ExZarr

Hex version Hex docs License CI Coverage Status

Elixir implementation of Zarr: compressed, chunked, N-dimensional arrays designed for parallel computing and scientific data storage.

Zarr v3.1 core support: ExZarr implements Zarr v2 and Zarr core v3.1 for the documented dtype/codec subset, with automatic version detection and Python interop tested in CI against zarr-python 2.18 / 3.2 / 3.3 / 3.4. See ZARR_V3_STATUS.md for details and known limitations.

Features

Installation

Add ex_zarr to your list of dependencies in mix.exs:

def deps do

  [

    {:ex_zarr, "~> 1.2"}

  ]

end

Quick Start

Creating an Array

# Create a Zarr v3 array (recommended for new projects)

{:ok, array} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  dtype: :float64,

  codecs: [

    %{name: "bytes"},

    %{name: "gzip", configuration: %{level: 5}}

  ],

  zarr_version: 3,

  storage: :memory

)



# Or use v2 format for compatibility with older tools

{:ok, array_v2} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  dtype: :float64,

  compressor: :zlib,

  zarr_version: 2,

  storage: :memory

)

Saving and Loading Arrays

# Save array to filesystem

:ok = ExZarr.save(array, path: "/tmp/my_array")



# Open existing array

{:ok, array} = ExZarr.open(path: "/tmp/my_array")



# Load entire array into memory

{:ok, data} = ExZarr.load(path: "/tmp/my_array")

Streaming Large Arrays (v1.1+)

Process arrays larger than memory with lazy chunk streaming:

{:ok, array} = ExZarr.open(path: "/data/large_dataset")



array

|> ExZarr.Array.stream_chunks(concurrency: 8, ordered: false)

|> Stream.map(fn {_index, data} -> process_chunk(data) end)

|> Stream.run()



# Row-wise slice streaming

array

|> ExZarr.Array.stream_slices(0, start: {0, 0}, stop: {100, 10})

|> Enum.each(fn {_start, row} -> process_row(row) end)

Attach telemetry handlers for production observability - see docs/guides/telemetry.md.

ExZarr.Array.write_stream(array, chunk_stream,

  batch_size: 4,

  checkpoint: fn stats -> save_progress(stats) end

)

See migration_guide_v1_1_0.md and docs/educational/v1_1_streaming_guide.md.

Performance

ExZarr v0.8+ includes major performance optimizations:

Benchmark results (400×400 array, 16 chunks):

See Performance Guide for tuning recommendations and Benchmarks for running your own tests.

# Run quick performance check (completes in 6 seconds)

mix run benchmarks/slicing_bench_quick.exs

Zarr Format Support

ExZarr provides production-ready support for both Zarr v2 and v3 specifications. Arrays can be created in either format, and opening arrays automatically detects the version.

Zarr v3 - Fully Supported (Recommended for New Projects)

Zarr v3 is fully implemented with a unified codec pipeline and improved metadata format:

# Create v3 array with unified codec pipeline

{:ok, array} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  dtype: :float64,

  codecs: [

    %{name: "bytes"},  # Required array-to-bytes codec

    %{name: "gzip", configuration: %{level: 5}}  # Optional compression

  ],

  zarr_version: 3,

  storage: :filesystem,

  path: "/tmp/my_v3_array"

)

Zarr v2 (Default for Compatibility)

Zarr v2 uses separate filters and compressor configuration:

# Create v2 array (explicit version)

{:ok, array} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  dtype: :float64,

  filters: [{:shuffle, [elementsize: 8]}],

  compressor: :zlib,

  zarr_version: 2,

  storage: :filesystem,

  path: "/tmp/my_v2_array"

)

Automatic Version Detection

When opening arrays, ExZarr automatically detects the format version:

# Opens v2 or v3 transparently

{:ok, array} = ExZarr.open(path: "/tmp/my_array")



# Check which version was detected

array.version  # Returns 2 or 3

Key Differences Between v2 and v3

Feature v2 v3
Metadata file .zarray zarr.json
Chunk keys Dot-separated (0.1.2) Slash-separated with prefix (c/0/1/2)
Codec organization Separate filters and compressor Unified codecs array
Data types NumPy-style strings (<f8) Simplified names (float64)
Groups Separate .zgroup files Unified zarr.json with node_type
Attributes Separate .zattrs files Embedded in zarr.json

Converting from v2 to v3

v2-style configuration is automatically converted when creating v3 arrays:

# This v2-style configuration

{:ok, array} = ExZarr.create(

  shape: {1000},

  chunks: {100},

  dtype: :int64,

  filters: [{:shuffle, [elementsize: 8]}],

  compressor: :zlib,

  zarr_version: 3  # Request v3 format

)



# Automatically converts to v3 codec pipeline:

# [

#   %{name: "shuffle", configuration: %{elementsize: 8}},

#   %{name: "bytes"},

#   %{name: "gzip", configuration: %{level: 5}}

# ]

For detailed migration guidance, see docs/V2_TO_V3_MIGRATION.md.

