spectrumx 0.1.13

SpectrumX Data System SDK

SpectrumX Data System | SDK

• SpectrumX Data System | SDK
  • Getting Started
    • Installation
    • Basic Usage
    • Error Handling
  • Concurrent Access

The SpectrumX Data System (SDS) SDK is a Python package that provides a simple interface for interacting with the SDS Gateway. The SDK is designed to be easy to use and to provide a high-level interface for common tasks, such as uploading and downloading files, searching for files, and managing RF datasets.

[!NOTE]

SDS is not meant for personal files or as a backup tool. Files may be rejected by the Gateway when uploaded, or deleted without warning. All uploaded files have an expiration date. Do not upload sensitive, personally identifiable, confidential information, or any file that you do not have permission to share. Do not upload binary executables.

If you own data in https://sds.crc.nd.edu that needs to be permanently deleted, please reach out to the team at crc-sds-list [·at·] nd.edu, as SDS may retain uploaded data for a period of time after deletion.

Getting Started

Installation

uv add spectrumx

# or one of:
#   poetry add spectrumx
#   pip install spectrumx
#   conda install spectrumx
#   ...

[!NOTE] When not using uv, make sure you are using Python 3.12 or higher (python --version).

Basic Usage

1. In a file named .env, enter the secret_token provided to you:

   SDS_SECRET_TOKEN=your-secret-token-no-quotes
   

   OR set the environment variable SDS_SECRET_TOKEN to your secret token:

   # the env var takes precedence over the .env file
   export SDS_SECRET_TOKEN=your-secret-token
   

2. Then, in your Python script or Jupyter notebook

   See ./tests/e2e_examples/check_build_acceptance.py for more examples.

   from pathlib import Path
   from random import randint, random
   from spectrumx.client import Client
   
   # Example of files upload, listing, and download from SDS.
   
   # NOTE: the SDS client-server interaction is stateless, so it is
   #   not recommended to have multiple clients writing to the same
   #   locations simultaneously, as they may overrule each other
   #   and cause loss of data. See "Concurrent Access" in the
   #   usage guide to learn more.
   sds = Client(
       host="sds.crc.nd.edu",
       # env_file=Path(".env"),  # default, recommended to keep tokens out of version control
       # env_config={"SDS_SECRET_TOKEN": "my-custom-token"},  # alternative way to pass the access token
   )
   
   # when in dry-run (default), no changes are made to the SDS or the local filesystem
   # to enable the changes, set dry_run to False, as in:
   # sds.dry_run = False
   
   # authenticate using either the token from
   # the .env file or in the config passed in
   sds.authenticate()
   
   # local_dir has your own local files that will be uploaded to the SDS
   reference_name: str = "my_spectrum_capture"
   local_dir: Path = Path(reference_name)
   
   # or, if the directory doesn't exist, let's create some fake data
   if not local_dir.exists():
       local_dir.mkdir(exist_ok=True)
       num_files = 10
       for file_idx in range(num_files):
           num_lines = randint(10, 100)  # noqa: S311
           file_name = f"capture_{file_idx}.csv"
           with (local_dir / file_name).open(mode="w", encoding="utf-8") as file_ptr:
               fake_nums = [random() for _ in range(num_lines)]  # noqa: S311
               file_ptr.write("\n".join(map(str, fake_nums)))
   
   # upload all files in a directory to the SDS
   # sds.dry_run = False   # uncomment to actually upload the files
   upload_results = sds.upload(
       local_path=local_dir,  # may be a single file or a directory
       sds_path=reference_name,  # files will be created under this virtual directory
       verbose=True,  # shows a progress bar (default)
   )
   success_results = [success for success in upload_results if success]
   failed_results = [success for success in upload_results if not success]
   assert len(failed_results) == 0, (
       f"No failed uploads should be present: {failed_results}"
   )
   log.debug(f"Uploaded {len(success_results)} assets.")
   
   # download the files from an SDS directory
   # sds.dry_run = False
   local_downloads = Path("sds-downloads") / "files" / reference_name
   sds.download(
       from_sds_path=reference_name,  # files will be downloaded from this virtual dir
       to_local_path=local_downloads,  # download to this location (it may be created)
       overwrite=False,  # do not overwrite local existing files (default)
       verbose=True,  # shows a progress bar (default)
   )
   
   if not sds.dry_run:
       print("Downloaded files:")
       for file_path in local_downloads.iterdir():
           print(file_path)
   else:
       print("Turn off dry-run to download and write files.")
   

