metabaiji-pod 1.1.1

The versioned cache

The versioned cache provides access to a repository of files. The changes to those files are tracked and identified with to a semver-like version number.

To use the versioned cache, you need a copy of a manifest file, which lists all the versioned paths and the latest version of each one. When you request a file from the cache, it consults this manifest file to determine the correct version. The versioned cache delegates loading to the underlying asset cache.

The versioned cache was designed for compute assets: chunks of data which are used in code. When the manifest is checked in with the code, it pins the version of each asset. If the asset is subsequently updated, that revision of the code will continue to get the version it’s expecting.

The bucket containing the versioned assets is intended to be immutable. Nothing there should ever be changed or deleted. Only new versions added.

The manifest looks like this:

{
    "/foo/bar.csv": "1.2.5",
    "/foo/bar.json": "0.1.6"
}

To load a versioned asset:

import json
from baiji.pod import AssetCache
from baiji.pod import Config
from baiji.pod import VersionedCache

config = Config()
# Improve performance by assuming the bucket is immutable.
config.IMMUTABLE_BUCKETS = ['my-versioned-assets']

vc = VersionedCache(
    cache=AssetCache(config),
    manifest_path='versioned_assets.json',
    bucket='my-versioned-assets')

with open(vc('/foo/bar.json'), 'r') as f:
    data = json.load(f)

Or, with `baiji-serialization <https://github.com/bodylabs/baiji-serialization>`__:

from baiji.serialization import json
data = json.load(vc('s3://example-bucket/example.json'))

To add a new versioned path, or update an existing one, use the vc command-line tool:

vc add /foo/bar.csv ~/Desktop/bar.csv
vc update --major /foo/bar.csv ~/Desktop/new_bar.csv
vc update --minor /foo/bar.csv ~/Desktop/new_bar.csv
vc update --patch /foo/bar.csv ~/Desktop/new_bar.csv

A VersionedCache object is specific to a manifest file and a bucket.

Though the version number uses semver-like semantics, the cache ignores version semantics. The manifest pins an exact version number.

The asset cache

The asset cache works at a lower level of abstraction. It holds local copies of arbitrary S3 assets. Calling the cache() function with an S3 path ensures that the file is available locally, and then returns a valid, local path.

On a cache miss, the file is downloaded to the cache and then its local path is returned. Subsequent calls will return the same local path. After a timeout, which defaults to one day, the validity of the local file is checked by comparing a local MD5 hash with the remote etag. This check is repeated once per day.

To gain a performance boost, you can configure immutable buckets, whose contents are never revalidated after download. The versioned cache uses this feature.

import json
from baiji.pod import AssetCache

cache = AssetCache.create_default()

with open(cache('s3://example-bucket/example.json'), 'r') as f:
    data = json.load(f)

Or, with `baiji-serialization <https://github.com/bodylabs/baiji-serialization>`__:

from baiji.serialization import json
data = json.load(cache('s3://example-bucket/example.json'))

It is safe to call cache multiple times: cache(cache('path')) will behave correctly.