CN1298515A

CN1298515A - Leasing for failure detection

Info

Publication number: CN1298515A
Application number: CN 99805492
Authority: CN
Inventors: J·H·瓦尔多; A·M·沃尔蕾斯; R·谢夫勒; K·C·R·C·阿诺德
Original assignee: Sun Microsystems Inc
Current assignee: Sun Microsystems Inc
Priority date: 1998-02-26
Filing date: 1999-02-17
Publication date: 2001-06-06
Also published as: EP1058882A1; JP2002505468A; AU2770499A; WO1999044128A1

Abstract

A system for using a lease to detect a failure and to perform tailure recovery is provided in using this system, a client requests a lease from a server to utilize a resource managed by the server for a period of time. Responsive to the request, the server grants the lease, and the client continually requests renewal of the lease. If the client fails to renew the lease, the server detects that an error has occurred to the client. Similarly, if the server fails to respond to a renew request, the client detects that an error has occurred to the server. As part of the lease establishment, the client and server exchange failure-recovery routines that each invokes if the other experiences a failure.

Description

Be used for renting of fault detect

Related application

The application is the part continuation application of the U.S. Patent application 08/729421 of submission on October 11st, 1996, and last application is quoted in this application by reference.

Following patented claim is relevant with the application.

The interim U.S. Patent application 60/076048 that is entitled as " distributed computing system " that on February 26th, 1998 submitted to.

With the U.S. Patent application 09/044923 that be entitled as " method and system of leasing storage " of the application's while with attorney docket 06502.0011-01000 submission.

With the U.S. Patent application 09/044838 that be entitled as " method that in distributed system the certificate of authority rent, device and product " of the application's while with attorney docket 06502.0011-02000 submission.

With the U.S. Patent application 09/044934 that be entitled as " method that in distributed system member of community rent, device and product " of the application's while with attorney docket 06502.0011-03000 submission.

With the U.S. Patent application 09/044933 that be entitled as " method that in based on the system of incident transmit behavior " of the application's while with attorney docket 06502.0054-00000 submission.

With the U.S. Patent application 09/044919 that be entitled as " object in distributed system postpone rebuild and event notice long-range loading " of the application's while with attorney docket 06502.0062-01000 submission.

With the U.S. Patent application 09/044938 that be entitled as " method and apparatus of remote method invocation (RMI) " of the application's while with attorney docket 06502.0102-00000 submission.

With the U.S. Patent application 09/045652 that be entitled as " method and system of the definite garbage of identify remote method " of the application's while with attorney docket 06502.0103-00000 submission.

With the U.S. Patent application 09/044790 that be entitled as " method and apparatus of in distributed system determining remote object situation " of the application's while with attorney docket 06502.0104-00000 submission.

With the U.S. Patent application 09/044930 that be entitled as " in distributed system carry out downloadable smart proxies with remote procedure call relevant treatment " of the application's while with attorney docket 06502.0105-00000 submission.

With the U.S. Patent application 09/044917 that be entitled as " remote method suspend and continue " of the application's while with attorney docket 06502.0106-00000 submission.

With the U.S. Patent application 09/044835 that be entitled as " in the database method and system of multiple entry and multi-template matching " of the application's while with attorney docket 06502.0107-00000 submission.

With the U.S. Patent application 09/044839 that be entitled as " method and system in database on the spot revised " of the application's while with attorney docket 06502.0108-00000 submission.

With the U.S. Patent application 09/044945 that be entitled as " in the database method and system of type safety attributes match " of the application's while with attorney docket 06502.0109-00000 submission.

With the U.S. Patent application 09/044931 that be entitled as " in distributed system dynamically search service " of the application's while with attorney docket 06502.0110-00000 submission.

With the application simultaneously with attorney docket 06502.0112-00000 being entitled as of submitting to " but provide in the distributed system with devices communicating in the apparatus and method of the download code used " U.S. Patent application 09/044939.

With the U.S. Patent application 09/044826 that be entitled as " be convenient to visit search service method and system " of the application's while with attorney docket 06502.0113-00000 submission.

With the U.S. Patent application 09/044932 that be entitled as " in the distributed system apparatus and method of dynamically verifying information " of the application's while with attorney docket 06502.0114-00000 submission.

The U.S. Patent application 09/030840 that is entitled as " dynamic distributed Calculation Method and device on the network " that on February 26th, 1998 submitted to.

With the U.S. Patent application 09/044936 that be entitled as " the mutual design tool in lasting shared storage space " of the application's while with attorney docket 06502.0116-00000 submission.

With the U.S. Patent application 09/044934 that be entitled as " based on the control of multiform token " of the application's while with attorney docket 06502.0117-00000 submission.

With the U.S. Patent application 09/044915 that be entitled as " based on the access control of storehouse " of the application's while with attorney docket 06502.0118-00000 submission.

With the U.S. Patent application 09/044944 that be entitled as " based on the safety requirements of storehouse " of the application's while with attorney docket 06502.0119-00000 submission.

With the U.S. Patent application 09/044837 that be entitled as " by method the specifying of safety requirements " of the application's while with attorney docket 06502.0120-00000 submission.

Background of invention

A. invention field

The present invention relates generally to data handling system, be specifically related to be used for renting of data handling system fault detect and reparation.

B. description of Related Art

Suitable resource management is effective computed importance.Usually resource management comprises according to request Resources allocation (for example storer) and carves (when for example the requestor no longer requires this resource) release resource in due course.Usually, resource comprises the data that the computational entity (for example application, program, applet (griggles) etc.) carried out in the computing machine is quoted.

In fact, when the application of carrying out in the computing machine was attempted to quote resource, computing machine must at first distribute or allocated resource, made this application can suitably visit them.And when this application was no longer quoted resource, computing machine can discharge or regain this resource to reuse.In computing machine, each resource all has one unique " handle " that can quote this resource.Handle can be by realizing such as variety of ways such as address, array index, unique value, pointers.

