hidebound 0.21.1

A local database service for converting directories of arbitrary files into validated assets and derived JSON metadata.

Introduction

A local database service for converting directories of arbitrary files into validated assets and derived metadata for export to databases like AWS S3 and MongoDB.

See documentation for details.

Installation

Python

pip install hidebound

Docker

1. Install docker
2. Install docker-machine (if running on macOS or Windows)
3. docker pull thenewflesh/hidebound:latest

Docker For Developers

1. Install docker
2. Install docker-machine (if running on macOS or Windows)
3. Ensure docker-machine has at least 4 GB of memory allocated to it.
4. git clone [e-mail]:theNewFlesh/hidebound.git
5. cd hidebound
6. chmod +x bin/hidebound
7. bin/hidebound start

The service should take a few minutes to start up.

Run bin/hidebound --help for more help on the command line tool.

Overview

Hidebound is an ephemeral database and asset framework used for generating, validating and exporting assets to various data stores. Hidebound enables developers to ingest arbitrary sets of files and output them as content and generated metadata, which has validated according to specifications they define.

Assets are placed into an ingress directory, typically reserved for Hidebound projects, and then processed by Hidebound. Hidebound extracts metadata from the files and directories that make each asset according to their name, location and file properties. This data comprises the entirety of Hidebound's database at any one time.

Dataflow

Data begins as files on disk. Hidebound creates a JSON-compatible dict from their name traits and file traits and then constructs an internal database table from them, one dict per row. All the rows are then aggregated by asset, and converted into JSON blobs. Those blobs are then validated according to their respective specifications. Files from valid assets are then copied or moved into Hidebound's content directory, according to their same directory structure and naming. Metadata is written to JSON files inside Hidebound's metadata directory. Each file's metadata is written as a JSON file in /hidebound/metadata/file, and each asset's metadata (the aggregate of its file metadata) is written to /hidebound/metadata/asset. From their exporters, can export the valid asset data and its accompanying metadata to various locations, like an AWS S3 bucket.

Workflow

The acronynm to remember for workflows is CRUDES: create, read, update, delete, export and search. Those operations constitue the main functionality that Hidebound supports.

Create Asset

For example, an asset could be an image sequence, such as a directory full of PNG files, all of which have a frame number, have 3 (RGB) channels, and are 1024 pixels wide by 1024 pixels tall. Let's call the specification for this type of asset "spec001". We create an image sequence of a cat running, and we move it into the Hidebound projects directory.

Update

We call the update function via Hidebound's web app. Hidebound creates a new database based upon the recursive listing of all the files within said directory. This database is displayed to us as a table, with one file per row. If we choose to group by asset in the app, the table will display one asset per row. Hidebound extracts metadata from each filename (not any directory name) as well as from the file itself. That metadata is called file_traits. Using only information derived from filename and file traits, Hidebound determines which files are grouped together as a single asset and the specification of that asset. Asset traits are then derived from this set of files (one or more). Finally, Hidebound validates each asset according to its determined specification. All of this data is displayed as a table within the web app. Importantly, all of the errors in filenames, file traits and asset traits are included.

Review Graph

If we click on the graph tab, we are greeted by a hierarchical graph of all our assets in our project directory. Our asset is red, meaning it's invalid. Valid asset's are green, and all other files and directories, including parent directories, are cyan.

Diagnose and Repair

We flip back to the data tab. Using table within it, we search (via SQL) for our asset within Hidebound's freshly created database. We see an error in one of the filenames, conveniently displayed in red text. The descriptor in one orf our filenames has capital letters in it. This violates Hidebound's naming convention, and so we get an error. We go and rename the file appropriately and call update again. Our asset is now valid. The filenames are correct and we can see in the height and width columns, that it's 1024 by 1024 and the channels column says it has three.

