thds.adls 4.5.20260121010416

ADLS tools

thds.adls

A high-performance Azure Data Lake Storage (ADLS Gen2) client for the THDS monorepo. It wraps the Azure SDK with hash-aware caching, azcopy acceleration, and shared client/credential plumbing so applications can transfer large blob datasets quickly and reliably.

Highlights

• Environment-aware paths first: Almost every consumer starts by importing fqn, AdlsFqn, and defaults.env_root() to build storage-account/container URIs that follow the current THDS environment.
• Cache-backed reads: download_to_cache is the standard entry point for pulling blobs down with a verified hash so local workflows, tests, and pipelines can operate on read-only copies.
• Bulk filesystem helpers: ADLSFileSystem powers scripts and jobs that need to walk directories, fetch batches of files, or mirror hive tables without re-implementing Azure SDK plumbing.
• Spark/Databricks bridges: abfss and uri conversions keep analytics code agnostic to whether it needs an adls://, abfss://, https://, or dbfs:// view of the same path.
• Composable utilities: Higher-level modules (cache, upload, copy, list) layer on top of those imports so teams can opt into more advanced behavior without leaving the public API surface.

Key Modules

Component	Typical usage in the monorepo
fqn	Parse, validate, and join ADLS paths; used when materializing model datasets and configuring pipelines.
AdlsFqn	Strongly typed value passed between tasks and tests to represent a single blob or directory.
defaults / named_roots	Resolve environment-specific storage roots (defaults.env_root(), named_roots.require(...)).
download_to_cache (cached module)	Bring a blob down to the shared read-only cache before analytics, feature builds, or test fixtures run.
ADLSFileSystem (impl module)	Fetch or list entire directory trees and integrate with caching inside scripts and notebooks.
abfss	Translate AdlsFqn objects into abfss:// URIs for Spark/Databricks jobs.
uri	Normalize adls://, abfss://, https://, and dbfs:// strings into AdlsFqn values (and vice versa).
global_client / shared_credential	Shared, fork-safe Azure clients and credentials backing the public helpers above.

Example Usage

1. Use the caching helpers and Source integration:

   from thds.adls import cached, upload, source
   
   cache_path = cached.download_to_cache("adls://acct/container/path/to/file")
   src = upload("adls://acct/container/path/out.parquet", cache_path)
   verified = source.get_with_hash(src.uri)
   

2. For CLI usage, run (from repo root):

   uv run python -m thds.adls.tools.download adls://acct/container/path/file
   

Operational Notes

• Hash metadata: Uploads attach hash_xxh3_128_b64 automatically when the bytes are known. Download completion back-fills missing hashes when permissions allow.
• Locks and concurrency: Large transfers acquire per-path file locks to keep azcopy instances cooperative. Global HTTP connection pools default to 100 but are configurable via thds.core.config.
• Error handling: BlobNotFoundError and other ADLS-specific exceptions translate into custom error types to simplify retries and diagnostics.
• Extensibility: Additional hash algorithms can be registered by importing dependent packages (e.g., blake3). Named roots can be populated dynamically via environment-specific modules (thds.adls._thds_defaults hook).