in which the computer receives the tuple from the input destination, inputs the tuple subjected to the predetermined processing in the query to the variable delay calculation, and adds the delay time to a time stamp of the tuple. And a fifth step in which the computer outputs a tuple to which the delay time has been added to the output destination.

  According to the present invention, it is possible to simplify the query definition and change the connection relationship between the queries dynamically without stopping the processing while maintaining the reproducibility of the stream processing.

FIG. 1 is a block diagram illustrating an example of a computer system in which a stream processing server operates according to a first embodiment of the present invention. 1 is a block diagram illustrating a first embodiment of the present invention, in which a computer system is applied to monitoring processing of a production line in a factory. 1 is a block diagram illustrating a first embodiment of the present invention, in which a computer system is applied to inter-vehicle communication of an automobile. 4 is a script showing a conventional example of the present invention and showing an example of a continuous query for monitoring a production line of a factory. It is a block diagram which shows the prior art example and which shows an example of the query graph of a continuous query which monitors the production line of a factory. It is a block diagram which shows the conventional example and which shows an example of cooperation between stream processing processes which monitors the production line of a factory. It is a block diagram which shows the prior art example and which shows an example of cooperation between stream processing processes which implement | achieves the communication between vehicles of a motor vehicle. FIG. 4 is a block diagram illustrating the first embodiment of the present invention, in which a recursive query into which a variable delay operation is inserted is applied to a process of monitoring a production line in a factory. 6 is a script illustrating the example 1 of the present invention and illustrating an example of a recursive query into which a variable delay operation is inserted. 6 is a script illustrating the example 1 of the present invention and illustrating an example of a recursive query into which a variable delay operation is inserted. FIG. 1 is a block diagram illustrating a stream processing process according to a first embodiment of the present invention. FIG. 4 is a block diagram illustrating Example 1 of the present invention and illustrating an example of a configuration change instruction. FIG. 7 illustrates the first embodiment of the present invention, and illustrates an example of a result tuple queue. FIG. 6 illustrates the first embodiment of the present invention, and illustrates an example of a configuration change instruction queue. FIG. 9 illustrates the first embodiment of the present invention, and illustrates an example of a delay time management table for each output destination. FIG. 3 illustrates the first embodiment of the present invention, and illustrates an example of a variable delay calculation execution unit. 5 is a flowchart illustrating an example of a process performed by the variable delay calculation execution unit according to the first exemplary embodiment of the present invention. FIG. 4 is a block diagram illustrating Example 1 of the present invention and illustrating an example of cooperation via a delay operation between stream processing processes for monitoring a production line in a factory. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram illustrating an example 1 of the present invention and illustrating an example of cooperation via a delay operation between stream processing processes for realizing inter-vehicle communication of an automobile. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram illustrating an example 1 of the present invention and illustrating an example of cooperation via a delay operation between stream processing processes for realizing inter-vehicle communication of an automobile. FIG. 10 is a block diagram illustrating a stream processing process according to a second embodiment of the present invention.

  Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings.

  FIG. 1 is a block diagram illustrating an example of a computer system (stream processing distributed execution system 801) on which a stream processing server operates. The computer system includes a plurality of stream processing servers 100-1 to 100-n, a stream processing distributed execution management server 1001 that distributes a query definition 132 to the stream processing servers 100-1 to 100-n, a query definition 132, and a configuration change. It includes a host computer 130 that issues instructions 815, a data source 120 that supplies tuples 121, a data receiver 140 that receives result tuples 141, and a network 150 that connects each computer.

  The stream processing server 100-1 is a computer including a CPU 101, a memory 103, a network interface 105, a storage 106, and a bus 104 connecting these. Note that the configuration of the stream processing servers 100-1 to 100-n is omitted, and thus redundant description will be omitted.

  On the memory 103, a stream processing process 110-1 that defines a logical procedure of the stream processing is arranged. The stream processing process 110-1 is an execution image (a program or the like) executable by the CPU 101.

  The stream processing server 100-1 is connected to the network 150 via the network interface 105. The stream processing server 100-1 receives the query definition 132 and the configuration change command 815 from the stream processing distributed execution management server 1001 connected to the network 150.

  On the host computer 130 connected to the network 150, the query registration interface 131 outputs the query definition 132 specified by the user to the stream processing distributed execution management server 1001.

  The host computer 130 has a CPU 12 and a memory 13 and an input / output device 133 including an input device such as a display device and a keyboard. The host computer 130 loads and executes a query registration interface 131 in the memory 13, and executes a query definition 132 and a configuration change instruction 815. Edit, etc., and assign a stream processing process.

  Next, the stream processing process 110-1 constructs a query graph that can execute stream processing according to the query definition 132. Thereafter, when the stream processing server 100-1 receives the tuple 121 transmitted from the data source 120 connected to the network 150, the stream processing server 100-1 processes the time-series tuple 121 with a query according to the query graph. Then, a result tuple 141 is generated.

  As a result, the tuple 141 is transmitted to the data receiver 140 connected to the network 150. The storage 106 stores the text file of the query definition 132 once received, in addition to the execution image of the stream processing process 110-1. The stream processing process 110-1 can also load the file of the query definition 132 from the storage 106 at the time of startup and construct a query graph.

  A plurality of stream processing servers 100-1 to 100-n are connected to the network 150. Each of the stream processing servers 100-2 to 100-n has the same components as the stream processing server 100-1, and holds an execution image of the stream processing process 110-1 in the memory 103 included in each of them.

  In the following description, when specifying each of the stream processing servers, reference numerals 100-1 to 100-n including "-" are used, and the entire stream processing server is indicated by reference numeral 100 without "-". The same applies to reference numerals of other components.

