Modifying HBase Parameters to Optimize Cluster Performance

Scenario

A CloudTable HBase cluster has many HBase parameters related to read/write performance. You can modify the HBase parameters to tune cluster performance under different read/write request loads. You need to restart the cluster to make changes take effect.

Constraints

Restart the CloudTable HBase cluster after parameter configuration modification. Otherwise, services will be interrupted.
Do not modify cluster parameters when the CloudTable HBase cluster is being restarted.

Prerequisites

No task is running in the cluster.

Procedure

Log in to the CloudTable console.
Select a region in the upper left corner.
In the navigation pane, click Cluster Management.
Click the name of a cluster for which you want to modify HBase parameters to access the cluster details page.
In the Parameter Configuration area, click the Parameter Configuration tab to modify HBase parameters.

For details about the HBase parameters you can modify, see HBase Parameters.
1. Select the target parameter and click in the Value column.
2. Enter a new value in the text box, and then click . The parameter value has been successfully changed if the system prompts "The parameter changed to xx successfully. Save the modified value." The new parameter value is marked with a red asterisk (*).
  If you want to cancel it, click .
3. Click Save Changes in the upper left corner above the parameter list. A dialog box is displayed.
4. In the Save Changes dialog box, verify whether the parameter settings are correct. Select Restart the cluster immediately and click OK.
  - After you select Restart the cluster immediately and click OK, the cluster restarts immediately. After the cluster restart is complete, the new parameter configurations take effect. Click above the parameter list. The system prompts "The new value is applied" in the upper right corner of the parameter list.
  - If you do not select Restart the cluster immediately and click OK in the Save Changes dialog box, the system prompts "The new value is not applied" above the parameter list. In this case, you need to restart the cluster for the changes to take effect. For details about how to restart a cluster, see Restarting an HBase Cluster.
  - If the new parameter value is incorrect, click Quick Restore to cancel the modification.
After modifying parameters, you can click the Change History tab to view the change history.

On the Change History tab page, you can view the following information.
- Name: Name of the changed parameter.
- Old Value: old parameter value
- New Value: new parameter value
- Modified: time when you modify a parameter value

HBase Parameters

Table 1 lists HBase parameters you can modify currently.

The sum of the values of hbase.regionserver.global.memstore.size and hfile.block.cache.size cannot exceed 0.8.