Working with Groups

# Create a hierarchical group structure

{:ok, root} = ExZarr.Group.create("/data",

  storage: :filesystem,

  path: "/tmp/zarr_data"

)



# Create arrays within the group

{:ok, measurements} = ExZarr.Group.create_array(root, "measurements",

  shape: {1000},

  chunks: {100},

  dtype: :float64

)



# Create subgroups

{:ok, subgroup} = ExZarr.Group.create_group(root, "experiments")

Interoperability with Python

ExZarr is fully compatible with Python's zarr library. Arrays created by one can be read by the other:

# Run the interoperability demo

elixir examples/python_interop_demo.exs

This demonstrates:

For detailed interoperability information, see INTEROPERABILITY.md which covers:

Custom Codecs Example

See how to create and use custom compression codecs:

# Run the custom codec example

mix run examples/custom_codec_example.exs

This demonstrates:

Custom Storage Backend Example

See the test suite for a complete example of implementing a custom storage backend:

# View the custom storage tests

cat test/ex_zarr_custom_storage_test.exs

The example demonstrates:

Supported Data Types

ExZarr supports the following data types:

All data types use little-endian byte order by default, consistent with the Zarr specification.

Compression Codecs

ExZarr provides the following built-in compression options:

:zlib uses Erlang's built-in :zlib module. The other codecs come from ExCodecs, pure-Rust NIFs that are downloaded precompiled for macOS, Linux (glibc and musl) and Windows, so no system libraries or compilers are needed.

Custom Codecs

ExZarr supports custom codecs through a behavior-based plugin system. You can create your own compression, checksum, or transformation codecs:

defmodule MyCustomCodec do

  @behaviour ExZarr.Codecs.Codec



  @impl true

  def codec_id, do: :my_codec



  @impl true

  def codec_info do

    %{

      name: "My Custom Codec",

      version: "1.0.0",

      type: :compression,  # or :transformation

      description: "My custom compression algorithm"

    }

  end



  @impl true

  def available?, do: true



  @impl true

  def encode(data, opts) when is_binary(data) do

    # Your encoding logic here

    {:ok, compressed_data}

  end



  @impl true

  def decode(data, opts) when is_binary(data) do

    # Your decoding logic here

    {:ok, decompressed_data}

  end



  @impl true

  def validate_config(opts) do

    # Validate options

    :ok

  end

end



# Register your codec

:ok = ExZarr.Codecs.register_codec(MyCustomCodec)



# Use it like any built-in codec

{:ok, array} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  compressor: :my_codec

)

For complete examples, see examples/custom_codec_example.exs which includes:

Custom codec features:

Storage Backends

ExZarr includes three built-in storage backends:

Arrays stored on the filesystem use the standard Zarr format:

Using Zip Storage

Zip storage stores the entire array (metadata + all chunks) in a single zip file:

# Create array with zip storage

{:ok, array} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  dtype: :float64,

  storage: :zip,

  path: "/tmp/my_array.zip"

)



# Write data

ExZarr.Array.set_slice(array, data, start: {0, 0}, stop: {100, 100})



# Save to zip file

:ok = ExZarr.save(array, path: "/tmp/my_array.zip")



# Open existing zip

{:ok, reopened} = ExZarr.open(path: "/tmp/my_array.zip", storage: :zip)

Custom Storage Backends

ExZarr supports custom storage backends through a behavior-based plugin system, similar to custom codecs. Create backends for S3, databases, cloud storage, or any other storage system:

defmodule MyApp.S3Storage do

  @behaviour ExZarr.Storage.Backend



  @impl true

  def backend_id, do: :s3



  @impl true

  def init(config) do

    # Initialize S3 connection

    bucket = Keyword.fetch!(config, :bucket)

    {:ok, %{bucket: bucket, client: setup_s3_client()}}

  end



  @impl true

  def read_chunk(state, chunk_index) do

    # Read chunk from S3

    key = build_s3_key(chunk_index)

    AWS.S3.get_object(state.client, state.bucket, key)

  end



  @impl true

  def write_chunk(state, chunk_index, data) do

    # Write chunk to S3

    key = build_s3_key(chunk_index)

    AWS.S3.put_object(state.client, state.bucket, key, data)

  end



  # Implement other required callbacks...

end



# Register your backend

:ok = ExZarr.Storage.Registry.register(MyApp.S3Storage)



# Use it like any built-in backend

{:ok, array} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  storage: :s3,

  bucket: "my-zarr-data"

)

Custom storage backend features:

Required callbacks:

Cloud and Database Storage Backends

ExZarr includes several pre-built storage backends for cloud services and databases:

AWS S3 Storage

# Add dependencies

{:ex_aws, "~> 2.5"},

{:ex_aws_s3, "~> 2.5"}



# Register and use

:ok = ExZarr.Storage.Registry.register(ExZarr.Storage.Backend.S3)



{:ok, array} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  storage: :s3,

  bucket: "my-zarr-bucket",

  prefix: "experiments/array1",

  region: "us-west-2"

)