  FIG. 2 is a block diagram illustrating an example in which the computer system is applied to monitoring of a production line in a factory. The stream processing servers 101-1 to 101-5 are computers that operate in addition to the manufacturing equipment, and execute stream processing processes 110-1 to 110-5, respectively. Sensors 200-1 to 200-4 are connected to the stream processing servers 101-1 to 101-4, respectively, and sensor data generated by each sensor is analyzed by stream processing processes 110-1 to 110-4, respectively. Is done.

  The sensors 200-1 and 200-4 have acquired data of the products 220 and 221 flowing on the belt conveyor 210, and the sensors 200-2 and 200-3 have data of the products 222 and 223 flowing on the belt conveyor 211. , 224 of the data.

  In the manufactured product 220, data is first acquired by the sensor 200-1 and then several seconds later, data is acquired by the sensor 200-4. The same applies to the product 221. Further, the product 222 is obtained by the sensor 200-2 several seconds after the data is first obtained by the sensor 200-2. The same applies to the products 223 and 224.

  The stream processing processes 110-1 to 110-5 calculate data analysis values such as an average value, a maximum value, and a minimum value of the sensor data, and collectively collect analysis values of the entire production line in cooperation with each other.

  The analysis values are sequentially aggregated from the stream processing processes 110-1 and 110-2 located upstream of the production line to the stream processing process 110-5 located downstream. Further, each stream processing process 110 monitors the occurrence of abnormal data that may lead to the occurrence of defective products. In view of the necessity of data analysis across a plurality of manufacturing devices, the connection relationship between the stream processing processes 110 is defined according to the flow of the production line. When the product is changed and the flow of the production line or the analysis method for detecting occurrence of abnormal data is changed, the connection relationship between the stream processing processes 110 is also changed. This connection relationship can be changed by a configuration change command 815 described later.

  FIG. 3 is a block diagram showing an example in which the computer system of FIG. 1 is applied to inter-vehicle communication of an automobile. The stream processing servers 101-1 to 101-4 included in each of the vehicles 300-1 to 300-4 are, for example, in-vehicle microcomputers, and execute stream processing processes 110-1 to 110-4, respectively. The stream processing processes cooperate with each other, acquire data such as position information from the GPS 230 and various sensors included in each vehicle 300, and exchange information such as position information and speed with each other to run each vehicle 300. Determine the planned route. The connection relationship between the vehicles is changed at any time according to the mutual positional relationship.

  FIG. 4 is a script showing a conventional example, and showing an example of a continuous query for monitoring a production line in a factory. A query 400 that defines the analysis of the entire production line is shown. At the beginning, register stream clauses S1 to S4 in the figure that define data acquired by the sensors 200-1 to 200-4 as an input stream are listed. I have.

  The query Q3 (query Q3) indicates an analysis query executed in the stream processing process 110-3, and the result of the analysis query Q2 (query Q2) operating in the stream processing process 110-2 and the sensor This means that the input stream S3 of 200-3 is matched and analyzed.

  The query Q12 (query Q12) merges the result of the analysis query Q1 operated in the stream processing process 110-1 and the result of the analysis query Q2. The query Q4 represents an analysis query that operates in the stream processing process 110-4, and means that the output of the query Q12 and the input stream S4 of the sensor 200-4 are matched and analyzed by the from clause. .

  The annotation “exec_on” inserted in the middle of the query is an execution node designation annotation that defines the identifier of the stream processing server that executes each query (and the stream processing process that runs there). The query Q3 is executed by the stream processing process 110-3, and the queries Q12 and Q4 are executed by the stream processing process 110-4.

  Although the description is omitted in the first embodiment, the definitions of the queries that operate in the other stream processing processes 110-1, 110-2, and 110-5 are similarly defined.

  FIG. 5 is a block diagram showing an example of a query graph of a continuous query for monitoring a production line of a factory. The input / output relationship between the input streams 501 to 503 and the analysis queries 511 to 515 is determined according to the definition of the analysis query that operates in each stream processing process in the query 400 illustrated in FIG.

  The input streams 501 to 504 correspond to the streams S1 to S4 of the query 400 in FIG. The analysis queries 511 to 515 represent queries defined to operate in the stream processing processes 110-1 to 110-5, respectively, by the annotation of the query 400.

  For example, the analysis query 513 indicates Q3, and the analysis query 514 indicates Q12 and Q4. The annotations 521 to 525, which are components of the execution node designation annotation 520, correspond to the execution processes (stream processing processes 110-1 to 110-5) of the analysis queries 511 to 515, respectively.

  FIG. 6 is a block diagram showing a conventional example, in which processing for monitoring a production line at a site is linked between stream processing processes.

  The stream processing processes 110-1, 110-2, and 110-3 operate on the stream processing servers 100-1, 100-2, and 100-3, respectively, and the sensors 200-1 and 200 included in the stream processing servers 100 respectively. -2 and 200-3 are analyzed. Each sensor 200 generates, as stream tuples 631 to 639, a combination of a time stamp indicating the data acquisition time and the sensor value of the detected product, and sends the combination to the stream processing processes 110-1, 110-2, and 110-3. input.

  The stream processing processes 110-1, 110-2, and 110-3 analyze the sensor data, generate analysis result tuples 641, 642, and 643, respectively, and transmit them to the stream processing process 110 located downstream thereof.

  The analysis result tuples 641, 642, and 643 are combinations of the time stamp of the sensor data and the analysis result. At this time, each stream processing process 110 analyzes not only the sensor data but also the analysis result tuple of the stream processing process 110 located upstream of itself. In addition, the past analysis result tuples received from the upstream process are secured in an amount necessary for the matching. The analysis result tuples 644 to 647 are tuples reserved for this purpose.

