CN109992373A - Resource regulating method, approaches to IM and device and task deployment system - Google Patents
Resource regulating method, approaches to IM and device and task deployment system Download PDFInfo
- Publication number
- CN109992373A CN109992373A CN201711487682.3A CN201711487682A CN109992373A CN 109992373 A CN109992373 A CN 109992373A CN 201711487682 A CN201711487682 A CN 201711487682A CN 109992373 A CN109992373 A CN 109992373A
- Authority
- CN
- China
- Prior art keywords
- data
- task
- data center
- mark
- node
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
- G06F9/485—Task life-cycle, e.g. stopping, restarting, resuming execution
- G06F9/4856—Task life-cycle, e.g. stopping, restarting, resuming execution resumption being on a different machine, e.g. task migration, virtual machine migration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
- G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
Abstract
This application discloses a kind of resource regulating method, approaches to IM and device and task deployment systems, belong to big data technical field.In the task deployment system, calculate node sends location information request to meta data server, the location information of the data for the required by task processing that the location information request is used to request the calculate node pending;The location information answer that the meta data server returns is received, the location information answer includes the location information of the data of the pending required by task processing;Task deployment request is sent to resource manager node, the task deployment request includes mark, the mark of the mark of data center and calculate node of the pending task;Resource manager node requests to determine the data center for disposing the pending task according to the task deployment, to realize the task deployment across data center.
Description
Technical field
This application involves field of computer technology, in particular to a kind of resource regulating method, approaches to IM and device
And task deployment system.
Background technique
In the industry cycle, there are many definition, such as research institution Gartner to provide for big data (big data) is defined as: big
Data be need new tupe could have stronger decision edge, see clearly discovery power and process optimization ability adapt to magnanimity,
High growth rate and diversified information assets.Big data can be applied to several scenes, such as MPP database,
Data mining power grid, distributed file system, distributed data base, cloud computing platform, internet and expansible storage system
Deng." processing " is carried out to data by technology relevant to big data, to improve the application value of data.
With the rapid development of big data business, the scale of data center (Data Center, DC) cluster also gradually expands
Greatly, it needs to dispatch system by cluster resource and carries out United Dispatching come the resource to DC cluster.In the related technology, DC cluster is every
An independent system of distributed resource scheduling and an independent distributed file system are provided in a DC, each DC is logical
The storage and management that distributed file system realizes local data are crossed, local resource is realized by system of distributed resource scheduling
It dispatches and manages, the data across DC are periodically carried out between each DC and are synchronized, guarantee that each DC can be got on other DC
Data are to execute task.
During realizing the application, the inventor finds that the existing technology has at least the following problems: sending out in a DC
In the case of rising for task needs the data across DC to be performed, the data of other DC are synchronized to the DC using timing by the prior art
On mode executed in the DC, the processing delay of the task in this way is larger, and synchronizing for data causes the waste of resource.
Summary of the invention
In order to solve the problems, such as that DC can only execute task using local resource in the prior art, the embodiment of the invention provides
A kind of resource regulating method, approaches to IM and device and task deployment system enable DC to utilize the resource of other DC
Execution task, it is not necessary that the data on other DC are first synchronized to local, to reduce task processing delay.The technical solution
It is as follows:
In a first aspect, provide a kind of task deployment system, the task deployment system includes the first calculate node and the
One resource manager node, first calculate node and first resource manager node belong to the first data center;
First calculate node, for sending first task deployment request to first resource manager node, described the
The request of one task deployment includes the mark of a calculate node in the mark, the mark of a data center and the data center of a task
Know, the mark of the data center, which is used to indicate, will dispose the data center of the task;The first resource manager node,
The first task deployment request sent for receiving first calculate node;It is removed in the mark instruction of the data center described
Resource in the case where data center except first data center, in the corresponding data center of mark of Xiang Suoshu data center
Manager node sends the request of the second task deployment, and the second task deployment request is for described in the data center deployment
Task, the second task deployment request include mark, the mark of the task and the data center of the data center
In a calculate node mark.
This application provides a kind of task deployment systems, including calculate node and resource manager node, calculate node to exist
When execution task, the Resource Management node into same data center sends task deployment request, carries in resource management request
The mark of data center and calculate node where mark and the task data to be dealt with of being executed for task, resource
Manager node receives task deployment request, when task deployment request includes the mark of other data centers, then by task portion
Administration's request is sent to other data centers, the data center where the data handled by other with pending required by task Lai
The task is disposed, in this way in the local not data of required by task, the strange land processing of task may be implemented, realize strange land number
According to Real-Time Sharing, it is not necessary that the data of other data centers are synchronized to local by data are synchronous, task processing delay is smaller,
It reduces resource waste.
Wherein, first task deployment request and the request of the second task deployment can be the same or different.For example, first
An only corresponding task is requested in business deployment, and it is also the task that the request of the second task deployment is corresponding, and the two can be identical at this time;
Corresponding two tasks of first task deployment request, and the corresponding task of the second task deployment request, the two is different at this time.
Wherein, the pending task of the first calculate node can be one or more, the data of each required by task processing
It may include one or more files, the one or more file can be stored in same data center, can also be stored in
In different data centers.
In the illustrative implementation of first aspect one, the task deployment system further includes meta data server;Institute
The mark for being identified as the second data center of data center is stated, a calculate node is identified as the second calculating in the data center
The mark of node, first calculate node are also used to send location information request, the position to the meta data server
Information request includes Data Identification, and the location information request is used to request the Data Identification pair to the meta data server
The information of data center and calculate node where the data answered, the corresponding data of the Data Identification are the execution task institute
The data needed;The meta data server, for receiving the location information request;And request according to the positional information,
Location information answer is sent to first calculate node, the location information answer includes the mark of second data center
With the mark of second calculate node;First calculate node, for receiving the location information answer.
In this implementation, task deployment system provides place information inquiry clothes by arrangement meta data server
Business, the calculate node of each data center can be by asking meta data server to send request, to obtain required by task data
Location information, to provide basis for the task deployment across data center.
Wherein, meta data server provides the interface of the location information for data center's inquiry data.For example, metadata
Server provides the location information that two application programming interfaces to obtain data for data center.One interface be Map <
String, FileStatus > getCrossDCFileStatus (Path path)-, the interface are used to obtain the number where data
According to the information at center, such as the mark of data center.Another interface be Map<String, LocatedFileStatus>
GetCrossDCLocatedFileStatus (Path path)-, the interface are used to obtain the letter of the data center where data
Breath and calculate node information, such as data center mark and calculate node mark.
In the illustrative implementation of first aspect one, the mark for being identified as the second data center of the data center
Know, the mark for being identified as the second calculate node of a calculate node in the data center, the first resource manager node,
For receiving the first resource assignment messages of the Secondary resource manager node in second data center, first money
Source assignment messages are used to indicate the task and can dispose to second calculate node;First calculate node, for
Second calculate node sends task execution request, and the task execution request includes the mark and data mark of the task
Know, the Data Identification is used to indicate the data that the required by task is executed in second calculate node.
In this implementation, during the task deployment across data center, the second data center passes through resource allocation
Information come indicate the first data center the first calculate node can to the second calculate node deployment task of the second data center,
For first calculate node after receiving resource allocation message, the second calculate node of Xiang Suoshu sends task execution request.
Wherein, resource allocation information may include the resource for task of distributing to, for example including container identification.
Optionally, first calculate node is also used to receive the task action result that second calculate node returns.
It include multiple data centers in the first task deployment request in the illustrative implementation of first aspect one
Mark, the mark of the multiple data center is used to indicate the multiple data centers that can dispose the task, described first
Resource Management node, the number of the task will be disposed by being also used to be determined from the multiple data center according to resource allocation policy
According to center.
Wherein, the resource allocation policy may include scheduling strategy, and the scheduling strategy is following any: described
In the case that task can not be deployed in first calculate node, the task is allowed to be deployed in any idling-resource that possesses
In the calculate node of data center;In the case where the task can not be deployed in first calculate node, allow described appoint
Business is preferentially deployed in saving with first calculate node positioned at other calculating in rack for first data center
Point on;In the case where the task can not be deployed in first calculate node, the task is allowed preferentially to be deployed in institute
It states in other calculate nodes of the first data center.
Optionally, the resource allocation policy can also be included in use the scheduling strategy determine two or two with
In the case where upper data center, according to the attribute information of following at least one data center selection data center: two or two
The capacity or ratio of the idling-resource of each data center in the priority above data center, two or more data centers
Example, each data center in two or more data centers resource manager node and the first resource manager
The size of network bandwidth between node.
Optionally, the task deployment system further includes the host node of distributed file system;The meta data server,
For sending place information inquiry request to the host node, the place information inquiry request includes Data Identification;The master
Node, the place information inquiry request sent for receiving the meta data server;Inquiry request according to the positional information
Data center where determining the host node whether there is the corresponding data of the Data Identification;To the meta data server
Return position information inquiry response, the place information inquiry response includes at least one of following information: the host node
The data center of the data center at place identifies the calculate node of calculate node locating for data corresponding with the Data Identification
Mark or the place information inquiry response are used to indicate in the data center where the host node that there is no the data
Identify corresponding data.
Second aspect, provides a kind of resource regulating method, and the method is used for the first resource pipe of the first data center
Device node is managed, further includes the first calculate node in first data center, which comprises receives described first and calculates section
The first task that point is sent disposes request, and the first task deployment request includes the mark of the mark of a task, a data center
Know the mark with a calculate node in the data center, the mark of the data center, which is used to indicate, will dispose the task
Data center;In the case where the mark of the data center indicates the data center in addition to first data center, to
Resource manager node in the corresponding data center of mark of the data center sends the request of the second task deployment, and described the
The request of two task deployments is for task described in the data center deployment, the second task deployment request to include the data
The mark of a calculate node in the mark at center, the mark of the task and the data center.
In this implementation, resource manager node receives the task deployment sent with the calculate node of data center and asks
It asks, is carried in resource management request in the mark of being executed for task and the data where task data to be dealt with
Task deployment then is requested to send out by the mark of the heart and calculate node when task deployment request includes the mark of other data centers
Other data centers are sent to, this is disposed by the data center where other data with pending required by task processing
Business may be implemented the strange land processing of task, realize the real-time of strange land data in this way in the local not data of required by task
Shared, it is not necessary that the data of other data centers are synchronized to local by data are synchronous, task processing delay is smaller, reduces money
Source waste.
It include multiple data centers in the first task deployment request in the illustrative implementation of second aspect one
Mark, the mark of the multiple data center is used to indicate the multiple data centers that can dispose the task, the method
Further include: the data center of the task will be disposed by being determined from the multiple data center according to resource allocation policy.
In this implementation, resource manager node determines the data for disposing the task by resource allocation policy
On the one hand center can meet different dispatching requirements by the way that different resource allocation policies is arranged, on the other hand, pass through money
Source allocation strategy can be realized the Automatic dispatching of resource.
Wherein, resource allocation policy can carry in task deployment request, so that resource manager node can be with
It is directly requested to determine resource allocation policy according to task deployment, process is simple and convenient.