When recyclable the resource management of single computing machine is simple relatively, because be easy to determine to show the incident of resource (for example when use no longer quote resource or power fail after).The resource management of distributed system that connects a plurality of computing machines is then much more difficult, because identical resource may be used in a plurality of application in several various computing machine.

Distributed system breaks and can cause inappropriate and too early resource recovery or cause reclaiming resource failed.For example, a plurality of application of working in the various computing machine of distributed system can be quoted the resource that is positioned on the miscellaneous equipment.If be connected interruption between the residing computing machine of resource and the application of quoting these resources, then computing machine can be regained resource prematurely.Perhaps, computing machine can for good and all keep this resource, and ignores the time lengthening that application can not be visited this resource.

These difficulties cause the system of development management Internet resources, and one of them is " distribution garbage (the rubbish utmost point) collection " known.This term is provided by a kind of function that is provided for distributed system by language or run-time system, can manage application or the employed resource of set of applications moved in the various computing machine of network automatically.

Usually, a kind of idea is adopted in garbage collection, and promptly when resource was no longer quoted by any part application, it can be free to using in the future.Distribution garbage collection expands to distributed computing spectrum to this idea, does not just reclaim them when the application on arbitrary computing machine is not all quoted this resource.

Distribution garbage collection must be in Resources allocation and being kept perfectly property between to these cell refernces.In other words, system must not allow to discharge or idle this resource when the application that moves in arbitrary computing machine in the network continues to quote resource.This quoting with combining of resource is called " referential integrity ", and it does not guarantee that this is quoted all the time and allows its resource of quoting of visit.For example, network failure can make this visit not carry out.But integrality guarantees, can be used for visiting any resource if this is quoted, and then exists and quotes the identical resource of given first resource with this.

The resource that no longer is cited when using the distributed system of garbage collection must regain in the near future some.In other words, system must provide the assurance that does not make " memory leakage ".When all application abandon quoting resource, and for example use when still quoting unrecovered this resource of this resource system, can produce memory leakage because of mistake is judged to be some.

Often, the disconnection between the garbage collection system of the application of quoting resource and this resources allocation of management and release lost efficacy and memory leakage owing to producing referential integrity.For example, the disconnection that network is connected between the garbage collection system of the application of quoting resource and this resource of management can make this garbage collection system can not determine and where reach when reclaim this resource.Perhaps, because the not middle at the fixed time access resources of application, it should collect this resource the judgement of garbage collection system possible errors ground.Some technology have been adopted, by attempting to guarantee that this mechanism no memory leaks ground and keeps referential integrity to improve distributed garbage collection mechanism.A conventional method is used the reference count form, wherein to quoting the application counting number of every kind of resource.When resource count is zero, recyclable this resource of garbage collection system.But have only when resource is created by corresponding reference counter, this reference count scheme just works.In this case, garbage collection system increases the quoting resource counting as the additional application (application) of quoting this resource, and reduces counting when this resource is no longer quoted in application.

But the reference count scheme especially can encounter problems in the face of the producible fault of distributed system.But forms such as this fault computer failure, application and trouble or network failure, their prevention information is transmitted, and this information notice garbage collection system resource no longer is cited.If because network disconnection information is not transmitted, then garbage collection system does not know when reclaim this resource.

For preventing this fault, some conventional reference count scheme comprises " keeping survival " (" keep-alive ") information, is also referred to as " loopback " (" ping back ").According to this scheme, the application in the network sends information and indicates this application still can communicate by letter to the garbage collection system of management resource.These information prevent that garbage collection system from abandoning quoting resource.The indicating fault garbage collection system that can not receive this " keep survival " information can be to a resource shrinkage reference count, thereby, when counting reaches zero, recyclable this resource of garbage collection system.But this still can fail to receive " keeping survival " because of network failure thereby information is regained resource too early along with reference count arrives zero.This has violated the referential integrity requirement.

The another kind of method that solves the referential integrity proposition in garbage collection system is the corresponding identifier of each computational entity that not only keeps reference count but also keep and quote resource, " distributed garbage collection of network object " with reference to A. Birrell etc., the digital display circuit research centre, the phase 15 days the 116th Dec in 1993.This method runs into the problem identical with the reference count scheme.And then this method requires the additional unique identifier of computational entity that each is quoted each resource, increases expense like this and can unnecessarily increase communication in the distributed system, increases memory requirement (promptly corresponding to the application table of quoting each resource) simultaneously.

Brief summary of the invention

According to the present invention, keep managing this resource by the each side in distributed system (for example application) to renting this resource and garbage collection system during the quoting resource, can not have expensive memory leakage and guarantee referential integrity, thereby guarantee this resource and quoting this resource.Rent the end of term at this, the assurance of quoting resource will disappear, thereby allow garbage collection system to reclaim this resource.Because keep the application of quoting resource and the garbage collection system of management resource are reached an agreement with regard to the limited guarantee lease period, thereby when both all know lease expiration, thereby should guarantee expiration.This has guaranteed referential integrity and has been avoided fault because of the network error slack resources quoting lease period.

In a conversion embodiment of the present invention, the technology of renting is used for fault detect and reparation.When the operational failure detection is rented, client requests rental service device, and the client carries out various processing to the resource of this server admin after renting approval.When renting expiration, the client upgrades this and rents.If because of any reason should upgrade lose efficacy, then its reason communication agency of being that server is made mistakes or transmitting data between client and server is made mistakes.Which kind of situation no matter, the client all detects and makes mistakes.In addition, if lease expiration and client do not upgrade and rent or explicitly call for cancellation to rent, then server knows that client or communication agency make mistakes, and in this case, server detects makes mistakes.

