JPH0690703B2 - Global detection system - Google Patents

Description

[Detailed Description of the Invention] [Table of Contents] Outline of Industrial Application Field of the Invention and Problems to be Solved by the Invention Means for Solving Problems Problems Working Example Lock Management Mechanism (Fig. 1) Waiting notification unit (Figs. 1 and 2) Waiting inquiry unit (Fig. 1) Deadlock detection processing example (Figs. 3 and 4) Effect of the invention [Outline] Execution in a computer network consisting of multiple sites This is a method of detecting so-called global deadlock that occurs across sites due to generated transactions. Queuing information between transactions is collected at sites that are defined as a function of the transaction identifier, and information is supplemented by inquiries to the sites that are also defined to detect the presence of global deadlocks.

[Industrial application field]

The present invention occurs across sites due to a transaction executed in a plurality of sites in a computer network including a plurality of sites in an information processing system. The present invention relates to a so-called global deadlock detection method.

When a plurality of shared resources of the computer system are exclusively used by a plurality of transactions, for example, in order to wait for the resources occupied by other transactions to wait for each other to be occupied by the other transactions, It is well known that a so-called deadlock state may occur in which the processing of the above transaction is stopped.

As is well known, when a deadlock occurs, all of the causal transactions are frozen in a state in which processing cannot proceed unless execution of at least one transaction among the causal transactions is interrupted.

For example, a database was distributed to multiple sites that were connected to each other by communication lines. In a computer network or the like, it may be necessary to exclusively access data in databases of a plurality of sites in one transaction.

When a plurality of such transactions exist, they are executed in parallel at each site, and are normally executed, each transaction takes exclusive possession of the data of all the sites when the access of all the required sites is successful. In the case of the solving method, a deadlock across a plurality of sites, a so-called global deadlock may occur.

The global deadlock cannot be detected until the queuing state of the transaction occurring at each site is combined and seen for two or more sites, and therefore special consideration is required for efficient detection.

[Problems to be solved by conventional technology and invention]

Figure 5 shows a computer network consisting of four sites.
A database system is dispersedly provided at each of the sites 1, 2, 3 and 4, and the sites are connected by a communication line 5.

In each of the sites 1 to 4, for example, the user's terminal device 6
Generate a transaction to process the specified database according to the command input from.

A transaction is typically a parent subtransaction that occurs first and the parent subtransaction
It consists of one or more child subtransactions that are generated according to the needs of the process. There is also a system in which a child transaction further causes a child transaction.

For example Site Transaction T ₁ generated in the 1, when it takes the process to access both database site 1 and site 2, sub-transactions T ₁ to Site 1
When sub-transaction T ₁ is created, sub-transaction T ₁ communicates with site 2 to generate a child sub-transaction T ₁ ′ at site 2.

In the following, assuming that all access to the database is required to exclusively occupy the required data, the sub-transactions T ₁ and T ₁ ′ are assigned to the lock management mechanism of the respective sites 1 and 2. Only when you request to exclusively occupy the required data, and if you can do so,
Access processing to the data is started.

The sub-transactions T ₁ and T ₁ ′ that have finished processing notify each other of the end and request the lock management mechanism to release the exclusive use when the access of both is completed.

If the other transaction has already occupied the data in the occupancy request, waiting is necessary, and the occupancy becomes possible after the processing of the other transaction of the previous request is completed.

Transactions as above, under control like this
The sub-transaction for T ₁ is at sites 1 and 2, and the sub-transaction for transaction T ₂ is at sites 2 and 3
And there are subtransactions of transaction T ₃ at sites 3 and 1 and the subtransactions at each site tried to occupy the same data, so that a queuing relationship as shown in FIG. 6 (a) occurred. .

In this description, all subtransactions of transaction Ti are represented as Ti. Also, Ti
→ Tj indicates that the sub-transaction Ti is waiting for the occupation end of the sub-transaction Tj.

If Figure 6 waiting relation shown in (a) occurs, for example, looking at the transaction T _1, even if the T ₁ of the site 1 played occupied, can not be released occupied until the processing end T ₁ of the site 2 This state cannot change because all transactions have the relation, and they wait for the other transaction to release their possession. That is, a global deadlock has occurred.

The occurrence of such a global deadlock cannot be detected only by looking at the waiting state in each site. In this example, the states of sites 1 to 3 are collected in one place, and the state of FIG. It can be detected for the first time by obtaining a queuing relation graph such as and having a loop there.

For this purpose, for example, there is a so-called centralized system in which waiting information of all sites is transferred to a specific one site and a global deadlock is detected at the specific site.