Error Handling

The SDK provides context-aware exceptions that can be caught and handled in your code.

# ======== Authentication ========
from pathlib import Path
from spectrumx.client import Client
from spectrumx.errors import AuthError, NetworkError

sds = Client(host="sds.crc.nd.edu")
try:
    sds.authenticate()
except NetworkError as err:
    print(f"Failed to connect to the SDS: {err}")
    # check your host= parameter and network connection
    # if you're hosting the SDS Gateway, make sure it is accessible
except AuthError as err:
    print(f"Failed to authenticate: {err}")
    # TODO: take action

# ======== File operations ========

from time import sleep
from spectrumx.errors import NetworkError
from spectrumx.errors import Result
from spectrumx.errors import SDSError
from spectrumx.errors import ServiceError
from loguru import logger as log

# ...
local_dir: Path = Path("my_spectrum_files")
reference_name: str = "my_spectrum_files"
retries_left: int = 5
is_success: bool = False
uploaded_files: list[File] = []
while not is_success and retries_left > 0:
    try:
        retries_left -= 1

        # `sds.upload()` will restart a partial file transfer from zero,
        # but it won't re-upload already finished files.
        upload_results: list[Result[File]] = sds.upload(
            local_path=local_dir,
            sds_path=reference_name,
            verbose=True,
        )

        # Since `upload()` is a batch operation, some files may succeed and some
        #   may fail. The return value of `sds.upload` stored in `upload_results`
        #   is a list of `Result` objects:
        # A `Result` wraps either the value of a variable (in this case the File
        #   object that was uploaded) or an exception. Here's how we can check if
        #   there were any failed uploads:
        success_results = [success for success in upload_results if success]
        failed_results = [success for success in upload_results if not success]

        log.debug(f"Uploaded {len(success_results)} assets.")
        log.warning(f"Failed to upload {len(failed_results)} assets")

        # calling a successful result will return the value it holds
        uploaded_files = [result() for result in success_results]

        # And calling a failed result will raise the exception it holds.
        # Here we re-raise it to handle retries with the except blocks below,
        #   based on the exception raised:
        for result in failed_results:
            result()  # will raise

    except (NetworkError, ServiceError) as err:
        # NetworkError refers to connection issues between client and SDS Gateway
        # ServiceError refers to issues with the SDS Gateway itself (e.g. HTTP 500)
        # sleep longer with each retry, at least 5s, up to 5min
        sleep_time = max(5, 5 / (retries_left**2) * 60)
        log.error(f"Error: {err}")
        log.warning(f"Failed to reach the gateway; sleeping {sleep_time}s")
        if retries_left > 0:
            sleep(sleep_time)
        continue
    except SDSError as err:
        log.error(f"Another SDS error occurred: {err}")
        # other errors might include e.g. OSError
        #   if listed files cannot be found.
        # TODO: take action or break
        break

log.debug(f"Uploaded files: {uploaded_files}")

Concurrent Access

The SDS client-server interaction is stateless, meaning that each request contains all the information needed to complete that request. One positive outcome is that it allows multiple clients to interact with the SDS Gateway at the same time. However, this opens up the possibility of having multiple clients writing to the same locations simultaneously, causing loss of data by overruling each other's writes (race condition).

For example, if two clients are uploading files with the same directory, file names, and at the same time, only the last file successfully uploaded (from the Gateway's perspective) is guaranteed to be kept, which might not be aligned with the user's expectations.

To avoid potential race conditions, it is not recommended to have multiple clients writing to the same locations simultaneously. Neither the SDK nor the Gateway currently take any measure to detect this, in part, because any measure towards it would either be incomplete, or it would make our APIs stateful and significantly increase code complexity.

If this is needed, SDK users have a few options:

1. Restructure their architecture to forward writes to a single centralized client responsible for them.
2. Restructure the code by writing to different locations and/or at different application stages. The latter assumes all conflicting clients are part of the same application.
3. Implement a custom locking mechanism for writes to serve their specific use case.

One writer (an SDK client that creates, updates, and/or deletes contents) and multiple readers are generally safe.