Create

Next we click the create button. For each valid asset, Hidebound generates file and asset metadata as JSON files within the hidebound/metadata directory. Hidebound also copies or moves, depending on the config write mode, valid files and directories into the hidebound/content directory. Hidebound/content and hidebound/metadata are both staging directories used for generating a valid ephemeral database. We now have a hidebound directory that looks like this (unmentioned assets are collapsed behind the ellipses):

/tmp/hidebound
├── hidebound_config.json
│
├── specifications
│   └── specifications.py
│
├── data
│   ...
│   └── p-cat001
│       └── spec001
│           └── p-cat001_s-spec001_d-running-cat_v001
│               ├── p-cat001_s-spec001_d-running-cat_v001_c0000-0005_f0001.png
│               ├── p-cat001_s-spec001_d-running-cat_v001_c0000-0005_f0002.png
│               └── p-cat001_s-spec001_d-running-cat_v001_c0000-0005_f0003.png
│
├── metadata
    ├── asset
    │   ...
    │   └── a9f3727c-cb9b-4eb1-bc84-a6bc3b756cc5.json
    │
    └── file
        ...
        ├── 279873a2-bfd0-4757-abf2-7dc4f771f992.json
        ├── e50160ae-8678-40b3-b766-ee8311b1f0c9.json
        └── ea95bd79-cb8f-4262-8489-efe734c5f65c.json

Export

The hidebound directories contain only valid assets. Thus, we are now free to export this data to various data stores, such as AWS S3, MongoDB, and Girder. Exporters are are defined within the exporters subpackage. They expect a populated hidebound directory and use the files and metadata therein to export hidebound data. Exporter configurations are stored in the hidebound config, under the "exporters" key. Currently supported exporters include, local disk, s3 and girder. Below we can see the results of an export to Girder in the Girder web app.

Delete

Once this export process is complete, we may click the delete button. Hidebound deletes the hidebound/content and hidebound/metdata directories and all their contents. If write_mode in the Hidebound configuration is set to "copy", then this step will merely delete data created by Hidebound. If it is set to "move", then Hidebound will presumably delete, the only existing copy of out asset data on the host machine. The delete stage in combination with the removal of assets from the ingress directory is what makes Hidebound's database ephemeral.

Workflow

/api/workflow is a API endpoint that initializes a database a with a given config, and then calls each method from a given list. For instance, if you send this data to /api/workflow:

{config={...}, workflow=['update', 'create', 'export', 'delete']}

A database instance will be created with the given config, and then that instance will call its update, create, export and delete methods, in that order.

Naming Convention

Hidebound is a highly opinionated framework that relies upon a strict but composable naming convention in order to extract metadata from filenames. All files and directories that are part of assets must conform to a naming convention defined within that asset's specification.

In an over-simplified sense; sentences are constructions of words. Syntax concerns how each word is formed, grammar concerns how to form words into a sentence, and semantics concerns what each word means. Similarly, filenames can be thought of as crude sentences. They are made of several words (ie fields). These words have distinct semantics (as determines by field indicators). Each word is constructed according to a syntax (ie indicator + token). All words are joined together by spaces (ie underscores) in a particular order as determined by grammar (as defined in each specification).

Syntax

• Names consist of a series of fields, each separated by a single underscore “_”, also called a field separator.
• Periods, ".", are the exception to this, as it indicates file extension.
• Legal characters include and only include:

Name	Characters	Use
Underscore	_	only for field separation
Period	.	only for file extensions
Lowercase letter	a to z	everything
Number	0 to 9	everything
Hyphen	-	token separator

Fields are comprised of two main parts:

Name	Use
Field indicator	determines metadata key
Field token	a set of 1+ characters that define the field's data

---

Example Diagrams

In our example filename: p-cat001_s-spec001_d-running-cat_v001_c0000-0005_f0003.png the metadata will be:

{
    "project": "cat001",
    "specification": "spec001",
    "descriptor": "running-cat",
    "version": 1,
    "coordinate": [0, 5],
    "frame": 3,
    "extension": "png",
}