  The sensors 200-1, 200-2, and 200-3 acquire data of products 651 to 655 flowing through the production line 650. The time difference between when the data is acquired by the sensor 200-1 and when the data is acquired by the sensor 200-2 is 3 seconds. After the data is acquired by the sensor 200-2, the data is acquired by the sensor 200-3. The time difference before the execution is 2 seconds.

  For example, the time when the sensor 200-1 acquires the data of the product 651 is 9:05:05, and at that time, a stream tuple 635 is generated. After that, the time when the sensor 200-2 acquires the data of the product 651 is 9:05:08, and at that time, a stream tuple 638 is generated. Thereafter, the time when the sensor 200-3 acquires the data of the product 651 is 9:05:10, and at that time, a stream tuple 639 is generated.

  At time 9:05:10, sensors 200-1, 200-2, and 200-3 acquire data of manufactured products 655, 653, and 651, respectively, and generate stream tuples 631, 636, and 639, respectively. The value indicated by the square attached to each stream tuple represents real time. For example, each sensor 200 acquires data at a timing of a second with an accuracy up to the second decimal place.

  It is assumed that it takes about 0.3 seconds in real time after each stream tuple is analyzed by the stream processing process 110 until the analysis result reaches the next stream processing process. For example, the stream tuple 631 generated by the sensor 200-1 at 9: 05: 10.00 is analyzed by the stream processing process 110-1, and the analysis result tuple 641 as a result reaches the stream processing process 110-2. The time when it was performed is 9: 05: 10.32. In this example, 0.32 seconds is the time required for the processing of the stream processing server 100-1.

  Here, in the stream processing process 110-2, the analysis of the stream tuple 636 generated by the sensor 200-2 at 9: 05: 10.00 cannot be started from 9: 05: 10.00. This is because the analysis result tuple from the stream processing process 110-1 including the time stamp 9:05:10 or later of the stream tuple 636 has not been received at 9: 05: 10.00. .

  The analysis result tuple received by the stream processing process 110-2 most recently from the stream processing process 110-1 is 645, and its time stamp is 9:05:09. Therefore, there is a possibility that an analysis result tuple including a time stamp of 9:05:09 to 9:05:10 may arrive from the stream processing process 110-1, and the stream tuple 636 is placed before the analysis result tuple. Must be banned. In order to realize strictly reproducible stream processing according to the time order, such execution control is required.

  The timing at which the stream processing server 100-2 starts the analysis at 9:05:10 is held until 9: 05: 10.32 when the analysis result tuple 641 arrives, so the analysis result tuple at 9:05:10 The timing when the 642 reaches the stream processing processes 110-2 to 110-3 is 9: 05: 10.61, which is about 0.3 seconds later. Similarly, the timing at which the analysis result tuple 643 of 9:05:10 by the stream processing process 110-3 reaches the subsequent stream processing process at 110-4 is 9: 05: 10.93, which is about 0.3 seconds later. become. As described above, as the production line becomes longer, the timing at which the analysis result is output is delayed.

  FIG. 7 is a block diagram showing an example of a conventional stream processing inter-process cooperation for realizing inter-vehicle communication of an automobile. In this conventional example, stream processing processes 110-1 and 110-2 operate on stream processing servers 100-1 and 100-2, which are vehicle-mounted microcomputers, respectively, in the same manner as in the present embodiment. The position and speed tuples 731 and 732 representing the position information and speed information obtained by various sensors of the vehicle on which the is mounted are analyzed, and the predicted position tuple 770 representing the predicted position of the own vehicle in a few seconds later is analyzed. , 771 and transfer them to another vehicle. At this time, each stream processing process 110 analyzes not only the position / speed tuples 731 and 732 but also the predicted position tuples 770 and 771 of other vehicles.

  The position / velocity tuples 731 and 732 are acquired every second, a time stamp at the time of acquisition is given, and the acquired time stamp is input to the stream processing process 110. Also, the time stamps having the same values as the position / velocity tuples 731 and 732 from which the predictions are made are given to the predicted position tuples 770 and 771.

  Here, the stream processing process 110-1 includes a position / velocity tuple 731 acquired at 9: 05: 10.00 and including a time stamp of 9:05:10, and a stream processing process 110-2. A predicted position tuple 770 including a time stamp of 9:05:10 is matched to generate a predicted position tuple, which is sent to the stream processing process 110-2.

  Therefore, this processing is suspended until the predicted position tuple 770 arrives at the stream processing processes 110-2 to 110-1. On the other hand, also in the stream processing process 110-2, the processing of the position / velocity tuple 732 is suspended until the predicted position tuple 771 arrives from the stream processing process 110-1. Therefore, the generation of the predicted position tuple 770 is suspended. .

  As described above, the stream processing process 110-1 and the stream processing process 110-2 of the conventional example need each other's analysis results of 9:05:10 to perform the analysis processing of 9:05:10. There is a problem that hang-up (or deadlock) occurs due to the interdependency.

  FIG. 8 is a block diagram showing an example in which a recursive query into which a variable delay operation has been inserted is applied to monitoring processing of a production line in a factory. The stream processing distributed execution system 801 manages the stream processing servers 100-1 to 100-5. In each stream processing server 100, stream processing processes 110-1 to 110-5 operate, respectively.

  The query definition 132 received by the stream processing distributed execution system 801 includes a query graph defined by a file including the query definition 132, including an input stream 811, an analysis query 812, and a variable delay operation 813. Then, the query definition 132 needs to form a recursive query (recursive loop 820) in which the output of the analysis query 812 becomes the input to the analysis query 812 again via the variable delay operation 813.