In the illustrative implementation of second aspect one, the resource allocation policy includes scheduling strategy, the scheduling
Strategy is following any: in the case where the task can not be deployed in first calculate node, allowing the task quilt
It is deployed in the calculate node of any data center for possessing idling-resource;Described first can not be deployed in the task to calculate
In the case where node, the task is allowed preferentially to be deployed in being located at first calculate node for first data center
With in other calculate nodes in rack;In the case where the task can not be deployed in first calculate node, allow
The task is preferentially deployed in other calculate nodes of first data center.
In this implementation, scheduling strategy can be arranged using various ways, to meet the diversified demand of user;
And in these strategies, preferentially handled using local resource, to guarantee the speed of performing task.
In the illustrative implementation of second aspect one, the resource allocation policy further includes using the scheduling plan
In the case where slightly determining two or more data centers, selected according to the attribute information of following at least one data center
Data center: each data center in the priority of two or more data centers, two or more data centers
Idling-resource capacity or each data center in ratio, two or more data centers resource manager node
The size of network bandwidth between the first resource manager node.
In this implementation, when scheduling strategy can not determine specific data center, according to priority, idle money
How many, network bandwidth in source etc. selects optimal data center to provide resource, guarantees the rational management of resource.
Optionally, resource allocation policy can also limit the range of choice or alternative condition of resource, determine data center
When chosen according to the range of choice or alternative condition.For example, the range of choice of resource is the 1-10 node of data center 1.Resource
Alternative condition be with graphics processor, big storage or calculate node of data center of big memory etc..
In the illustrative implementation of second aspect one, the second task deployment request is the first resource management
Heartbeat message between device node and the Secondary resource manager node, the heartbeat message include the mark of the data center
The mark of a calculate node in knowledge, the mark of the task and the data center.
In this implementation, task deployment request is realized simple by the heartbeat message transmission between data center.
In the illustrative implementation of second aspect one, the mark for being identified as the second data center of the data center
Know, the mark for being identified as the second calculate node of a calculate node in the data center, the method also includes: reception comes from
The first resource assignment messages of Secondary resource manager node in second data center, the first resource assignment messages are used
It can be disposed to second calculate node in the instruction task;Task execution request is sent to second calculate node,
The task execution request includes the mark and Data Identification of the task, and the Data Identification is used to indicate described second and calculates
The data of the required by task are executed in node.
In this implementation, data center resource manager is in addition to sending task deployment request to other data centers
Outside, the task deployment request of other data centers transmission can be also received, to be other data centers according to task deployment request
Calculate node distribute resource.
In the illustrative implementation of second aspect one, the method also includes: receive information resources manager node
The third task deployment of transmission is requested, and the third task deployment request includes the mark, first of first data center
The mark of third calculate node in the mark of business and first data center, the information resources manager node belong to the
Three data centers;Secondary resource assignment messages, the first resource assignment messages are sent to the information resources manager node
Being used to indicate the first task can dispose to the third calculate node.
In this implementation, the result of resource allocation returns to the transmission task deployment by resource allocation message and asks
The data center asked, the resource for allowing the data center for sending the task deployment request that resource allocation message is used to specify
Execution task.
The third aspect, provides a kind of resource regulating method, and the method calculates section for the first of the first data center
Point further includes first resource manager node in first data center, which comprises is sent to meta data server
Location information request, the location information request include Data Identification, and the location information request is used to take to the metadata
Business device requests the data center where the corresponding data of the Data Identification and the information of calculate node, and the Data Identification is corresponding
Data be to execute the data of the required by task;Receive the location information answer that the meta data server is sent, institute's rheme
Confidence breath response includes the mark of the second data center and the mark of the second calculate node, and second calculate node belongs to described
Second data center;First task deployment request is sent to the first resource manager node, the first task deployment is asked
Ask mark including a task, in the mark of a data center and the data center calculate node mark, the data
The mark at center, which is used to indicate, will dispose the data center of the task;Receive that the first resource manager node sends the
One resource allocation message, the first resource assignment messages are used to indicate the task can be to second calculate node portion
Administration.
In the illustrative implementation of the third aspect one, the method also includes: the second calculate node of Xiang Suoshu is sent
Task execution request, the task execution request include the mark and Data Identification of the task, and the Data Identification is for referring to
Show the data that the required by task is executed in second calculate node.
Fourth aspect provides a kind of approaches to IM, which comprises meta data server receives first and calculates
The location information request that node is sent, the location information request includes Data Identification, and the location information request is used for institute
It states meta data server and requests the data center where the corresponding data of the Data Identification and the information of calculate node, the number
It is to execute the data of the required by task according to corresponding data are identified, first calculate node belongs to the first data center;Root
According to the location information request, the first calculate node of Xiang Suoshu sends location information answer, and the location information answer includes institute
State the mark of the data center where the corresponding data of Data Identification and the mark of calculate node.
In the illustrative implementation of fourth aspect one, the method also includes: is determined according to the Data Identification
Two data centers, second data center are the data center where the corresponding data of the Data Identification;To described second
The host node of data center sends place information inquiry request, and the place information inquiry request includes Data Identification, and described the
The host node of two data centers is the host node of the distributed file system of second data center;Receive second data
The place information inquiry response that the host node at center returns, the place information inquiry response includes that the Data Identification is corresponding
The mark of calculate node where data;Alternatively, the method also includes: position, which is sent, to the host node of multiple data centers believes
Inquiry request is ceased, the place information inquiry request includes Data Identification, each data center in the multiple data center
Host node be each data center distributed file system host node;Receive the position letter that first host node returns
Inquiry response is ceased, the place information inquiry response includes at least one of following information: the corresponding number of the Data Identification
The described first main section is used to indicate according to the data center at place and the mark of calculate node or the place information inquiry response
The corresponding data of the Data Identification are not present in data center where point, first host node is in the multiple data
The host node of the distributed file system of any one data center in the heart.
In the illustrative implementation of fourth aspect one, the method also includes: receive the main section of third data center
The data synchronization message that point is sent, the data synchronization message are used to indicate in the third data center through data manipulation
The first data changed, the data manipulation comprises at least one of the following: creation data delete data and to other data centers
Synchrodata, the host node of the third data center are the host node of the distributed file system of the third data center;
According to the data synchronization message, first data corresponding data center after variation is recorded.
5th aspect, the embodiment of the invention provides a kind of resource scheduling device, the resource scheduling device includes being used for
Realize the unit for the method that the possible embodiment of any one in second aspect provides, such as receiving unit and transmission unit.
6th aspect, the embodiment of the invention provides a kind of resource scheduling device, the resource scheduling device includes being used for
Realize the unit for the method that the possible embodiment of any one in the third aspect provides, such as transmission unit and receiving unit.
7th aspect, the embodiment of the invention provides a kind of apparatus for management of information, the resource scheduling device includes being used for
Realize the unit for the method that the possible embodiment of any one in fourth aspect provides, such as receiving unit and transmission unit.
Eighth aspect, the embodiment of the invention provides a kind of resource scheduling device, described device includes: processor, storage
Device and communication interface;The processor, memory and communication interface are coupled by bus, and the memory is soft for storing
Part program can be connect when the processor is used to running or executing the software program in the memory by the communication
Mouth executes the method that the possible embodiment of any one in second aspect provides.
9th aspect, the embodiment of the invention provides a kind of resource scheduling device, described device includes: processor, storage
Device and communication interface;The processor, memory and communication interface are coupled by bus, and the memory is soft for storing
Part program can be connect when the processor is used to running or executing the software program in the memory by the communication
Mouth executes the method that the possible embodiment of any one in the third aspect provides.
Tenth aspect, the embodiment of the invention provides a kind of apparatus for management of information, described device includes: processor, storage
Device and communication interface;The processor, memory and communication interface are coupled by bus, and the memory is soft for storing
Part program can be connect when the processor is used to running or executing the software program in the memory by the communication
Mouth executes the method that the possible embodiment of any one in fourth aspect provides.
Tenth on the one hand, and the embodiment of the invention also provides a kind of computer-readable mediums, for storing for scheduling of resource
The program code that device executes, said program code include executing any one possible embodiment in second aspect to provide
The instruction of method.
12nd aspect, the embodiment of the invention also provides a kind of computer-readable mediums, for storing for scheduling of resource
The program code that device executes, said program code include executing any one possible embodiment in the third aspect to provide
The instruction of method.
13rd aspect, the embodiment of the invention also provides a kind of computer-readable mediums, for storing for information management
The program code that device executes, said program code include executing any one possible embodiment in fourth aspect to provide
The instruction of method.
Detailed description of the invention
Fig. 1 is a kind of structural schematic diagram of DC cluster provided in an embodiment of the present invention;
Fig. 2 is a kind of structural schematic diagram of device provided in an embodiment of the present invention;
Fig. 3 A- Fig. 3 C is a kind of flow chart of resource regulating method provided in an embodiment of the present invention;
Fig. 4 A- Fig. 4 B is the method flow diagram for the location information that a kind of DC provided in an embodiment of the present invention obtains data;
Fig. 5 is the method flow diagram for the location information that another kind DC provided in an embodiment of the present invention obtains data;
Fig. 6 is the method flow diagram for the location information that another kind DC provided in an embodiment of the present invention obtains data;
Fig. 7 is a kind of structural schematic diagram of resource scheduling device provided in an embodiment of the present invention;
Fig. 8 is the structural schematic diagram of another resource scheduling device provided in an embodiment of the present invention;
Fig. 9 is a kind of structural schematic diagram of apparatus for management of information provided in an embodiment of the present invention.
Specific embodiment
To keep the purposes, technical schemes and advantages of the application clearer, below in conjunction with attached drawing to the application embodiment party
Formula is described in further detail.
Fig. 1 is a kind of structural schematic diagram of DC cluster provided in an embodiment of the present invention, and referring to Fig. 1, DC cluster includes at least
Two DC, DC cluster shown in FIG. 1 are illustrated by taking 3 DC as an example, in other embodiments can also include two DC or
Four or more DC.
As shown in Figure 1, each DC includes resource allocation manager (Resource Manager, RM) node 40 and calculates
Node 20.
Calculate node 20 is also referred to as back end (DataNode) or memory node, and calculate node 20 is for storing data
(file), a file may include multiple pieces, and multiple pieces can store in same or different calculate node 20.Simultaneously
Calculate node 20 is also used to run application program to execute task, and provides the resource for executing required by task, and resource includes but not
It is limited to the resources such as CPU, storage, network interface, GPU.
Each application program in calculate node 20 includes an AM, and AM is a software module, and AM is used for as using journey
Sequence obtains resource and allocates resources to task performed by application program, and the work (Job) that application program executes can be split
For multiple tasks, the resource that multiple tasks can be provided using different calculate nodes is executed, and provides the meter of resource for multiple tasks
Operator node may belong to different DC also and may belong to identical DC.
Resource broker (Resource Agent, RA) is provided in each calculate node 20 (not show in figure
Out), RA is the resource and task management device in calculate node.On the one hand, RM node 40 of the RA timing into this DC reports resource
The operating status usage amount of each dimension resource (such as in container) of service condition and each container (Container), container
It is that resource is abstract, container encapsulates the various dimensions resource (CPU, storage, network interface, GPU etc.) in some calculate node, certainly
Resource can also be by other packing forms, such as virtual machine;On the other hand, RA receives the application program in calculate node 20
The request of Master (Application Master, AM), starts or stops operation to control container.