The spec001 specification is derived from the second field of this filename:

      field   field
  indicator   token
          | __|__
         | |     |
p-cat001_s-spec001_d-running-cat_v001_c0000-0005_f0003.png
         |_______|
             |
           field

Part	Value
Field	s-spec001
Field indicator	s-
Field token	spec001
Derived metadata	{specification: spec001}

Special Field Syntax

• Projects begin with 3 or 4 letters followed by 1 to 4 numbers
• Specifications begin with 3 or 4 letters followed by 3 numbers
• Descriptors begin with a letter or number and may also contain hyphens
• Descriptors may not begin with the words master, final or last
• Versions are triple-padded with zeros and must be greater than 0
• Coordinates may contain up to 3 quadruple-padded numbers, separated by hyphens
• Coordinates are always evaluated in XYZ order. For example: c0001-0002-0003 produces {x: 1, y: 2, z: 3}.
• Each element of a coordinate may be equal to or greater than zero
• Frames are quadruple-padded and are greater than or equal to 0
• Extensions may only contain upper and lower case letters a to z and numbers 0 to 9

Semantics

Hidebound is highly opionated, especially with regards to its semantics. It contains exactly seven field types, as indicated by their field indicators. They are:

Field	Indicator
project	p-
specification	s-
descriptor	d-
version	v
coordinate	c
frame	f
extension	.

Grammar

The grammar is fairly simple:

• Names are comprised of an ordered set of fields drawn from the seven above
• All names must contain the specification field
• All specification must define a field order
• All fields of a name under that specification must occcur in its defined field order

Its is highly encouraged that fields be defined in the following order:

project specification descriptor version coordinate frame extension

The grammatical concept of field order here is one of rough encapsulation:

• Projects contain assets
• Assets are grouped by specification
• A set of assets of the same content is grouped by a descriptor
• That set of assets consists of multiple versions of the same content
• A single asset may broken into chunks, identified by 1, 2 or 3 coordinates
• Each chunk may consist of a series of files seperated by frame number
• Each file has an extension

Encouraged Lexical Conventions

• Specifications end with a triple padded number so that they may be explicitely versioned. You redefine an asset specification to something slightly different, by copying its specification class, adding one to its name and change the class attributes in some way. That way you always maintain backwards compatibility with legacy assets.
• Descriptors are not a dumping ground for useless terms like wtf, junk, stuff, wip and test.
• Descriptors should not specify information known at the asset specification level, such as the project name, the generic content of the asset (ie image, mask, png, etc).
• Descriptors should not include information that can be known from the preceding tokens, such as version, frame or extension.
• A descriptor should be applicable to every version of the asset it designates.
• Use of hyphens in descriptors is encouraged.
• When in doubt, hyphenate and put into the descriptor.

Project Structure

Hidebound does not formally define a project structure. It merely stipulates that assets must exist under some particular root directory. Each asset specification does define a directory structure for the files that make up that asset. Assets are divided into 3 types: file, sequence and complex. File defines an asset that consists of a single file. Sequence is defined to be a single directory containing one or more files. Complex is for assets that consist of an arbitrarily complex layout of directories and files.

The following project structure is recommended:

project
    |-- specification
        |-- descriptor
            |-- asset      # either a file or directory of files and directories
                |- file