Although such a centralized method is simple as a control configuration, a high reliability is especially required for a specific site that monitors the global deadlock of the entire network. Is not a simple method, and a load of receiving and processing a large amount of waiting information is imposed on a specific site. Therefore, this method is suitable only for a relatively small local area network.

To that end, one specific global deadlock detection
A so-called distributed method is considered, which does not concentrate on the site.

As one of the methods, for example, by detecting a state of waiting for a sub-transaction having a child sub-transaction at another site and repeating the operation of transferring the waiting information to the other site, the graph of the waiting relationship is sequentially grown. , Depending on the transaction occurrence status
There is a method to detect deadlocks at some unspecified site.

However, this method has problems that the control is complicated and that the transfer amount of the waiting information between sites is increased.

[Means for solving problems]

FIG. 1 is a block diagram showing the configuration of the present invention. 10 is a communication control unit that controls communication with other sites, 11 is a sub-transaction, and 12 is a site address function unit that generates a site address from a transaction identifier. , 13 is a queuing notification unit that sends information about queuing relationships to a predetermined site, and 14 is a queuing inquiry unit that sends a query for required queuing information from a queuing relationship graph configured by the deadlock detection unit 15.

[Action]

The subtransaction is the lock management mechanism as described above.
16 request the occupation of shared resources such as required data.

When an exclusive wait occurs, the lock management mechanism 16 passes information indicating a wait relationship between subtransactions to the wait notification unit 13.

The queuing notification unit 13 passes the transaction identifier of the subtransaction related to the queuing to the site address function unit 12, obtains the site address, and transmits the queuing information to the site via the communication control unit 10.

The deadlock detection unit 15 receives queuing information from other sites and its own site, combines them, and forms a global queuing relationship graph between transactions.

The queuing inquiry unit 14 determines the necessity of the inquiry for each node on the configured queuing relationship graph under a predetermined condition, and when the inquiry is necessary, passes the transaction identifier to the site address function unit 12. Decide the contact destination site and send the inquiry.

The inquiry is received by the deadlock detection unit 15 of the destination site, and returns information of a predetermined part of the queuing relationship leading to the designated transaction.

The request source deadlock detection unit 15 receives this and supplements the preceding waiting relation graph.

If necessary, this replenishment operation is repeated, and when a loop is formed on the graph, one of the transactions on the loop is selected and the execution of the transaction is suspended by a known appropriate method. The global deadlock is resolved by ending it.

With the control of the above configuration, the notification of the waiting information may be sent to a predetermined site (generally two sites) once every time the waiting occurs.

After that, in some cases, an inquiry may be required for a required transaction, but since the site that always holds the required information is determined by the above notification method, useless inquiry or relay of inquiry may occur. There is no.

Therefore, it becomes possible to efficiently control the detection of the global deadlock.

The function of the site address function unit 12 that determines the site address from the transaction identifier can generally be appropriately distributed by using, for example, the site address of the site that has generated the transaction as the function value.

However, depending on the needs of the system, for example, if the function value is a fixed value, it becomes a centralized method, and if an appropriate hash function of the value of the transaction identifier is used, a plurality of specific sites will share the global deadlock detection. A highly flexible operation is possible, such as a centralized system.

〔Example〕

Lock Management Mechanism In FIG. 1, the sub-transaction 11 is a parent sub-transaction of a transaction generated at this site or a child sub-transaction of a transaction generated at another site as described above by a mechanism not shown. However, in either case, it has a transaction identifier attached at the site where the transaction occurred.

The transaction identifier must be able to uniquely identify the transaction within this computer network. For this reason, the transaction identifier has a structure in which, for example, a site address for identifying each site and a serial number generated without duplication for each site are connected.

When the sub-transaction 11 requests the lock management mechanism 16 to occupy the specified data in the database,
If the data is not occupied, it is set to the occupied state and the subtransaction 11 is notified of the occupation, so that the subtransaction 11 starts accessing the data.

However, when the request is already occupied by another sub-transaction, the request is connected to the queue and the sub-transaction 11 is placed in the waiting state.

In this case, the lock management mechanism 16 uses the sequence of transaction identifiers of the request subtransaction 11 and the subtransaction that is to be waited by the subtransaction 11 and immediately preceding on the queue as the queuing notification unit 13 as queuing information. Notify to.

Waiting Notification Unit The waiting notification unit 13 passes both transaction identifiers of this waiting information to the site address function unit 12, and obtains the site address as a function value corresponding to each transaction identifier.

