tencent cloud

Overview

The unified namespace capability of GooseFS integrates the access semantics of different underlying storage systems through a transparent naming mechanism, providing users with a unified interactive view for data management.

Leveraging this capability, GooseFS connects and communicates with different underlying storage systems, such as local file systems, Tencent Cloud Object Storage (COS), and Tencent Cloud HDFS (CHDFS), and provides unified access APIs and file protocols for upper-layer businesses. In this way, the business side only needs to call the access APIs provided by GooseFS to access data stored in different underlying storage systems.

The figure above shows the working principle of the unified namespace. You can use GooseFS's namespace creation instruction create ns to mount specified file directories of COS and CHDFS to GooseFS, and then use the gfs:// unified schema to access data. Details are as follows:

• COS contains 3 buckets in total: bucket-1, bucket-2, and bucket-3. bucket-1 contains the BU_A and BU_B directories, and both bucket-1 and bucket-2 are mounted to GooseFS.
• CHDFS contains 4 directories: BU_E, BU_F, BU_G, and BU_H. All of them are mounted to GooseFS, except BU_H.
• Among GooseFS file operations, the BU_A and BU_E directories can be accessed using the unified schema gfs://, and the files are cached in the local file system of GooseFS.
• The BU_A and BU_E directories stored in the underlying file systems (COS and HDFS) have been mounted to GooseFS. If files are cached on GooseFS, they can be accessed through the unified schema gfs:// (for example, hadoop fs ls gfs://BU_A), and they can also be accessed through the namespace of each remote file system (for example, hadoop fs ls cosn://bucket-1/BU_A).
• If the files are not cached on GooseFS, they cannot be accessed through the unified schema gfs:// because the files are not cached in the local file system of GooseFS, but they can still be accessed through the namespace of the underlying storage systems.

Using the Unified Namespace Capability

You can use the create ns instruction to create a namespace in GooseFS and map underlying storage systems to GooseFS. Currently supported underlying storage systems include COS, CHDFS, and local HDFS. The procedure for creating a namespace is similar to that for mounting a file volume disk in a Linux file system. With the namespace created, GooseFS can provide clients with a file system with uniform access semantics. The current operation instruction set for GooseFS namespaces is as follows:

$ goosefs ns
Usage: goosefs ns [generic options]
       [create <namespace> <CosN/Chdfs path> <--wPolicy <1-6>> <--rPolicy <1-5>> [--readonly] [--shared] [--secret fs.cosn.userinfo.secretId=<AKIDxxxxxxx>] [--secret fs.cosn.userinfo.secretKey=<xxxxxxxxxx>] [--attribute fs.ofs.userinfo.appid=1200000000][--attribute fs.cosn.bucket.region=<ap-xxx>/fs.cosn.bucket.endpoint_suffix=<cos.ap-xxx.myqcloud.com>]]
       [delete <namespace>]                                      
       [help [<command>]]                                        
       [ls [-r|--sort=option|--timestamp=option]]                
       [setPolicy [--wPolicy <1-6>] [--rPolicy <1-5>] <namespace>]
       [setTtl [--action delete|free] <namespace> <time to live>]
       [stat <namespace>]                                        
       [unsetPolicy <namespace>]                                 
       [unsetTtl <namespace>] 

The instructions are described as follows:

Instruction	Description
create	Creates a namespace and maps a remote storage system (UFS) to the namespace. When creating the namespace, you can set cache read and write policies. You need to pass in an authorized key (secretId and secretKey).
delete	Deletes a specified namespace.
ls	Lists the detailed information of a specified namespace, including the UFS path, creation time, cache policy, and TTL information.
setPolicy	Sets the cache policy of a specified namespace.
setTtl	Sets TTL for a specified namespace.
stat	Provides the description of a specified namespace, including the mount point, UFS path, creation time, cache policy, TTL information, persistence status, user group, ACL, last access time, and modification time.
unsetPolicy	Resets the cache policy of a specified namespace.
unsetTtl	Resets the TTL of a specified namespace.

Creating or Deleting Namespaces

By creating a namespace in GooseFS, you can cache frequently accessed hot data from a remote storage system to a local high-performance storage node to provide high-performance data access for local computing businesses. The following shows how to map the COS bucket example-bucket, the example-prefix directory in the bucket, and the CHDFS to the test_cos, test_cos_prefix, and test_chdfs namespaces respectively.