Except that fault detect, conversion embodiment also provides fault restoration.During renting foundation, the client provides fault restoration program (routine) to server, but also server provides the fault restoration program to the client.Like this, in case the fault of detecting, client and server all can call the opposing party's fault restoration program separately with mutual reparation fault.After fault restoration, client and server are all got back to the state of consulting in advance.That is, through negotiation in advance, client and server are determined the state of staggering the time locating, and for example return the institute that resource is done and change.As a result, client and server are all known the state of system after the fault, thereby can proceed to handle.

Summary of drawings

Be combined in wherein and constitute the description of drawings embodiments of the invention of an instructions part, and be used from explanation advantage of the present invention and principle with instructions one.In the accompanying drawings:

Fig. 1 is the process flow diagram of the example application call processing device execution in step according to the present invention.

Fig. 2 is the process flow diagram according to the performed step of the dirty calling of the invention process form server calls processor.

Fig. 3 handles the process flow diagram that cleaning is called out performed step according to the invention process form server call processing device.

Fig. 4 handles the process flow diagram of performed step according to the invention process form server call processing device for starting garbage collection.

Fig. 5 calls out best process flow diagram in distributed processing system(DPS).

Fig. 6 is the enforcement formation block diagram according to server calls method of the present invention.

Fig. 7 can be used for distributed processing system(DPS) figure of the invention process.

Fig. 8 is the block diagram according to each software ingredient in the distributed processing system(DPS) platform of the invention process form.

Fig. 9 is the data handling system block diagram that is used for a conversion embodiment of the present invention.

Figure 10 is conversion embodiment according to the present invention, the process flow diagram of client's execution in step when server requests is rented.

Figure 11 is conversion embodiment according to the present invention, the process flow diagram of server execution in step when client requests is rented.

Describe in detail

With reference to accompanying drawing in detail example of the present invention is described in detail.Accompanying drawing with hereinafter the explanation in, same numeral is represented same or similar part.

The present invention can be realized by the computing machine that conventional distributed processing system(DPS) architecture is formed.But implement architecture of the present invention and process is not conventional, because they provide the distributed garbage collection scheme, can guarantee referential integrity and avoid memory leakage.

A. general introduction

Method call (MI) part in each computing machine of distributed processing system(DPS) is implemented distributed garbage collection scheme of the present invention.This MI part can comprise some last software modules with the establishment of Java programming language.

Usually, no matter when, search as to the rreturn value or the other method of some other calling by name, the application in the distributed processing system(DPS) obtains the quoting and attempt to visit this resource of distributed resource, and this applications is called out this resource or called the MI part of managing this resource.This MI part (being called management MI part) is followed the tracks of uncompleted quoting resource number.When the quoting resource number is zero, recyclable this resource of management MI part.The counting of quoting resource number is commonly referred to " reference count ", and the calling of increase reference count is called " dirty calling ".

When an application no longer required distributed resource, it partly sent different callings to this resource or management MI.In case receive this calling, management MI partly reduces the reference count of resource.The calling that this reduction is quoted can be described as " cleaning is called out ".

According to example of the present invention, dirty calling can comprise quoting the time interval that requires of this resource, be referred to as lease duration.In case receive dirty calling, management MI partly sends one and returns calling, indication rent approval during.Management MI partly follows the tracks of these lease durations of quoting and does not finish number of references.Thereby, when quoting resource counting is zero or when the resource lease duration expires, manage recyclable this resource of MI part.

B. process

Application call processing device in the MI part is carried out the step of application calling procedure 100 shown in Figure 1.Process 200,300 shown in the server difference execution graph 2-4 of the call processing device in the management MI part and 400 step.The garbage collector of this management MI part is carried out conventional process and was originally depended on the resource that this is quoted to regain according to the instruction from the server of call processing device.Thereby, the conventional process of garbage collector no longer is described.

1. application call processing device

Fig. 1 is the process flow diagram of process 100, and the application of call processing device uses this process to handle the quote request of the resource of being managed by another MI part in this MI part or the distributed processing system(DPS) in the MI part.

Obtained after the quoting of resource an application, the application of this call processing device sends dirty calling, partly indicates the lease duration (step 110) of quoting and asking of this resource to the management MI of resource.This dirty calling can be at resource itself or management MI part.

The application of call processing device is waited for then and is received and returns calling (step 120) from management MI part.This returns to call out and comprises the mandate lease duration, manages MI in this period and guarantees that partly quoting of dirty calling will depend on its resource.In other words, the management MI partly agree not collect with authorize during dirty calling quote corresponding resource.If management MI part does not provide during the mandate, or the refusal lease request, then the application of call processing device will have to send another dirty calling during receiving mandate.

The application monitors of call processing device is used the use to quoting, when the application of using clear and definite notification call processor no longer requires to quote, or this decision (step 130) is made in the application oneself of working as call processing device, then the application of call processing device partly sends cleaning to management MI and calls out (step 140), with to the similar mode of the employed method of dirty calling, this cleaning is called out and can will be handled the cleaning calling at resource of quoting and management MI part.Then, (step 150) quoted in the application of call processing device cancellation from use the reference list that is just using.

If use and to quote (step 130) as yet, but the application of call processing device will expire (step 160) during determining to quote mandate, and then the application repeating step 110 of call processing device and 120 is used maintenance to guarantee that quoting resource is partly represented by management MI.

2. server call processing device

The server of the call processing device of MI part is carried out 3 main processes: (1) handles dirty calling; (2) handling the incoming call cleaning calls out; (3) start the garbage collection cycle to carve the withdrawal resource in due course.

(ⅰ) dirty calling

Fig. 2 is the process flow diagram of process 200, and what the server of the call processing device of MI part used that this process handles the management of MI software section quotes resource request, promptly dirty calling.These requests are from the application of the call processing device of MI part in the distributed processing system(DPS), and this system comprises the application of asking the call processing device of identical MI part with the server of call processing device.

