This section describes HBase multi-tenant

In HBase, creating a namespace is a lightweight operation. Isolating tables of different services in different namespaces is the simplest method for resource isolation. In addition, common resource isolation methods, such as ACL, quota, and RSGroup, can be set on the namespace.

Users are classified into common users and super users. Super users include users who start the HBase service and hbase.superuser users who can manage clusters. Common users can access or operate HBase only after being authorized. Scope can be understood as the granularity of a resource.

Various operation requires the Action of HBase can look up in the official documentation of HBase: http://hbase.apache.org/book.html#appendix_acl_matrix

Combining with the user’s access or operation requirements, the user can set a reasonable action in a reasonable scope, which is the best way to achieve user permission control.

Set or cancel the permission in HBase Shell or by invoking the HBase API. The operations in the shell are shown below:

To set the namespace permission, prefix it with @ :

Set Cell permissions:

It is stored in hbase: ACL table. The rowkey is calculated by scope. The acl table structure is as follows:

Cell permission is stored using tags of HFile V3.

Authentication permission is used to determine whether a user has the permission to perform an operation. This is done in AccessController, a coprocessor that implements MasterObserver, RegionServerObserver, RegionObserver, and so on. Check permissions in hooks of operations such as Master, RegionServer, and Region. Since a full PermissionCache is maintained on each RS, check that the PermissionCache contains the required permissions and throw an AccessDeniedException if the permissions are insufficient.

Because there are limits on the resources and services a cluster can provide, Quota limits the amount of data and the speed at which each resource can be accessed.

To enable the HBase quota function, perform the following operations:

The following figure shows HBase concepts about Quota and their relationships:

Throttle specifies the number of resources or data to be accessed in a specified period of time.

• The supported time units include SEC, min, hour, and day.

• Use REQ to limit the number of requests;

• Use B, K, M, G, T, P to limit the amount of data requested;

• CU limits the read/write capacity unit. A read/write capacity unit refers to the read/write volume of a request less than 1KB. If a request reads 2.5 KB of data, three capacity units are required. Can use hbase. Quota. Read. Capacity. The unit or hbase. Quota. Write. Capacity. The unit equipped with a capacity of unit volume of data.

• Machine scope stands for Throttle configured on a single RS. Cluster indicates that the THROTTLE quota is shared by all RS in the Cluster. If the QuotaScope is not specified, the default is Machine.

The shell command for configuring Throttle is as follows:

You can throttle the RegionServer. The Quota of the RS indicates the service upper limit of the RS. You are advised to set the parameter in seconds.

Set quota of Cluster scope:

How the quota of the Cluster scope is allocated to each RS:

• TableMachineLimit = ClusterLimit/TotalTableRegionNum * MachineTableRegionNum;

• For namespace quota, NamespaceMachineLimit = ClusterLimit/RsNum. Note that RSGroup is not considered. The throttle limit assigned to RS is too small, and this calculation method needs to be improved in the future.

GlobalBypass is configured for users globally, bypassing all throttle.

The Space is used to limit the data amount of resources. The Space is configured on the namespace or table. When the amount of data reaches the upper limit, the configured violation policies are executed, including:

Disable: Disable table/ the tables of namespace

NoInserts: Disables Mutation operations other than Delete, allowing Compaction

NoWrites: disables the Mutation operation and allows Compaction

NoWritesCompactions: disables Mutation operations and compactions

View the snapshot of the current Space quota (not HBase snapshots), but the size of the current table, the configured limit, and the status of the triggered policy:

Limit the number of tables or regions in a namespace:

hbase.namespace.quota.maxtables/hbase.namespace.quota.maxregionsCopy the code

If the limit is exceeded, a QuotaExceededException is thrown.

Space Quota is implemented by:

(1) the Region of the RS cycle the size information is sent to the master: RegionSizeReportingChoreMaster

(2) the statistics table size and triggered policies are stored in the quota table: Quota observerchorers

(3) Read quota table periodically, execute Policy: SpaceQuotaRefresherChore

You can configure throttle limit to soft limit when the cluster resources are insufficient. To enable or disable oversending, run the following command:

Note that oversending allows users to request more user/namespace/table quotas when RS quotas are too large. Therefore, the RS quota must be set before oversending can be enabled. It is recommended to set RS quotas in seconds, because if RS quotas are consumed first by requests from other users, restoring them will take a long time, possibly affecting subsequent requests. Even if these later requests did not exceed their configured User /namespace/table quota.

Quota information is stored in the hbase: Quota table.

Row keys have the following types:

N.namespace: indicates the quota of the namespace

T. table: indicates the quota of the table

U.user: User quota

R.ll: Indicates the quota of RegionServer

ExceedThrottleQuota: Whether exceedThrottleQuota is allowed

Quotas related to Throttle are stored in the Q CF, and quotas related to Space are stored in the U CF.

Throttle Specifies whether to enable the ZK node stored in /hbase/ Rpc-throttle. The value can be true or false. Because Throttle is on and off in real time, other quota configurations are delayed by RS periodically reading the quota table.

RSGroup allocates RS to different groups, and then allocates a namespace or table to an RSGroup to achieve isolation. It can be interpreted as that each RSGroup forms a small cluster.

To use RSGroup, you need to add the following configuration:

When RSGroup is enabled, all RS are in the default group by default.

After creating a new group, you must first move the RS to the group before you can move the namespace or table to the group.

Add a new RSGroup:

First move RS to this group, then move namespace to this group:

The function of RSGroup is mainly realized in RSGroupAdminEndpoint, which is an Endponit that implements MasterObserver. In the hook of master operation, the region of the table is moved to the corresponding RSGroup.

RSGroup information is stored in hbase: RSGroup table. At the same time, the RSGroup information is also stored in the ZK. When the cluster starts and the RSGroup table is not online, the RSGroup information is read from the ZK.

In summary, this section describes HBase multi-tenant functions. You are encouraged to use them in the production environment and provide suggestions to the community to further improve HBase multi-tenant functions.

About the author

Meyi, the youngest beauty of Xiaomi, is an HBase Committer. The HBase ecological team to which Meyi belongs has a strong technical atmosphere, and has cultivated 9 HBase committers and 2 PMCS, as well as several open source projects under Contributer. Open source enthusiasts and HBase enthusiasts are welcome to join us and grow together.

PS: their team offer oh, poke recruit to | looking for talented engineers for details.