The waiting notification unit 13 notifies the waiting information by requesting the communication control unit 10 to transmit the waiting information addressed to the site of the address obtained from the site address function unit 12.

It should be noted that the destination may include the destination of the own site, in which case the waiting information is interrupted from the communication control unit 10 to the deadlock detection unit 15.

Here, the address function value of the site address function unit 12 is
If the value of the site address portion forming a part of the input transaction identifier is taken, the above-mentioned queuing information is collected at the transaction origination site.

Usually, considering that the system scale of each site is determined by using the transaction generation amount as the main data, it can be expected that generally the appropriate load sharing will be performed by the address function.

The deadlock detector 15 combines the waiting information received from the communication controller 10 to obtain a global waiting relation graph. That is, individual waiting information as shown in FIG. 6 (a) is combined as shown in FIG. 6 (b) to form a waiting relationship graph.

As a result, even if the waiting functions T ₁ → T ₂ , T ₂ → T ₃ , ... all occur at different sites, as shown in Fig. 2, at least every other transaction is executed on the deadlock loop. If the notification destination sites determined by the same are the same, the waiting information is gathered at that site (site S _{1 in} the case of the figure), and the loop can be immediately detected.

In the figure, S () indicates a function of the site address function unit 12, and Sm indicates a site address.

However, if there is no such special condition, the deadlock detection unit 15, the deadlock detection unit 15, the deadline detection unit 15, the deadlock detection unit 15 by , Send inquiries to other sites and supplement the meeting information.

Here, except for redundant queries, in order to avoid duplication of deadlock resolution processing, for example, when there is a deadlock loop in the queuing relationship graph, the transaction identifier We will establish a convention that the site of the transaction with the highest weight function value will process the loop.

In this way, one site has to keep track of the presence or absence of a loop on the waiting information graph from the transaction of another site on the waiting information graph at the terminal on the waiting side to the terminal on the waiting side. Among the paths leading to the transactions of other sites, the one with the largest weight function value of the transaction on that path is only the path that is the transaction of the own site (this path is called the responsible path of the site).

Therefore, the waiting inquiry unit 14 specifies the transaction of the terminal on the waiting side of the responsible path that matches this condition, and sends the inquiry to the site determined by the address function of the transaction identifier of the transaction.

From the logic of the notification processing of the above-mentioned waiting information, if there is a waiting for a specified transaction,
The destination site always has information regarding the waiting of the designated transaction, and the inquiry source site can receive the reply of the valid waiting information.

The deadlock detection unit 15 of the inquiry source site replenishes the waiting relationship graph with the response information to the inquiry, and thus extends the branches of the waiting relationship graph for the responsible path as long as the waiting relationship continues. As a result, if a loop exists, the loop is detected.

Deadlock Detection Processing Example The process of detecting a loop by the above processing will be described with reference to the examples shown in FIGS. 3 and 4.

In Figs. 3 and 4, the site addresses are 1 to 4 (indicated by to distinguish them from other numbers below).
The two-digit number indicates the value of the transaction identifier, and the first digit thereof indicates the site address. That is,
When the transaction identifier is ik, S (ik) = k. Also, the transaction identifier in the square indicates the transaction of the site where it is shown.

FIG. 3 shows a situation assuming a waiting relationship between transactions, and FIGS. 4 (a), (b), and (c) show deadlocks at any site when the waiting relationship shown in FIG. 3 exists. The process of detecting a loop will be described.

FIGS. 4 (a), (b), and (c) show a waiting relation graph in each site at the beginning of each processing phase and the first to third phases, and messages transmitted / received based on the graph. , FIG. 4 (a) is the first phase of notifying the waiting information, FIG. 4 (b) is the second phase of supplementing the information by inquiry, and FIG. 4 (c) is the third phase of the judgment processing.

In this example, the weighting function takes the value of the transaction identifier as its function value.

In the first phase, the site appointment notification unit 13
The transaction information '1' is used as the waiting information 13 → 32.
Send to the site that is the transaction site of 3'and the transaction site of transaction identifier '32'. Similarly, the waiting information 34 → 33 is notified to the site by transmitting the waiting information 31 → 12 → 22 of the waiting information 31 → 12 → 22 to the site and the portion 12 → 22 to the site.

Here, the destination of the site is the transfer from the waiting notification unit 13 of itself to the deadlock detection unit 15, and the destination is shown in parentheses in the figure for distinction.

The site, too, as shown in the first phase of the figure,
Similar to the case of the site, it sends the notification of the meeting information to each site.