  In the stream processing distributed execution system 801, for example, when the stream processing distributed execution management server 1001 receives the query definition 132, the query distribution unit 1002 distributes the stream processing to the stream processing processes 110-1 to 110-5. At this point, the query graphs on each stream processing process 110 operate independently.

  In this state, the configuration change instruction 815 including the input server identifier (or data source identifier), output server identifier (or data receiver 140 identifier), delay time, and time stamp (execution start time). For example, when the stream processing server 100 is input from the host computer 130 to the stream processing distributed execution system 801, the query graph is reset in the stream processing server 100.

  That is, when the stream processing process 110 receives the configuration change instruction 815, the recursive loop 820 of the query graph is disconnected, and the input stream 811, the analysis query 812, and the variable delay operation 813 are arranged in one way. Then, the output of the variable delay operation 813 is input to the analysis query 812 of the downstream query graph.

  The tuple transmission path is from the stream processing process 110 running on the stream processing server 100 indicated by the input server identifier to the stream processing process 110 running on the stream processing server 100 indicated by the output server identifier. Is set.

  When transmitting the analysis result tuple generated by the stream processing process 110 on the input side to the stream processing process 110 on the output side, the time stamp of the tuple is changed to a time stamp obtained by adding the value of the delay time. The effect of the variable delay calculation will be described later in comparison with the conventional example shown in FIGS.

  A query graph that performs an analysis equivalent to the query graph shown in FIG. 5 can be defined by the query definition 132. This definition states that each analytics query 812 processes the input stream and the result tuple of another analytics query 812 to generate an analytic result and outputs it to another analytics query 812 (downstream analytics query). doing. The connection relationship between the analysis queries 812 can be dynamically defined or changed by the configuration change instruction 815.

  As described above, the query definition 132 including the recursive query is distributed to each stream processing process 110, and the input stream 811 given to the analysis query 812 by giving the configuration change instruction 815 and the output destination of the variable delay operation 813 are It is possible to decide.

  Accordingly, the input / output relationship and the delay time of the analysis query 812 can be dynamically changed without stopping the stream processing process 110, and the availability of the stream processing system can be improved. The configuration change instruction 815 makes it possible to change the input / output relationship and the delay time of the analysis queries 812 of many stream processing processes 110 at any time, so that it is possible to easily deal with changes in production equipment and products. It becomes possible.

  Further, since it is sufficient that at least one query definition 132 is distributed to each stream processing process 110, real-time analysis application can be easily developed.

  Note that the input-side server identifier of the configuration change instruction 815 can specify one or more computers or devices such as another stream processing server 100, sensor 200, or data source 120. The output server identifier of the configuration change instruction 815 can specify one or more computers such as the stream processing server 100 and the data receiver 140 that transmit tuples. The data source 120 may be configured by a device that outputs tuples such as a computer and a sensor network.

  9A and 9B are scripts illustrating an example of a recursive query into which a variable delay operation has been inserted. The beginning of the query 900 in FIG. 9A describes an SX register stream clause that defines data acquired by the sensors 200-1 to 200-4 as an input stream.

  The query QX defines an analysis query that operates in the stream processing processes 110-1 to 110-5. The query QXD defines a process of adding a variable delay operation to the result of the query QX, and the description <variable> defines a variable delay operation that does not have a fixed delay time definition. ing.

  The query QX receives the result of the query QXD, that is, both the stream in which the result of the query QX itself is delayed and the stream SX as input, generates an analysis process, and again uses the query QXD as the input of the query QX itself. Recursive processing is defined. A query that performs an analysis that is equivalent to query 400 can be defined by query 900.

  The query 901 of FIG. 9B shows an example in which the analysis query is composed of many register query clauses. An analysis query is formed by a series of queries from the query QA to the query QX, and the result of the last query QX is delayed by the query QXD and input again as the result of the query QA. Such a query can also be input to the stream processing distributed execution system 801 as the query definition 132.

  FIG. 10 is a block diagram showing a configuration of the stream processing process 110. The stream processing distributed execution system 801 includes a stream processing distributed execution management server 1001 and a plurality of stream processing servers 100 that execute a stream processing process. The stream processing distributed execution management server 1001 receives, for example, the query definition 132 from the host computer 130 via the query registration interface 131 and the like, and distributes the query definition 132 to the stream processing process 110, and, for example, from the host computer 130. It includes a configuration change command distribution unit 1003 that receives the configuration change command 815 and distributes it to each stream processing process 110, and a server table 1004 in which an identifier and an address of the stream processing server 100 are set in advance.

  It is assumed that the stream processing server 100-1 and the stream processing server 100-2 are specified in the input server identifier and the output server identifier of the configuration change instruction 815, respectively.

  The stream processing processes 110-1 and 110-2 are respectively composed of query processing engines (query processing units) 1011 and 1031, query analysis units 1012 and 1032, tuple input adapters 1016 and 1036, and a configuration change instruction receiving unit. 1022, 1041.

  Each of the query processing engines 1011 and 1031 has query graphs 1013 and 1033, result tuple queues 1014 and 1034 for storing the result tuples of the query graph, and execution time management units 1015 and 1035 for managing the execution time of the query graph. .

  The query analysis units 1012 and 1032 receive the query definition described in the query definition 132 from the query distribution unit 1002, parse the query definition and construct a query graph, and generate a query graph in the query processing engines 1011 and 1031 respectively. Input as query graphs 1013 and 1033.

  The configuration change instruction receiving units 1022 and 1041 are composed of a configuration change instruction receiving unit 1022 located in the input stream processing process 110-1 and a configuration change instruction receiving unit 1041 located in the output stream processing process 110-2. Processing contents are different.