RM node 40 includes application manager (Applications Manager, ASM) (not shown), DC tune
Device (Datacenter Scheduler) 41 and local scheduler (Local Scheduler) 42 are spent, is mainly responsible for entire DC's
Scheduling of resource and management.ASM is responsible for managing the application program in entire DC, including application program is submitted, negotiated with DC scheduler
Resource is to start AM, monitoring AM operating status and restarting AM etc. when running failure.DC scheduler 41 and local scheduler
42 are all made of software module, are created by RM node 40.DC scheduler 42 includes across DC communicator (Cross DC
Communicator) 411 and across DC scheduler (Cross DC Scheduler) 412.Wherein, across DC scheduler 412 is for connecing
It receives and handles the task deployment request (for obtaining resource for task (Task)) from the AM in DC in calculate node 20, and sentence
It is disconnected that the task deployment request scheduling is still distributed to other DC processing to processing locality;Across DC communicator 411 is used for and it
The DC communicator 411 of his DC transmits heartbeat message, and heartbeat message is the information of cyclical transmission between DC, and heartbeat message is in addition to can
Outside confirming whether the DC connecting with this DC survives, the resource utilization messages and resource point of DC can also be carried out by heartbeat message
Synchronization with message.Resource utilization messages are used to indicate the service condition of resource in DC, for example, resource utilization messages may include
Used resource, allocated resource and workable resource.Resource allocation message, which is used to indicate, distributes to a task
Resource.When realization, it can be set two RM nodes 40 in each DC, two RM nodes 40 main and standby relation each other.
Each DC further includes namenode (NameNode) 10, and namenode is also referred to as host node, and host node 10 is for depositing
The metadata information of the file stored in each calculate node 20 of DC where storing up and managing.When realization, it can be set in each DC
Set two host nodes 10, the main and standby relation each other of host node 10.RM node 40 and host node 10 both can be real using identical hardware
It is existing, mutually independent two sets of hardware realizations can also be used.
In embodiments of the present invention, calculate node 20 has the resource for providing and executing required by task, while passing through deployment RA
To the function of the resource and task management device in calculate node, at this point, 40 groups of ingredients of the calculate node 20 of each DC and RM node
Cloth dispatches system, and distributed scheduling system can use Hadoop another kind resource coordination person (Yet Another
Resource Negotiator, YARN) framework realization.Calculate node 20 has the function of storing data, at this point, each DC
Calculate node 20 and host node 10 form distributed file system (Distributed File System, DFS), distributed text
Part system can be real using Hadoop distributed file system (Hadoop Distributed File System, HDFS) framework
It is existing.
Further, DC cluster further includes metadata (Metadata Service) server 30, meta data server 30
It is connect simultaneously with the host node 10 of each DC in DC cluster, the file in each DC can either be obtained from the host node 10 of each DC
Metadata information, the metadata information for the file in each DC that can also be will acquire stores, the metadata of file
Information includes Name & Location information of file etc..The metadata information of file where host node 10 can obtain in DC, when
When host node 10 will obtain the metadata information of other DC, then it can be realized by meta data server 30.Metadata Service
Device 30 can be independent server, on the node that also can integrate a DC in DC cluster.In addition, due in DC cluster
Each DC can be distributed in the different location in the whole world, therefore the meta data server 30 or the whole world (Global) DC are first
Data server.When realization, two meta data servers 30 can be set in each DC cluster, two meta data servers 30 are mutual
For main and standby relation.
In embodiments of the present invention, meta data server 30, host node 10 and RM node 40 can use legal log
Manager (Quorum Journal Manager, QJM) mechanism realizes active and standby synchronization.In addition, meta data server 30 can be with
High availability is guaranteed using failure branch controller (ZooKeeperFailoverController, ZKFC) mechanism.
It is worth noting that calculate node is used as back end and distribution in distributed file system simultaneously in the present invention
Calculate node in formula scheduling system, back end and distributed scheduling in other implementations in distributed file system
Calculate node in system can also independently be provided separately as two nodes.
In the prior art, since the distributed file system of each DC in DC cluster is independently disposed, so each DC
In the task of execution, the local data for whether being stored with and executing the required by task can only be judged, and be not aware that the data at it
Specific location on his DC, so when the local not stored data for having execution required by task, it can only be by regularly across DC number
The data for executing required by task are obtained according to synchronizing, and cause task processing delay larger.
The task that the application is collectively formed by resource manager node, calculate node, meta data server and host node
Deployment system realizes across DC deployment of task, so that the problems of the prior art are solved, each node in task deployment system
And the detailed functions of meta data server are referring to being hereinafter described.
First realization device provided in an embodiment of the present invention is illustrated in conjunction with specific hardware configuration below.
Fig. 2 shows the structural block diagram of a device 140 provided in an embodiment of the present invention, which can be resource tune
Device or apparatus for management of information are spent, which can be aforementioned resource manager node or calculate node, the letter
Ceasing managing device can be aforementioned metadata server.Referring to fig. 2, device 140 may include one or more than one core
The portions such as processor 31, the memory 32 including one or more computer readable storage mediums and communication interface 33
Part, processor 31 can be connected with bus with memory 32 and communication interface 33.It will be understood by those skilled in the art that in Fig. 2
The restriction of the structure shown not structure twin installation 140 may include than illustrating more or fewer components, or combination is certain
Component or different component layouts.Wherein:
Processor 31 is the control centre of device 140, utilizes each portion of various interfaces and connection whole device 140
Point, by running or executing the software program being stored in memory 32, and the data that calling is stored in memory 32, hold
Luggage sets 140 various functions and processing data, to carry out integral monitoring to device 140.Optionally, processor 31 can wrap
One or more than one processing unit are included, which can be central processing unit (English: Central
Processing Unit, referred to as: CPU) or network processing unit (English: Network Processor, referred to as: NP) etc..
Memory 32 can be used for storing various data, such as various configuration parameters and computer instruction, which refers to
Order can be executed by processor 31.Memory 32 may include high-speed random access memory, can also include non-volatile deposit
Reservoir, for example, at least disk, a flash memory, are also possible to other volatile solid-state parts.Correspondingly, memory 32 may be used also
To include Memory Controller, to provide access of the processor 31 to memory 32.
Communication interface 33 is connect by the way of wired or wireless with the other equipment in DC cluster, with other equipment into
Row communication, to carry out data transmission.
When resource scheduling device is resource manager node, processor 31 is used to receive described the by communication interface 33
The first task that one calculate node is sent disposes request, and the first task deployment request includes mark, a data for a task
The mark of a calculate node, the mark of the data center are used to indicate disposing in the mark at center and the data center
State the data center of task;The feelings of the data center in addition to first data center are indicated in the mark of the data center
Under condition, the is sent by resource manager node of the communication interface 33 into the corresponding data center of mark of the data center
The request of two task deployments, second task deployment request is for task described in the data center deployment, and described second
Business deployment request includes the mark of a calculate node in the mark, the mark of the task and the data center of the data center
Know.
Resource manager node receives the task deployment request sent with the calculate node of data center, which asks
The mark of data center and calculate node where mark and the task data to be dealt with for task of middle carrying being asked to be executed
Know, when task deployment request includes the mark of other data centers, then send other data centers for task deployment request,
The task is disposed by the data center where other data with pending required by task processing, is not had in this way locally
When the data of required by task, the strange land processing of task may be implemented, the Real-Time Sharing of strange land data is realized, without relying on data
Synchronous that the data of other data centers are synchronized to local, task processing delay is smaller, reduces resource waste.
When resource scheduling device is resource manager node, processor 31 is also used to execute following resource regulating method
Step performed by middle resource manager node.
When resource scheduling device is calculate node, processor 31 is used to send out by communication interface 33 to meta data server
Location information request is sent, the location information request includes Data Identification, and the location information request is used for the metadata
Server requests the data center where the corresponding data of the Data Identification and the information of calculate node, the Data Identification pair
The data answered are to execute the data of the required by task;The position that the meta data server is sent is received by communication interface 33
Information answer, the location information answer include the second data center mark and the second calculate node mark, described second
Calculate node belongs to second data center;It is first to first resource manager node transmission by communication interface 33
Business deployment request, first task deployment request include the mark of a task, in the mark and the data of a data center
The mark of a calculate node in the heart, the mark of the data center, which is used to indicate, will dispose the data center of the task;Pass through
Communication interface 33 receives the first resource assignment messages that the first resource manager node is sent, and the first resource distribution disappears
Breath is used to indicate the task can dispose to second calculate node.
Calculate node sends task deployment to the resource manager node of same data center and requests, in resource management request
The mark of being executed for task and the mark of data center and calculate node where task data to be dealt with are carried,
When task deployment request includes the mark of other data centers, then other data centers are sent by task deployment request, by
Data center where other data handled with pending required by task disposes the task, in this way locally without appointing
When data needed for business, the strange land processing of task may be implemented, realize the Real-Time Sharing of strange land data, without same by data
The data of other data centers are synchronized to local by step, and task processing delay is smaller, are reduced resource waste.
When resource scheduling device is calculate node, processor 31 is also used to execute to be calculated in following resource regulating method
Step performed by node.
When apparatus for management of information is meta data server, processor 31 is used to receive first by communication interface 33 and calculate
The location information request that node is sent, the location information request includes Data Identification, and the location information request is used for institute
It states meta data server and requests the data center where the corresponding data of the Data Identification and the information of calculate node, the number
It is to execute the data of the required by task according to corresponding data are identified, first calculate node belongs to the first data center;Root
According to the location information request, location information answer, the position are sent to first calculate node by communication interface 33
Information answer includes the mark of the data center where the corresponding data of the Data Identification and the mark of calculate node.
In this implementation, task deployment system provides place information inquiry clothes by arrangement meta data server
Business, the calculate node of each data center can be by asking meta data server to send request, to obtain required by task data
Location information, to provide basis for the task deployment across data center.
When resource scheduling device is meta data server, processor 31 is also used to execute in following resource regulating method
Step performed by meta data server.
Wherein, processor 31 is referred to by 33 messaging of communication interface or request, and processor 31 is sent out to communication interface 33
Control instruction is sent, so that 33 messaging of communication interface or request.
Fig. 3 A- Fig. 3 C is a kind of flow chart of resource regulating method provided in an embodiment of the present invention, and this method is based on Fig. 1 institute
The DC cluster shown realizes that wherein Fig. 3 B and Fig. 3 C illustrates only the interactive process in DC cluster between each node (equipment), ginseng
See that Fig. 3 A- Fig. 3 C, method flow include:
The RM node of 200: the one DC receives the application program that client is submitted, and being in the calculate node of the first DC should
Application assigned resource is to run the application program.
Step 200 is executed by the ASM in RM node, and ASM is receiving applications client (Application Client) (letter
Claim client) after the application program submitted, negotiate to distribute resource to run the application program with the DC scheduler in RM node, and
AM is created for the application program.
The calculate node of 201: the one DC determines the pending task of the application program.
Each application program includes an AM, and the work that application program executes is split as multiple by the AM in application program
Task, multiple tasks both can be parallel or serial, and AM is the scheduler administrator for the task that application program currently needs to be implemented.