As a result, for example, the deadlock detection unit 15 of the site
As shown in the message of the first phase, the appointment information 31 → 12 from the appointment notification section 13 of the own site, 21 from the site →
Since 14 → 31, 41 → 13 and 12 → 11, 21 → 41 from the site, and 24 → 11 → 21 from the site are received, they are combined and the waiting relationship shown in the (b) Phase 2 site in the figure Get the graph.

On the site, on this graph, 11 → 21 → 14 → 31 → 12 → 11
Is detected, and the maximum weight function value on this loop is the transaction identifier '31'. Since it is the transaction of the own site, the transaction identifier '31' is used to cancel this deadlock. Suspend execution of the transaction and notify each site.

Further, the waiting inquiry unit 14 detects that the following paths remain on the waiting relation graph.

(A) 24 → 11 → 21 → 41 → 13 (b) 24 → 11 → 21 → 14 (c) 12 → 11 → 21 → 41 → 13 (d) 12 → 11 → 21 → 14 Of these, the above conditions The responsibility path by (a),
Since it is (c) and (d), the inquiry message that specifies transactions '13' and '14' on the terminal on the waiting side of those paths (shown as 13 ?, 14? In the figure)
Are sent to the site and, which are determined by the address function of the site address function unit 12, respectively.

Similarly, each of the other sites detects a path that needs an inquiry and sends an inquiry message. For example, since the site sends the message 11? With the destination as the site as shown in the figure, the deadlock detection unit 15 of the site responds to this inquiry message with the transaction '11 as shown in the figure. The information of the waiting path starting with 'is returned.

When each site receives a response to the inquiry sent to the other site from each site, the second information is used as the response information.
Fig. 4 (c), supplementing the waiting relationship graph of the phase
The waiting relation graph of the third phase shown in FIG.

In the case of the site, the response from the site does not add new information, but the response from the site causes the transaction '13' to wait for the transaction '32', and goes through the waiting relationship 32 → 23 → 34 → 33, '1
By adding the path to 2 ', it is possible to detect that there is a deadlock loop as shown.

Therefore, the site can cancel this deadlock by interrupting the execution of the transaction with the transaction identifier '41', for example.

〔The invention's effect〕

As is clear from the above description, according to the present invention, it is possible to efficiently detect a global deadlock that occurs in transactions of a computer network composed of a plurality of sites and that spans the sites. There is a significant industrial effect of improving performance.

[Brief description of drawings]

 FIG. 1 is a block diagram of a configuration of an embodiment of the present invention, FIG. 2 is a diagram showing an example of a deadlock loop, FIG. 3 is a diagram showing an example of a waiting relationship, and FIG. 4 is an explanatory diagram of global deadlock detection processing. FIG. 5 is a configuration block diagram of the computer network, and FIG. 6 is an explanatory diagram of a deadlock occurrence state. In the figure, 1 to 4 are sites, 5 is a communication line, 6 is a terminal device, 10 is a communication control unit, 11 is a sub-transaction, 12 is a site address function unit, 13 is a waiting notification unit, 14 is a waiting inquiry unit, 15 Indicates a deadlock detector, and 16 indicates a lock management mechanism.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (56) References: Riichiro Take et al., “Distributed Global Deadlock Detection Method with Optimal Load Distribution”, IPSJ 31st National Convention 2B-2 (1985)

Claims (2)

[Claims] 1. A plurality of sub-transactions (11) each having a means (10) for communicating with each other and comprising a plurality of sites each comprising an individual computer system and constituting a transaction are parallel to each other at the plurality of sites. In the computer network to be executed, each of the sub-transactions (11) has a transaction identifier of a transaction constituted by the sub-transaction, and each of the sites has exclusive control of shared resources between the sub-transactions executed by the site. When a waiting for a specific use occurs, go to the site uniquely defined as a function (12) of only the transaction identifier of the sub-transaction,
A means (13) for notifying information about the queuing, and a part which can reach the waiting side from a certain sub-transaction is called a start point side, and a part which can also reach the waiting side is reached by the end point side. , The sub-transactions of the other site exist on the start point side and the end point side, the sub-transaction of the own site exists between the start point side and the end point side, and Means (14) for detecting queuing information having the maximum function value of the subtransaction of the site, and inquiring of the site determined by the function (12) for the subtransaction on the end point side of the detected queuing information
A global deadlock detection method characterized by having. 2. A transaction identifier includes a site address of a site that has generated the transaction, and the function of the transaction identifier takes a value of the site address included in the transaction identifier. The global deadlock detection method described in the first item of the range.