At first, the server of call processing device receives dirty calling (step 210).During the definite then acceptable mandate of the server of call processing device (step 220).Can be during this is authorized or section is identical At All Other Times with the request lease duration.According to some conditions, the server of call processing device determines during the suitable mandate that these conditions comprise the stock number of requirement and before counted during other mandate to the same asset mandate.

When the server of call processing device determine as yet not to dirty calling quote Resources allocation the time (step 230), then the server-assignment requested resource (step 240) of call processing device.

The server of call processing device increases then with dirty calling quotes corresponding reference count (step 250), for quoting and (step 260) during combining of resource is provided with acceptable mandate, and send to the application of call processing device and to return calling (step 270) during having mandate.By this way, the dirty calling of server controls incoming call of call processing device is quoted resource under its control.

Renting by preceding dirty crying by transmission prolongation request at present, application can prolong rents.Shown in process 200, the processing that prolongs lease request is similar to rent and starts request.Prolongation is meant that simply this resource is not recovered in some additional time interval, unless reference count is zero.

(ⅱ) cleaning is called out

The server of the call processing device of MI part also can be handled from the cleaning of the application of call processing device and call out.When the application of distributed processing system(DPS) no longer required to quote resource, its notice was used to re-use to reclaim this resource for this MI part of quoting management resource.Fig. 3 is the process flow diagram of process 300 steps, and the server of the call processing device in the MI part uses this process to handle cleaning and calls out.

When the server of call processing device receives the cleaning of quoting the resource that MI partly manages and calls out (step 310), the server of call processing device reduces corresponding reference count (step 320).Can send cleaning to this resource and call out, the server monitoring resource of call processing device and implementation 300 are called out to handle.Then, the server of call processing device partly sends to the MI that sends the cleaning calling and returns calling to confirm to receive (step 330).According to example of the present invention, the cleaning of abandoning quoting is called out and is not rejected but necessary affirmation.

(ⅲ) garbage collection

The server of call processing device also can start the garbage collection cycle to regain resource, for this reason, determine this resource is not more quoted, or this resource authorization is rented and ended.The process 400 that is shown in Fig. 4 comprises that the server of call processing device is used to start the process flow diagram of the step in garbage collection cycle.

The server monitoring reference count of call processing device and authorize lease duration, and the reference count of definite MI resource of partly managing whether be zero or the mandate of quoting during whether expire (step 410).When arbitrary condition is set up, the garbage collection (step 420) of the startup of server resource of call processing device.Otherwise the server of call processing device continues to monitor reference count and authorizes lease duration.

C. call flow

Fig. 5 is the call flow diagram between the MI part in the distributed processing system(DPS).By monitoring resource 530 quoted management MI part 525 management these resources 530 (referring to garbage collection 505).Because management MI part 525 these resources of management, the server of the call processing device of management MI part 525 is carried out the operation of this call flow diagram explanation.

Fig. 5 also display application 510 and 540 has corresponding M I part 515 and 545 respectively.Each is used 510 and 540 and obtains the quoting and attempt one of access resources 530 an of resource in the resource 530 combined thereby quote with respective resources.In order to obtain visit, use 510 and 540 and call its corresponding MI part 515 and 545 respectively, to send dirty calling 551 and 571 to MI part 525 respectively.Because MI part 515 and 525 is handled the application request of visit by the resource 530 that waits another MI partly to manage such as management MI part 525, the operation of this call flow explanation is carried out in the application of MI part 515 and 545 call processing device.

Respond dirty calling 551 and 571, management MI part 525 each transmission in MI part 515 and 545 is respectively returned calling.Dirty calling comprises the mandate lease duration of quoting to dirty calling 551 and 571.

Equally, Fig. 5 shows that also MI part 515 and 545 sends cleaning to management MI part 525 respectively and calls out 561 and 581.Cleaning is called out 561 and 581 and is notified management MI part 525 respectively, and application 510 and 540 all no longer requires visit to clean and calls out specified resource in 561 and 581.Management MI part 525 is called out 562 and 582 response cleaning callings 561 and 581 with returning respectively.Return and call out 562 and 582 and call out 552 and 572 differences and be with returning, returning and calling out 562 and 582 is simple affirmations from the MI part 525 that receives cleaning calling 561 and 581.

Use 510 and all can ask to visit identical resource with 540.For example, use 510 and can require visit " resource (1) ", and before used 540 these resources of granted access.MI part 525 is handled this situation by making this resource can be employed 510 and 540 both acquisitions at the mandate lease duration.Like this, MI part 525 will not start the garbage collection cycle to regain " resource (1) ", up to using 510 and 540 lease duration of quoting or the agreeing expirations of all abandoning this resource (no matter that incident takes place earlier).

Visit same resource simultaneously by allowing more than one to use, the present invention also can allow one to be applied in and partly to send cleaning to management MI and call out and abandon visiting again this resource to after the quoting of resource.Thereby this situation is not end management MI part 525 still unrecovered these resources as yet because this resource is still quoted or quote by Another Application to rent.But rent or during last lease expiration, this resource will be recovered after the limited duration when no longer including to use.

The D.MI part

Fig. 6 is the module frame chart according to the MI part 100 of the invention process form.MI part 600 can comprise server 650 and the garbage collector 660 that is used for each application of quoting part 605, call processing device 640 of quoting supervision, call processing device.

Quote part 605 preferably constitute a table or be similar to reference data part 610, reference counter 620 and authorize during the structure of register 630.630 to determine when that starting garbage collector 660 regains respective resources during the reference count of respectively quoting 620 of the corresponding reference data part of MI part 600 uses appointment and the mandate.

The application 640 of call processing device is a software module, the step of process 100 in the execution graph 1.The server 650 of call processing device is process 200,300 and 400 a software module in the execution graph 2～4.Garbage collector 660 is software modules, as mentioned above, reclaims resource in response to the instruction from the server 650 of call processing device.