  The configuration change instruction receiving unit 1041 of the output-side stream processing process 110-2 operates between the input adapter 1036 and the output adapter 1017 so as to receive the tuple output by the server 100-1 specified by the input-side server identifier. Establish a communication connection. On the other hand, the configuration change instruction receiving unit 1022 of the input-side stream processing process 110-1 also outputs the tuple to the server 100-2 specified by the output-side server identifier, so that the tuple is output to the server 100-2. Establish a communication connection with

  The stream processing process 110-1 on the input side stores the configuration change commands 815 received by the configuration change command receiving unit 1022 in the order of time stamps, and the configuration tuple queue 1023 and the result tuple stored in the result tuple queue 1014. , A variable delay calculation execution unit 1024 that requests the output adapter 1017 to transmit to another server, and a delay time management table 1025 that stores different delay time widths for each output destination. Have. Note that the variable delay calculation execution unit 1024 corresponds to the variable delay calculation 813 shown in the query definition 132 in FIG.

  The settings of the output adapter 1017 and the input adapters 1016 and 1036 may be obtained by referring to the server table 1004 and acquiring the address corresponding to the server identifier. The server table 1004 may be configured with the server identifier and the IP address as one entry. Also, the stream processing distributed execution management server 1001 may distribute the server table 1004 to each stream processing server 100 so that each stream processing process 110 refers to the server table 1004.

  Further, the stream processing process 110 does not receive, for example, a configuration change command 815 input from the host computer 130, but autonomously determines a configuration change following a change in an analysis target. A portion 1021.

  The configuration change autonomous determination unit 1021 generates the determined configuration change as the configuration change command 815, and transmits it to the configuration change command receiving units 1022 and 1041 provided on both the input side and the output side. The configuration change autonomous determination unit 1021 may be arranged in any of the stream processing processes 110 on the input side and the output side.

  The algorithm used by the configuration change autonomous determination unit 1021 to determine the configuration change is not fixed to a specific procedure, and can be exchanged according to the analysis target. Therefore, the configuration change autonomous determination unit 1021 has an API for a configuration change determination algorithm, and can incorporate a predetermined algorithm implemented according to the API.

  With the above configuration, the stream processing server 100-1 processes the received input stream with the analysis query, and outputs the tuple on which the delay operation has been performed to the stream processing server 100-2 as the result tuple 141. The stream processing server 100-2 receives the result tuple 141 from the stream processing server 100-1 in addition to the input stream, performs processing using the analysis query, and outputs the result tuple 141.

  In the illustrated example, the delay calculation execution unit and the output adapter of the stream processing server 100-2 do not function and are not illustrated. However, when an analysis query is further executed downstream of the stream processing server 100-2, the variable delay calculation execution unit and the output adapter are enabled.

  Each functional unit of the query analysis unit 1012, the configuration change autonomous determination unit 1021, the configuration change instruction receiving unit 1022, the variable delay calculation execution unit 1024, and the query processing engine 1026 of the stream processing process 110 is a memory 103 as a program. Is loaded.

  The CPU 101 operates as a functional unit that provides a predetermined function by performing processing according to a program of each functional unit. For example, the CPU 101 functions as a device selection query analysis unit 1012 by performing processing according to a query analysis program. The same applies to other programs. Further, the CPU 101 also operates as a functional unit that provides each function of a plurality of processes executed by each program. The computer and the computer system are devices and systems including these functional units.

  FIG. 11 is a block diagram illustrating an example of the configuration change instruction 815. The configuration change instruction 815 indicates that the instruction is to be applied to a result tuple including a time stamp after 9:05:10, and that the input side server and the output side server are 100-1 and 100-2, respectively. And a delay time of 3 seconds.

  The configuration change instruction distribution unit 1003 of the stream processing distributed execution management server 1001 includes an input side configuration change instruction 1101 for delivering the configuration change instruction 815 to the configuration change instruction reception unit 1022 of the input side stream processing server 100-1; Is decomposed into an output-side configuration change instruction 1102 to be delivered to the configuration change instruction receiving unit 1041 of the stream processing server 100-2 on the side.

  Upon receiving the configuration change instruction 815, the stream processing servers 100-1 and 100-2 can dynamically switch the input / output relationship and the delay time at the designated time.

  12A to 12D are diagrams illustrating an example of data processed by the variable delay calculation execution unit. FIG. 12A is a diagram illustrating an example of the result tuple queue 1014. The result tuple queue 1014 indicates a state in which the result tuples 1211, 1212, 1213, and 1214 are stored.

  FIG. 12B is a diagram illustrating an example of the configuration change instruction queue 1023. The example in which the input side configuration change instructions 1101 and 1231 are stored in the configuration change instruction queue 1023 is shown. In the illustrated example, the output side is switched to the stream processing server 100-2 at 9:05:10, the delay time is set to 3 seconds, and the output side is switched to the stream processing server 100-3 at 9:05:20. An example in which the delay time is changed to 4 seconds is shown.

  FIG. 12C is a diagram illustrating an example of the delay time management table 1025 for each output destination. The output destination-specific delay time management table 1025 has two entries, and the values of the output destination 1201 and the delay time 1202 are as shown in the figure.

  FIG. 12D is a diagram illustrating a state of the variable delay calculation execution unit 1024. The variable delay calculation execution unit 1024 takes out the first tuple 1211 of the result tuple queue 1014 and sets the time stamp of the tuple 1211 to 3 seconds in accordance with the value of the column of the delay time 1202 of the output destination-specific delay time management table 1025. Two tuples 1221 and 1222 with 5 seconds added are generated and sent to the stream processing servers 100-3 and 100-4 (server 3 and server 4 in the figure) indicated by the column of the output destination 1201 respectively.