**Table 1** HBase parameters
Parameter	Value Range	Default Value	Description
hbase.regionserver.thread.compaction.small	[1,20]	1	Indicates the number of HFile compaction threads. You can increase the parameter value in heavy-put-load scenarios.
hbase.regionserver.global.memstore.size	(0,0.8)	0.4	It is recommended that you set this parameter to "hbase.hregion.memstore.flush.size x Number of regions with active writes/RegionServer GC -Xmx". The default value is 0.4, indicating that 40% of RegionServer GC -Xmx is used.
hbase.hstore.blockingStoreFiles	[1,2147483647]	60	When the HFile number in the column cluster reaches this threshold, all operations in the region are blocked until the compaction is complete. You can increase the parameter value in heavy-put-load scenarios.
hbase.client.scanner.timeout.period	[1,2147483647]	60000	A client and RegionServer parameter, indicating the scan lease period. It is recommended that you set this parameter to an integral multiple of 60,000 ms, and increase the parameter value in heavy-read-load scenarios. Unit: millisecond
hfile.block.cache.size	(0,0.8)	0.2	Indicates the data cache percentage in the RegionServer GC -Xmx. You can increase the parameter value in heavy-read-load scenarios, in order to improve cache hit ratio and performance. The default value is 0.2, indicating that 20% of RegionServer GC -Xmx is used.
hbase.regionserver.handler.count	[1,300]	100	Indicates the number of RPC server instances on the RegionServer. The recommended value ranges from 100 to 300.
hbase.regionserver.metahandler.count	[1,100]	50	Indicates the number of program instances for processing prioritized requests. The recommended value ranges from 20 to 100.
hbase.hstore.flusher.count	[1,10]	2	Indicates the number of memstore flush threads. You can increase the parameter value in heavy-put-load scenarios.
hbase.ipc.server.callqueue.read.ratio	[0,1]	0.5	When used under different load models, it controls the ratio between the numbers of read and write RPC queues. The value ranges from 0 to 1.0, and the default value is 0.5. 0 indicates the total RPC queues of read and write operations. If the value is less than 0.5, it indicates that the read load is less than the write load. 0.5 indicates that the read load equals to the write load. If the value is greater than 0.5, it indicates that the read load is greater than the write load. 1.0 indicates that all RPC queues except one are used for read operations.
hbase.regionserver.hotregion.handler.count	[1,65535]	66	Number of RPC listener instances started on RegionServers for hotspot regions.
hbase.ipc.server.hotregion.max.callqueue.length	[1,65535]	330	Maximum length of the queue for RegionServers to process requests of hotspot regions. Upon receiving a new request, the system checks whether the length of the queue exceeds the threshold. If so, the request is discarded.
hbase.metric.controller.analysis.period	[1,2147483647]	60	Hotspot analysis period of MetricController, in seconds.
hbase.metric.controller.analysis.threads.max	[1,100]	10	Maximum number of threads for hotspot analysis in the thread pool.
hbase.metric.controller.collect.threads.max	[1,100]	16	Maximum number of threads for hotspot analysis in the traffic collection thread pool.
hbase.metric.regionserver.hotspot.threshold	[1,2147483647]	20000	Hotspot threshold of a RegionServer. Unit: requests per second.
hbase.metric.region.hotspot.threshold	[1,2147483647]	10000	Hotspot threshold of a single region. Unit: requests per second.
hbase.hotspot.enable	[true,false]	true	Whether to enable hotspot self-healing. The value true means to enable it, and false means to disable it. After this function is enabled, access hotspots will be automatically processed.
hbase.tries.cache.enabled	[true,false]	false	If this parameter is set to true, LoudsTriesLruBlockCache is used to cache index blocks and data blocks.
hbase.write.tries	[true,false]	false	If this parameter is set to true, the succinct tries feature is enabled. In this case, a new data structure is used to improve the utilization of index blocks.
hbase.hfile.hsync	[true,false]	false	Specifies whether to enable the HFile durability to make data persistence on disks. If this parameter is set to true, the performance is affected because each Hfile file is synchronized to the disk.
hbase.wal.hsync	[true,false]	false	Specifies whether to enable WAL file durability to make the WAL data persistence on disks. If this parameter is set to true, the performance is affected because each WAL file is synchronized to the disk.
hbase.hot.cold.data.separator.chore.enabled	[true,false]	false	Specifies whether to enable the CDS function. If this parameter is set to true, asynchronous dump is enabled for cold data.
hbase.hot.cold.data.separator.chore.interval	[60000, 2147483647]	86400000	Specifies the CDS check period. Unlike common compaction tasks, compaction tasks triggered by CDS will write hot and cold data to separate data files. Unit: millisecond
hbase.regionserver.cold.data.shipper.chore.interval	[60000, 2147483647]	300000	Specifies the period for checking cold data files to be dumped. If there are cold data files that can be dumped, a task is triggered to migrate the cold data files to the cold storage. Unit: millisecond
hbase.regionserver.shipper.cold.file.verification.enabled	[true, false]	true	Specifies whether to check the consistency of data files migrated to the cold storage.
hbase.cold.data.shipper.thread.count	[1, 20]	3	Specifies the number of threads used to copy data files to the cold storage.
hbase.cold.data.separator.offpeak.end.hour	[-1, 23]	-1	Specifies the end time of the off-peak period. The value is an integer hour in 24-hour format. During off-peak hours, CDS uses more resources to speed up the process of changing hot data to cold data. If this parameter is set to -1, there is no off-peak hour.
hbase.cold.data.separator.offpeak.start.hour	[-1, 23]	-1	Specifies the start time of the off-peak period. The value is an integer hour in 24-hour format. During off-peak hours, CDS uses more resources to speed up the process of changing hot data to cold data. If this parameter is set to -1, there is no off-peak hour.
hbase.compaction.skip.unnecessary.cold.files.enabled	[true, false]	false	Specifies whether to skip cold data files during major compaction.
hbase.regionserver.thread.compaction.cold	[1, 20]	1	Specifies the number of threads for merging cold data files.
hbase.regionserver.compaction.cold.check.period	[60000, 2147483647]	30000	Specifies the period for checking whether cold files need to be merged. Unit: millisecond