Configuring Centralized Cache Management in HDFS
Centralized cache management in HDFS is an explicit caching mechanism that allows users to specify paths to be cached by HDFS. The NameNode communicates with DataNodes and instructs them to cache specific blocks in off-heap caches.
Centralized and explicit caching has several advantages:
- Frequently used data is pinned in memory. This is important when the size of the working set exceeds the size of main memory, which is common for many HDFS workloads.
- Cluster memory is optimized because you can pin m of n block replicas, saving n-m memory. Before centralized pinning, repeated reads of a block caused all n replicas to be pulled into each DataNode buffer cache.
- Tasks are co-located with cached block replicas, improving read performance. Because the NameNode manages DataNode caches, applications can query the set of cached block locations when making task placement decisions.
- Clients can use the zero-copy read API, and incur almost no overhead, because each DataNode does a checksum verification of cached data only once.
Centralized cache management is most useful for files that are accessed repeatedly. For example, a fact table in Hive that is often used in joins is a good candidate for caching. On the other hand, caching the input of a one-year reporting query is probably less useful, as the historical data might be read only once.
Centralized cache management is also useful for mixed workloads with performance service-level agreements (SLAs). Caching the working set of a high-priority workload insures that it does not contend for disk I/O with a low-priority workload.
In this architecture, the NameNode is responsible for coordinating all the DataNode off-heap caches in the cluster. The NameNode periodically receives a "cache report" from each DataNode which describes all the blocks cached on a given DataNode. The NameNode manages DataNode caches by piggybacking cache and uncache commands on the DataNode heartbeat.
The NameNode queries its set of cache directives to determine which paths should be cached. Cache directives are persistently stored in the fsimage and edit log, and can be added, removed, and modified using Java and command-line APIs. The NameNode also stores a set of cache pools, which are administrative entities used to group cache directives together for resource management and enforcing permissions.
The NameNode periodically rescans the namespace and active cache directories to determine which blocks need to be cached or uncached and assigns caching to DataNodes. Rescans can also be triggered by user actions such as adding or removing a cache directive or removing a cache pool.
Currently, blocks that are under construction, corrupt, or otherwise incomplete are not cached. If a cache directive covers a symlink, the symlink target is not cached. Caching is currently done on a per-file basis (and not at the block-level).
A cache directive defines a path that should be cached. Paths can be either directories or files. Directories are cached non-recursively, meaning only files in the first-level listing of the directory are cached.
Directives have parameters, such as the cache replication factor and expiration time. Replication factor specifies the number of block replicas to cache. If multiple cache directives refer to the same file, the maximum cache replication factor is applied. Expiration time is specified on the command line as a time-to-live (TTL), a relative expiration time in the future. After a cache directive expires, it is no longer considered by the NameNode when making caching decisions.
A cache pool is an administrative entity used to manage groups of cache directives. Cache pools have UNIX-like permissions that restrict which users and groups have access to the pool. Write permissions allow users to add and remove cache directives to the pool. Read permissions allow users to list the cache directives in a pool, as well as additional metadata. Execute permissions are not used.
Cache pools are also used for resource management. Pools can enforce a maximum limit that restricts the aggregate number of bytes that can be cached by directives in the pool. Normally, the sum of the pool limits roughly equals the amount of aggregate memory reserved for HDFS caching on the cluster. Cache pools also track a number of statistics to help cluster users determine what is and should be cached.
Pools also enforce a maximum time-to-live. This restricts the maximum expiration time of directives being added to the pool.
cacheadmin Command-Line Interface
On the command-line, administrators and users can interact with cache pools and directives using the hdfs cacheadmin subcommand. Cache directives are identified by a unique, non-repeating 64-bit integer ID. IDs are not reused even if a cache directive is later removed. Cache pools are identified by a unique string name.
Cache Directive Commands
Description: Add a new cache directive.
Usage: hdfs cacheadmin -addDirective -path <path> -pool <pool-name> [-force] [-replication <replication>] [-ttl <time-to-live>]
Where, path: A path to cache. The path can be a directory or a file.
pool-name: The pool to which the directive will be added. You must have write permission on the cache pool in order to add new directives.
force: Skips checking of cache pool resource limits.
replication: The cache replication factor to use. Defaults to 1.
time-to-live: Time period for which the directive is valid. Can be specified in seconds, minutes, hours, and days, for example: 30m, 4h, 2d. The value never indicates a directive that never expires. If unspecified, the directive never expires.
Description: Remove a cache directive.Usage: hdfs cacheadmin -removeDirective <id>
Where, id: The id of the cache directive to remove. You must have write permission on the pool of the directive in order to remove it. To see a list of PathBasedCache directive IDs, use the -listDirectives command.