  Here, the variable delay calculation execution unit 1024 compares the time stamp 9:05:10 of the configuration change instruction 1101 and the time stamp 9:05:09 of the result tuple 1211 located at the head of the configuration change instruction queue 1023. Since the former is the future, the configuration change instruction 1101 is held in the queue. Thereafter, when the variable delay calculation execution unit 1024 processes the result tuple 1212, the time stamp 9:05:10 of the tuple is after the time stamp 9:05:10 of the configuration change instruction 1101, In order to store the contents of the configuration change instruction 1101 in the output destination-specific delay time management table 1025, an entry of the server 2 (stream processing server 100-2) is added as an output destination, and the number is made three. As a result, the tuple 1212 is sent to the three stream processing servers 100-2 to 100-4.

  FIG. 13 is a flowchart illustrating an example of a process performed by the variable delay calculation execution unit 1024. The processing of FIG. 13 shows the flow of processing of one tuple in the variable delay calculation execution unit 1024 of the stream processing server 100-1, starting from step 1300 and ending with step 1311. This process is executed each time a tuple arrives.

First, in step 1301, the variable delay calculation execution unit 1024 acquires a head tuple from the result tuple queue 1014. In subsequent step 1302, variable delay calculation execution section 1024 determines whether or not configuration change instruction 815 exists in configuration change instruction queue 1023.
If there is no configuration change instruction 815, the process proceeds to step 1307.

  On the other hand, if the configuration change instruction 815 exists, the variable delay calculation execution unit 1024 determines in step 1303 the timestamp of the tuple acquired from the result tuple queue 1014 in step 1301 and the position at the top of the configuration change instruction queue 1023. Compare the time stamps of the configuration change instructions. If the former is in the past, the process proceeds to step 1307.

  In other cases, in step 1304, the variable delay calculation execution unit 1024 determines whether an entry matching the server identifier on the output side of the configuration change instruction exists in the output destination-specific delay time management table 1025. It is determined, and if it exists, the process proceeds to step 1305. On the other hand, if the entry matching the server identifier on the output side does not exist in the output destination-specific delay time management table 1025, the process proceeds to step 1306.

  In step 1305, the variable delay calculation execution unit 1024 updates the entry with the contents of the configuration change instruction. On the other hand, if it does not exist in step 1306, the variable delay calculation execution unit 1024 registers the contents of the configuration change instruction as a new entry in the output destination-specific delay time management table 1025. When the processing of steps 1305 and 1306 is completed, the process returns to step 1302 and the above processing is repeated.

  The processing from step 1307 to step 1310 is repeated by the variable delay calculation execution unit 1024 by the number of entries existing in the output destination-specific delay time management table 1025. First, in step 1308, the variable delay calculation execution unit 1024 generates a copy of the tuple obtained from the result tuple queue 1014 in step 1301, and shifts the time stamp by the delay time 1202 of the entry.

  Then, in step 1309, the variable delay calculation execution unit 1024 transmits the tuple copied above to the output destination 1201 of the entry. After exiting the loop from step 1307 to step 1310, the variable delay calculation execution unit 1024 reaches step 1311 and ends the processing.

  According to the above processing, if the time stamp of the result tuple 141 of the result tuple queue 1014 is the time stamp of the configuration change instruction in the configuration change instruction queue 1023, the variable delay calculation execution unit 1024 adds the output destination of the configuration change instruction, and The tuple 141 is transmitted. As a result, the input / output relationship can be dynamically changed without stopping the stream processing process 110.

  FIG. 14 is a block diagram illustrating an example of cooperation between the stream processing processes 110 for monitoring a production line of a factory via delay calculation. In FIG. 6 of the conventional example, delay operations 1401 and 1402 are inserted in stream processing processes 110-1 and 110-2, respectively.

  The delay time of 3 seconds and 2 seconds set for each delay calculation is the time difference between when the data is acquired by the sensor 200-1 and when the data is acquired by the sensor 200-2, and by the sensor 200-2. It is determined according to the time difference from when the data is obtained to when the data is obtained by the sensor 200-3.

  The tuple 633 generated by the sensor 200-1 at 9:05:07 is analyzed by the stream processing process 110-1 and output as a result tuple 644. At this time, the time stamp of the analysis result tuple 644 is determined to be 9:05:10 three seconds after 9:05:07 due to the delay operation. The tuple 644 reaches the stream processing process 110-2 at 9: 05: 07.33, which is about 0.3 seconds after 9: 05: 07.00 in real time.

  Thereafter, at 9: 05: 10.00, when the sensor 200-2 generates the stream tuple 636, both the analysis result tuple 644 including the time stamp of 9:05:10 and the stream tuple 636 are aligned. Therefore, the stream processing process 110-2 can immediately start the analysis at 9: 05: 10.00. Accordingly, this analysis result tuple 642 can reach the stream processing process 110-3 at approximately 9: 05: 10.31, approximately 0.3 seconds later.

  In addition, by adjusting the delay time of the delay calculation 1401 to 3 seconds, which is the time difference of data acquisition between the stream processing processes 110, the analysis result of the same product c (653) in the stream processing process 110-1 Since the stream processing process 110-2 outputs the 9:05:10 analysis result tuple 642, the stream processing process 110-2 does not need to hold the tuple 644.

  Since the delay calculation 1402 of the stream processing process 110-2 has the same effect in the analysis processing of the stream processing process 110-3, the analysis result tuple 643 also has the stream tuple 639 generated at 9:05:10. At about 9: 05: 10.30, about 0.3 seconds after 00, the downstream process of stream processing process 110-3 can be reached.

  Thus, according to the present embodiment, even when the production line becomes longer and the time difference for detecting the same object increases, it is possible to avoid an increase in the delay in the output timing of the analysis result. Then, even when a delay time is added, the time order of tuples can be maintained, so that it is possible to ensure both consistency of stream processing and reduction of latency.