E. distributed processing system(DPS)

Fig. 7 explanation can be used for implementing distributed processing system(DPS) 50 of the present invention.Among Fig. 7, distributed processing system(DPS) 50 comprises 3 independent heterogeneous platforms 100,200 and 300 that connect in the represented grid configuration of the cloud form frame 55 of network.Among Fig. 7 by the represented network configuration of cloud form frame 55 form and agreement inessential, as long as it allows 700,800 and 900 information communications of platform.In addition, only using 3 platforms only is in order to illustrate, rather than restriction the present invention uses the platform of specific quantity.And then concrete network architecture is not crucial to the present invention.For example, the network controller that spendable another network architecture can use a platform to be attached thereto as all other platforms according to the present invention.

In the example of distributed processing system(DPS) 50, each in the platform 700,800 and 900 comprises processor 710,810 and 910 and storer 750,850 and 950 respectively.Each processor 710,810 and 910 comprises respectively uses 720,820 and 920, operating system 740,840 and 940, MI part 730,830 and 930.

Use 720,820 and 920 programs, write in advance and revise utilizing the present invention, or number to obtain service advantage provided by the invention specially.Use 720,820 and 920 and call operation of the present invention.

MI part 730,830 and 930 is corresponding to the above-mentioned MI part of discussing with reference to Fig. 6 600.

Operating system 740,840 and 940 is the standard operation systems that interrelate with respective processor 710,810 and 910 respectively.Platform 700,800 is different with 900.For example the Ultra Sparc microprocessor made with Sun Microsystems, Inc. of platform 700 is as processor 710 and use solaris operating system 740.The MIPS microprocessor that platform 800 is made with Silicon Graphics company is as processor 810 and use Unix operating system.At last, the Pentium microprocessor made with Intel Company of platform 900 is as processor 910 and use the Windows95 of Microsoft operating system.The present invention does not limit, and can adapt to similar platform.

Sun, Sun Micro systems, Solaris, Java and Sun Logo are the trade marks of Sun Micro systems company or in the U.S. and other national registered trademark.In the U.S. and other country, Ultra Sparc and all licensed uses of other SPARC trade mark and be the trade mark of SPARC International company.Have the architecture of the product of SPARC trade mark based on the development of Sun Micro system company.

Storer 750,850 and 950 has some functions, for example is used for the common store of related platform.Other function is storage application 720,820 and 920 before being dealt with separately device 710,810 and 910 and carrying out, MI part 730,830 and 930 and operating system 740,840 and 940.In addition, partial memory 750,850 and 950 can constitute all platforms 700,800 and 900 storeies of sharing in the network 50.

The E.MI service

The present invention uses CLIENT to implement.The client produces requests such as dirty calling and cleaning calling, server acknowledge request.

Each the MI part 730,830 and 930 that is shown in Fig. 7 preferably includes client portion and server section.Fig. 8 is the block diagram of customer platform 1000 and server platform 1100, is applied in platform 700,800 shown in Figure 7 and 900 any two.

Platform 1000 and 1100 comprises storer 1050 and 1150 and processor 1010 and 1110 respectively.Its function of parts in the platform 1000 and 1100 is identical with the function of the above-mentioned like of discussing with reference to Fig. 7.In this example, processor 1010 is carried out the client and is used 1020, and processor 1110 is carried out server and used 1120.Processor 1010 and 1110 also can be distinguished executive operating system 1040,1140 and MI part 1030,1130.

MI part 1030 and 1130 comprises server 1031 and 1131, the application 1032 and 1132 of call processing device, the garbage collector 1033 and 1133 of call processing device separately.But MI part 1030 and 1130 is each self-contained part of quoting also, both quote part each self-contained reference data part 1034 and 1134, reference count 1035 and 1135, authorize during register 1036 and 1136, what be used for that MI part 1030 or 1130 monitors respectively respectively quotes.

The application 1032 and 1132 of call processing device is represented customer service and is communicated by letter with 1131 with the server 1031 of the call processing device of representative server service respectively.Because platform 1000 and 1100 comprises the server of call processing device, the application of call processing device, garbage collector and quote part, arbitrary platform all can play client or server effect.

For discussing conveniently, hereinafter platform 1000 is represented customer platform and platform 1100 representative server platforms.In this example, the client uses 1020 and obtains the resource of quoting and using MI part 1030 partly to manage to the MI of server platform 1100 of distributed resource is sent dirty calling.

In addition, server platform 1100 may just be carried out server and use 1120.Server is used 1120 and also can be used MI part 1130 to send dirty calling, and when these dirty call resources were managed by MI part 1130, this was called out by MI part 1130 and handles.Perhaps, server is used 1120 resources that can use MI part 1130 to manage to MI part 1030 and is sent dirty calling.

Corresponding therewith, the server 1031, garbage collector 1033 and the reference counter 1035 that are used for the MI call processing device partly of customer platform 1000 are not worked, shown in Fig. 8 dash area.Equally, the application 1132 of the call processing device of the MI part 1130 of server platform 1100 is also because of being in dormant state with shadow representation.

When the client uses 1020 when obtaining to quote accordingly with resource, the application 1032 of call processing device sends dirty calling, is received by the server 1131 of call processing device.This dirty calling comprises the request lease duration.Quoting of 1131 pairs of dirty callings of server of call processing device increases during reference count and definite the mandate.In response, the server 1131 of call processing device sends the calling of returning during having mandate to the application 1030 of call processing device.The application 1032 of call processing device use during these mandates with during the mandate of new record more 1035 and determine when and this dirty calling to quote corresponding resource recyclable.

The server 1131 of call processing device also monitors during reference count corresponding with quoting of its management resource and the mandate.When one of this reference count 1135 is zero, maybe when during the mandate of quoting 1135 when having expired, no matter which incident at first takes place, the server 1131 of call processing device can start garbage collector 1131 with reclaim with counting serve as zero or mandate during end quote corresponding resource.