AM needs to obtain resource for each task, to execute each task, detailed process is referring to hereinafter after splitting out multiple tasks
Description.
Wherein, the calculate node of step 201 is to run the calculate node of the application program.
The calculate node of 202: the one DC sends location information request to meta data server.
Wherein, the location information request includes Data Identification, and the location information request is used to take to the metadata
Business device requests the information of DC and calculate node where the corresponding data of the Data Identification, the corresponding data of the Data Identification
For the data for executing the required by task.
Correspondingly, meta data server receives the location information request that the calculate node of the first DC is sent.Such as Fig. 3 B, 3C institute
Show, DC1 is aforementioned first DC.
In embodiments of the present invention, the pending task of the calculate node of the first DC can be one or more, Mei Geren
The data handled needed for business may include one or more files, and the one or more file can be stored in same DC,
It can be stored in different DC.Data Identification in the location information request can be file name, such as file x, file
Title can be specified by user by interface parameters when step 200 submits application program.
In embodiments of the present invention, meta data server provides the interface of the location information for DC inquiry data.For example,
Meta data server provides the location information that two api interfaces to obtain data for DC.
One interface is Map<String, and FileStatus>getCrossDCFileStatus (Path path)-, this connects
Mouthful for obtaining the information of DC where data, returns Value Types are Map<String, FileStatus>.In the interface,
String indicates filename, and FileStatus is object, including file attribute, such as when number of copies, locating DC, document creation
Between, file status etc., MAP indicates set, can have multiple files and the corresponding file attribute of file, (Path in a MAP
Path it) is used to indicate whether specified DC, or is used to indicate specified DC.
Another interface is Map<String, LocatedFileStatus>getCrossDCLocatedFileStatus
(Path path)-, the interface are used to obtain the information of the DC where data and the information of calculate node, and return Value Types are Map
<String,LocatedFileStatus>.In the interface, the difference of LocatedFileStatus and FileStatus are also
It may include the information of the calculate node where file.
In practical applications, DC and calculate node where second interface polls also can be directly used in calculate node;
Calculate node first can also first inquire the DC at place using first interface, then using where second interface polls file
Calculate node.The letter of the calculate node of data can be inquired simultaneously to avoid a large amount of location information requests using the second various ways
Breath, leads to data flow excessive the problem of causing system crash.
203: meta data server determines the pending required by task of the calculate node of the first DC according to location information request
The location information of the data of processing.
Wherein, the location information of data includes the information of the DC where data and the information of calculate node, wherein data institute
The information of DC and the information of calculate node can be the mark of the DC where data and the mark of calculate node.For example, number
Information according to the DC at place is DC1, and the information of calculate node is N3, then it represents that the data are located in the N3 calculate node in DC1.
For meta data server in the location information for determining data, there are the following two kinds situations:
The first situation is stored with the full location information of the data in meta data server, including the DC where data
Information and calculate node information.Second situation, the location information of the not stored data, Huo Zhewei in meta data server
The full location information of the data is stored, such as is only stored with the information of the DC where data, the calculate node of not stored data
Information.When the information stored in meta data server is not enough to carry out response to location information request, then need to each
DC further requests the location information of the data.
In above-mentioned two situations, the location information of stored data can pass through two kinds of sides in meta data server
Formula obtains, and first way is each DC active reporting (uploading to meta data server), and the second way is metadata clothes
Business location information of the device to each DC request data, and the location information for the data requested is stored to meta data server
In.
The detailed process of first way is illustrated below by the flow chart that Fig. 4 A and Fig. 4 B are provided, wherein Fig. 4 B
The interactive process between each equipment is illustrated only, referring to fig. 4 A and Fig. 4 B, which includes:
2031: distributed type file system client side is connected to the host node of the distributed file system of DC, creates data
Metadata.
Wherein, distributed type file system client side is the client for accessing distributed file system in DC, distributed document
System client is usually a linux machine, is equipped with distributed file system client on distributed type file system client side
Software is held, distributed file system management role can be logged in by the software, to realize step 2031 and 2032.
In this application, distributed type file system client side and aforementioned applications client can be two stand-alone machines,
It can be realized using a machine by installing different client-side program.
The metadata of data includes the location information of data in step 2031, for example, distributed file system visitor in Fig. 4 B
Family end includes the location information of data :/opt/data/x to the metadata of write-in.
2032: data namely file are written to the calculate node of DC in distributed type file system client side.
The calculate node of 2033:DC writes the data into other by Data Migration Tools (Replication Tool)
DC, such as the first DC write data into the 2nd DC connecting with the first DC.
The Data Migration Tools can also pass through design by manually starting by distributed type file system client side
Calling interface calls and starts.
Need to illustrate when, the process of above-mentioned Data Migration is optional step, whether can carry out data with configured in advance and move
It moves.
In the present embodiment, the data are synchronized to other DC after creating data, guarantee the real-time of each DC data sharing
Property.
For example, Data Migration Tools are from DC1 to DC2 migrating data in Fig. 4 B, position of the data in DC1 are as follows:/opt/
Data/x, moving to the position in DC2 is /opt/dc2data/x.
The data synchronization message of the data is reported to meta data server, data synchronization message by the host node of 2034:DC
Path and the data before being synchronized including the data synchronize after path.Pass through the data synchronization message got, metadata
Server obtains the information of the DC at the place in the location information of the data.
For example, data are migrated and (replicated) to DC2 from DC1, path of the data in DC1 is /opt/data/x, in DC2
In path be /opt/dc2data/x.Path and the data before being synchronized by data in the synchronizing information synchronize after road
Diameter, can store the information of the DC where this document x in meta data server includes DC1 and DC2.
Wherein, the host node of DC is provided with PUT REST api interface, which is used for metadata operation each time
Data synchronization message be sent to meta data server, data synchronization message here, which is used to indicate in the 3rd DC, passes through number
The data changed according to operation, data synchronization message can also refer to other than it can indicate the operation to other DC synchrodatas
Show the operations such as creation data, deletion data.
In addition, can also directly be uploaded to by Data Migration Tools for the data synchronization message that synchrodata generates
Meta data server.
The detailed process of the second way is illustrated below by the flow chart that Fig. 4 B, Fig. 5 and Fig. 6 are provided, referring to
Fig. 4 B, Fig. 5 and Fig. 6, the second way include two kinds of situations: the first situation, the DC where meta data server given data,
Process is as shown in Fig. 4 B, Fig. 5 at this time:
2035: meta data server is determined according to location information request where the calculate node data to be processed of the first DC
DC information.
2036: the corresponding DC of the information of from meta data server to the DC determined host node sends place information inquiry and asks
It asks.
In embodiments of the present invention, the data of the pending required by task processing of the calculate node of the first DC can be one
Or multiple files, place information inquiry request include Data Identification, which can be the title of file.
2037: the host node of the corresponding DC of the information of the DC determined is answered to the inquiry of meta data server return position information
It answers, place information inquiry response includes the information of calculate node, such as the mark of calculate node.
Second case, meta data server do not know the DC where track data, and process is as shown in Figure 6 at this time:
2038: meta data server sends place information inquiry request to the host node of each DC.
In embodiments of the present invention, the data of the pending required by task processing of the calculate node of the first DC may include one
A or multiple files, place information inquiry request include Data Identification.
2039: for the host node of each DC to meta data server return position information inquiry response, which includes data
The information of DC where identifying corresponding data, such as the mark of DC;Or the response includes DC where the corresponding data of Data Identification
Information and calculate node information, such as the mark of DC and the mark of calculate node;Or the response is used to indicate in the DC not
There are the corresponding data of Data Identification.
In above two situation, meta data server can be asked by REST api interface to the transmission of the host node of DC
It asks.
By the above process, the location information of data is obtained in meta data server.
In embodiments of the present invention, meta data server can realize master metadata server, standby metadata using QJM mechanism
Synchronizing information between server.For example, master metadata server is recorded in each metadata operation by master metadata server
In Journal node (journal node) editor (edit) file, while standby meta data server can read the letter at any time
Breath.Active and standby meta data server can all realize regularly checkpoint (checkpoint) mechanism, i.e., active and standby meta data server will
After record flushes to the image file in file system in memory, checkpoint operation can be all carried out.Standby meta data server
Edit file can regularly load to memory, when guarantee active-standby switch being capable of quick start.
204: meta data server sends location information answer to the calculate node of the first DC, which includes
The information of DC where the data of the pending required by task processing of the calculate node of first DC and the information of calculate node.
Correspondingly, the calculate node of the first DC receives the location information answer that meta data server is sent.
In embodiments of the present invention, the interface type that meta data server is called according to the first DC returns to corresponding return
Value is returned, which can use java script (JavaScript, JS) object tag (JavaScript Object
Notation, JSON) format.
For example, location information answer can carry the location information of file x: DC1={ B1 in the step 204 of Fig. 3 B
N1, N4, B2@.. }, DC2={ B1@N3, N5, B2@.. }, the location information indicates N1 node, N4 of the B1 block in DC1 of file x
The positions such as node, in positions such as the N3 node of DC2, N5 nodes.
The calculate node of 205: the one DC sends first task deployment request, the first task portion to the RM node of the first DC
The pending mark of task of administration's calculate node of the request including the first DC, the mark of the mark of DC and calculate node.
Correspondingly, the first task that the calculate node that the RM node of the first DC receives the first DC is sent disposes request.
Wherein, the mark of at least one task can be carried in first task deployment request, the mark of each task is corresponding
The mark of several DC and the mark of calculate node.When the mark of the corresponding multiple DC of the mark of a task, need according to resource
A DC is selected in the corresponding DC of the mark of allocation strategy from multiple DC to dispose the task.
It optionally, can also include resource allocation policy in first task deployment request, resource allocation policy can be
It submits and is specified by user by interface parameters when the work.The resource allocation policy is used to indicate from the pending task
The corresponding DC of mark in, determine the pending required by task of the calculate node of the DC for disposing the task namely the first DC
The data of processing may be stored in simultaneously in multiple DC, select one to be used to dispose this from multiple DC using resource allocation policy
Task.The resource allocation policy may include scheduling strategy, and scheduling strategy can be following any:
1) feelings of the calculate node of the first DC can not, free time DC (RELAX_DC) strategy: be deployed in the task
Under condition, the task is allowed to be deployed in the calculate node of any DC for possessing idling-resource.Wherein, the calculating section of the first DC
Point is the calculate node for sending the corresponding task deployment request of the task, above-mentioned strategy namely if the resource that task deployment is requested
Localization is unable to satisfy, then the task deployment request is allowed to be scheduled for any DC for possessing idling-resource.
In the present embodiment, the resource localization refers in the calculate node for sending the task deployment request for institute
State pending task distribution resource.When resource localization is unable to satisfy, then can permit the task be scheduled for it is arbitrarily free
The DC of not busy resource is the calculate node place DC for needing to obtain resource without limiting.
Optionally, a time-out time can also be set in the strategy, after more than a time-out time, if the task deployment
When the resource localization of request is still unable to satisfy, task deployment request is allowed to be scheduled for any idling-resource that possesses
DC。
2) the first DC can not, local rack preferential (LOCAL_RACK_FIRST) strategy: be deployed in the task
Calculate node in the case where, allow the task to be preferentially deployed in being located at first calculate node for the first DC
With in other calculate nodes in rack.It is above-mentioned strategy namely if task deployment request resource localization be unable to satisfy,
The task is then allowed preferentially to be scheduled in other calculate nodes in same a rack with DC.