  FIG. 15 and FIG. 16 are block diagrams showing an example of cooperation between the stream processing processes 110 for realizing the inter-vehicle communication of the automobile through delay calculation. 15 and 16, delay operations 1501 and 1502 are inserted into the stream processing processes 110-1 and 110-2, respectively, in contrast to FIG. The delay time of 2 seconds set in FIG. 15 is updated to 1 second in FIG. 16 because the distance between the vehicles is reduced.

  In the stream processing process 110-1, the position / speed tuples 720, 721, and 722 of the own vehicle are respectively set at timings of 9: 05: 06.00, 9: 05: 07.00, and 9: 05: 08.00. Is entered. From the stream processing process 110-2, predicted position tuples 743, 744, and 745 indicating the predicted positions of the other vehicles at times t0, t0, and t1, respectively, are 9: 05: 06.21 and 9:05:07. .24, 9: 08.22 (it takes about 0.2 seconds until the position / velocity tuple is generated and analyzed in the stream processing process 110-2 and reaches the stream processing process 110-1) . The time stamps are changed to 9:05:08, 9:05:09, and 9:05:10 by the delay operation 1502.

  Here, the same time stamp 9:05:08 is set at the timing of 9: 05: 08.00 when the position / velocity tuple 722 of the time stamp 9:05:08 is generated in the stream processing process 110-1. The predicted position tuple 743 from the stream processing process 110-2 has already arrived. Therefore, the stream processing process 110-1 calculates the predicted position at the time t1 from both tuples as the analysis processing of the time stamp 9:05:08, shifts the time stamp to 9:05:10 by the delay operation 1501, and It can be sent as a position tuple 742 to the stream processing process 110-2. Here, the more future the time t1 at which the predicted position is determined from the time stamp 9:05:10 after the delay, the higher the value as the predicted value.

  Since the same effect is obtained in the analysis on the stream processing process 110-2 side, the processing can be performed without causing a hang-up as in the conventional example shown in FIG. When the vehicles 300 approach each other, it is necessary to reduce the delay time in order to keep the time difference between the predicted time tx and the time stamp after the delay added to the predicted position tuple so as not to impair the effect of the prediction. .

  In the example of FIG. 16, the delay time is reduced from 2 seconds to 1 second, and a predicted position tuple 746 with a time stamp 9:05:10 is generated from a position / velocity tuple 726 with a time stamp 9:05:09. . That is, although the time stamp overlaps with the previously output predicted position tuple 742, the duplication does not pose a particular problem because it can be handled as a correction of the predicted value in the analysis query.

  As described above, according to the first embodiment, the query definition can be easily performed while maintaining the reproducibility of the stream processing, and the connection relation between the dynamic queries without stopping the processing can be achieved. Can be changed. Furthermore, by setting the delay time in anticipation of the time at which the acquired sensor data is used according to the position, it is possible to realize a query including suppression of an increase in latency and interdependency.

  FIG. 17 is a block diagram of the stream processing distributed execution system 801 according to the second embodiment of the present invention. In the second embodiment, the function of the stream processing distributed execution management server 1001 of the first embodiment is incorporated in the stream processing server 100-1, and the other configuration is the same as that of the first embodiment.

  In the stream processing server 100-1, a query distribution unit 1002, a configuration change command distribution unit 1003, and a server table 1004 operate in addition to the stream processing process 110-1.

  The query distribution unit 1002 distributes a query to another stream processing process 110 according to the content upon receiving the query definition 132 from the host computer 130, as in the first embodiment. Also, similarly to the first embodiment, when the configuration change command distribution unit 1003 receives the configuration change command 815 from the host computer 130, the configuration change command distribution unit 815 sends the configuration change command 815 to the specified stream processing server 100-1 with reference to the server table 1004. Distribute.

<Summary>
As described above, in the first and second embodiments, in the distributed processing, it is possible to simplify the query definition while maintaining the reproducibility of the stream processing and to change the connection relationship between the dynamic queries without stopping the processing. , Suppressing latency increase, and realizing queries including interdependencies.

  Accordingly, stream processing, which can be applied to distributed execution of analysis processing between edges (stream processing servers 100), such as monitoring and control of devices based on IoT sensor data analysis and cooperative processing between moving objects, etc. It is easy to develop a real-time analytics application that ensures both consistency and low-latency processing. Also, in order to make it possible to change the optimal configuration of the distributed processing at the time of execution, the transparency of the position in the distributed execution of the query is guaranteed, the process of the application development and the SI (System Integration) work are separated, and the application Portability (or versatility) can be improved.

  The embodiments of the present invention have been described above. However, these are exemplifications for explaining the present invention, and the application range of the present invention is not limited to only the exemplified form. Also, any combination of the above-described embodiments can be an embodiment of the present invention.

  Note that the present invention is not limited to the above-described embodiment, and includes various modifications. For example, the above-described embodiments have been described in detail in order to explain the present invention in an easy-to-understand manner, and are not necessarily limited to those having all the configurations described above. Further, a part of the configuration of one embodiment can be replaced with the configuration of another embodiment, and the configuration of one embodiment can be added to the configuration of another embodiment. In addition, for a part of the configuration of each embodiment, addition, deletion, or replacement of another configuration can be applied alone or in combination.

  In addition, each of the above-described configurations, functions, processing units, processing units, and the like may be realized by hardware, for example, by designing a part or all of them using, for example, an integrated circuit. In addition, the above-described configurations, functions, and the like may be implemented by software by a processor interpreting and executing a program that implements each function. Information such as a program, a table, and a file for realizing each function can be stored in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.