The clock synchronization on the platform 1000 and 1100 that the scheme of quoting do not need to comprise in the agreement of renting according to the invention process form.This scheme only requires the increase of time period between platform to compare.Rent not in the special time expiration, and behind specified time interval, expire.As long as an interval agreement is roughly arranged, platform 1000 and 1100 pairs of mandate lease durations will roughly be reached an agreement.And then, from the computing machine aspect, rent regularly quite longly, thereby a spot of clock rate monitors difference influence or influence is very not little.

The transmission of dirty calling can influence agreement.Rent and wait until always and rent and to upgrade with request before the expiration that this is rented, and meeting expires before MI part 1130 request of receiving if MI part 1030 keeps quoting.If like this, when receiving update request, MI part 1130 recoverable resources.Like this, when sending dirty calling, consider the time of transmitting to the platform of handling dirty call resources, transmit leg should add time factor to the lease duration of request, so that before the resource lease duration ends, and the dirty calling that can upgrade.

F. conclusion

According to the present invention, by provide with distributed processing system(DPS) in quote resource and authorize lease duration accordingly, the distributed garbage collection scheme can guarantee referential integrity and avoid memory leakage, thus when expiring during authorizing, quoting resource also expires.This resource is recovered then.When resource no longer is distributed processing in the formula disposal system and quotes, can collect this resource according to the counter of distributing to this quoting resource.

Conversion embodiment

The above-mentioned technology of renting relates to garbage collection.But a conversion embodiment of the present invention can be used for to rent detection failure and to make mistakes reparation.

Many systems are used for detection failure in the client-server environment, as heart beat (heart beat) or overtime.Use heart beat system, the client sends message with periodic intervals to server and indicates this client's survival.If have one at interval server do not receive message, then server is known the client or is transmitted between client and server in the communication agency (for example network) of data and breaks down.Use overtime system, set predetermined amount of time, if in this time interval, server does not receive any communication information from the client, and then server learns that client or communication agency break down.

Though these conventional systems enough point out when produce fault, client and server are not all known the state of system after the fault.For example, when the client is program and server when being the file system management program, the client may ask a specific file of server admin is carried out write operation.Though when producing fault, conventional fault detection system can detect this fault, the client does not know that fault occurs in after before the file execution write operation still being.In this case, the client can not determine system state.

Conversion embodiment of the present invention has solved the problems referred to above by the technology of renting that use is used for fault detect and reparation.When use was used for renting of fault detect, the client rented and is authorizing lease duration that the resource of server admin is carried out various processing to server requests.When this rented expiration, the client upgraded this and rents.If because of any reason, no matter be server failure or communication agency fault, this upgrades inefficacy.The client detects fault under above-mentioned arbitrary situation.At server side, if both not upgrade to rent, lease expiration and client clearly do not carry out deletion, then server learns that client or communication agency produce fault, thus server detects fault.

In case detection failure, by forwarding the state of consulting in advance to, server is repaired.That is, client and server are consulted one in advance in case the state that suffers or will shift when detecting fault.For example, in the said system example,, client and server just carry out rollback (roll back) if can consulting to detect fault in advance." rollback " is meant and places other any related entities such as client, server or files fault that preceding state takes place.Like this, in this example, if write operation is crossed in the server executed, then server returns to write operation to file and has just carried out preceding state, and write operation was not carried out after the client learnt detection failure, thereby the client can continue to handle.

Perhaps, client and server even further rollback.For example, client and server can be consulted in advance, no matter when when producing fault during the file processing, rollback make client and server get back to the client rent before the state of (for example before the document creation).Another kind of conversion is that rollback can replace the checkpoint that retreats in the file processing.Determine between this client and the server that the negotiation in advance of system state can several modes be carried out after the fault, but comprise signal exchange, read predetermined file or simple instruction client and server and in the time subsequently, be transferred to system state after the given fault all the time.

In addition, setting up lease duration, the client can provide the server failure repair procedure, and server also can provide client's fault restoration program equally.Like this, in case the fault of detecting, client and server all can call the other side's fault restoration program to carry out fault restoration.In this case, if server produces fault, in case the client detects this fault, then the repair procedure of client's invoking server is repaired server.For example, repair procedure can restart server and send message to the system manager.Similarly, if the client produces fault, then server calls client repair procedure is to carry out fault restoration to the client.

Because client and server are repaired mutually, carry out system management on distributed basis.Promptly, the technology of renting that is used for fault detect and reparation by use, be not centralized management, but replace conversion embodiment dispersant system management processing, thereby the client can carry out reparation to its client to server execution reparation and server as the sort of executive system management in some conventional system.

Conversion embodiment can be used for client-server relation, comprises that client and server are arranged in through the separate device of network service or are arranged in the operation of the distributed system of identical device.This be suitable for distributed system that conversion embodiment uses be a kind of Application No. as No.____, be entitled as the exemplary distribution formula system described in the patented claim text of " dynamically searching in the distributed system ", this application is quoted as a reference by the application.For clarity sake, conversion embodiment is that this situation of file system manager of renting the auxilary unit memory location is illustrated for server hereinafter.

The general introduction of leasing storage position

Memory storage has many memory locations that hold various data logic groups, and these data can be used by an above program.These logic groups can be taken from file, database or document.Renting the time that allows visit (for example read and write) memory location one predetermined length of memory location.Whether the data type of storing in the memory location or this memory location store data is unessential to renting the memory location.The memory location is rented and be can be used for the different memory level, for example Database field, file, storage block or actual storage locations.

In computer system or distributed system, many programs can be at war with to the file that is stored in the various sets of memory locations.Like this, sets of memory locations can have the program of many contention access.The technology of renting is used in the use of arbitration memory location under this environment.