# Map the COS bucket `example-bucket` to the `test_cos` namespace
$ goosefs ns create test_cos cosn://example-bucket-1250000000/ --wPolicy 1 --rPolicy 1 --secret fs.cosn.userinfo.secretId=AKIDxxxxxxx --secret fs.cosn.userinfo.secretKey=xxxxxxxxxx --attribute fs.cosn.bucket.region=ap-guangzhou --attribute fs.cosn.bucket.endpoint_suffix=cos.ap-guangzhou.myqcloud.com 
# Map the `example-prefix` directory in the COS bucket `example-bucket` to the `test_cos_prefix` namespace
$ goosefs ns create test_cos_prefix cosn://example-bucket-1250000000/example-prefix/ --wPolicy 1 --rPolicy 1 --secret fs.cosn.userinfo.secretId=AKIDxxxxxxx --secret fs.cosn.userinfo.secretKey=xxxxxxxxxx --attribute fs.cosn.bucket.region=ap-guangzhou --attribute fs.cosn.bucket.endpoint_suffix=cos.ap-guangzhou.myqcloud.com
# Map the CHDFS `f4ma0l3qabc-Xy3` to the `test_chdfs` namespace
$ goosefs ns create test_chdfs ofs://f4ma0l3qabc-Xy3/ --wPolicy 1 --rPolicy 1 --attribute fs.ofs.userinfo.appid=1250000000

After successful creation, you can use the goosefs fs ls instruction to view the directory details:

$ goosefs fs ls /test_cos

You can use the delete instruction to delete unwanted namespaces:

$ goosefs ns delete test_cos
Delete the namespace: test_cos

Setting the cache policy

You can use setPolicy and unsetPolicy to set the cache policy of a namespace. The instruction set is as follows:

$goosefs ns setPolicy [--wPolicy <1-6>] [--rPolicy <1-5>] <namespace>

The parameters are described as follows:

• wPolicy: write cache policy. Up to 6 write cache policies are supported.
• rPolicy: read cache policy. Up to 5 read cache policies are supported.
• namespace: specified namespace

The read and write cache policies currently supported by GooseFS are as follows:

Write cache policies

Policy Name	Behavior	Corresponding Write_Type	Data Security	Write Efficiency
MUST_CACHE (1)	Data is stored only in GooseFS and is not written to the remote storage system.	MUST_CACHE	Unreliable	High
TRY_CACHE (2)	If the cache has space, data is written to GooseFS. Otherwise, data is written directly to underlying storage systems.	TRY_CACHE	Unreliable	Medium
CACHE_THROUGH (3)	Data is cached as much as possible and simultaneously written to remote storage systems.	CACHE_THROUGH	Reliable	Low
THROUGH (4)	Data is not stored in GooseFS, but written directly to the remote storage system.	THROUGH	Reliable	Medium
ASYNC_THROUGH (5)	Data is written to GooseFS and asynchronously purged to remote storage systems.	ASYNC_THROUGH	Weak reliability	High

Note：

Write_Type indicates the file cache policy specified when the user calls the SDK or API to write data to GooseFS. It takes effect only for a single file.

Read cache policies

Policy Name	Behavior	Metadata Sync	Corresponding Read_Type	Data Consistency	Read Efficiency	Whether to Cache Data
NO_CACHE (1)	Data is not cached and is directly read from remote storage systems instead.	NO	NO_CACHE	Strong	Low	No
CACHE (2)	Metadata access behavior: if a cache is hit, the metadata in the master shall prevail. Metadata is not proactively synchronized from the underlying layer. Data stream access behavior: the data stream Read_Type is CACHE.	Once	CACHE	Weak	Hit: high Not hit: low	Yes
CACHE_PROMOTE (3)	Metadata access behavior: same as CACHE. Data stream access behavior: the data stream Read_Type is CACHE_PROMOTE.	Once	CACHE_PROMOTE	Weak	Hit: high Not hit: low	Yes
CACHE_CONSISTENT_PROMOTE (4)	Metadata access behavior: sync the metadata in the remote storage system (UFS) each time before a read operation. If the data does not exist in the UFS, report the Not Exists exception. Data stream access behavior: the data stream Read_Type is CACHE_PROMOTE. If a cache is hit, data is cached to the hottest cache medium.	Always	CACHE	Strong	Hit: medium Not hit: low	Yes
CACHE_CONSISTENT (5)	Metadata access behavior: same as CACHE_CONSISTENT_PROMOTE. Data stream access behavior: the data stream Read_Type is CACHE. That is, when a cache is hit, data is not moved between different media layers.	Always	CACHE_PROMOTE	Strong	Hit: medium Not hit: low	Yes

Note：

Read_Type indicates the file cache policy specified when the user calls the SDK or API to read data from GooseFS. It takes effect only for a single file.

Based on current big data business practices, we recommend the following combinations of read and write cache policies:

Write Cache Policy	Read Cache Policy	Policy Combination Performance
CACHE_THROUGH (3)	CACHE_CONSISTENT (5)	Strong data consistency between the cache and remote storage systems
CACHE_THROUGH (3)	CACHE (2)	Write: strong consistency; read: eventual consistency
ASYNC_THROUGH (5)	CACHE_CONSISTENT (5)	Write: eventual consistency; read: strong consistency
ASYNC_THROUGH (5)	CACHE (2)	Read/Write: eventual consistency
MUST_CACHE (1)	CACHE (2)	Data is read from the cache only.