Description: Remove every cache directive with the specified path.
Usage: hdfs cacheadmin -removeDirectives <path>
Where, path: The path of the cache directives to remove. You must have write permission on the pool of the directive in order to remove it.
Description: List PathBasedCache directives.
Usage: hdfs cacheadmin -listDirectives [-stats] [-path <path>] [-pool <pool>]
Where, path: List only PathBasedCache directives with this path. Note that if there is a PathBasedCache directive for path in a cache pool that we do not have read access for, it will not be listed.
pool: List only path cache directives in that pool.
stats: List path-based cache directive statistics.
Cache Pool Commands
Description: Add a new cache pool.
Usage: hdfs cacheadmin -addPool <name> [-owner <owner>] [-group <group>] [-mode <mode>] [-limit <limit>] [-maxTtl <maxTtl>]
Where, name: Name of the new pool.
owner: Username of the owner of the pool. Defaults to the current user.
group: Group of the pool. Defaults to the primary group name of the current user.
mode: UNIX-style permissions for the pool. Permissions are specified in octal, for example: 0755. By default, this is set to 0755.
limit: The maximum number of bytes that can be cached by directives in this pool, in aggregate. By default, no limit is set.
maxTtl: The maximum allowed time-to-live for directives being added to the pool. This can be specified in seconds, minutes, hours, and days, for example: 120s, 30m, 4h, 2d. By default, no maximum is set. A value of never specifies that there is no limit.
Description: Modify the metadata of an existing cache pool.
Usage: hdfs cacheadmin -modifyPool <name> [-owner <owner>] [-group <group>] [-mode <mode>] [-limit <limit>] [-maxTtl <maxTtl>]
Where, name: Name of the pool to modify.
owner: Username of the owner of the pool.
group: Groupname of the group of the pool.
mode: Unix-style permissions of the pool in octal.
limit: Maximum number of bytes that can be cached by this pool.
maxTtl: The maximum allowed time-to-live for directives being added to the pool.
Description: Remove a cache pool. This also uncaches paths associated with the pool.
Usage: hdfs cacheadmin -removePool <name>
Where, name: Name of the cache pool to remove.
Description: Display information about one or more cache pools, for example: name, owner, group, permissions, and so on.
Usage: hdfs cacheadmin -listPools [-stats] [<name>]
Where, name: If specified, list only the named cache pool.
stats: Display additional cache pool statistics.
To lock block files into memory, the DataNode relies on native JNI code found in libhadoop.so. Be sure to enable JNI if you are using HDFS centralized cache management.
- dfs.datanode.max.locked.memory: The maximum amount of memory a DataNode uses for caching (in bytes). The "locked-in-memory size" ulimit (ulimit -l) of the DataNode user also needs to be increased to match this parameter (see OS Limits). When setting this value, remember that you need space in memory for other things as well, such as the DataNode and application JVM heaps and the operating system page cache.
- dfs.namenode.path.based.cache.refresh.interval.ms: The NameNode uses this as the amount of milliseconds between subsequent path cache rescans. This calculates the blocks to cache and each DataNode containing a replica of the block that should cache it. By default, this parameter is set to 300000, which is five minutes.
- dfs.datanode.fsdatasetcache.max.threads.per.volume: The DataNode uses this as the maximum number of threads per volume to use for caching new data. By default, this parameter is set to 4.
- dfs.cachereport.intervalMsec: The DataNode uses this as the amount of milliseconds between sending a full report of its cache state to the NameNode. By default, this parameter is set to 10000, which is 10 seconds.
- dfs.namenode.path.based.cache.block.map.allocation.percent: The percentage of the Java heap which we will allocate to the cached blocks map. The cached blocks map is a hash map which uses chained hashing. Smaller maps may be accessed more slowly if the number of cached blocks is large; larger maps will consume more memory. By default, this parameter is set to 0.25 percent.
If you get the error, Cannot start datanode because the configured max locked memory size... is more than the datanode's available RLIMIT_MEMLOCK ulimit, the operating system is imposing a lower limit on the amount of memory that you can lock than what you have configured. To fix this, adjust the DataNode ulimit -l value. Usually, this value is configured in /etc/security/limits.conf; but varies depending on your operating system and distribution.
You have correctly configured this value when you can run ulimit -l from the shell and get back either a higher value than what you have configured with dfs.datanode.max.locked.memory, or the string unlimited, indicating that there is no limit. It is typical for ulimit -l to output the memory lock limit in KB, but dfs.datanode.max.locked.memory must be specified in bytes.