  In addition, control lines and information lines are shown as necessary for the description, and do not necessarily indicate all control lines and information lines on a product. In fact, it can be considered that almost all components are connected to each other.

Claims (12)

A distributed processing method of stream data for processing a tuple to which a time stamp is added by a computer having a processor and a memory,
The computer includes a query that processes an input tuple, a variable delay operation that adds a delay time to a tuple output by the query, and a recursive loop that recursively inputs an output of the variable delay operation to the query. A first step of accepting a query definition;
A second step in which the computer receives an input destination of the tuple, an output destination of the tuple, and a configuration change instruction including a delay time;
The computer cancels the recursive loop, changes the output of the variable delay operation to the output destination of the configuration change instruction, changes the input destination of the tuple to the input destination of the configuration change instruction, and changes the configuration change instruction. A third step of setting the delay time of
A fourth step in which the computer receives the tuple from the input destination, inputs the tuple obtained by performing a predetermined process in the query to the variable delay operation, and adds the delay time to a time stamp of the tuple;
A fifth step in which the calculator outputs a tuple to which the delay time has been added to the output destination;
A distributed processing method for stream data. The stream data distributed processing method according to claim 1, wherein
The second step is
The relationship between the output destination and the delay time is stored in output destination-specific delay time management information,
The third step is
A stream data distribution processing method, wherein a delay time corresponding to an output destination is added to the time stamp from output destination-specific delay time management information in which a delay time is stored for each output destination. The stream data distributed processing method according to claim 1, wherein
The configuration change instruction includes an execution start time of the configuration change instruction,
The second step is
Storing the configuration change instruction in a configuration change instruction queue;
The third step is
If the input timestamp of the tuple has passed the execution start time of the configuration change instruction at the head of the configuration change instruction queue, setting the input destination, the output destination, and the delay time is performed. Characterized stream data distribution processing method. The stream data distributed processing method according to claim 1, wherein
The distributed query processing method of stream data, wherein the query definition defines a delay time of the variable delay operation as a variable. A computer having a processor and a memory;
A management computer that manages a plurality of the computers,
A data source that supplies a tuple with a time stamp to the plurality of computers, and a distributed processing system for stream data,
The management computer,
A query that includes a query that processes an input tuple, a variable delay operation that adds a delay time to a tuple output by the query, and a recursive loop that recursively inputs the output of the variable delay operation to the query. A query distribution unit for distributing to a plurality of computers;
An input destination of the tuple, an output destination of the tuple, and a configuration change instruction distribution unit that distributes a configuration change instruction including a delay time to the plurality of computers,
The calculator is:
After accepting the query definition, accepting the configuration change instruction, canceling the recursive loop, changing the output of the variable delay operation to the output destination of the configuration change instruction, and changing the input destination of the tuple to the configuration change A query processing unit for changing to an instruction input destination,
A variable delay calculation execution unit that adds the delay time of the configuration change instruction,
The query processing unit receives a tuple from the input destination, inputs a tuple that has been subjected to predetermined processing in the query to the variable delay operation,
The stream data distributed processing system, wherein the variable delay calculation execution unit adds the delay time to a time stamp of the tuple and outputs the result to the output destination. The stream data distributed processing system according to claim 5, wherein
The calculator is:
The relationship between the output destination and the delay time is stored in output destination-specific delay time management information,
The variable delay calculation execution unit,
A distributed processing system for stream data, wherein a delay time corresponding to an output destination is added to the time stamp from delay time management information for each output destination in which a delay time is stored for each output destination. The stream data distributed processing system according to claim 5, wherein
The configuration change instruction includes an execution start time of the configuration change instruction,
The calculator is:
Storing the configuration change instruction in a configuration change instruction queue;
The query processing unit,
If the input time stamp of the tuple has passed the execution start time of the configuration change instruction at the head of the configuration change instruction queue, the input destination, the output destination, and the delay time are set. Stream data distributed processing system. The stream data distributed processing system according to claim 5, wherein
The distributed query processing system for stream data, wherein the query definition defines a delay time of the variable delay operation as a variable. A storage medium storing a program for controlling a computer having a processor and a memory,
A query that includes a query that processes the input tuple, a variable delay operation that adds a delay time to a tuple output by the query, and a recursive loop that recursively inputs the output of the variable delay operation to the query A first step;
An input destination of the tuple, an output destination of the tuple, and a second step of receiving a configuration change instruction including a delay time;
Release the recursive loop, change the output of the variable delay operation to the output destination of the configuration change instruction, change the input destination of the tuple to the input destination of the configuration change instruction, and change the delay time of the configuration change instruction. A third step of setting the variable delay operation;
A fourth step of receiving a tuple from the input destination, inputting a tuple obtained by performing a predetermined process in the query to the variable delay operation, and adding the delay time to a time stamp of the tuple;
A fifth step of outputting a tuple to which the delay time has been added to the output destination;
A non-temporary computer-readable storage medium storing a program for causing the computer to execute the program. The storage medium according to claim 9, wherein
The second step is
The relationship between the output destination and the delay time is stored in output destination-specific delay time management information,
The third step is
A storage medium characterized by adding a delay time corresponding to an output destination to the time stamp from output destination-specific delay time management information storing a delay time for each output destination. The storage medium according to claim 9, wherein
The configuration change instruction includes an execution start time of the configuration change instruction,
The second step is
Storing the configuration change instruction in a configuration change instruction queue;
The third step is
If the input time stamp of the tuple has passed the execution start time of the configuration change instruction at the head of the configuration change instruction queue, the input destination, the output destination, and the delay time are set. Storage medium. The storage medium according to claim 9, wherein
The storage medium, wherein the query definition defines a delay time of the variable delay operation by using a variable.

Images