In the embodiment of the present invention, each DC includes multiple racks, and multiple calculate nodes, Mei Geji are disposed in each rack
Operator node can have multiple hard disks.
In the strategy, when resource localization is unable to satisfy, then it can permit the task and be preferentially scheduled for DC's
Task portion is unable to satisfy with other calculate nodes in other calculate nodes in rack, but on same a rack with DC
When administration, permission task is scheduled for on the calculate node of other racks of DC or other DC.
3) the calculating section of the first DC can not, local DC (DC_LOCALITY) preference strategy: be deployed in the task
In the case where point, the task is allowed preferentially to be deployed in other calculate nodes of the first DC.It is above-mentioned strategy namely such as
The resource localization of fruit task deployment request is unable to satisfy, then the task is allowed preferentially to be scheduled for other calculating with DC
On node.
In the strategy, when resource localization is unable to satisfy, then it can permit the task and be preferentially scheduled for DC's
In other calculate nodes, but when other calculate nodes of same a DC are unable to satisfy task deployment, permission task is scheduled for it
On his DC.
In other embodiments, resource localization can also refer to the rack where needing to obtain the calculate node of resource
Calculate node be task distribute resource, or the DC where needing to obtain the calculate node of resource calculate node be task
Distribute resource.
Further, for the ease of realizing across the resource localization between DC, can by extending the AM agreement of HADOOP,
Localization information is specified in first task deployment request, such as new field indicates localization information, e.g., node in AM agreement
It indicates there was only local computing node it can be selected that rack indicates all calculate nodes of this rack it can be selected that any is indicated in DC
Any calculate node can be used as selecting object.
Secondly, can allow to refer in first task deployment request by string expression by extending AM agreement
Determine scheduling strategy, such as DC_LOCALITY.
In addition to this, resource allocation policy can also limit the range of choice or alternative condition of resource, when determining DC by
It is chosen according to the range of choice or alternative condition.For example, the range of choice of resource is the 1-10 node of DC1.The alternative condition of resource
For with GPU, big storage or calculate node of DC of big memory etc..
Resource allocation policy can also include other rules, such as the resource allocation policy can also be included in using institute
It states in the case that scheduling strategy determines two or more DC, DC is selected according to the attribute information of following at least one DC:
The capacity of the idling-resource of each DC or ratio, two in the priority of two or more DC, two or more DC
Or the size of the network bandwidth between the RM node and the first RM node of each DC in two or more DC.Other above-mentioned rule
The part that then can be used as resource allocation policy carries in first task deployment request, can also be by the way of arranging in advance
Definition.
When not including resource allocation policy in first task deployment request, RM node can distribute plan using default resource
Slightly handled.
For example, first resource request can carry following information: task1:N3 DC1-P:0 in the step 205 of Fig. 3 B,
N8@DC2–P:0;Task2:N1@DC3-P:0, Policy=RELAX_DC, above- mentioned information indicate first resource request for for
Two task requests resources of task1 and task2, the location information of task1 data to be dealt with are the N3 node and DC2 of DC1
N8 node, the location informations of task2 data to be dealt with is the N1 node of DC3, while specifying the resource allocation of task2
Strategy is free time DC strategy, and the priority of DC1, DC2, DC3 are P:0.
The RM node of 206: the one DC is disposed according to first task and is requested, and determines to dispose in first task deployment request
Task DC, namely the pending corresponding DC of task in first task deployment request is determined, to pass through pending task
Corresponding DC is that pending task distributes resource.It is pending in request when determining to dispose by local DC for first task
When task distributes resource, local DC is that task pending in first task deployment request distributes the process of resource referring to step
207;When determining is that task pending in first task deployment request distributes resource by other DC, in the mark of the DC
In the case where knowing DC of the instruction in addition to the first DC, other DC are task distribution pending in first task deployment request
The process of resource is referring to step 208 and subsequent step.
Since first task deployment request can request to distribute resource, step 207 for two or more tasks simultaneously
It may be performed simultaneously with step 208.
It wherein, include the mark of multiple DC in the first task deployment request, the mark of the multiple DC is used to indicate
Multiple DC that the task can be disposed, the method also includes:
The DC of the task will be disposed by being determined from the multiple DC according to resource allocation policy.
In embodiments of the present invention, the RM node of the first DC determines the place of first task deployment request in the following way
Reason mode:
The RM node of first DC filters out satisfactory DC according to scheduling strategy in aforementioned resource allocation policy.
When filtering out the satisfactory DC of two or more, the RM node of the first DC is idle according to the priority of DC, DC
Network bandwidth etc. between the capacity or ratio of resource, the DC filtered out and the first DC is ranked up the DC filtered out, and selects
Sequence is selected in most preceding DC as the DC for handling first task deployment request.
When sorting in the manner described above, positioned at the priority of two or more most preceding DC, the capacity of idling-resource or
Person's ratio randomly chooses one as the processing first task from described two or multiple DC when perhaps network bandwidth is identical
The DC for disposing request either selects the DC of first response in described two or multiple DC as the first task is handled and disposes
The DC of request.Wherein, first response refers to the corresponding DC of first heartbeat message got in the assignment procedure.
It certainly,, can also be directly from institute when filtering out the satisfactory DC of two or more in other implementations
It states and randomly chooses a DC in the satisfactory DC of two or more as the DC, Huo Zhezhi for handling first task deployment request
The DC that first response in described two or more satisfactory DC is selected in selecting disposes request as the first task is handled
DC。
A DC is finally determined in the manner described above, when the DC is a DC, thens follow the steps 207;When the DC is
When other DC other than the first DC, 208 are thened follow the steps.
In above process, if will be according to the size of available resources come when carrying out DC sequence, the RM node of the first DC
Need first to obtain the resource utilization messages of each DC, resource utilization messages are used to indicate the service condition of resource in each DC, should
Resource utilization messages are transmitted by the heartbeat message between DC.
The RM node of 207: the one DC is that the pending task distributes resource according to first task deployment request.
It in embodiments of the present invention, directly should when first task deployment request is handled by local computing node
First task deployment request is sent to local scheduler and is handled.Local scheduler is first task deployment request pair
The task distribution resource answered, the resource can be a container, and the task can only use resource described in the container at this time.
The RM node of 208: the one DC sends the request of the second task deployment to the RM node of the 2nd DC, second task
For disposing the task to the 2nd DC, the second task deployment request includes the mark of the 2nd DC for deployment request
The mark of calculate node in knowledge, the mark of the task and the 2nd DC.
Correspondingly, the second task deployment request that the RM node that the RM node of the 2nd DC receives the first DC is sent.Second
Business deployment request may be the same or different with first task deployment request.For example, first task deployment request is two corresponding
Task, and only a corresponding task or first task deployment request carry resource allocation policy to the request of the second task deployment,
And the request of the second task deployment does not carry resource allocation policy.
Wherein, the 2nd DC is the DC that step 206 is determined, the 2nd DC can be for one, two or multiple, when
When 2nd DC includes multiple, each 2nd DC is performed both by the method flow of step 208-210.As shown in Fig. 3 B, 3C, DC2 and DC3
It is the 2nd DC.
In embodiments of the present invention, the second task deployment request is the first RM node and the 2nd RM node
Between heartbeat message, the heartbeat message include in the mark of the DC, the mark of the task and the DC one calculate section
The mark of point.The RM node of first DC can be saved the RM that second task deployment request is sent to the 2nd DC by heartbeat message
Point.
The RM node of 209: the two DC is scheduled the second task deployment request requested resource.
Across DC scheduler in the RM node of 2nd DC when receiving heartbeat message, can check in the heartbeat message whether
Carry task deployment request;If task deployment request (i.e. the second task deployment request) is carried, by the task deployment
Local scheduler is requested assignment to be handled.Local scheduler is the pending task according to task deployment request
Distribute resource;Resource allocation message is generated according to the resource of distribution, resource allocation message is used to indicate the task can be to institute
State the deployment of the second calculate node.Wherein, the resource allocation message can carry the resource for distributing to the task, such as container
Mark.
The RM node of 210: the two DC sends resource allocation message to the RM node of the first DC.
Correspondingly, the RM node of the first DC receives the resource allocation message that the RM node of the 2nd DC is sent.
In embodiments of the present invention, which is transmitted by heartbeat message.
For example, the RM node of DC2 sends resource allocation message, the resource allocation message packet to the RM node of DC1 in Fig. 3 C
Include the position where the resource for the calculate node that DC1 is distributed in N8@DC2, N8@DC2 expression.
The RM node of 211: the one DC sends resource allocation message to the calculate node of the first DC.
Correspondingly, the calculate node of the first DC receives the resource allocation message that the RM node of the first DC is sent.
The calculate node of 212: the one DC sends task execution request to the calculate node of the 2nd DC, and the task execution is asked
Mark and Data Identification including the task are asked, the Data Identification is used to indicate in second calculate node described in execution
The data of required by task.
Correspondingly, the task execution request that the calculate node that the calculate node of the 2nd DC receives the first DC is sent.
Optionally, the task execution request can also include container identification.
In embodiments of the present invention, the calculate node of the first DC requests the meter where resource according to the resource allocation message
Operator node executes the pending task, and the calculate node where the resource distributes to the pending task to be described
Resource where calculate node namely the first DC calculate node request the 2nd DC calculate node starting distribution container hold
Row task.
The calculate node of 213: the two DC executes institute using the resource that the task execution requests corresponding task is distributed to
It states task execution and requests corresponding task, and generate task action result.
Wherein, task action result is the result that calculate node executes that task obtains.
During execution task, the RM node of the 2nd DC sends running state information to the RM node of the first DC, should
Running state information is the running state information of the container of execution task;The RM node of first DC receives the RM node hair of the 2nd DC
The running state information sent.
Task action result is returned to the calculate node of the first DC by the calculate node of 214: the two DC.
Correspondingly, the calculate node of the first DC receives the task action result that the calculate node of the 2nd DC is sent.
The embodiment of the invention provides a kind of resource scheduling devices, the first resource manager applied to the first data center
Node further includes the first calculate node in first data center, and referring to Fig. 7, resource scheduling device includes:
The resource scheduling device can be by special hardware circuit, alternatively, software and hardware is implemented in combination with as resource management
Device node all or part of.The resource scheduling device includes: receiving unit 301 and transmission unit 302.Wherein, it receives single
Member 301, the first task deployment request sent for receiving first calculate node, the first task deployment request include
The mark of a calculate node in the mark of one task, the mark of a data center and the data center, the data center
Mark, which is used to indicate, will dispose the data center of the task;Transmission unit 302, for the mark instruction in the data center
In the case where data center in addition to first data center, in the corresponding data center of the mark of Xiang Suoshu data center
Resource manager node send the request of the second task deployment, second task deployment request is for the data center section
The task is affixed one's name to, the second task deployment request includes mark, the mark of the task and the number of the data center
According to the mark of a calculate node in center.
It optionally, include the mark of multiple data centers, the multiple data center in the first task deployment request
Mark be used to indicate the multiple data centers that can dispose the task, described device further includes determination unit 303, be used for root
The data center of the task will be disposed by determining from the multiple data center according to resource allocation policy.