For Example

/tmp/projects
└── p-cat001
    ├── s-spec002
    │   ├── d-calico-jumping
    │   │   └── p-cat001_s-spec002_d-calico-jumping_v001
    │   │       ├── p-cat001_s-spec002_d-calico-jumping_v001_f0001.png
    │   │       ├── p-cat001_s-spec002_d-calico-jumping_v001_f0002.png
    │   │       └── p-cat001_s-spec002_d-calico-jumping_v001_f0003.png
    │   │
    │   └── d-tabby-playing
    │       ├── p-cat001_s-spec002_d-tabby-playing_v001
    │       │   ├── p-cat001_s-spec002_d-tabby-playing_v001_f0001.png
    │       │   ├── p-cat001_s-spec002_d-tabby-playing_v001_f0002.png
    │       │   └── p-cat001_s-spec002_d-tabby-playing_v001_f0003.png
    │       │
    │       └── p-cat001_s-spec002_d-tabby-playing_v002
    │           ├── p-cat001_s-spec002_d-tabby-playing_v002_f0001.png
    │           ├── p-cat001_s-spec002_d-tabby-playing_v002_f0002.png
    │           └── p-cat001_s-spec002_d-tabby-playing_v002_f0003.png
    │
    └── spec001
        └── p-cat001_s-spec001_d-running-cat_v001
            ├── p-cat001_s-spec001_d-Running-Cat_v001_c0000-0005_f0002.png
            ├── p-cat001_s-spec001_d-running-cat_v001_c0000-0005_f0001.png
            └── p-cat001_s-spec001_d-running-cat_v001_c0000-0005_f0003.png

Application

The Hidebound web application has five sections: data, graph, config, api and docs.

Data

The data tab is the workhorse of the Hidebound app.

Its functions are as follows:

• Search - Search the updated database's data via SQL
• Dropdown - Groups search results by file or asset
• Init - Initialized the database with the current config
• Update - Initializes and updates the database with the current config
• Create - Copies or moves valid assets to hidebound/content directory and creates JSON files in hidebound/metadata directory
• Delete - Deletes hidebound/content and hidebound/metadata directories

Prior to calling update, the application will look like this:

Graph

The graph tab is used for visualizing the state of all the assets within a root directory.

It's color code is as follows:

Color	Meaning
Cyan	Non-asset file or directory
Green	Valid asset
Red	Invalid asset

Config

The config tab is used for uploading and writing Hidebound's configuration file.

Its functions are as follows:

Name	Function
Upload	Upload a config JSON file
Write	Write config to hidebound/hidebound_config.json

API

The API tab is really a link to Hidebound's REST API documentation.

Docs

The API tab is really a link to Hidebound's github documentation.

Errors

Hidebound is oriented towards developers and technically proficient users. It displays errors in their entirety within the application.

Configuration

Hidebound is configured via the hidebound/config/hidebound_config.json file.

Hidebound configs consist of three main sections:

Base

• root_directory - the directory hidebound parses for assets that comprise its database
• hidebound_directory - the staging directory valid assets are created in
• specification_files - a list of python specification files
• include_regex - filepaths in the root that match this are included in the database
• exclude_regex - filepaths in the root that match this are excluded from the database
• write_mode - whether to copy or move files from root to staging
• dask_enabled - whether to enable Dask distributed computing
• dask_partitions - number of Dask partitions to use

Exporters

Which exporters to us in the workflow. Options include:

• s3
• local_disk
• girder

Webhooks

Webhooks to call after the export phase has completed.

---

Here is a full example config with comments:

