Struct opendal::services::Hdfs

pub struct Hdfs { /* private fields */ }

Available on crate feature services-hdfs only.

A distributed file system that provides high-throughput access to application data.

Capabilities

This service can be used to:

• stat
• read
• write
• create_dir
• delete
• copy
• rename
• list
• scan
• presign
• blocking
• append

Differences with webhdfs

Webhdfs is powered by hdfs’s RESTful HTTP API.

Features

HDFS support needs to enable feature services-hdfs.

Configuration

• root: Set the work dir for backend.
• name_node: Set the name node for backend.
• kerberos_ticket_cache_path: Set the kerberos ticket cache path for backend, this should be gotten by klist after kinit
• user: Set the user for backend
• enable_append: enable the append capacity. Default is false.

Refer to HdfsBuilder’s public API docs for more information.

Environment

HDFS needs some environment set correctly.

• JAVA_HOME: the path to java home, could be found via java -XshowSettings:properties -version
• HADOOP_HOME: the path to hadoop home, opendal relays on this env to discover hadoop jars and set CLASSPATH automatically.

Most of the time, setting JAVA_HOME and HADOOP_HOME is enough. But there are some edge cases:

• If meeting errors like the following:

error while loading shared libraries: libjvm.so: cannot open shared object file: No such file or directory

Java’s lib are not including in pkg-config find path, please set LD_LIBRARY_PATH:

export LD_LIBRARY_PATH=${JAVA_HOME}/lib/server:${LD_LIBRARY_PATH}

The path of libjvm.so could be different, please keep an eye on it.

• If meeting errors like the following:

(unable to get stack trace for java.lang.NoClassDefFoundError exception: ExceptionUtils::getStackTrace error.)

CLASSPATH is not set correctly or your hadoop installation is incorrect.

To set CLASSPATH:

export CLASSPATH=$(find $HADOOP_HOME -iname "*.jar" | xargs echo | tr ' ' ':'):${CLASSPATH}

• If HDFS has High Availability (HA) enabled with multiple available NameNodes, some configuration is required:

1. Obtain the entire HDFS config folder (usually located at HADOOP_HOME/etc/hadoop).
2. Set the environment variable HADOOP_CONF_DIR to the path of this folder.

export HADOOP_CONF_DIR=<path of the config folder>

3. Append the HADOOP_CONF_DIR to the CLASSPATH

export CLASSPATH=$HADOOP_CONF_DIR:$HADOOP_CLASSPATH:$CLASSPATH

4. Use the cluster_name specified in the core-site.xml file (located in the HADOOP_CONF_DIR folder) to replace namenode:port.

builder.name_node("hdfs://cluster_name");

macOS Specific Note

If you encounter an issue during the build process on macOS with an error message similar to:

ld: unknown file type in $HADOOP_HOME/lib/native/libhdfs.so.0.0.0
clang: error: linker command failed with exit code 1 (use -v to see invocation)

This error is likely due to the fact that the official Hadoop build includes the libhdfs.so file for the x86-64 architecture, which is not compatible with aarch64 architecture required for MacOS.

To resolve this issue, you can add hdrs as a dependency in your Rust application’s Cargo.toml file, and enable the vendored feature:

[dependencies]
hdrs = { version = "<version_number>", features = ["vendored"] }

Enabling the vendored feature ensures that hdrs includes the necessary libhdfs.so library built for the correct architecture.

Example

Via Builder

use std::sync::Arc;

use anyhow::Result;
use opendal::services::Hdfs;
use opendal::Operator;

#[tokio::main]
async fn main() -> Result<()> {
    // Create fs backend builder.
    let mut builder = Hdfs::default();
    // Set the name node for hdfs.
    // If the string starts with a protocol type such as file://, hdfs://, or gs://, this protocol type will be used.
    builder.name_node("hdfs://127.0.0.1:9000");
    // Set the root for hdfs, all operations will happen under this root.
    //
    // NOTE: the root must be absolute path.
    builder.root("/tmp");
    
    // Enable the append capacity for hdfs. 
    // 
    // Note: HDFS run in non-distributed mode doesn't support append.
    builder.enable_append(true);

    // `Accessor` provides the low level APIs, we will use `Operator` normally.
    let op: Operator = Operator::new(builder)?.finish();

    Ok(())
}

Implementations

impl HdfsBuilder

pub fn root(&mut self, root: &str) -> &mut Self

Set root of this backend.

All operations will happen under this root.

pub fn name_node(&mut self, name_node: &str) -> &mut Self

Set name_node of this backend.

Valid format including:

• default: using the default setting based on hadoop config.
• hdfs://127.0.0.1:9000: connect to hdfs cluster.

pub fn kerberos_ticket_cache_path( &mut self, kerberos_ticket_cache_path: &str ) -> &mut Self

Set kerberos_ticket_cache_path of this backend

This should be configured when kerberos is enabled.

pub fn user(&mut self, user: &str) -> &mut Self

Set user of this backend

pub fn enable_append(&mut self, enable_append: bool) -> &mut Self

Enable append capacity of this backend.

This should be disabled when HDFS runs in non-distributed mode.

pub fn atomic_write_dir(&mut self, dir: &str) -> &mut Self

Set temp dir for atomic write.

Notes

• When append is enabled, we will not use atomic write to avoid data loss and performance issue.

Trait Implementations

impl Builder for HdfsBuilder

const SCHEME: Scheme = Scheme::Hdfs

Associated scheme for this builder. It indicates what underlying service is.

type Accessor = HdfsBackend

The accessor that built by this builder.

fn from_map(map: HashMap<String, String>) -> Self

Construct a builder from given map which contains several parameters needed by underlying service.

fn build(&mut self) -> Result<Self::Accessor>

Consume the accessor builder to build a service.

impl Debug for HdfsBuilder

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for HdfsBuilder

fn default() -> HdfsBuilder

Returns the “default value” for a type.