Azure Blob Storage

# Add dependency

{:azure_sdk, "~> 0.4.1"}



# Register and use

:ok = ExZarr.Storage.Registry.register(ExZarr.Storage.Backend.AzureBlob)



{:ok, array} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  storage: :azure_blob,

  account_name: "mystorageaccount",

  account_key: System.get_env("AZURE_STORAGE_KEY"),

  container: "zarr-data",

  prefix: "experiments/array1"

)

Google Cloud Storage

# Add dependencies

{:goth, "~> 1.4"},

{:req, "~> 0.6.1"}



# Register and use

:ok = ExZarr.Storage.Registry.register(ExZarr.Storage.Backend.GCS)



{:ok, array} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  storage: :gcs,

  bucket: "my-zarr-bucket",

  prefix: "experiments/array1",

  credentials: "/path/to/service-account.json"

)

Mnesia (Distributed Database)

# No external dependencies - Mnesia is built into Erlang/OTP



# Initialize Mnesia

:mnesia.create_schema([node()])

:mnesia.start()



# Register and use

:ok = ExZarr.Storage.Registry.register(ExZarr.Storage.Backend.Mnesia)



{:ok, array} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  storage: :mnesia,

  array_id: "experiment_001",

  table_name: :zarr_storage

)

MongoDB GridFS

# Add dependency

{:mongodb_driver, "~> 1.4"}



# Register and use

:ok = ExZarr.Storage.Registry.register(ExZarr.Storage.Backend.MongoGridFS)



{:ok, array} = ExZarr.create(

  shape: {1000, 1000},

  chunks: {100, 100},

  storage: :mongo_gridfs,

  url: "mongodb://localhost:27017",

  database: "zarr_db",

  bucket: "arrays",

  array_id: "experiment_001"

)

Mock Storage (Testing)

# No dependencies - built-in for testing



:ok = ExZarr.Storage.Registry.register(ExZarr.Storage.Backend.Mock)



# Test with error simulation

{:ok, array} = ExZarr.create(

  shape: {100},

  chunks: {10},

  storage: :mock,

  pid: self(),

  error_mode: :random,

  delay: 50  # Simulate 50ms latency

)



# Verify operations

assert_received {:mock_storage, :write_chunk, _}

Cloud Storage Features:

Database Storage Features:

Mock Storage Features:

Architecture

ExZarr uses:

Development

Requires Elixir ~> 1.17 and OTP 26+. No native toolchain or system libraries are needed; the codec NIFs come precompiled with ExCodecs.

# Install dependencies

mix deps.get



# Compile

mix compile



# Run tests

mix test



# Run tests with coverage

mix coveralls



# Run specific test suites

mix test test/ex_zarr_property_test.exs              # Property-based tests

mix test test/ex_zarr_python_integration_test.exs    # Python integration tests



# Run static analysis

mix credo



# Run type checking

mix dialyzer



# Generate documentation

mix docs

Quality Checks

Before committing, ensure all quality checks pass:

# Run all tests

mix test



# Check code style

mix credo --strict



# Run type checker

mix dialyzer



# Verify test coverage

mix coveralls

CI/CD

The project uses GitHub Actions for continuous integration. The CI pipeline:

Testing

ExZarr includes comprehensive test coverage:

Key testing areas:

Python Integration Tests

ExZarr includes integration tests that verify compatibility with Python's zarr library. Full local setup (v2 on Python 3.11, v3 on Python ≥ 3.12, fixtures, Mix tags) is in Testing Python Interoperability.

# Zarr v3 example (Python ≥ 3.12)

python3.12 -m venv .venv-zarr3 && source .venv-zarr3/bin/activate

pip install 'zarr==3.4.0' numpy

OUT=/tmp/zarr_fixtures_v3

python test/support/python_fixtures/generate_fixtures.py --out "$OUT" --zarr-major 3

export EXZARR_PYTHON_FIXTURES="$OUT"

mix test --only python_fixtures

mix test --include python test/ex_zarr_v3_python_interop_test.exs

These tests verify that:

Requirements: Python 3.11+ (v2) or 3.12+ (v3), matching zarr-python, numpy

Documentation

Guides

Comprehensive guides for all skill levels:

Examples

Practical examples demonstrating real-world usage:

API Documentation

Full API documentation is available at hexdocs.pm/ex_zarr.

Key modules:

Roadmap

See ROADMAP.md for the full release plan.

v1.2.0 (released 2026-10-03) - Zarr 3.1 interoperability & range-aware cloud I/O: spec sharding_indexed, optional byte-range reads, version-aware cloud keys, AzureSDK migration, Python fixture/CI matrix.

Upcoming (high level):

Contributing

Contributions are welcome. Please:

  1. Fork the repository
  2. Create a feature branch
  3. Add tests for new functionality
  4. Ensure all tests pass with mix test
  5. Run code quality checks with mix credo and mix dialyzer
  6. Submit a pull request

License

MIT

Credits

Inspired by zarr-python. Implements both Zarr v2 and v3 specifications for full compatibility with the broader Zarr ecosystem.