The following example shows how to set the read and write cache policies of the test_cos namespace to CACHE_THROUGH and CACHE_CONSISTENT` respectively:

$ goosefs ns setPolicy --wPolicy 3 --rPolicy 5 test_cos

Note：

In addition to specifying cache policies when creating namespaces, you can also configure global cache policies by setting Read_Type or Write_Type for specific files when reading or writing files, or by using the Properties configuration file. If multiple policies exist at the same time, their priority order is as follows: custom priority > namespace read and write policies > global cache policy configured in the configuration file. For the read policy, the combination of the custom Read_Type and the namespace's DirReadPolicy takes effect. That is, the custom Read_Type is used as the data stream read policy, and the namespace policy is used for metadata.

For example, GooseFS contains a COSN namespace whose read policy is CACHE_CONSISTENT and the namespace contains a test.txt file. When the client reads the test.txt file, Read_Type is specified as CACHE_PROMOTE. Then the entire read behavior is to sync metadata and perform CACHE_PROMOTE.

To reset the read and write cache policies, you can use the unsetPolicy instruction. The following shows how to reset the read and write cache policies for the test_cos namespace:

$ goosefs ns unsetPolicy test_cos

Setting TTL

Time to Live (TTL) is used to manage data cached on the local nodes of GooseFS. Setting TTL allows a specified operation, such as delete or free, to be performed on the cached data after a specified period of time. The instruction for setting TTL is as follows:

$ goosefs ns setTtl [--action delete|free] <namespace> <time to live>

The parameters are described as follows:

• action: operation performed after the cache duration expires. Currently, delete and free are supported. The delete operation deletes data from the cache and UFS, while the free operation deletes data only from the cache.
• namespace: specified namespace
• time to live: data cache duration, in milliseconds

The following example shows how to set the policy of the test_cos namespace to delete data only from the cache after 60 seconds:

$ goosefs ns setTtl --action free test_cos 60000

Metadata management

This section describes how GooseFS manages metadata, including metadata synchronization and updates. GooseFS provides users with unified namespace capability. Users can access files on different underlying storage systems using a unified gfs:// path. You only need to specify the paths of the underlying storage systems. We recommend that you use GooseFS as a unified data access layer to uniformly read and write data from GooseFS to ensure metadata consistency.

Metadata synchronization overview

You can configure the metadata synchronization interval in the conf/goosefs-site.properties configuration file:

goosefs.user.file.metadata.sync.interval=<INTERVAL>

The metadata synchronization interval parameter supports 3 types of input values:

• -1: metadata is not updated after it is first loaded into GooseFS.
• 0: GooseFS updates metadata after each read/write operation.
• Positive integer: GooseFS periodically updates metadata at a specified interval.

You can choose an appropriate synchronization interval based on your number of nodes, the I/O distance between your GooseFS cluster and the underlying storage system, and the type of the underlying storage system. Usually:

• The greater the number of nodes in a GooseFS cluster, the greater the metadata synchronization delay.
• The greater the distance between the GooseFS cluster and the underlying storage system, the greater the metadata synchronization delay.
• The impact of the underlying storage system on the metadata synchronization delay depends on the QPS load. The higher the QPS load, the less the synchronization delay.

Metadata synchronization management

Configuration methods

1. Configuration via CLI
   You can set the metadata synchronization interval in command line interface (CLI) mode:

   goosefs fs ls -R -Dgoosefs.user.file.metadata.sync.interval=0 <path to sync>
   

2. Configuration via the configuration file
   For a large-scale GooseFS cluster, you can use the goosefs-site.properties configuration file to batch configure the metadata synchronization interval for the master nodes in the cluster, and other nodes will adopt this interval by default.

   goosefs.user.file.metadata.sync.interval=1m
   

Note：

Many businesses choose to distinguish the purpose of data by directory, and the data access frequencies of different directories are not all the same. You can set different metadata synchronization intervals for different directories. For some directories that change frequently, the metadata synchronization interval can be set to a shorter time (such as 5 minutes). For directories that change little or do not change, the synchronization interval can be set to -1, so that GooseFS will not automatically synchronize the metadata of the directories.

Recommended configuration

You can set different metadata synchronization intervals based on business access modes:

Access Mode		Metadata Synchronization Interval	Remarks
All file requests go through GooseFS		-1	-
Most file requests go through GooseFS	HDFS is used as UFS	Hot update or update by path is recommended	If the HDFS updates frequently, you are advised to set the update interval to `-1` to prohibit updates.
	COS is used as UFS	Configuring update intervals by path is recommended	Configuring different update intervals for different directories can alleviate the pressure of metadata synchronization.
File upload requests generally do not go through GooseFS	HDFS is used as UFS	Configuring update intervals by path is recommended
	COS is used as UFS	Configuring update intervals by path is recommended