E-MapReduce:Use the transparent caching feature of JindoCache to accelerate access to OSS

How it works

JindoCache organizes caching behavior into CacheSets. Each CacheSet binds a caching policy to an OSS path prefix: any data read from that prefix is cached according to the policy you define. You can configure multiple CacheSets to apply different policies to different data sets.

Once you configure a CacheSet and set fs.xengine=jindocache in Hadoop, all jobs that access the covered OSS paths automatically benefit from caching. Jobs don't need to call any cache API—the caching is transparent.

Prerequisites

Before you begin, ensure that you have:

• An EMR cluster with JindoCache enabled (selected at cluster creation time). For details, see Create a cluster.

• An activated OSS instance. For details, see Activate OSS.

Usage notes

• OSS stores files as objects. The terms "file" and "object" refer to the same data when the context involves OSS.

• If metaPolicy is set to ONCE, JindoCache reads metadata from OSS once and caches it locally. If the underlying OSS data changes after the initial read, jobs may read stale metadata until the cache is refreshed. Use ALWAYS if your data changes frequently and strong consistency is required.

• Transparent caching provides the most benefit for read-heavy workloads where the same data is accessed multiple times. Write-intensive workloads that rarely re-read the same data gain little from caching.

Configure caching policies

1. Log on to your cluster. For details, see Log on to a cluster.

2. Create a cacheset.xml file. The following example stores the file in the /path directory.

   <?xml version="1.0" encoding="UTF-8"?>
   <cachesets>
       <cacheset>
           <name>name1</name>
           <path>oss://emr-test/dir1</path>
           <cacheStrategy>DISTRIBUTED</cacheStrategy>
           <metaPolicy>
               <type>ALWAYS</type>
           </metaPolicy>
           <readPolicy>CACHE_ASIDE</readPolicy>
           <writePolicy>WRITE_AROUND</writePolicy>
       </cacheset>
       <cacheset>
           <name>name2</name>
           <path>oss://emr-test/dir2</path>
           <cacheStrategy>DHT</cacheStrategy>
           <metaPolicy>
               <type>ONCE</type>
           </metaPolicy>
           <readPolicy>CACHE_ASIDE</readPolicy>
           <writePolicy>WRITE_AROUND</writePolicy>
       </cacheset>
   </cachesets>

   The following table describes the parameters.

   Parameter	Description	Example
   name	A unique name for the CacheSet. If a CacheSet with the same name already exists, its configuration is overwritten.	name1
   path	The OSS path prefix covered by this CacheSet. The policy applies to all objects under this path.	oss://emr-test/dir1
   cacheStrategy	The cache distribution strategy. DISTRIBUTED: general-purpose, suitable for most workloads. DHT (distributed hash table): optimized for small-sized, read-only files.	DISTRIBUTED
   metaPolicy type	How JindoCache handles object metadata. ALWAYS: metadata is always read from OSS, providing strong consistency at the cost of higher latency. ONCE: metadata is read from OSS once and cached locally, reducing latency but risking stale reads if OSS data changes. If cacheStrategy is DHT, set this to ONCE.	ALWAYS
   readPolicy	The read strategy. Must be CACHE_ASIDE: reads are served from cache when available, and fetched from OSS on a cache miss.	CACHE_ASIDE
   writePolicy	The write strategy. WRITE_AROUND: writes go directly to OSS, bypassing the cache—use this for data written once and rarely re-read. WRITE_THROUGH: writes go to both the cache and OSS simultaneously—use this when you need cache consistency with slightly higher write latency. CACHE_ONLY: writes go only to the cache and are not persisted to OSS; requires metaPolicy to be ONCE.	WRITE_AROUND

3. Refresh JindoCache with the new CacheSet configuration.

   jindocache -refreshCacheSet -path /path/cacheset.xml

   If the command succeeds, the output contains Successfully refresh cacheset !!!. For a full list of JindoCache commands, see JindoCache CLI usage notes.

4. Verify that the CacheSets are registered.

   jindocache -listCacheSet

Enable transparent caching in Hadoop

Set the fs.xengine configuration item in Hadoop-Common so that JindoCache intercepts OSS read requests.

In the EMR console, go to the Hadoop-Common service page, click the Configure tab, and then click the core-site.xml tab. Add or update the following configuration item. For details on managing configuration items, see Manage configuration items.

This configuration takes effect immediately—no JindoCache restart is required.

After you save the configuration, all jobs that access OSS paths covered by your CacheSets automatically use the cache on subsequent reads.

FAQ

What's next

• JindoCache CLI usage notes

• Manage configuration items