Optionally, the resource allocation policy includes scheduling strategy, and the scheduling strategy is following any:
In the case where the task can not be deployed in first calculate node, the task is allowed to be deployed in any
In the calculate node for possessing the data center of idling-resource;
In the case where the task can not be deployed in first calculate node, the task is allowed preferentially to be deployed in
First data center is located at first calculate node in other calculate nodes in rack;
In the case where the task can not be deployed in first calculate node, the task is allowed preferentially to be deployed in
In other calculate nodes of first data center.
Optionally, the resource allocation policy further includes determining two or more numbers using the scheduling strategy
In the case where according to center, data center is selected according to the attribute information of following at least one data center: two or more
The capacity of the idling-resource of each data center or ratio in the priority of data center, two or more data centers,
The resource manager node of each data center in two or more data centers and the first resource manager section
The size of network bandwidth between point.
Optionally, the second task deployment request is the first resource manager node and the Secondary resource management
Heartbeat message between device node, the heartbeat message include the mark of the data center, the mark of the task and described
The mark of a calculate node in data center.
Optionally, the mark for being identified as the second data center of the data center, one calculates section in the data center
The mark for being identified as the second calculate node of point,
The receiving unit 301 is also used to receive the Secondary resource manager node in second data center
First resource assignment messages, the first resource assignment messages are used to indicate the task can be to second calculate node portion
Administration;
The transmission unit 302 is also used to send task execution request, the task execution to second calculate node
Request includes the mark and Data Identification of the task, and the Data Identification is used to indicate in second calculate node and executes institute
State the data of required by task.
Optionally, the receiving unit 301 is also used to receive the third task deployment of information resources manager node transmission
Request, the third task deployment request include mark, the mark of first task and described first of first data center
The mark of third calculate node in data center, the information resources manager node belong to third data center;
The transmission unit 302 is also used to send Secondary resource assignment messages, institute to the information resources manager node
Stating first resource assignment messages and being used to indicate the first task can dispose to the third calculate node.
Correlative detail is in combination with the embodiment of the method with reference to Fig. 3 A- Fig. 6.
It should be noted that above-mentioned determination unit 303 can be realized by processor alternatively, processor executes in memory
Program instruction is realized.Receiving unit 301, transmission unit 302 can be realized by communication interface alternatively, communication interface combination processing
Device is realized.
The embodiment of the invention provides a kind of resource scheduling device, applied to the first calculate node of the first data center,
It further include first resource manager node in first data center, referring to Fig. 8, resource scheduling device includes:
The resource scheduling device can be by special hardware circuit, alternatively, software and hardware is implemented in combination with as calculate node
All or part of.The resource scheduling device includes: transmission unit 401 and receiving unit 402.Wherein, transmission unit 401,
For sending location information request to meta data server, the location information request includes Data Identification, the location information
Request for where requesting data center from the corresponding data of the Data Identification to the meta data server and calculate node
Information, the corresponding data of the Data Identification are to execute the data of the required by task;Receiving unit 402, for receiving
The location information answer of meta data server transmission is stated, the location information answer includes the mark and second of the second data center
The mark of calculate node, second calculate node belong to second data center;The transmission unit 401, be also used to
The first resource manager node sends first task deployment request, and the first task deployment request includes the mark of a task
The mark of a calculate node in knowledge, the mark of a data center and the data center, the mark of the data center is for referring to
The data center of the task will be disposed by showing;The receiving unit 402 is also used to receive the first resource manager node hair
The first resource assignment messages sent, the first resource assignment messages, which are used to indicate the task, can calculate section to described second
Point deployment.
Optionally, the transmission unit 401 is also used to send task execution request to second calculate node, described
Task execution request includes the mark and Data Identification of the task, and the Data Identification is used to indicate second calculate node
The middle data for executing the required by task.
Correlative detail is in combination with the embodiment of the method with reference to Fig. 3 A- Fig. 6.
It should be noted that receiving unit 402 and transmission unit 401 can be realized by communication interface alternatively, communication interface
Combination processing device is realized.
The embodiment of the invention provides a kind of resource scheduling device, which can be aforementioned metadata service
Device, referring to Fig. 9, resource scheduling device includes:
The resource scheduling device can be by special hardware circuit, alternatively, being implemented in combination with for software and hardware takes as metadata
Be engaged in device all or part of.The resource scheduling device includes: receiving unit 501 and transmission unit 502.Wherein, receiving unit
501, for receiving the location information request of the first calculate node transmission, the location information request includes Data Identification, described
Location information request be used for request the corresponding data of the Data Identification to the meta data server where data center and
The information of calculate node, the corresponding data of the Data Identification are to execute the data of the required by task, and described first calculates section
Point belongs to the first data center;Transmission unit 502, for requesting according to the positional information, the first calculate node of Xiang Suoshu hair
Location information answer is sent, the location information answer includes the mark of the data center where the corresponding data of the Data Identification
With the mark of calculate node.
Optionally, described device further include: the determination unit 503, for determining the second number according to the Data Identification
According to center, second data center is the data center where the corresponding data of the Data Identification;The transmission unit
502, it is also used to send place information inquiry request, the place information inquiry request to the host node of second data center
Including Data Identification, the host node of second data center is the main section of the distributed file system of second data center
Point;The receiving unit 501 is also used to receive the place information inquiry response that the host node of second data center returns,
The place information inquiry response includes the mark of the calculate node where the corresponding data of the Data Identification;Alternatively, described
Transmission unit 502 is also used to send place information inquiry request, the place information inquiry to the host node of multiple data centers
Request includes Data Identification, and the host node of each data center in the multiple data center is the distribution of each data center
The host node of formula file system;The receiving unit 501 is also used to receive the place information inquiry that first host node returns
Response, the place information inquiry response includes at least one of following information: where the corresponding data of the Data Identification
Data center and calculate node mark or the place information inquiry response be used to indicate where first host node
Data center in be not present the corresponding data of the Data Identification, first host node be the multiple data center in
The host node of the distributed file system of any one data center.
Optionally, the receiving unit 501, the data for being also used to receive the host node transmission of third data center, which synchronize, to disappear
Breath, the data synchronization message are used to indicate the first data changed in the third data center by data manipulation, institute
State data manipulation to comprise at least one of the following: creation data delete data and to other data center's synchrodatas, the third
The host node of data center is the host node of the distributed file system of the third data center;
Described device further include: processing unit 504, for recording first data according to the data synchronization message
The corresponding data center after variation.
Correlative detail is in combination with the embodiment of the method with reference to Fig. 3 A- Fig. 6.
It should be noted that above-mentioned determination unit 503, processing unit 504 can be realized by processor alternatively, processor is held
Program instruction in line storage is realized.Receiving unit 501 and transmission unit 502 can be realized by communication interface alternatively, logical
Interface combination processing device is believed to realize.
In the above-described embodiments, it can be realized wholly or partly by software, hardware or a combination thereof.When using soft
When part program is realized, can entirely or partly it realize in the form of a computer program product.The computer program product packet
Include one or more computer instructions.When loading on computers and executing the computer instruction, entirely or partly generate
According to process or function described in the embodiment of the present invention.The computer can be general purpose computer, special purpose computer, computer
Network or other programmable devices.The computer instruction can be stored in computer readable storage medium, Huo Zhecong
One computer readable storage medium is transmitted to another computer readable storage medium, for example, the computer instruction can be with
Pass through wired (such as coaxial cable, twisted pair, optical fiber) or nothing from a web-site, computer, server or data center
Line (such as infrared, microwave etc.) mode is transmitted to another web-site, computer, server or data center.It is described
Computer readable storage medium can be any usable medium that computer can access or include one or more available
The data storage devices such as medium integrated server, data center.The usable medium can be magnetic medium (for example, floppy disk,
Hard disk, tape), optical medium or semiconductor medium (such as solid state hard disk (SSD)) etc..
Those of ordinary skill in the art will appreciate that realizing that all or part of the steps of above-described embodiment can pass through hardware
It completes, relevant hardware can also be instructed to complete by program, the program can store in a kind of computer-readable
In storage medium, storage medium mentioned above can be read-only memory, disk or CD etc..
The foregoing is merely the alternative embodiments of the application, not to limit the application, it is all in spirit herein and
Within principle, any modification, equivalent replacement, improvement and so on be should be included within the scope of protection of this application.
Claims (34)
1. a kind of task deployment system, which is characterized in that the task deployment system includes the first calculate node and first resource
Manager node, first calculate node and first resource manager node belong to the first data center;
First calculate node, for sending first task deployment request to the first resource manager node, described the
The request of one task deployment includes the mark of a calculate node in the mark, the mark of a data center and the data center of a task
Know, the mark of the data center, which is used to indicate, will dispose the data center of the task;
The first resource manager node, the first task deployment request sent for receiving first calculate node;?
In the case where data center of the mark instruction of the data center in addition to first data center, Xiang Suoshu data center
The corresponding data center of mark in resource manager node send the request of the second task deployment, second task deployment is asked
Ask for task described in the data center deployment, the second task deployment request include the data center mark,
The mark of a calculate node in the mark of the task and the data center.
2. task deployment system according to claim 1, which is characterized in that the task deployment system further includes metadata
Server;The mark for being identified as the second data center of the data center, the mark of a calculate node in the data center
For the mark of the second calculate node,
First calculate node is also used to send location information request to the meta data server, and the location information is asked
It asks including Data Identification, the location information request is used to request the corresponding number of the Data Identification to the meta data server
According to the data center at place and the information of calculate node, the corresponding data of the Data Identification are the number for executing the required by task
According to;
The meta data server, for receiving the location information request;And it requests according to the positional information, Xiang Suoshu
First calculate node sends location information answer, and the location information answer includes the mark of second data center and described
The mark of second calculate node;
First calculate node, for receiving the location information answer.
3. task deployment system according to claim 1, which is characterized in that the data center is identified as the second data
The mark at center, the mark for being identified as the second calculate node of a calculate node, the first resource pipe in the data center
Device node is managed, for receiving the first resource assignment messages of the Secondary resource manager node in second data center,
The first resource assignment messages are used to indicate the task and can dispose to second calculate node;
First calculate node, for sending task execution request, the task execution request to second calculate node
Mark and Data Identification including the task, the Data Identification, which is used to indicate in second calculate node, executes described appoint
Data needed for business.
4. task deployment system according to claim 1-3, which is characterized in that the first task deployment request
In include multiple data centers mark, the mark of the multiple data center, which is used to indicate, can dispose the multiple of the task
Data center, the first resource management node are also used to be determined from the multiple data center according to resource allocation policy
The data center of the task will be disposed.
5. a kind of resource regulating method, which is characterized in that the method is used for the first resource manager section of the first data center
Point further includes the first calculate node in first data center, which comprises
The first task deployment request that first calculate node is sent is received, the first task deployment request includes a task
Mark, in the mark of a data center and the data center calculate node mark, the mark of the data center uses
The data center of the task will be disposed in instruction;
In the case where the mark of the data center indicates the data center in addition to first data center, to the number
The request of the second task deployment, second task are sent according to the resource manager node in the corresponding data center of mark at center
Deployment request is for task described in the data center deployment, the second task deployment request to include the data center
The mark of a calculate node in mark, the mark of the task and the data center.