{
    "root_directory": "/mnt/storage/projects",                         // where hb looks for assets
    "hidebound_directory": "/mnt/storage/hidebound",                   // hb staging directory
    "specification_files": [                                           // list of spec files
        "/mnt/storage/specs/image_specs.py",
        "/mnt/storage/specs/video_specs.py"
    ],
    "include_regex": "",                                               // include files that match
    "exclude_regex": "\\.DS_Store",                                    // exclude files that match
    "write_mode": "copy",                                              // copy files from root to staging
                                                                       // options: copy, move
    "dask_enabled": true,                                              // enable Dask distributed computing
    "dask_partitions": 16,                                             // number of Dask partitions to use
    "exporters": {                                                     // list of exporter configs
        "s3": {                                                        // export to s3
            "access_key": "ABCDEFGHIJKLMNOPQRST",                      // aws access key
            "secret_key": "abcdefghijklmnopqrstuvwxyz1234567890abcd",  // aws secret key
            "bucket": "prod-data",                                     // s3 bucket
            "region": "us-west-2",                                     // bucket region
            "metadata_types": ["asset", "asset-chunk", "file-chunk"]   // drop file metadata
                                                                       // options: asset, file, asset-chunk, file-chunk
        },
        "local_disk": {                                                // export to local disk
            "target_directory": "/mnt/storage/archive",                // target location
            "metadata_types": ["asset", "file"]                        // only asset and file metadata
                                                                       // options: asset, file, asset-chunk, file-chunk
        },
        "girder": {                                                    // export to girder
            "api_key": "eyS0nj9qPC5E7yK5l7nhGVPqDOBKPdA3EC60Rs9h",     // girder api key
            "root_id": "5ed735c8d8dd6242642406e5",                     // root resource id
            "root_type": "collection",                                 // root resource type
            "host": "http://prod.girder.com",                          // girder server url
            "port": 8180,                                              // girder server port
            "metadata_types": ["asset"]                                // only export asset metadata
                                                                       // options: asset, file
        }
    },
    "webhooks": [                                                      // call these after export
        {
            "url": "https://hooks.slack.com/services/ABCDEFGHI/JKLMNOPQRST/UVWXYZ1234567890abcdefgh",
            "method": "post",                                          // post this to slack
            "timeout": 60,                                             // timeout after 60 seconds
            // "params": {},                                           // params to post (NA here)
            // "json": {},                                             // json to post (NA here)
            "data": {                                                  // data to post
                "channel": "#hidebound",                               // slack data
                "text": "export complete",                             // slack data
                "username": "hidebound"                                // slack data
            },
            "headers": {
                "Content-type": "application/json"                     // request headers
            }
        }
    ]
}

Specification

Asset specifications are defined in python using the base classes found in specification_base.py. The base classes are defined using the schematics framework. Hidebound generates a single JSON blob of metadata for each file of an asset, and then combines blob into a single blob with a list values per key. Thus every class member defined with schematics is encapsulated with ListType.

Example asset

Suppose we have an image sequence asset that we wish to define a specificqtion for. Our image sequences consist of a directory containing 1 or 3 channel png with frame numbers in the filename.

projects
    └── cat001
        └── raw001
            └── p-cat001_s-raw001_d-calico-jumping_v001
                ├── p-cat001_s-raw001_d-calico-jumping_v001_f0001.png
                ├── p-cat001_s-raw001_d-calico-jumping_v001_f0002.png
                └── p-cat001_s-raw001_d-calico-jumping_v001_f0003.png

Example specification

We would write the following specification for such an asset.

from schematics.types import IntType, ListType, StringType
import hidebound.core.validators as vd  # validates traits
import hidebound.core.traits as tr      # gets properties of files and file names
from hidebound.core.specification_base import SequenceSpecificationBase

class Raw001(SequenceSpecificationBase):
    asset_name_fields = [  # naming convention for asset directory
        'project', 'specification', 'descriptor', 'version'
    ]
    filename_fields = [    # naming convention for asset files
        'project', 'specification', 'descriptor', 'version', 'frame',
        'extension'
    ]
    height = ListType(IntType(), required=True)  # heights of png images
    width = ListType(IntType(), required=True)   # widths of png images
    frame = ListType(
        IntType(),
        required=True,
        validators=[vd.is_frame]  # validates that frame is between 0 and 9999
    )
    channels = ListType(
        IntType(),
        required=True,
        validators=[lambda x: vd.is_in(x, [1, 3])]  # validates that png is 1 or 3 channel
    )
    extension = ListType(
        StringType(),
        required=True,
        validators=[
            vd.is_extension,
            lambda x: vd.is_eq(x, 'png')  # validates that image is png
        ]
    )
    file_traits = dict(
        width=tr.get_image_width,            # retrieves image width from file
        height=tr.get_image_height,          # retrieves image height from file
        channels=tr.get_num_image_channels,  # retrieves image channel number from file
    )