When the renting of the sets of memory locations of using store file data, a program (" client ") is rented with in a time period (" lease duration ") visit sets of memory locations to system administration manager (" server ") request.Depend on availability, right of priority or other factor, lease duration can be refused or authorize to server.The lease duration of authorizing is lease duration or its part of whole request.In case the client is rented, then the client is in lease duration accessible storage set of locations.

During the request lease duration, the client can ask an accurate lease duration.In this case, whether be whole lease duration rather than its part of request to server if only authorizing this to rent.

When renting when effective, guarantee the client access sets of memory locations and can carry out the read and write operation it.Equally, at this effective lease duration, server also keeps the integrality of memory location.For example, at lease duration, it is deleted that server does not allow to rent file, it write or is polluted by other entity beyond the client, unless this entity is also rented.But after lease expiration, server no longer guarantees the integrality of file to the client, and like this, but this server deleted file or do the change of other essence also can be rented another client to carry out above-mentioned same operation.Server is regained and is not finished the memory location of renting.

Can there be relevant limiting parameter each memory location, as access parameter or franchise parameter.Access parameter determines that server supports the access type of this memory location.For example, the memory location can be decided to be read-only access.At this moment, server only allows thereafter mandate to rent this particular memory location is read.Otherwise the attempt that the client writes this memory location allows not serviced device.Other possible memory location access parameter can comprise write access, assigns access, heavy assigns access and sub-block access (for big storage block).

Relevant franchise parameter is stipulated the level of privilege that the client must have before mandate is rented.It is preferential that server can use the level of privilege parameter that the lease request of competition is obtained.In other words, when server is equipped with a lot of given lease request to same bank bit, according to the be ranked priority of these requests of the client's who makes request level of privilege.

Authorize by a plurality of visits simultaneously to same memory location, conversion embodiment also supports sets of memory locations is conducted interviews simultaneously.For example, if specific memory set of locations parameter regulation " reading " visit, server can be authorized this memory location is carried out a plurality of visits simultaneously and do not destroyed the integrality of this memory location.Rent simultaneously and can for example be used for mass file.Server is the integrality of not damaging big file than boy's piece of granted access file only.

In case client requests is rented, server is promptly sent an object back to the client, comprises definite lease duration, upgrades and rent, cancel the method for renting and carry out fault restoration.Should to as if the class example, it can number of ways expansion so that greater functionality to be provided, but basic class limits as follows with the Java programming language:

interface Lease{     obj FileHandle；     public long getDuration( )；     public void cancel( ) throws UnknownLeaseException，                                  RemoteException；     Public void renew (long renewDuration) throws                         LeaseDeniedException，                         UnknownLeaseException，                         RemoteException；     public void recover( )；}

Such comprises certain methods, comprises getDuration method, renew (renewal) method and recover (reparation) method.Segment length when " getDuration " method provides the mandate lease duration for the client.Represent during this period renting recently of server authorizes.Yet client's responsibility is segment length when determining remaining renting.

" renew " method allows the client to upgrade to rent, and can require more rental periods and needn't restart original lease request.The client wishes to upgrade the situation of renting and comprises that former renting proves not enough (that is, the additional memory location of using of customer requirement), or only authorization portions is rented (lease duration that promptly is shorter than request).

The client can use the renew method asking additional lease duration, or the client repeatedly calls the renew method continuously up to authorizing a plurality of additional lease durations.This renew method does not have rreturn value.If agree to upgrade, new lease duration is reflected in the object of making calling.Rent if server can not or be reluctant to upgrade, then its reason also illustrates in sending the object of calling.

When the client wishes that cancellation is rented, can call " cancel " method.Like this, thus this method call allow server regain other program of memory location addressable they.The cancel method guarantees that server can optimize the use of memory location in the distributed system.Be noted that if lease expiration and the indeterminate cancellation of client, then the server supposition is broken down.

" recover " method is provided by server, makes the client carry out fault restoration to server.For example, this fault restoration can comprise and restarts server.

It is No.____ that the memory location is rented at application number, be entitled as in the U.S. Patent application text of " method and system of leasing storage " and describe in detail, and this application is quoted as a reference by the application.

Implement to describe in detail

Fig. 9 represents to apply to the data handling system 9000 of conversion embodiment of the present invention.This data handling system 9000 comprises the computer system 9001 that is connected with the Internet 9002.Computing system 9001 comprises storer 9003, auxilary unit 9004, CPU (central processing unit) (CPU) 9006, input media 9008 and video display 9010.Storer 9003 further comprises operating system 9012 and program 9014, i.e. client.Operating system 9012 include file system administration managers 9016 are server, the file 9018 in its management auxilary unit.Auxilary unit 9004 also comprises Java ^TMSpace 9019.By rent the one or more files 9018 of client's 9014 request visits to server 9016 requests.In response, server 9016 can selection mandate as described below or is refused this and rent.

Java space 9019 is object resources banks that the program in the data handling system 9000 is used for storage object.Program is used Java space 9019 storage object and make them can be by other device visit on the network for good and all.Describe in detail in the U.S. Patent application text that submitted on November 17th, 1997 in the Java space, application number is No.08/971529, be entitled as " using the Database Systems of many types of project and project coupling ", this application transfers same assignee, is quoted as a reference in this application.It will be understood by those skilled in the art that computer system 9000 can comprise other or different parts.

Though conversion embodiment is described as being stored in the storer 9003, those skilled in the art's understanding also can be stored or read from other computer-readable media, as auxilary units such as hard disk, floppy disk or CD-ROM; Carrier wave from the Internet; Or the RAM of other form or ROM.In addition, it will be appreciated by those skilled in the art that conversion embodiment can be used for renting the data of other form in the supplementary storage, as database, electrical form and document.

Figure 10 is the process flow diagram of client's execution in step when server requests is rented.The 1st step of this step is to send lease request (step 10002) to server.This request is to have the function of a plurality of parameters to call out, and comprising: (1) client wishes the request memory location rented, the lease duration of (2) expectation, (3) accurately rent indicator, (4) client's access type of wishing, (5) client's privilege, (6) contain the object of restorative procedure.This method is included as the code that the client carries out fault restoration.

The memory location that the memory location indication of request will be rented.The lease duration of expectation comprises the period that the client wishes to utilize this memory location.Accurately lease request is just much of that during comprising the indication accurate lease request of execution or being shorter than request.The type of the memory location visit of the access type indication client requests of request.Access type comprises read access, write access, assigns access, heavy assigns access and sub-block access (to large-scale storage block).The privilege field is represented user or client's level of privilege.For forming effective request, client requests must be protected and be contained request memory location and expectation lease duration.

Usually there are two kinds of situations to produce the memory location lease request.The 1st kind of situation appears at when creating file, creates file and server is produced lease request with " establishment " order to be used for access file.When the client wishes to visit existing memory location or the file of having rented (situation of promptly renting simultaneously), produce the 2nd kind of situation.

After sending request, the client receives from server and rents object (step 10004).This is rented object and comprises various information, as mentioned above, comprises file handle (handle), getDuration method, renew (renewal) method, cancel (cancellation) method and recover (reparation) method.

After object was rented in reception, the client utilized this document (step 10005).Then, the client determines whether to finish file and uses (step 10006).If then the client rents (step 10007) to renting the object reference canceling method clearly to cancel this.By calling this method, the server cancellation is rented and can not made server think the generation fault.

If the client does not finish file and uses, then the client determines to rent whether will expire (step 10008).By calling the getDuration method and determining whether in predetermined threshold, carry out this step excess time.Not and to expire if rent, handle proceeding to step 10005.If but rent expiration, then the client sends update request (step 10009) to server.In this step, the client is to renting the object reference update method.After calling this method, whether successfully the client determines update request (step 10010).In this step, whether successfully return the client by update method and determine whether update request is successful.If success handles proceeding to step 10005.If but update method do not complete successfully, then the client is to renting object reference restorative procedure (step 10012).Because of update request does not complete successfully, the client knows the generation fault, thereby need call restorative procedure to carry out fault restoration.Restorative procedure is repaired server then.

Figure 11 is the process flow diagram of the conversion embodiment server execution in step according to the present invention.The 1st step of carrying out is visit Java space 9019 (steps 11002).Server keeps the Java space wherein store all objects during the lease request.These object storage in the Java space, thereby, if server detects fault, its addressable this Java space and by the object reference restorative procedure is carried out reparation.In addition, object is stored for a long time, if thereby server produce fault and lost efficacy, server can be to each the object reference restorative procedure in the Java space when then restarting, these objects reflection server failures all uncompleted renting in the time.In step 11002, server access comprises the Java space of all objects (if any), this to as if receive from the client as a part of renting.If in the Java space, object is arranged, then during server process, must produce fault.

Then, server is to each the object reference restorative procedure (step 11004) in the Java space.In this step, if in the Java space object is arranged, then it shows and hinders the server end process for some reason and must carry out repair.By each there being the client who rents call restorative procedure, server is carried out and is repaired.This restorative procedure for example can be restarted the client and make it to return the state of consulting in advance, the state before the request for example returned is rented.After calling used restorative procedure, server is deleted all objects (step 11006) from the Java space.After carrying out reparation, no longer need this object.

Behind the deletion object, server receives lease request (step 11008) from a client.After receiving lease request, server the object storage that receives in this request in Java space (step 11010).By the Java space storage object at permanent storage object, if produce fault, then server can be visited the Java space and the object reference restorative procedure is thought that the client carries out fault restoration.

Behind the storage object of Java space, by sending the object that has methods such as above-mentioned server restorative procedure, server authorizes lease request (step 11012) back to.The point that in server process some is later, server determine whether to receive update request (step 11014) from the client.If received update request, then server update is rented (step 11017).If but do not receive update request, then server would determine whether to receive that the client calls canceling method and the cancellation request (step 11015) of sending.If the client calls canceling method, then by from Java space deletion objects stored step 11010, the server cancellation is rented, and if this be to last uncompleted renting in the file, then server deletion this document (step 11016).

If do not receive the cancellation request, then server determines to rent whether expire (step 11018).If rent not expiration, then handle proceeding to step 11014.If expire but rent, then server is learnt the generation fault, thereby for having the client that finishes to rent to the object reference restorative procedure (step 11020) in the Java space.After calling restorative procedure, server is deleted this object, because it no longer needs (step 11022).

Though narrated method and system of the present invention with reference to most preferred embodiment, one skilled in the art will appreciate that the scope that does not break away from spirit of the present invention and claims, can make various changes to form and details.

Claims

1. the method in the data handling system with client and server is characterized in that it comprises the steps:

Described client rents to use the resource of described server admin to described server requests, and described lease request comprises the 1st repair procedure;

Server authorizes is described rents and makes described client use described resource and server to send the 2nd repair procedure to described client in can be during one;

Described client utilizes described resource;

Described client determines when will expire during described;

When described client determines describedly when renting expiration, send request to upgrade described renting to described server;

Determine whether described update request is successful;

When described client determined that update request gets nowhere, the client called described the 2nd repair procedure and thinks described server execution fault restoration;

Described server is determined to rent and when is expired;

When lease expiration is carried out fault restoration to the client, described the 1st repair procedure of described server calls.

2. the fault detection method of a distributed system is characterized in that processor carries out following step:

Provide and rent to visit the equipment of described distributed system;

Determine whether described renting requires to prolong;

Send update request to prolong described renting;

According to failing to receive, detect the incident that stops the described equipment of visit in response to newly the renting of described update request.

3. method as claimed in claim 2 is characterized in that, in case also comprise the incident that detects the described equipment of prevention visit, promptly carries out the step of repair procedure.