6. according to the method described in claim 5, it is characterized in that, including in multiple data in first task deployment request
The mark of the heart, the mark of the multiple data center are used to indicate the multiple data centers that can dispose the task, the side
Method further include:
The data center of the task will be disposed by being determined from the multiple data center according to resource allocation policy.
7. according to the method described in claim 6, it is characterized in that, the resource allocation policy includes scheduling strategy, the tune
Degree strategy is following any:
In the case where the task can not be deployed in first calculate node, the task is allowed to be deployed in any possess
In the calculate node of the data center of idling-resource;
In the case where the task can not be deployed in first calculate node, the task is allowed preferentially to be deployed in described
First data center is located at first calculate node in other calculate nodes in rack;
In the case where the task can not be deployed in first calculate node, the task is allowed preferentially to be deployed in described
In other calculate nodes of first data center.
8. the method according to the description of claim 7 is characterized in that the resource allocation policy further includes using the scheduling
In the case that strategy determines two or more data centers, selected according to the attribute information of following at least one data center
Select data center: in the priority of two or more data centers, two or more data centers in each data
The capacity or ratio of the idling-resource of the heart, each data center in two or more data centers resource manager section
The size of network bandwidth between point and the first resource manager node.
9. according to the described in any item methods of claim 5-8, which is characterized in that second task deployment request is described the
Heartbeat message between one resource manager node and the Secondary resource manager node, the heartbeat message include the number
According to the mark of a calculate node in the mark at center, the mark of the task and the data center.
10. according to the described in any item methods of claim 5-8, which is characterized in that the data center is identified as the second number
According to the mark at center, the mark for being identified as the second calculate node of a calculate node, the method are also wrapped in the data center
It includes:
Receive the first resource assignment messages of the Secondary resource manager node in second data center, first money
Source assignment messages are used to indicate the task and can dispose to second calculate node;
Task execution request is sent to second calculate node, the task execution request includes the mark sum number of the task
According to mark, the Data Identification is used to indicate the data that the required by task is executed in second calculate node.
11. according to the described in any item methods of claim 5-8, which is characterized in that the method also includes:
The third task deployment request that information resources manager node is sent is received, the third task deployment request includes described
The mark of the mark of first data center, the mark of first task and the third calculate node in first data center, institute
It states information resources manager node and belongs to third data center;
Secondary resource assignment messages are sent to the information resources manager node, the first resource assignment messages are used to indicate
The first task can be disposed to the third calculate node.
12. a kind of resource regulating method, which is characterized in that the method is used for the first calculate node of the first data center, institute
Stating further includes first resource manager node in the first data center, which comprises
Location information request is sent to meta data server, the location information request includes Data Identification, the location information
Request for where requesting data center from the corresponding data of the Data Identification to the meta data server and calculate node
Information, the corresponding data of the Data Identification are to execute the data of the required by task;
The location information answer that the meta data server is sent is received, the location information answer includes the second data center
The mark of mark and the second calculate node, second calculate node belong to second data center;
First task deployment request is sent to the first resource manager node, the first task deployment request includes one
The mark of a calculate node, the mark of the data center in the mark of business, the mark of a data center and the data center
The data center of the task will be disposed by being used to indicate;
The first resource assignment messages that the first resource manager node is sent are received, the first resource assignment messages are used for
Indicate that the task can be disposed to second calculate node.
13. according to the method for claim 12, which is characterized in that the method also includes:
Task execution request is sent to second calculate node, the task execution request includes the mark sum number of the task
According to mark, the Data Identification is used to indicate the data that the required by task is executed in second calculate node.
14. a kind of approaches to IM, which is characterized in that the described method includes:
Meta data server receives the location information request that the first calculate node is sent, and the location information request includes data mark
Know, the location information request is for the data where requesting the corresponding data of the Data Identification to the meta data server
The information at center and calculate node, the corresponding data of the Data Identification are to execute the data of the required by task, described first
Calculate node belongs to the first data center;
It requests according to the positional information, the first calculate node of Xiang Suoshu sends location information answer, the location information answer
The mark of mark and calculate node including the data center where the corresponding data of the Data Identification.
15. according to the method for claim 14, which is characterized in that the method also includes:
Determine that the second data center, second data center are the corresponding data of the Data Identification according to the Data Identification
The data center at place;
Place information inquiry request is sent to the host node of second data center, the place information inquiry request includes number
According to mark, the host node of second data center is the host node of the distributed file system of second data center;
Receive the place information inquiry response that the host node of second data center returns, the place information inquiry response bag
Include the mark of the calculate node where the corresponding data of the Data Identification;
Alternatively, the method also includes:
Place information inquiry request is sent to the host node of multiple data centers, the place information inquiry request includes data mark
Know, the host node of each data center in the multiple data center is the master of the distributed file system of each data center
Node;
The place information inquiry response that first host node returns is received, the place information inquiry response includes following information
At least one of: the mark or institute's rheme of data center and calculate node where the corresponding data of the Data Identification
Confidence breath inquiry response is used to indicate in the data center where first host node that there is no the Data Identification is corresponding
Data, first host node are the main section of the distributed file system of any one data center in the multiple data center
Point.
16. method according to claim 14 or 15, which is characterized in that the method also includes:
The data synchronization message that the host node of third data center is sent is received, the data synchronization message is used to indicate described the
The first data changed in three data centers by data manipulation, the data manipulation comprise at least one of the following: creation number
According to, delete data and to other data center's synchrodatas, the host node of the third data center is in the third data
The host node of the distributed file system of the heart;
According to the data synchronization message, first data corresponding data center after variation is recorded.
17. a kind of resource scheduling device, which is characterized in that applied to the first resource manager node of the first data center, institute
Stating further includes the first calculate node in the first data center, and described device includes:
Receiving unit, the first task deployment request sent for receiving first calculate node, the first task deployment
Request includes the mark of a calculate node in the mark, the mark of a data center and the data center of a task, the number
The data center of the task will be disposed by being used to indicate according to the mark at center;
Transmission unit, the feelings for data center of the mark instruction in the data center in addition to first data center
Under condition, the resource manager node in the corresponding data center of mark of Xiang Suoshu data center, which sends the second task deployment, is asked
It asks, second task deployment request is for task described in the data center deployment, the second task deployment request packet
Include the mark of the data center, in the mark of the task and the data center calculate node mark.
18. device according to claim 17, which is characterized in that include multiple data in the first task deployment request
The mark at center, the mark of the multiple data center is used to indicate the multiple data centers that can dispose the task, described
Device further includes determination unit, for according to resource allocation policy, determination will to dispose the task from the multiple data center
Data center.
19. device according to claim 18, which is characterized in that the resource allocation policy includes scheduling strategy, described
Scheduling strategy is following any:
In the case where the task can not be deployed in first calculate node, the task is allowed to be deployed in any possess
In the calculate node of the data center of idling-resource;
In the case where the task can not be deployed in first calculate node, the task is allowed preferentially to be deployed in described
First data center is located at first calculate node in other calculate nodes in rack;
In the case where the task can not be deployed in first calculate node, the task is allowed preferentially to be deployed in described
In other calculate nodes of first data center.
20. device according to claim 19, which is characterized in that the resource allocation policy further includes using the tune
In the case that degree strategy determines two or more data centers, according to the attribute information of following at least one data center
Select data center: each data in the priority of two or more data centers, two or more data centers
The capacity or ratio of the idling-resource at center, each data center in two or more data centers resource manager
The size of network bandwidth between node and the first resource manager node.
21. the described in any item devices of 7-20 according to claim 1, which is characterized in that the second task deployment request is institute
The heartbeat message between first resource manager node and the Secondary resource manager node is stated, the heartbeat message includes institute
State the mark of data center, in the mark of the task and the data center calculate node mark.
22. the described in any item devices of 7-20 according to claim 1, which is characterized in that the data center is identified as second
The mark of data center, the mark for being identified as the second calculate node of a calculate node in the data center,
The receiving unit is also used to receive the first resource of the Secondary resource manager node in second data center
Assignment messages, the first resource assignment messages are used to indicate the task and can dispose to second calculate node;
The transmission unit is also used to send task execution request, the task execution request packet to second calculate node
The mark and Data Identification of the task are included, the Data Identification is used to indicate in second calculate node and executes the task
Required data.
23. the described in any item devices of 7-20 according to claim 1, which is characterized in that the receiving unit is also used to receive
The third task deployment request that three resource manager nodes are sent, the third task deployment request includes in first data
The mark of the mark of the heart, the mark of first task and the third calculate node in first data center, the information resources
Manager node belongs to third data center;
The transmission unit, be also used to the information resources manager node send Secondary resource assignment messages, described first
Resource allocation message is used to indicate the first task and can dispose to the third calculate node.
24. a kind of resource scheduling device, which is characterized in that applied to the first calculate node of the first data center, described first
It further include first resource manager node in data center, described device includes:
Transmission unit, for sending location information request to meta data server, the location information request includes Data Identification,
The location information request is in the data where requesting the corresponding data of the Data Identification to the meta data server
The information of the heart and calculate node, the corresponding data of the Data Identification are to execute the data of the required by task;
Receiving unit, the location information answer sent for receiving the meta data server, the location information answer include
The mark of second data center and the mark of the second calculate node, second calculate node belong to second data center;
The transmission unit is also used to send first task deployment to the first resource manager node and request, and described first
Task deployment request includes the mark of a calculate node in the mark, the mark of a data center and the data center of a task
Know, the mark of the data center, which is used to indicate, will dispose the data center of the task;
The receiving unit is also used to receive the first resource assignment messages that the first resource manager node is sent, described
First resource assignment messages are used to indicate the task and can dispose to second calculate node.
25. device according to claim 24, which is characterized in that the transmission unit is also used to calculate to described second
Node sends task execution request, and the task execution request includes the mark and Data Identification of the task, the data mark
Knowledge is used to indicate the data that the required by task is executed in second calculate node.
26. a kind of apparatus for management of information, which is characterized in that be applied to meta data server, described device includes:
Receiving unit, for receiving the location information request of the first calculate node transmission, the location information request includes data
Mark, the location information request is for the number where requesting the corresponding data of the Data Identification to the meta data server
According to the information at center and calculate node, the corresponding data of the Data Identification are to execute the data of the required by task, described
One calculate node belongs to the first data center;
Transmission unit, for requesting according to the positional information, the first calculate node of Xiang Suoshu sends location information answer, described
Location information answer includes the mark of the data center where the corresponding data of the Data Identification and the mark of calculate node.
27. device according to claim 26, which is characterized in that described device further include: determination unit, for according to institute
It states Data Identification and determines that the second data center, second data center are the data where the corresponding data of the Data Identification
Center;
The transmission unit is also used to send place information inquiry request, institute's rheme to the host node of second data center
Setting information inquiring request includes Data Identification, and the host node of second data center is the distribution of second data center
The host node of file system;
The receiving unit is also used to receive the place information inquiry response that the host node of second data center returns, institute
State the mark that location information inquiry response includes the calculate node where the corresponding data of the Data Identification;
Alternatively, the transmission unit, is also used to send place information inquiry request, institute's rheme to the host node of multiple data centers
Setting information inquiring request includes Data Identification, and the host node of each data center in the multiple data center is each data
The host node of the distributed file system at center;
The receiving unit is also used to receive the place information inquiry response that first host node returns, the location information
Inquiry response includes at least one of following information: data center where the corresponding data of the Data Identification and calculating section
The mark or the place information inquiry response of point are used to indicate in the data center where first host node and are not present
The corresponding data of the Data Identification, first host node are point of any one data center in the multiple data center
The host node of cloth file system.
28. the device according to claim 26 or 27, which is characterized in that the receiving unit is also used to receive third number
According to the data synchronization message that the host node at center is sent, the data synchronization message, which is used to indicate in the third data center, leads to
The first data crossing data manipulation and changing, the data manipulation comprise at least one of the following: creation data, delete data and to
Other data center's synchrodatas, the host node of the third data center are the distributed field system of the third data center
The host node of system;
Described device further include: processing unit, for recording first data after variation according to the data synchronization message
Corresponding data center.
29. a kind of resource scheduling device, described device includes: processor, memory and communication interface;The processor is deposited
Reservoir and communication interface are coupled by bus, and the memory is for storing software program, which is characterized in that when the processing
When device is used to running or executing the software program in the memory, such as claim 5- can be executed by the communication interface
11 described in any item methods.
30. a kind of resource scheduling device, described device includes: processor, memory and communication interface;The processor is deposited
Reservoir and communication interface are coupled by bus, and the memory is for storing software program, which is characterized in that when the processing
When device is used to running or executing the software program in the memory, such as claim 12 can be executed by the communication interface
Or method described in 13.
31. a kind of apparatus for management of information, described device includes: processor, memory and communication interface;The processor is deposited
Reservoir and communication interface are coupled by bus, and the memory is for storing software program, which is characterized in that when the processing
When device is used to running or executing the software program in the memory, such as claim can be executed by the communication interface
The described in any item methods of 14-16.
32. a kind of computer readable storage medium, including instruction, which is characterized in that when run on a computer, make to succeed in one's scheme
Calculation machine executes such as the described in any item methods of claim 5-11.
33. a kind of computer readable storage medium, including instruction, which is characterized in that when run on a computer, make to succeed in one's scheme
Calculation machine executes method as described in claim 12 or 13.
34. a kind of computer readable storage medium, including instruction, which is characterized in that when run on a computer, make to succeed in one's scheme
Calculation machine executes such as the described in any item methods of claim 14-16.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711487682.3A CN109992373B (en) | 2017-12-29 | 2017-12-29 | Resource scheduling method, information management method and device and task deployment system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711487682.3A CN109992373B (en) | 2017-12-29 | 2017-12-29 | Resource scheduling method, information management method and device and task deployment system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109992373A true CN109992373A (en) | 2019-07-09 |
CN109992373B CN109992373B (en) | 2021-04-09 |
Family
ID=67111407
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711487682.3A Active CN109992373B (en) | 2017-12-29 | 2017-12-29 | Resource scheduling method, information management method and device and task deployment system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109992373B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111340613A (en) * | 2020-02-26 | 2020-06-26 | 中国邮政储蓄银行股份有限公司 | Job processing method, system and storage medium |
CN112085378A (en) * | 2020-09-04 | 2020-12-15 | 中国平安财产保险股份有限公司 | Resource allocation method, device, computer equipment and storage medium |
CN112383878A (en) * | 2020-09-27 | 2021-02-19 | 中国信息通信研究院 | Collaborative computing method and electronic device |
WO2022088612A1 (en) * | 2020-10-27 | 2022-05-05 | 上海商汤临港智能科技有限公司 | Task processing method, apparatus, device and system, and storage medium |
Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101167054A (en) * | 2005-05-27 | 2008-04-23 | 国际商业机器公司 | Methods and apparatus for selective workload off-loading across multiple data centers |
US20110055604A1 (en) * | 2008-04-21 | 2011-03-03 | Adaptive Computing Enterprises Inc. formerly known as Cluster Resources, Inc. | System and method for managing energy consumption in a compute environment |
CN101997929A (en) * | 2010-11-29 | 2011-03-30 | 北京卓微天成科技咨询有限公司 | Data access method, device and system for cloud storage |
CN102426542A (en) * | 2011-10-28 | 2012-04-25 | 中国科学院计算技术研究所 | Resource management system for data center and operation calling method |
CN102445978A (en) * | 2010-10-12 | 2012-05-09 | 深圳市金蝶中间件有限公司 | Method and device for managing data center |
CN102567851A (en) * | 2011-12-29 | 2012-07-11 | 武汉理工大学 | Safely-sensed scientific workflow data layout method under cloud computing environment |
CN102739785A (en) * | 2012-06-20 | 2012-10-17 | 东南大学 | Method for scheduling cloud computing tasks based on network bandwidth estimation |
CN103530182A (en) * | 2013-10-22 | 2014-01-22 | 海南大学 | Working scheduling method and device |
CN104104655A (en) * | 2013-04-07 | 2014-10-15 | 华为技术有限公司 | Resource release method, device and system |
CN104683161A (en) * | 2015-03-18 | 2015-06-03 | 杭州华三通信技术有限公司 | Network management method and device based on SaaS (software as a service) |
CN106201698A (en) * | 2016-07-15 | 2016-12-07 | 北京金山安全软件有限公司 | Method and device for managing application program and electronic equipment |
CN106648464A (en) * | 2016-12-22 | 2017-05-10 | 柏域信息科技(上海)有限公司 | Multi-node mixed block cache data read-writing method and system based on cloud storage |
CN106921977A (en) * | 2015-12-26 | 2017-07-04 | 华为技术有限公司 | A kind of service quality planing method, apparatus and system based on Business Stream |
CN107291746A (en) * | 2016-03-31 | 2017-10-24 | 阿里巴巴集团控股有限公司 | A kind of method and apparatus for storing and reading data |
-
2017
- 2017-12-29 CN CN201711487682.3A patent/CN109992373B/en active Active
Patent Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101167054A (en) * | 2005-05-27 | 2008-04-23 | 国际商业机器公司 | Methods and apparatus for selective workload off-loading across multiple data centers |
US20110055604A1 (en) * | 2008-04-21 | 2011-03-03 | Adaptive Computing Enterprises Inc. formerly known as Cluster Resources, Inc. | System and method for managing energy consumption in a compute environment |
CN102445978A (en) * | 2010-10-12 | 2012-05-09 | 深圳市金蝶中间件有限公司 | Method and device for managing data center |
CN101997929A (en) * | 2010-11-29 | 2011-03-30 | 北京卓微天成科技咨询有限公司 | Data access method, device and system for cloud storage |
CN102426542A (en) * | 2011-10-28 | 2012-04-25 | 中国科学院计算技术研究所 | Resource management system for data center and operation calling method |
CN102567851A (en) * | 2011-12-29 | 2012-07-11 | 武汉理工大学 | Safely-sensed scientific workflow data layout method under cloud computing environment |
CN102739785A (en) * | 2012-06-20 | 2012-10-17 | 东南大学 | Method for scheduling cloud computing tasks based on network bandwidth estimation |
CN104104655A (en) * | 2013-04-07 | 2014-10-15 | 华为技术有限公司 | Resource release method, device and system |
CN103530182A (en) * | 2013-10-22 | 2014-01-22 | 海南大学 | Working scheduling method and device |
CN104683161A (en) * | 2015-03-18 | 2015-06-03 | 杭州华三通信技术有限公司 | Network management method and device based on SaaS (software as a service) |
CN106921977A (en) * | 2015-12-26 | 2017-07-04 | 华为技术有限公司 | A kind of service quality planing method, apparatus and system based on Business Stream |
CN107291746A (en) * | 2016-03-31 | 2017-10-24 | 阿里巴巴集团控股有限公司 | A kind of method and apparatus for storing and reading data |
CN106201698A (en) * | 2016-07-15 | 2016-12-07 | 北京金山安全软件有限公司 | Method and device for managing application program and electronic equipment |
CN106648464A (en) * | 2016-12-22 | 2017-05-10 | 柏域信息科技(上海)有限公司 | Multi-node mixed block cache data read-writing method and system based on cloud storage |
Non-Patent Citations (1)
Title |
---|
荀亚玲 等: "MapReduce集群环境下的数据放置策略", 《软件学报》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111340613A (en) * | 2020-02-26 | 2020-06-26 | 中国邮政储蓄银行股份有限公司 | Job processing method, system and storage medium |
CN111340613B (en) * | 2020-02-26 | 2023-10-03 | 中国邮政储蓄银行股份有限公司 | Job processing method, job processing system and storage medium |
CN112085378A (en) * | 2020-09-04 | 2020-12-15 | 中国平安财产保险股份有限公司 | Resource allocation method, device, computer equipment and storage medium |
CN112383878A (en) * | 2020-09-27 | 2021-02-19 | 中国信息通信研究院 | Collaborative computing method and electronic device |
WO2022088612A1 (en) * | 2020-10-27 | 2022-05-05 | 上海商汤临港智能科技有限公司 | Task processing method, apparatus, device and system, and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN109992373B (en) | 2021-04-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11249815B2 (en) | Maintaining two-site configuration for workload availability between sites at unlimited distances for products and services | |
CN107066319B (en) | Multi-dimensional scheduling system for heterogeneous resources | |
US10873623B2 (en) | Dynamically modifying a cluster of computing nodes used for distributed execution of a program | |
US9460185B2 (en) | Storage device selection for database partition replicas | |
US10084858B2 (en) | Managing continuous priority workload availability and general workload availability between sites at unlimited distances for products and services | |
EP3069274B1 (en) | Managed service for acquisition, storage and consumption of large-scale data streams | |
EP3069228B1 (en) | Partition-based data stream processing framework | |
US9276959B2 (en) | Client-configurable security options for data streams | |
US9858322B2 (en) | Data stream ingestion and persistence techniques | |
CN102103518B (en) | System for managing resources in virtual environment and implementation method thereof | |
CA2930026C (en) | Data stream ingestion and persistence techniques | |
CN102981929B (en) | The management method of disk mirroring and system | |
US9716746B2 (en) | System and method using software defined continuity (SDC) and application defined continuity (ADC) for achieving business continuity and application continuity on massively scalable entities like entire datacenters, entire clouds etc. in a computing system environment | |
CN104113597A (en) | Multi- data-centre hadoop distributed file system (HDFS) data read-write system and method | |
CN109992373A (en) | Resource regulating method, approaches to IM and device and task deployment system | |
CN108132775B (en) | Tenant management system and method | |
CN106980678A (en) | Data analysing method and system based on zookeeper technologies | |
JP2015510187A (en) | Partition management in a scalable environment | |
CN108491163B (en) | Big data processing method and device and storage medium | |
CN107528871A (en) | Data analysis in storage system | |
CN116881012A (en) | Container application vertical capacity expansion method, device, equipment and readable storage medium | |
CN114745377A (en) | Edge cloud cluster service system and implementation method | |
CN112328388B (en) | Parallel computing method and system fusing multithreading and distributed technology | |
CN113918297A (en) | Distributed scheduling system, distributed scheduling method, device and medium | |
KR20230174150A (en) | Distributed cloud system, distributed cloud system data processing method and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |