All good! going to merge now.

Fixed various code problems, going to merge soon.

.env

@@ -75,3 +75,8 @@ TTS_HIFIGAN_SUPERRES_VOCODER_MODEL_FILENAME=hifigan_superres_twilight_33000.pt
 
 # Development key. Permissible to check in.
 NEWRELIC_API_KEY=c402be1c4c4216ec83d0cfc0851520cfFFFFNRAL
+
+STORYTELLER_ROOT = "/home/tensor/code/storyteller" # your path here!
+STORYTELLER_FRONTEND = "storyteller-frontend"
+STORYTELLER_ML = "storyteller-ml"
+STORYTELLER_RUST = "storyteller-rust"

.gitattributes

@@ -1,5 +1,6 @@
 # Ignore certain code for Github language statistics
 # https://github.com/github/linguist/blob/master/docs/overrides.md
 old_code/trumpet-trumped.com/dictionary/cmudict-0.7b linguist-vendored
-old_code/voder-vocodes-1.0/frontends/admin/node_modules linguist-vendored
-windows/sqlite linguist-vendored
+old_code/voder-vocodes-1.0/frontends/admin/node_modules/** linguist-vendored
+test_data/html_scraping/** linguist-vendored
+windows/sqlite/** linguist-vendored

.github/dependabot.yml

@@ -0,0 +1,11 @@
+# To get started with Dependabot version updates, you'll need to specify which
+# package ecosystems to update and where the package manifests are located.
+# Please see the documentation for all configuration options:
+# https://docs.github.com/github/administering-a-repository/configuration-options-for-dependency-updates
+
+version: 2
+updates:
+  - package-ecosystem: "cargo" # See documentation for possible values
+    directory: "/" # Location of package manifests
+    schedule:
+      interval: "weekly"

.idea/dictionaries/bt.xml

@@ -3,6 +3,8 @@
     <words>
       <w>arpabet</w>
       <w>backfill</w>
+      <w>canonicalized</w>
+      <w>canonicalizing</w>
       <w>cashapp</w>
       <w>cheermote</w>
       <w>cidrs</w>
@@ -11,7 +13,9 @@
       <w>dalle</w>
       <w>demuxer</w>
       <w>dereferencing</w>
+      <w>downloaders</w>
       <w>downvote</w>
+      <w>encodec</w>
       <w>fakeyou</w>
       <w>ffprobe</w>
       <w>greenlight</w>
@@ -22,6 +26,8 @@
       <w>hparams</w>
       <w>huggingface</w>
       <w>lipsync</w>
+      <w>mochi</w>
+      <w>nestedness</w>
       <w>pathbuf</w>
       <w>replayable</w>
       <w>retryably</w>

Cargo.toml

@@ -11,28 +11,32 @@ members = [
   # CLI tools for operations, etc.
   "crates/cli/database_backup",
   "crates/cli/database_cleanup",
+  "crates/cli/dev_database_seed",
+  "crates/cli/dev_upload_media_file",
+  "crates/cli/kubectl_pod_cleanup",
   "crates/cli/scrape_test",
 
   # Shared libraries
   "crates/lib/actix_helpers",
   "crates/lib/api_clients/cloud_storage",
-  #"crates/lib/api_clients/fakeyou_client",
   "crates/lib/api_clients/google_drive_common",
   "crates/lib/api_clients/twitch_common",
   "crates/lib/bootstrap",
   "crates/lib/caching/memory_caching",
   "crates/lib/caching/redis_caching",
   "crates/lib/concurrency",
+  "crates/lib/cookies",
   "crates/lib/crockford",
   "crates/lib/datetimes",
   "crates/lib/deprecated/config", # TODO(bt, 2023-02-13): Deprecate or repurpose
   "crates/lib/deprecated/container_common", # TODO(bt, 2023-02-13): Pull apart and rehome pieces
   "crates/lib/deprecated/http_server_common", # TODO(bt, 2023-03-20): Move useful pieces to actix_helpers
-  "crates/lib/envvar",
+  "crates/lib/easyenv", # NB: Vendored version of https://github.com/echelon/easyenv.rs that we'll improve.
   "crates/lib/errors",
   "crates/lib/files/archives",
   "crates/lib/files/filesys",
   "crates/lib/files/hashing",
+  "crates/lib/files/images",
   "crates/lib/files/media",
   "crates/lib/files/mimetypes",
   "crates/lib/idempotency",
@@ -40,13 +44,16 @@ members = [
   "crates/lib/mysql_queries",
   "crates/lib/redis_common",
   "crates/lib/server_environment",
-  #"crates/lib/sqlite_queries",
   "crates/lib/sqlx_mysql_helpers",
+  "crates/lib/storyteller_root",
   "crates/lib/subprocess_common",
+  "crates/lib/testing",
   "crates/lib/tts_common",
   "crates/lib/url_config",
   "crates/lib/user_input_common",
   "crates/lib/web_scrapers",
+  #"crates/lib/api_clients/fakeyou_client",
+  #"crates/lib/sqlite_queries",
 
   # Programs: Desktop apps
   #"crates/programs/aichatbot-sidecar",

Cranky.toml

@@ -39,6 +39,9 @@ deny = [
   "clippy::redundant_field_names", # Lint: Duplicate names is a burden when reading: you wonder why they're dupes.
   "clippy::redundant_static_lifetimes", # Lint: feels more modern to elide static lifetimes
 
+  # Rust
+  "unused_imports", # These are easy to automatically cull
+
   # Possible:
   # "clippy::ptr_arg",
   # "clippy::too_many_lines" (note: this is configurable)

_develop/docs/local_development/general_setup.md

@@ -1,6 +1,14 @@
 FakeYou Development Setup (bare metal)
 ======================================
 
+Set your Environment .env File with the following paths, copy this in
+The root path and the folders associated with the path.
+This will change the base paths for development.
+STORYTELLER_ROOT = "/home/tensor/code/storyteller" # change this
+STORYTELLER_FRONTEND = "storyteller-frontend" 
+STORYTELLER_ML = "storyteller-ml"
+STORYTELLER_RUST = "storyteller-rust"
+
 This document describes how to run the FakeYou infrastructure on your development 
 machine's bare metal (no containers).
 
@@ -32,8 +40,7 @@ Mac: (Lost the list, but it should match Linux. Install Homebrew.)
 Ubuntu 22.04:
 
 ```bash
-sudo apt install \ 
-  jq \
+sudo apt install jq \
   libmysqlclient-dev \
   libsqlite3-dev \
   pkgconf
@@ -109,7 +116,9 @@ error (see below).
 ### (7) (optional) Install the SQLx CLI (if doing database development)
 
 ```bash
-cargo install sqlx-cli --no-default-features --features rustls,mysql,sqlite
+# NB(bt,2023-10-04): "--no-default-features" not working
+# cargo install sqlx-cli --no-default-features --features rustls,mysql,sqlite
+cargo install sqlx-cli --features rustls,mysql,sqlite
 ```
 
 We'll be using diesel to manage the migrations, but sqlx within the app to actually perform queries.

_sql/actions/generic_inference_jobs.sql

@@ -10,4 +10,9 @@ select count(*) from generic_inference_jobs where status = 'pending';
 
 update generic_inference_jobs
 set status = 'dead'
-where status IN ('pending', 'started', 'attempt_failed');
+where status IN ('pending', 'started', 'attempt_failed');
+
+update generic_inference_jobs
+set status = 'dead'
+where status IN ('pending', 'started', 'attempt_failed')
+and maybe_model_type IN ('sad_talker', 'so_vits_svc');

_sql/actions/voice_conversion_models.sql

@@ -302,7 +302,10 @@ where token IN (
     'vcm_p7m9jw1krrkr',
     'vcm_x7v3m9mcvten',
     'vcm_6r057m7594bv',
-    'vcm_wkm9e42d3spm'
+    'vcm_wkm9e42d3spm',
+    'vcm_cvz78weydh57',
+    'vcm_7g5q89m75n1p',
+    'vcm_fg3175bxy7kj'
 );
 
 

_sql/analytics/generic_inference/generic_inference_jobs.sql

@@ -2,6 +2,14 @@
 -- noinspection SqlNoDataSourceInspectionForFile
 -- noinspection SqlResolveForFile
 
+-- See how many jobs of each time were created in the last period
+select
+    distinct maybe_model_type as model_type,
+             count(*) as created
+from generic_inference_jobs
+where created_at > (CURDATE() - INTERVAL 60 MINUTE)
+group by maybe_model_type;
+
 -- See how long recent jobs are taking
 select
     id,
@@ -15,6 +23,19 @@ and success_execution_millis IS NOT NULL
 order by id desc
 limit 50;
 
+-- See who is uploading SadTalker jobs
+select
+    maybe_creator_user_token,
+    users.username as maybe_creator_username,
+    creator_ip_address,
+    count(*) as attempts
+from generic_inference_jobs
+left outer join users
+on users.token = generic_inference_jobs.maybe_creator_user_token
+where maybe_model_type = 'sad_talker'
+and status IN ('pending', 'started', 'complete_failure', 'attempt_failed')
+group by maybe_creator_user_token, maybe_creator_username, creator_ip_address
+order by attempts desc;
 
 -- Detailed report on most recent jobs, ordered by worst performing.
 -- TODO: Determine if the problem is in downloading models. Make sure the job timer doesn't include
@@ -49,7 +70,8 @@ from (
                  m.token = jobs.maybe_input_source_token
          where
              jobs.status != 'pending'
-        AND jobs.created_at > NOW() - INTERVAL 20 MINUTE
+         AND jobs.created_at > NOW() - INTERVAL 20 MINUTE
+         AND jobs.maybe_model_type IN ('so_vits_svc')
          order by id desc
              limit 200
      ) as t

_sql/analytics/tts/tts_inference_jobs.sql

@@ -35,7 +35,7 @@ order by attempts desc
 select *
 from tts_inference_jobs
 where status = 'pending'
-and creator_ip_address = '186.188.28.73'
+and creator_ip_address = '191.102.120.182'
 limit 1;
 
 -- Top voices in requests

_sql/migrations/2023-05-23-090105_password_resets/up.sql

@@ -2,6 +2,20 @@
 -- noinspection SqlNoDataSourceInspectionForFile
 -- noinspection SqlResolveForFile
 
+-- Per-user password resets.
+-- Password resets are valid until any of the following are true:
+--
+--   1) database NOW() > password_rests.expires_at
+--   2) password_rests.is_redeemed == TRUE
+--   3) users.password_version > password_resets.current_password_version
+--
+-- When creating a new password reset, send the user an email with the `secret_key` in the URL.
+-- The user will have to supply their username and/or email address, plus the secret key to
+-- redeem the password reset.
+--
+-- Also when creating the password hash, be sure to set `expires_at` to be "NOW() + 3 hours", or
+-- some other small, reasonable time frame.
+--
 CREATE TABLE password_resets (
   -- Not used for anything except replication.
   id BIGINT(20) NOT NULL AUTO_INCREMENT,
@@ -22,7 +36,7 @@ CREATE TABLE password_resets (
   -- reset is no longer valid.
   current_password_version INT NOT NULL DEFAULT 0,
 
-  -- Username is a lookup key; display_name allows the user to add custom case.
+  -- Whether the password reset has been consumed.
   is_redeemed BOOLEAN NOT NULL DEFAULT false,
 
   -- For abuse tracking.

_sql/migrations/2023-09-26-070239_zs_voices/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+drop table zs_voices;

_sql/migrations/2023-09-26-070239_zs_voices/up.sql

@@ -0,0 +1,154 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- This table contains both zero shot TTS voices (speaker embeddings) and zero shot V2V voices (speaker embeddings).
+-- These are the zero shot voices that are usable by our users.
+CREATE TABLE zs_voices (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" for joins, API, etc. (PUBLIC)
+  token VARCHAR(32) NOT NULL,
+
+  -- NB: No idempotency token.
+  -- Store it in another table to prevent duplicate record creation from rapid user inputs.
+
+  -- ========== ZERO SHOT FOREIGN KEYS AND STATE==========
+
+  -- The zero shot dataset that was used to create this voice.
+  -- The dataset will only be accessible to the original author of the voice.
+  -- The original author can create multiple voices from the same dataset.
+  dataset_token VARCHAR(32) NOT NULL,
+
+  -- When datasets are edited, we bump their version.
+  -- This tells us which version of the dataset created this voice (not that we should preserve
+  -- record of which samples are in which version.)
+  -- We can surface this to the users as well.
+  dataset_version INT NOT NULL DEFAULT 0,
+
+  -- ========== MODEL DETAILS ==========
+
+  -- The class or category of model/algorithm this voice is intended for.
+  --   * 'tts' for zero shot TTS
+  --   * 'vc' for zero shot voice conversion
+  --   * 'pitch' for pitch correction models (hypothetical)
+  --   * (Since this isn't an enum, the list of values can be expanded.)
+  model_category VARCHAR(16) NOT NULL,
+
+  -- The model used for the voice:
+  --   * 'vall-e-x' for Vall-E-X
+  --   * 'ns2vc' for NaturalSpeech2 voice conversion (hypothetical)
+  --   * (Since this isn't an enum, the list of values can be expanded.)
+  model_type VARCHAR(16) NOT NULL,
+
+  -- The version of the model.
+  -- If we deploy new model weights for the model, we may need to regenerate
+  -- all of the corresponding speaker embeddings. This lets us know which voices
+  -- are the old version.
+  -- In some cases, we may need to preserve, hide, or roll back to older versions.
+  model_version INTEGER UNSIGNED NOT NULL DEFAULT 0,
+
+  -- What kind of features these are: spectrogram, encodec, etc.
+  -- Technically we shouldn't need this, as the 2-tuple (model_type, model_version) implicitly knows
+  -- what its features should be, but this is probably useful for our own analysis and may be easier
+  -- to deal with.
+  --
+  -- Example values:
+  --   * 'spectrogram'
+  --   * 'encodec'
+  --   * (Since this isn't an enum, the list of values can be expanded.)
+  encoding_type VARCHAR(16) NOT NULL,
+
+  -- ========== VOICE DETAILS ==========
+
+  -- The "title" of the voice. This can be the name and additional information as supplied by the user.
+  title VARCHAR(255) NOT NULL,
+
+  -- The full IETF BCP47 language tag (eg. en, en-US, es-419, ja-JP, pt, etc.)
+  -- The wide varchar space is because these language tags can become cumbersome, eg "hy-Latn-IT-arevela"
+  ietf_language_tag VARCHAR(64) NOT NULL DEFAULT 'en',
+
+  -- The IETF BCP47 language tag's primary language subtag (eg. "es-419" becomes "es" and "en-US" becomes "en")
+  ietf_primary_language_subtag VARCHAR(12) NOT NULL DEFAULT 'en',
+
+  -- TODO: misc speaker metadata
+  -- We'll want rich metadata for search
+  -- series_name VARCHAR(255) DEFAULT NULL,
+  -- speaker_name VARCHAR(255) DEFAULT NULL,
+  -- speaker_gender VARCHAR(255) DEFAULT NULL,
+  -- speaker_age VARCHAR(255) DEFAULT NULL,
+  -- speaker_voice_style VARCHAR(255) DEFAULT NULL,
+
+  -- ========== STORAGE DETAILS ==========
+
+  -- The hash for the bucket directory that contains the original upload
+  -- eg. /speaker_vectors/{bucket_hash}/{voice_model_version}.bin
+  bucket_hash  VARCHAR(32) NOT NULL,
+
+  -- ========== CREATOR DETAILS AND PREFERENCES ==========
+
+  -- Foreign key to user
+  maybe_creator_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- Based on a cookie sent by the frontend.
+  -- We'll save this even if the user is logged in.
+  maybe_creator_anonymous_visitor_token VARCHAR(32) DEFAULT NULL,
+
+  -- For abuse tracking.
+  -- Wide enough for IPv4/6
+  creator_ip_address VARCHAR(40) NOT NULL,
+
+  -- The creator can set a desired visibility for their data.
+  -- This does not always apply to every upload type.
+  -- Additionally, some upload types may require moderator approval prior
+  -- to being publicly listed, and this field has no bearing on that.
+  -- NB: DO NOT CHANGE THE ORDER OF THE ENUM VALUES DURING SCHEMA MIGRATIONS.
+  creator_set_visibility ENUM(
+    'public',
+    'hidden',
+    'private'
+  ) NOT NULL DEFAULT 'public',
+
+  -- The synthetic id associated with this result.
+  -- These ids are incremented on a per-user basis to help users
+  -- sequence their own work. They serve no other purpose.
+  maybe_creator_synthetic_id BIGINT(20) DEFAULT NULL,
+
+  -- ========== MODERATION DETAILS ==========
+
+  -- The last moderator that made changes last.
+  -- Fine details of moderator actions are kept in a separate `audit_logs` table.
+  maybe_mod_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- If this is removed by a mod or the creator.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+
+  -- TODO: These are not correct. Fix these before landing / running migrations.
+  KEY index_model_category (model_category),
+  KEY index_model_type (model_type),
+  KEY index_model_type_and_version (model_type, model_version),
+  KEY index_encoding_type (encoding_type),
+  KEY index_ietf_primary_language_subtag (ietf_primary_language_subtag),
+  KEY fk_maybe_creator_user_token (maybe_creator_user_token),
+  KEY fk_maybe_mod_user_token (maybe_mod_user_token),
+  KEY index_creator_ip_address (creator_ip_address),
+  KEY index_user_deleted_at (user_deleted_at),
+  KEY index_mod_deleted_at (mod_deleted_at)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations/2023-09-26-070244_zs_voice_datasets/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+drop table zs_voice_datasets;

_sql/migrations/2023-09-26-070244_zs_voice_datasets/up.sql

@@ -0,0 +1,106 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- This table describes a zero shot voice in the process of being created (or later reused / edited / enhanced)
+-- These are not usable voices until their vector embeddings have been created, and this table does not serve
+-- the production voice list.
+--
+-- Multiple voices can be created from a single "dataset" as the data set is mutable over time. Users can add or
+-- remove new samples to the data set at any time.
+CREATE TABLE zs_voice_datasets (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" for joins, API, etc. (PUBLIC)
+  token VARCHAR(32) NOT NULL,
+
+  -- NB: No idempotency token.
+  -- Store it in another table to prevent duplicate record creation from rapid user inputs.
+
+  -- ========== VOICE DETAILS ==========
+
+  -- The "title" of the voice. This can be the name and additional information as supplied by the user.
+  -- When the voice embeddings are created, this will be copied to the produced "voice" (embeddings) record.
+  title VARCHAR(255) NOT NULL,
+
+  -- The full IETF BCP47 language tag (eg. en, en-US, es-419, ja-JP, pt, etc.)
+  -- The wide varchar space is because these language tags can become cumbersome, eg "hy-Latn-IT-arevela"
+  ietf_language_tag VARCHAR(64) NOT NULL DEFAULT 'en',
+
+  -- The IETF BCP47 language tag's primary language subtag (eg. "es-419" becomes "es" and "en-US" becomes "en")
+  ietf_primary_language_subtag VARCHAR(12) NOT NULL DEFAULT 'en',
+
+  -- TODO: misc speaker metadata
+  -- We'll want rich metadata for search
+  -- When the voice embeddings are created, this will be copied to the produced "voice" (embeddings) record.
+  -- We don't want a hard link between the dataset and the produced voice as the voice is static once
+  -- created (with the exception of regenerating embeddings), but the dataset my drift as the user experiments,
+  -- adds/removes samples, etc.
+  -- series_name VARCHAR(255) DEFAULT NULL,
+  -- speaker_name VARCHAR(255) DEFAULT NULL,
+  -- speaker_gender VARCHAR(255) DEFAULT NULL,
+  -- speaker_age VARCHAR(255) DEFAULT NULL,
+  -- speaker_voice_style VARCHAR(255) DEFAULT NULL,
+
+  -- ========== CREATOR DETAILS AND PREFERENCES ==========
+
+  -- Foreign key to user
+  maybe_creator_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- Based on a cookie sent by the frontend.
+  -- We'll save this even if the user is logged in.
+  maybe_creator_anonymous_visitor_token VARCHAR(32) DEFAULT NULL,
+
+  -- For abuse tracking.
+  -- Wide enough for IPv4/6
+  creator_ip_address VARCHAR(40) NOT NULL,
+
+  -- The creator can set a desired visibility for their data.
+  -- This does not always apply to every upload type.
+  -- Additionally, some upload types may require moderator approval prior
+  -- to being publicly listed, and this field has no bearing on that.
+  -- NB: DO NOT CHANGE THE ORDER OF THE ENUM VALUES DURING SCHEMA MIGRATIONS.
+  creator_set_visibility ENUM(
+    'public',
+    'hidden',
+    'private'
+  ) NOT NULL DEFAULT 'public',
+
+  -- The synthetic id associated with this result.
+  -- These ids are incremented on a per-user basis to help users
+  -- sequence their own work. They serve no other purpose.
+  maybe_creator_synthetic_id BIGINT(20) DEFAULT NULL,
+
+  -- ========== MODERATION DETAILS ==========
+
+  -- The last moderator that made changes last.
+  -- Fine details of moderator actions are kept in a separate `audit_logs` table.
+  maybe_mod_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- If this is removed by a mod or the creator.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+
+  KEY index_ietf_primary_language_subtag (ietf_primary_language_subtag),
+  KEY fk_maybe_creator_user_token (maybe_creator_user_token),
+  KEY fk_maybe_mod_user_token (maybe_mod_user_token),
+  KEY index_creator_ip_address (creator_ip_address),
+  KEY index_user_deleted_at (user_deleted_at),
+  KEY index_mod_deleted_at (mod_deleted_at)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations/2023-09-26-070252_zs_voice_dataset_samples/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+drop table zs_voice_dataset_samples;

_sql/migrations/2023-09-26-070252_zs_voice_dataset_samples/up.sql

@@ -0,0 +1,84 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- These are the individual files in each dataset.
+-- Files can be added and removed from a dataset at any time.
+CREATE TABLE zs_voice_dataset_samples (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" for joins, API, etc. (PUBLIC)
+  token VARCHAR(32) NOT NULL,
+
+  -- NB: No idempotency token.
+  -- Store it in another table to prevent duplicate record creation from rapid user inputs.
+
+  -- ========== ZERO SHOT FOREIGN KEYS AND STATE==========
+
+  -- The zero shot dataset that was used to create this voice.
+  -- The dataset will only be accessible to the original author of the voice.
+  -- The original author can create multiple voices from the same dataset.
+  dataset_token VARCHAR(32) NOT NULL,
+
+  -- Foreign key to the uploaded media file.
+  -- There was an option of handling uploads and file management separately, but by electing to store the dataset
+  -- samples in the "media_files" table, they can be reused by our other code, queries, machinery, etc.
+  -- Note that this requires that the "media_files" record must be an audio (not video) file (which our code will
+  -- enforce), and that it is given a category of "voice_zero_shot" so it isn't co-mingled with other uploads/outputs
+  media_file_token VARCHAR(32) NOT NULL,
+
+  -- Some data sets require raw transcripts for the audio.
+  -- If required, they'll live here.
+  maybe_raw_transcription TEXT DEFAULT NULL,
+
+  -- A way to order the data set samples
+  -- We don't necessarily need to reimplement file reordering.
+  -- order_id INT DEFAULT NULL,
+
+  -- ========== CREATOR DETAILS AND PREFERENCES ==========
+
+  -- Foreign key to user
+  maybe_creator_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- Based on a cookie sent by the frontend.
+  -- We'll save this even if the user is logged in.
+  maybe_creator_anonymous_visitor_token VARCHAR(32) DEFAULT NULL,
+
+  -- For abuse tracking.
+  -- Wide enough for IPv4/6
+  creator_ip_address VARCHAR(40) NOT NULL,
+
+  -- ========== MODERATION DETAILS ==========
+
+  -- The last moderator that made changes last.
+  -- Fine details of moderator actions are kept in a separate `audit_logs` table.
+  maybe_mod_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- If this is removed by a mod or the creator.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+
+  KEY fk_dataset_token (dataset_token),
+  KEY fk_media_file_token (media_file_token),
+  KEY fk_maybe_creator_user_token (maybe_creator_user_token),
+  KEY fk_maybe_mod_user_token (maybe_mod_user_token),
+  KEY index_creator_ip_address (creator_ip_address),
+  KEY index_user_deleted_at (user_deleted_at),
+  KEY index_mod_deleted_at (mod_deleted_at)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations/2023-09-29-033922_zs_voice_included_samples/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+drop table zs_voice_included_samples;

_sql/migrations/2023-09-29-033922_zs_voice_included_samples/up.sql

@@ -0,0 +1,26 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- In order to rebuild any "zs_voice", we need to know which dataset samples were included.
+-- Since users can change the dataset at any time after a voice was originally created, these
+-- records are generated to retain a record of which samples fed into a given voice.
+-- These records will probably not be used for any product features, just as a means to study
+-- poor-performing voices, perform migrations, etc.
+CREATE TABLE zs_voice_included_samples (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- First member of composite effective "primary key"
+  -- The "zs_voice" that was created.
+  voice_token VARCHAR(32) NOT NULL,
+
+  -- Second member of composite effective "primary key"
+  -- The "zs_voice_dataset_sample" that was used.
+  dataset_sample_token VARCHAR(32) NOT NULL,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (voice_token, dataset_sample_token)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations/2023-10-01-212745_alter_media_files_add_bucket_extension/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE media_files DROP COLUMN maybe_public_bucket_extension;

_sql/migrations/2023-10-01-212745_alter_media_files_add_bucket_extension/up.sql

@@ -0,0 +1,7 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE media_files
+    ADD COLUMN maybe_public_bucket_extension VARCHAR(16) DEFAULT NULL
+    AFTER public_bucket_directory_hash;

_sql/migrations/2023-10-04-062503_alter_media_files_add_bucket_prefix/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE media_files DROP COLUMN maybe_public_bucket_prefix;

_sql/migrations/2023-10-04-062503_alter_media_files_add_bucket_prefix/up.sql

@@ -0,0 +1,7 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE media_files
+    ADD COLUMN maybe_public_bucket_prefix VARCHAR(16) DEFAULT NULL
+    AFTER public_bucket_directory_hash;

_sql/migrations/2023-10-11-235742_alter_media_files_add_origin_product_category/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE media_files DROP COLUMN origin_product_category;

_sql/migrations/2023-10-11-235742_alter_media_files_add_origin_product_category/up.sql

@@ -0,0 +1,10 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE media_files
+    ADD COLUMN origin_product_category VARCHAR(16) NOT NULL DEFAULT "unknown"
+    AFTER origin_category;
+
+ALTER TABLE media_files
+    ADD INDEX index_origin_product_category (origin_product_category);

_sql/migrations/2023-10-11-235800_alter_generic_inference_jobs_add_frontend_failure_category/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs DROP COLUMN frontend_failure_category;

_sql/migrations/2023-10-11-235800_alter_generic_inference_jobs_add_frontend_failure_category/up.sql

@@ -0,0 +1,7 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN frontend_failure_category VARCHAR(32) DEFAULT NULL
+    AFTER internal_debugging_failure_reason;

_sql/migrations/2023-10-13-093804_alter_zs_voice_dataset_samples_add_idempotency/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE zs_voice_dataset_samples DROP COLUMN uuid_idempotency_token;

_sql/migrations/2023-10-13-093804_alter_zs_voice_dataset_samples_add_idempotency/up.sql

@@ -0,0 +1,10 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE zs_voice_dataset_samples
+    ADD COLUMN uuid_idempotency_token VARCHAR(36) NOT NULL
+    AFTER token;
+
+ALTER TABLE zs_voice_dataset_samples
+    ADD UNIQUE INDEX unique_index_uuid_idempotency_token (uuid_idempotency_token);

_sql/migrations/2023-10-18-061204_alter_password_resets_add_unique_index/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE password_resets DROP INDEX unique_user_token_and_secret_key;

_sql/migrations/2023-10-18-061204_alter_password_resets_add_unique_index/up.sql

@@ -0,0 +1,6 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE password_resets
+    ADD UNIQUE INDEX unique_user_token_and_secret_key (user_token, secret_key);

_sql/migrations/2023-10-24-020831_alter_password_resets_ip_address_redemption_nullable/down.sql

@@ -0,0 +1,6 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE password_resets
+    MODIFY ip_address_redemption VARCHAR(40) NOT NULL;

_sql/migrations/2023-10-24-020831_alter_password_resets_ip_address_redemption_nullable/up.sql

@@ -0,0 +1,6 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE password_resets
+    MODIFY ip_address_redemption VARCHAR(40) DEFAULT NULL;

_sql/migrations/2023-10-24-044500_alter_generic_inference_jobs_add_status_values/down.sql

@@ -0,0 +1,13 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs
+    MODIFY status ENUM(
+        'pending',
+        'started',
+        'complete_success',
+        'complete_failure',
+        'attempt_failed',
+        'dead'
+  ) NOT NULL DEFAULT 'pending';

_sql/migrations/2023-10-24-044500_alter_generic_inference_jobs_add_status_values/up.sql

@@ -0,0 +1,16 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Add two states to allow us to distinguish user-cancelled jobs and system-cancelled jobs.
+ALTER TABLE generic_inference_jobs
+    MODIFY status ENUM(
+        'pending',
+        'started',
+        'complete_success',
+        'complete_failure',
+        'attempt_failed',
+        'dead',
+        'cancelled_by_user',
+        'cancelled_by_system'
+) NOT NULL DEFAULT 'pending';

_sql/migrations_squashed/generic_inference_jobs.sql

@@ -123,11 +123,19 @@ CREATE TABLE generic_inference_jobs (
   -- ========== DEVELOPMENT AND DEBUGGING METADATA ==========
 
   -- If true, the request gets a "debug" flag, which may do different
-  -- things depending on the type of work.
+  -- things depending on the type of work. This doesn't really have any
+  -- specific meaning -- it can be used to flag for anything -- but should
+  -- typically be used for one-off debugging and not a part of the normal
+  -- product surface area. Contrast this with `maybe_routing_tag`, which
+  -- forces the job onto a specific physical server: this might be a flag
+  -- deployed to handle debugging across the entire fleet.
   is_debug_request BOOLEAN NOT NULL DEFAULT FALSE,
 
   -- If set, the request gets processed by a special "tagged" worker
   -- that matches this tag. The ordinary workers will ignore this work.
+  -- The tag is typically added via a special HTTP header when the work
+  -- is enqueued. In practice, this means we can canary deploy or cordon
+  -- off special workers to handle certain jobs.
   maybe_routing_tag VARCHAR(32) DEFAULT NULL,
 
   -- ========== JOB SYSTEM DETAILS ==========
@@ -137,9 +145,11 @@ CREATE TABLE generic_inference_jobs (
   --  * Pending = job is ready to go
   --  * Started = job is running
   --  * Complete_Success = job is done (success)
-  --  * Complete_Failure = job is done (failure)
+  --  * Complete_Failure = job is done (permanent failure, eg. a Face Animator job without a face)
   --  * Attempt_Failed = job failed but may retry.
-  --  * Dead = job failed permanently.
+  --  * Dead = job failed permanently (ie. exhausted retries)
+  --  * Cancelled_By_User = the user canceled their own job
+  --  * Cancelled_By_System = the system terminated the job (ie. load shedding, user account deletion, etc.)
   --
   -- Pending -> Started -> Complete_Success
   --                    |-> Complete_Failure
@@ -150,17 +160,41 @@ CREATE TABLE generic_inference_jobs (
     'complete_success',
     'complete_failure',
     'attempt_failed',
-    'dead') NOT NULL DEFAULT 'pending',
+    'dead',
+    'cancelled_by_user',
+    'cancelled_by_system'
+  ) NOT NULL DEFAULT 'pending',
 
   -- We can track this against a "max_attempt_count"
   attempt_count SMALLINT UNSIGNED NOT NULL DEFAULT 0,
 
   -- If there is a failure, tell the user why.
+  -- This isn't localized or very friendly to present to the user.
   failure_reason VARCHAR(512) DEFAULT NULL,
 
   -- Optional internal-only debugging information in the case of failure.
   internal_debugging_failure_reason VARCHAR(512) DEFAULT NULL,
 
+  -- An enum-like key to present the frontend with a failure class in an
+  -- i18n-friendly way. The javascript/frontend can use these as indices
+  -- into descriptive, broadly-localized error messages.
+  --
+  --   * 'face_not_detected' for images or videos that do not have a detectable
+  --                         face (SadTalker, Wav2Lip, etc.)
+  --
+  --   * 'keep_alive_elapsed' when users walk away from their inference job and
+  --                          the workload class expects for them to remain
+  --
+  --   * 'not_yet_implemented' error for developers during feature development
+  --                           and prototyping. Should ideally never happen in
+  --                           production
+  --
+  --   * 'retryable_worker_error' for generic worker errors (full filesystem,
+  --                              etc.) that can be retried. (We don't need to
+  --                              tell the user why the job failed.)
+  --
+  frontend_failure_category VARCHAR(32) DEFAULT NULL,
+
   -- On success, we populate how long the job execution took in milliseconds.
   -- We have this because 1) the job system timestamps are second-resolution,
   -- and 2) the structure of the jobs executor may change, potentially changing

_sql/migrations_squashed/media_files.sql

@@ -19,12 +19,28 @@ CREATE TABLE media_files (
 
   -- ========== FOREIGN KEY TO ORIGIN ==========
 
-  -- Where the file came from:
+  -- Broad category for where the file came from:
   --   * 'inference' for inference output
-  --   * 'processed' for processed file
-  --   * 'upload' for direct user upload
+  --   * 'processed' for processed file (eg. mp3 encoding, stem splitting, etc.)
+  --   * 'upload' for direct user upload (from the filesystem)
+  --   * 'device_api' for direct user uploads recorded using Browser/Device APIs.
   origin_category VARCHAR(16) NOT NULL,
 
+  -- TODO: Remove default value once records backfilled and queries updated.
+  -- Product area where the media file originated.
+  -- This is not the *model* that created the thing, this is the *product*.
+  -- (The underlying models can change over time.)
+  --
+  -- This value indicates what product originally created the media file. (Not the ML model or
+  -- user upload process.) This will let us scope media files to the product that generated them
+  -- and filter them out of unrelated products if necessary (eg. a user probably doesn't want
+  -- "Voice Designer" dataset samples in a video generation flow.)
+  --
+  -- Possible values:
+  --   * 'face_animator' for uploads and outputs
+  --   * 'unknown' for legacy records without an associated product (TODO: This should be temporary)
+  origin_product_category VARCHAR(16) NOT NULL DEFAULT "unknown",
+
   -- For inference that can be tied back to a model, the type of model.
   --   * 'sad_talker', which doesn't have a value for maybe_origin_model_token (!!!)
   --   * (more tome come)
@@ -88,8 +104,24 @@ CREATE TABLE media_files (
 
   -- The hash for the bucket directory that contains the original upload
   -- as well as any transcodings, downsamplings, etc.
+  -- The bucket filename for the primary file (not including the directory path) is given by:
+  -- `[{maybe_public_bucket_prefix}]{public_bucket_directory_hash}[{maybe_public_bucket_extension}]`
   public_bucket_directory_hash  VARCHAR(32) NOT NULL,
 
+  -- An optional prefix on the bucket filename.
+  -- If present, this will be prepended to the beginning of the bucket filename to access the file.
+  -- The bucket filename for the primary file (not including the directory path) is given by:
+  -- `[{maybe_public_bucket_prefix}]{public_bucket_directory_hash}[{maybe_public_bucket_extension}]`
+  maybe_public_bucket_prefix VARCHAR(16) DEFAULT NULL,
+
+  -- An optional appended extension on the bucket filename.
+  -- If present, this will be appended to the end of the bucket filename to access the file.
+  -- To allow for flexibility, this extension typically will contain a leading period if
+  -- the file needs it (eg ".mp4" rather than "mp4")!
+  -- The bucket filename for the primary file (not including the directory path) is given by:
+  -- `[{maybe_public_bucket_prefix}]{public_bucket_directory_hash}[{maybe_public_bucket_extension}]`
+  maybe_public_bucket_extension VARCHAR(16) DEFAULT NULL,
+
   -- NB: Removed, since this can be derived.
   -- The directory this media is uploaded to will be exclusive for this file.
   -- Only this given record will live in this bucket, but the directory may include

_sql/migrations_squashed/media_uploads.sql

@@ -98,7 +98,7 @@ CREATE TABLE media_uploads (
     'public',
     'hidden',
     'private'
-    ) NOT NULL DEFAULT 'public',
+  ) NOT NULL DEFAULT 'public',
 
   -- The synthetic id associated with this result.
   -- These ids are incremented on a per-user basis to help users

_sql/migrations_squashed/tts_tables/tts_models.sql

@@ -83,9 +83,10 @@ CREATE TABLE tts_models (
   -- ========== UNSTRUCTURED METADATA ==========
 
   -- The full IETF BCP47 language tag (eg. en, en-US, es-419, ja-JP, pt, etc.)
+  -- The wide varchar space is because these language tags can become cumbersome, eg "hy-Latn-IT-arevela"
   ietf_language_tag VARCHAR(64) NOT NULL DEFAULT 'en',
 
-  -- The IETF BCP47 language tag's primary language subtag (eg. "es-419" becomes "es")
+  -- The IETF BCP47 language tag's primary language subtag (eg. "es-419" becomes "es" and "en-US" becomes "en")
   ietf_primary_language_subtag VARCHAR(12) NOT NULL DEFAULT 'en',
 
   -- The "name" of the voice model, which might be complicated.

_sql/migrations_squashed/user_tables/password_resets.sql

@@ -0,0 +1,68 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Per-user password resets.
+-- Password resets are valid until any of the following are true:
+--
+--   1) database NOW() > password_rests.expires_at
+--   2) password_rests.is_redeemed == TRUE
+--   3) users.password_version > password_resets.current_password_version
+--
+-- When creating a new password reset, send the user an email with the `secret_key` in the URL.
+-- The user will have to supply their username and/or email address, plus the secret key to
+-- redeem the password reset.
+--
+-- Also when creating the password hash, be sure to set `expires_at` to be "NOW() + 3 hours", or
+-- some other small, reasonable time frame.
+--
+CREATE TABLE password_resets (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Password hash "primary key"
+  -- (ie. for the future when we have a support dashboard)
+  token VARCHAR(32) NOT NULL,
+
+  -- Foreign key to user
+  user_token VARCHAR(32) NOT NULL,
+
+  -- Secret "key" for the password reset
+  -- This will be shared via URL or user text input.
+  secret_key VARCHAR(32) NOT NULL,
+
+  -- Copied from the user record at the time of password reset issuance.
+  -- If the user's password version is greater than this value, then this
+  -- reset is no longer valid.
+  current_password_version INT NOT NULL DEFAULT 0,
+
+  -- Whether the password reset has been consumed.
+  is_redeemed BOOLEAN NOT NULL DEFAULT false,
+
+  -- For abuse tracking.
+  -- Wide enough for IPv4/6
+  ip_address_creation VARCHAR(40) NOT NULL,
+  ip_address_redemption VARCHAR(40) DEFAULT NULL,
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- When the password reset expires
+  expires_at TIMESTAMP NOT NULL,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+
+  UNIQUE KEY (user_token, secret_key),
+
+  KEY fk_user_token (user_token),
+  KEY index_secret_key (secret_key),
+  KEY index_expires_at (expires_at)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/zs_voices/zs_voice_dataset_samples.sql

@@ -0,0 +1,89 @@
+-- NB: This is a manually squashed view of all the CREATE and ALTER statements,
+-- with comments attached to the fields for centralized documentation.
+
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- These are the individual files in each dataset.
+-- Files can be added and removed from a dataset at any time.
+CREATE TABLE zs_voice_dataset_samples (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" for joins, API, etc. (PUBLIC)
+  token VARCHAR(32) NOT NULL,
+
+  -- Idempotency token from client
+  -- This is so the frontend client doesn't submit duplicate uploads.
+  uuid_idempotency_token VARCHAR(36) NOT NULL,
+
+  -- ========== ZERO SHOT FOREIGN KEYS AND STATE==========
+
+  -- The zero shot dataset that was used to create this voice.
+  -- The dataset will only be accessible to the original author of the voice.
+  -- The original author can create multiple voices from the same dataset.
+  dataset_token VARCHAR(32) NOT NULL,
+
+  -- Foreign key to the uploaded media file.
+  -- There was an option of handling uploads and file management separately, but by electing to store the dataset
+  -- samples in the "media_files" table, they can be reused by our other code, queries, machinery, etc.
+  -- Note that this requires that the "media_files" record must be an audio (not video) file (which our code will
+  -- enforce), and that it is given a category of "voice_zero_shot" so it isn't co-mingled with other uploads/outputs
+  media_file_token VARCHAR(32) NOT NULL,
+
+  -- Some data sets require raw transcripts for the audio.
+  -- If required, they'll live here.
+  maybe_raw_transcription TEXT DEFAULT NULL,
+
+  -- A way to order the data set samples
+  -- We don't necessarily need to reimplement file reordering.
+  -- order_id INT DEFAULT NULL,
+
+  -- ========== CREATOR DETAILS AND PREFERENCES ==========
+
+  -- Foreign key to user
+  maybe_creator_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- Based on a cookie sent by the frontend.
+  -- We'll save this even if the user is logged in.
+  maybe_creator_anonymous_visitor_token VARCHAR(32) DEFAULT NULL,
+
+  -- For abuse tracking.
+  -- Wide enough for IPv4/6
+  creator_ip_address VARCHAR(40) NOT NULL,
+
+  -- ========== MODERATION DETAILS ==========
+
+  -- The last moderator that made changes last.
+  -- Fine details of moderator actions are kept in a separate `audit_logs` table.
+  maybe_mod_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- If this is removed by a mod or the creator.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+  UNIQUE KEY (uuid_idempotency_token),
+
+  KEY fk_dataset_token (dataset_token),
+  KEY fk_media_file_token (media_file_token),
+  KEY fk_maybe_creator_user_token (maybe_creator_user_token),
+  KEY fk_maybe_mod_user_token (maybe_mod_user_token),
+  KEY index_creator_ip_address (creator_ip_address),
+  KEY index_user_deleted_at (user_deleted_at),
+  KEY index_mod_deleted_at (mod_deleted_at)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/zs_voices/zs_voice_datasets.sql

@@ -0,0 +1,109 @@
+-- NB: This is a manually squashed view of all the CREATE and ALTER statements,
+-- with comments attached to the fields for centralized documentation.
+
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- This table describes a zero shot voice in the process of being created (or later reused / edited / enhanced)
+-- These are not usable voices until their vector embeddings have been created, and this table does not serve
+-- the production voice list.
+--
+-- Multiple voices can be created from a single "dataset" as the data set is mutable over time. Users can add or
+-- remove new samples to the data set at any time.
+CREATE TABLE zs_voice_datasets (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" for joins, API, etc. (PUBLIC)
+  token VARCHAR(32) NOT NULL,
+
+  -- NB: No idempotency token.
+  -- Store it in another table to prevent duplicate record creation from rapid user inputs.
+
+  -- ========== VOICE DETAILS ==========
+
+  -- The "title" of the voice. This can be the name and additional information as supplied by the user.
+  -- When the voice embeddings are created, this will be copied to the produced "voice" (embeddings) record.
+  title VARCHAR(255) NOT NULL,
+
+  -- The full IETF BCP47 language tag (eg. en, en-US, es-419, ja-JP, pt, etc.)
+  -- The wide varchar space is because these language tags can become cumbersome, eg "hy-Latn-IT-arevela"
+  ietf_language_tag VARCHAR(64) NOT NULL DEFAULT 'en',
+
+  -- The IETF BCP47 language tag's primary language subtag (eg. "es-419" becomes "es" and "en-US" becomes "en")
+  ietf_primary_language_subtag VARCHAR(12) NOT NULL DEFAULT 'en',
+
+  -- TODO: misc speaker metadata
+  -- We'll want rich metadata for search
+  -- When the voice embeddings are created, this will be copied to the produced "voice" (embeddings) record.
+  -- We don't want a hard link between the dataset and the produced voice as the voice is static once
+  -- created (with the exception of regenerating embeddings), but the dataset my drift as the user experiments,
+  -- adds/removes samples, etc.
+  -- series_name VARCHAR(255) DEFAULT NULL,
+  -- speaker_name VARCHAR(255) DEFAULT NULL,
+  -- speaker_gender VARCHAR(255) DEFAULT NULL,
+  -- speaker_age VARCHAR(255) DEFAULT NULL,
+  -- speaker_voice_style VARCHAR(255) DEFAULT NULL,
+
+  -- ========== CREATOR DETAILS AND PREFERENCES ==========
+
+  -- Foreign key to user
+  maybe_creator_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- Based on a cookie sent by the frontend.
+  -- We'll save this even if the user is logged in.
+  maybe_creator_anonymous_visitor_token VARCHAR(32) DEFAULT NULL,
+
+  -- For abuse tracking.
+  -- Wide enough for IPv4/6
+  creator_ip_address VARCHAR(40) NOT NULL,
+
+  -- The creator can set a desired visibility for their data.
+  -- This does not always apply to every upload type.
+  -- Additionally, some upload types may require moderator approval prior
+  -- to being publicly listed, and this field has no bearing on that.
+  -- NB: DO NOT CHANGE THE ORDER OF THE ENUM VALUES DURING SCHEMA MIGRATIONS.
+  creator_set_visibility ENUM(
+    'public',
+    'hidden',
+    'private'
+  ) NOT NULL DEFAULT 'public',
+
+  -- The synthetic id associated with this result.
+  -- These ids are incremented on a per-user basis to help users
+  -- sequence their own work. They serve no other purpose.
+  maybe_creator_synthetic_id BIGINT(20) DEFAULT NULL,
+
+  -- ========== MODERATION DETAILS ==========
+
+  -- The last moderator that made changes last.
+  -- Fine details of moderator actions are kept in a separate `audit_logs` table.
+  maybe_mod_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- If this is removed by a mod or the creator.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+
+  KEY index_ietf_primary_language_subtag (ietf_primary_language_subtag),
+  KEY fk_maybe_creator_user_token (maybe_creator_user_token),
+  KEY fk_maybe_mod_user_token (maybe_mod_user_token),
+  KEY index_creator_ip_address (creator_ip_address),
+  KEY index_user_deleted_at (user_deleted_at),
+  KEY index_mod_deleted_at (mod_deleted_at)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/zs_voices/zs_voices.sql

@@ -0,0 +1,157 @@
+-- NB: This is a manually squashed view of all the CREATE and ALTER statements,
+-- with comments attached to the fields for centralized documentation.
+
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- This table contains both zero shot TTS voices (speaker embeddings) and zero shot V2V voices (speaker embeddings).
+-- These are the zero shot voices that are usable by our users.
+CREATE TABLE zs_voices (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" for joins, API, etc. (PUBLIC)
+  token VARCHAR(32) NOT NULL,
+
+  -- NB: No idempotency token.
+  -- Store it in another table to prevent duplicate record creation from rapid user inputs.
+
+  -- ========== ZERO SHOT FOREIGN KEYS AND STATE==========
+
+  -- The zero shot dataset that was used to create this voice.
+  -- The dataset will only be accessible to the original author of the voice.
+  -- The original author can create multiple voices from the same dataset.
+  dataset_token VARCHAR(32) NOT NULL,
+
+  -- When datasets are edited, we bump their version.
+  -- This tells us which version of the dataset created this voice (not that we should preserve
+  -- record of which samples are in which version.)
+  -- We can surface this to the users as well.
+  dataset_version INT NOT NULL DEFAULT 0,
+
+  -- ========== MODEL DETAILS ==========
+
+  -- The class or category of model/algorithm this voice is intended for.
+  --   * 'tts' for zero shot TTS
+  --   * 'vc' for zero shot voice conversion
+  --   * 'pitch' for pitch correction models (hypothetical)
+  --   * (Since this isn't an enum, the list of values can be expanded.)
+  model_category VARCHAR(16) NOT NULL,
+
+  -- The model used for the voice:
+  --   * 'vall-e-x' for Vall-E-X
+  --   * 'ns2vc' for NaturalSpeech2 voice conversion (hypothetical)
+  --   * (Since this isn't an enum, the list of values can be expanded.)
+  model_type VARCHAR(16) NOT NULL,
+
+  -- The version of the model.
+  -- If we deploy new model weights for the model, we may need to regenerate
+  -- all of the corresponding speaker embeddings. This lets us know which voices
+  -- are the old version.
+  -- In some cases, we may need to preserve, hide, or roll back to older versions.
+  model_version INTEGER UNSIGNED NOT NULL DEFAULT 0,
+
+  -- What kind of features these are: spectrogram, encodec, etc.
+  -- Technically we shouldn't need this, as the 2-tuple (model_type, model_version) implicitly knows
+  -- what its features should be, but this is probably useful for our own analysis and may be easier
+  -- to deal with.
+  --
+  -- Example values:
+  --   * 'spectrogram'
+  --   * 'encodec'
+  --   * (Since this isn't an enum, the list of values can be expanded.)
+  encoding_type VARCHAR(16) NOT NULL,
+
+  -- ========== VOICE DETAILS ==========
+
+  -- The "title" of the voice. This can be the name and additional information as supplied by the user.
+  title VARCHAR(255) NOT NULL,
+
+  -- The full IETF BCP47 language tag (eg. en, en-US, es-419, ja-JP, pt, etc.)
+  -- The wide varchar space is because these language tags can become cumbersome, eg "hy-Latn-IT-arevela"
+  ietf_language_tag VARCHAR(64) NOT NULL DEFAULT 'en',
+
+  -- The IETF BCP47 language tag's primary language subtag (eg. "es-419" becomes "es" and "en-US" becomes "en")
+  ietf_primary_language_subtag VARCHAR(12) NOT NULL DEFAULT 'en',
+
+  -- TODO: misc speaker metadata
+  -- We'll want rich metadata for search
+  -- series_name VARCHAR(255) DEFAULT NULL,
+  -- speaker_name VARCHAR(255) DEFAULT NULL,
+  -- speaker_gender VARCHAR(255) DEFAULT NULL,
+  -- speaker_age VARCHAR(255) DEFAULT NULL,
+  -- speaker_voice_style VARCHAR(255) DEFAULT NULL,
+
+  -- ========== STORAGE DETAILS ==========
+
+  -- The hash for the bucket directory that contains the original upload
+  -- eg. /speaker_vectors/{bucket_hash}/{voice_model_version}.bin
+  bucket_hash  VARCHAR(32) NOT NULL,
+
+  -- ========== CREATOR DETAILS AND PREFERENCES ==========
+
+  -- Foreign key to user
+  maybe_creator_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- Based on a cookie sent by the frontend.
+  -- We'll save this even if the user is logged in.
+  maybe_creator_anonymous_visitor_token VARCHAR(32) DEFAULT NULL,
+
+  -- For abuse tracking.
+  -- Wide enough for IPv4/6
+  creator_ip_address VARCHAR(40) NOT NULL,
+
+  -- The creator can set a desired visibility for their data.
+  -- This does not always apply to every upload type.
+  -- Additionally, some upload types may require moderator approval prior
+  -- to being publicly listed, and this field has no bearing on that.
+  -- NB: DO NOT CHANGE THE ORDER OF THE ENUM VALUES DURING SCHEMA MIGRATIONS.
+  creator_set_visibility ENUM(
+    'public',
+    'hidden',
+    'private'
+  ) NOT NULL DEFAULT 'public',
+
+  -- The synthetic id associated with this result.
+  -- These ids are incremented on a per-user basis to help users
+  -- sequence their own work. They serve no other purpose.
+  maybe_creator_synthetic_id BIGINT(20) DEFAULT NULL,
+
+  -- ========== MODERATION DETAILS ==========
+
+  -- The last moderator that made changes last.
+  -- Fine details of moderator actions are kept in a separate `audit_logs` table.
+  maybe_mod_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- If this is removed by a mod or the creator.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+
+  -- TODO: These are not correct. Fix these before landing / running migrations.
+  KEY index_model_category (model_category),
+  KEY index_model_type (model_type),
+  KEY index_model_type_and_version (model_type, model_version),
+  KEY index_encoding_type (encoding_type),
+  KEY index_ietf_primary_language_subtag (ietf_primary_language_subtag),
+  KEY fk_maybe_creator_user_token (maybe_creator_user_token),
+  KEY fk_maybe_mod_user_token (maybe_mod_user_token),
+  KEY index_creator_ip_address (creator_ip_address),
+  KEY index_user_deleted_at (user_deleted_at),
+  KEY index_mod_deleted_at (mod_deleted_at)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

codegen_database.sh

@@ -18,7 +18,7 @@ build_shared_mysql_database_library() {
   # NB: For now, this is our monolithic DB that serves all of our microservices (gross)
   # It's a single package so that queries can be shared (again, gross)
   pushd crates/lib/mysql_queries
-  SQLX_OFFLINE=true cargo sqlx prepare --merged
+  SQLX_OFFLINE=true cargo sqlx prepare
   popd
 }
 
@@ -35,6 +35,9 @@ build_shared_sqlite_database_library() {
 echo 'mysql prepare'
 build_shared_mysql_database_library
 
-echo 'sqlite prepare'
-build_shared_sqlite_database_library
+#echo 'sqlite prepare'
+#build_shared_sqlite_database_library
+
+echo 'done'
+
 

crates/_types/buckets/src/public/media_files/directory.rs

@@ -1,6 +1,7 @@
+use crockford::crockford_entropy_lower;
+
 use crate::public::public_path::PublicPath;
 use crate::util::hashed_directory_path_long_string::hashed_directory_path_long_string;
-use crockford::crockford_entropy_lower;
 
 // TODO: Generate these from a macro.
 

crates/_types/buckets/src/public/media_files/original_file.rs

@@ -1,7 +1,9 @@
+use std::path::PathBuf;
+
+use crockford::crockford_entropy_lower;
+
 use crate::public::media_files::directory::MediaFileBucketDirectory;
 use crate::public::public_path::PublicPath;
-use crockford::crockford_entropy_lower;
-use std::path::PathBuf;
 
 const ORIGINAL_FILE_BASENAME : &str = "file.bin";
 
@@ -12,25 +14,43 @@ const ORIGINAL_FILE_BASENAME : &str = "file.bin";
 #[derive(Clone)]
 pub struct MediaFileBucketPath {
   directory: MediaFileBucketDirectory,
+
+  filename: String,
   full_object_path: String,
+
+  optional_prefix: Option<String>,
+  /// NB: Extension contains the leading period, eg. ".mp4"
+  optional_extension: Option<String>,
 }
 
 impl PublicPath for MediaFileBucketPath {}
 
 impl MediaFileBucketPath {
 
-  pub fn generate_new() -> Self {
+  pub fn generate_new(optional_prefix: Option<&str>, optional_extension: Option<&str>) -> Self {
     let entropy = crockford_entropy_lower(32);
-    Self::from_object_hash(&entropy)
+    Self::from_object_hash(&entropy, optional_prefix, optional_extension)
   }
 
-  pub fn from_object_hash(hash: &str) -> Self {
+  pub fn from_object_hash(hash: &str, optional_prefix: Option<&str>, optional_extension: Option<&str>) -> Self {
     // TODO: Path construction could be cleaner.
     let directory = MediaFileBucketDirectory::from_object_hash(hash);
-    let full_object_path = format!("{}/{}", directory.get_directory_path_str(), ORIGINAL_FILE_BASENAME);
+
+    let filename = match (optional_prefix, optional_extension) {
+      (None, None) => ORIGINAL_FILE_BASENAME.to_string(),
+      (None, Some(ext)) => format!("{}{}", hash, ext),
+      (Some(pre), None) => format!("{}{}", pre, hash),
+      (Some(pre), Some(ext)) => format!("{}{}{}", pre, hash, ext),
+    };
+
+    let full_object_path = format!("{}/{}", directory.get_directory_path_str(), filename);
+
     Self {
       directory,
+      filename,
       full_object_path,
+      optional_prefix: optional_prefix.map(|p| p.to_string()),
+      optional_extension: optional_extension.map(|e| e.to_string()),
     }
   }
 
@@ -50,52 +70,76 @@ impl MediaFileBucketPath {
     &self.directory.get_object_hash()
   }
 
-  pub fn get_basename() -> &'static str {
-    ORIGINAL_FILE_BASENAME
+  pub fn get_basename(&self) -> &str {
+    &self.filename
   }
 }
 
 #[cfg(test)]
 mod tests {
-  use crate::public::media_files::original_file::MediaFileBucketPath;
   use std::path::PathBuf;
 
-  #[test]
-  pub fn generate_new_entropy() {
-    let file = MediaFileBucketPath::generate_new();
-    assert_eq!(file.get_object_hash().len(), 32);
-    assert_eq!(file.get_directory().get_object_hash().len(), 32);
+  use crate::public::media_files::original_file::MediaFileBucketPath;
+
+  mod with_prefix_and_extension {
+    use super::*;
+
+    #[test]
+    pub fn generate_new_entropy() {
+      let file = MediaFileBucketPath::generate_new(Some("pre_"), Some(".mp4"));
+      assert_eq!(file.get_object_hash().len(), 32);
+      assert_eq!(file.get_object_hash().len(), 32);
+      assert_eq!(file.get_directory().get_object_hash().len(), 32);
+    }
+
+    #[test]
+    pub fn get_full_object_path_str() {
+      let file = MediaFileBucketPath::from_object_hash("abcdefghijk", Some("pre_"), Some(".mp4"));
+      assert_eq!(file.get_full_object_path_str(), "/media/a/b/c/d/e/abcdefghijk/pre_abcdefghijk.mp4");
+      assert_eq!(file.get_basename(), "pre_abcdefghijk.mp4");
+    }
   }
 
-  #[test]
-  pub fn get_full_object_path_str() {
-    let file = MediaFileBucketPath::from_object_hash("abcdefghijk");
-    assert_eq!(file.get_full_object_path_str(), "/media/a/b/c/d/e/abcdefghijk/file.bin");
-  }
+  mod without_prefix_and_extension {
+    use super::*;
 
-  #[test]
-  pub fn to_full_object_pathbuf() {
-    let file = MediaFileBucketPath::from_object_hash("abcdefghijk");
-    assert_eq!(file.to_full_object_pathbuf(), PathBuf::from("/media/a/b/c/d/e/abcdefghijk/file.bin"));
-  }
+    #[test]
+    pub fn generate_new_entropy() {
+      let file = MediaFileBucketPath::generate_new(None, None);
+      assert_eq!(file.get_object_hash().len(), 32);
+      assert_eq!(file.get_directory().get_object_hash().len(), 32);
+    }
 
-  #[test]
-  pub fn get_full_object_path_str_short_name() {
-    let file = MediaFileBucketPath::from_object_hash("foo");
-    assert_eq!(file.get_full_object_path_str(), "/media/f/o/foo/file.bin");
-  }
+    #[test]
+    pub fn get_full_object_path_str() {
+      let file = MediaFileBucketPath::from_object_hash("abcdefghijk", None, None);
+      assert_eq!(file.get_full_object_path_str(), "/media/a/b/c/d/e/abcdefghijk/file.bin");
+    }
 
-  #[test]
-  pub fn get_full_object_path_str_starts_with_directory() {
-    let file = MediaFileBucketPath::from_object_hash("abcdefghijk");
-    let full_path = file.get_full_object_path_str();
-    assert!(full_path.starts_with(file.get_directory().get_directory_path_str()))
-  }
+    #[test]
+    pub fn to_full_object_pathbuf() {
+      let file = MediaFileBucketPath::from_object_hash("abcdefghijk", None, None);
+      assert_eq!(file.to_full_object_pathbuf(), PathBuf::from("/media/a/b/c/d/e/abcdefghijk/file.bin"));
+    }
 
-  #[test]
-  pub fn get_object_hash() {
-    let hash = "abcdefghijk";
-    let file = MediaFileBucketPath::from_object_hash(hash);
-    assert_eq!(file.get_object_hash(), hash);
+    #[test]
+    pub fn get_full_object_path_str_short_name() {
+      let file = MediaFileBucketPath::from_object_hash("foo", None, None);
+      assert_eq!(file.get_full_object_path_str(), "/media/f/o/foo/file.bin");
+    }
+
+    #[test]
+    pub fn get_full_object_path_str_starts_with_directory() {
+      let file = MediaFileBucketPath::from_object_hash("abcdefghijk", None, None);
+      let full_path = file.get_full_object_path_str();
+      assert!(full_path.starts_with(file.get_directory().get_directory_path_str()))
+    }
+
+    #[test]
+    pub fn get_object_hash() {
+      let hash = "abcdefghijk";
+      let file = MediaFileBucketPath::from_object_hash(hash, None, None);
+      assert_eq!(file.get_object_hash(), hash);
+    }
   }
 }

crates/_types/buckets/src/public/media_uploads/directory.rs

@@ -1,6 +1,7 @@
+use crockford::crockford_entropy_lower;
+
 use crate::public::public_path::PublicPath;
 use crate::util::hashed_directory_path_long_string::hashed_directory_path_long_string;
-use crockford::crockford_entropy_lower;
 
 // TODO: Generate these from a macro.
 

crates/_types/buckets/src/public/media_uploads/original_file.rs

@@ -1,7 +1,9 @@
+use std::path::PathBuf;
+
+use crockford::crockford_entropy_lower;
+
 use crate::public::media_uploads::directory::MediaUploadDirectory;
 use crate::public::public_path::PublicPath;
-use crockford::crockford_entropy_lower;
-use std::path::PathBuf;
 
 const ORIGINAL_FILE_BASENAME : &str = "original_upload.bin";
 
@@ -57,9 +59,10 @@ impl MediaUploadOriginalFilePath {
 
 #[cfg(test)]
 mod tests {
-  use crate::public::media_uploads::original_file::MediaUploadOriginalFilePath;
   use std::path::PathBuf;
 
+  use crate::public::media_uploads::original_file::MediaUploadOriginalFilePath;
+
   #[test]
   pub fn generate_new_entropy() {
     let file = MediaUploadOriginalFilePath::generate_new();

crates/_types/buckets/src/public/voice_conversion_results/original_file.rs

@@ -1,7 +1,9 @@
+use std::path::PathBuf;
+
+use crockford::crockford_entropy_lower;
+
 use crate::public::public_path::PublicPath;
 use crate::public::voice_conversion_results::directory::VoiceConversionResultDirectory;
-use crockford::crockford_entropy_lower;
-use std::path::PathBuf;
 
 const ORIGINAL_FILE_PREFIX : &str = "fakeyou_";
 const ORIGINAL_FILE_SUFFIX : &str = ".wav";
@@ -61,9 +63,10 @@ impl VoiceConversionResultOriginalFilePath {
 
 #[cfg(test)]
 mod tests {
-  use crate::public::voice_conversion_results::original_file::VoiceConversionResultOriginalFilePath;
   use std::path::PathBuf;
 
+  use crate::public::voice_conversion_results::original_file::VoiceConversionResultOriginalFilePath;
+
   #[test]
   pub fn generate_new_entropy() {
     let file = VoiceConversionResultOriginalFilePath::generate_new();

crates/_types/buckets/src/util/hash_to_bucket_path.rs

@@ -1,6 +1,7 @@
+use std::path::PathBuf;
+
 use errors::AnyhowResult;
 use errors::bail;
-use std::path::PathBuf;
 
 #[deprecated(note = "this is a messy function and should be avoided")]
 pub (crate) fn hash_to_bucket_path(file_hash: &str,
@@ -28,9 +29,10 @@ pub (crate) fn hash_to_bucket_path(file_hash: &str,
 
 #[cfg(test)]
 mod tests {
-  use crate::util::hash_to_bucket_path::hash_to_bucket_path;
   use std::path::PathBuf;
 
+  use crate::util::hash_to_bucket_path::hash_to_bucket_path;
+
   #[test]
   fn test_bucket_without_root() {
     assert_eq!(

crates/_types/buckets/src/util/hash_to_bucket_path_string.rs

@@ -1,6 +1,7 @@
-use crate::util::hash_to_bucket_path::hash_to_bucket_path;
-use errors::AnyhowResult;
 use errors::anyhow;
+use errors::AnyhowResult;
+
+use crate::util::hash_to_bucket_path::hash_to_bucket_path;
 
 #[deprecated(note = "this is a messy function and should be avoided")]
 pub fn hash_to_bucket_path_string(file_hash: &str,

crates/_types/enums/src/by_table/audit_logs/audit_log_entity_action.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/audit_logs/audit_log_entity_type.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/comments/comment_entity_type.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/generic_inference_jobs/frontend_failure_category.rs

@@ -0,0 +1,145 @@
+use std::collections::BTreeSet;
+
+#[cfg(test)]
+use strum::EnumCount;
+#[cfg(test)]
+use strum::EnumIter;
+
+/// Used in the `generic_inference_jobs` table in `VARCHAR(32)` field `frontend_failure_category`.
+///
+/// When jobs fail (permanently or transiently), we need to inform the frontend of the reason,
+/// because perhaps there's something the user can do to change their input.
+///
+/// The previous "VARCHAR(32) failure_reason" column was a text-based message that could not be
+/// localized or made user friendly. This `frontend_failure_category` exists to provide well-defined
+/// failure categories to the frontend that can easily be localized and indicated consistently in
+/// the UI.
+///
+/// Another benefit is that we'll surface all of the various types of failure and perhaps eventually
+/// come to handle some in a cross-cutting way.
+///
+/// DO NOT CHANGE VALUES WITHOUT A MIGRATION STRATEGY.
+#[cfg_attr(test, derive(EnumIter, EnumCount))]
+#[derive(Clone, Copy, Eq, PartialEq, Hash, Ord, PartialOrd, Deserialize, Serialize)]
+pub enum FrontendFailureCategory {
+  /// When a face is not detected in the image used for animation.
+  /// For SadTalker (and possibly Wav2Lip)
+  #[serde(rename = "face_not_detected")]
+  FaceNotDetected,
+
+  /// The user stepped away from their device and expected the workload to finish.
+  /// Some workloads require that the user keep their browser open.
+  #[serde(rename = "keep_alive_elapsed")]
+  KeepAliveElapsed,
+
+  /// This is mostly for developers -- a feature isn't complete somewhere in the code.
+  /// Big oops if errors of this class make it to production.
+  #[serde(rename = "not_yet_implemented")]
+  NotYetImplemented,
+
+  /// Tell the user that some kind of transient error happened. They don't need to know
+  /// exactly what happened. We'll retry their workload in any case.
+  #[serde(rename = "retryable_worker_error")]
+  RetryableWorkerError,
+}
+
+// TODO(bt, 2022-12-21): This desperately needs MySQL integration tests!
+impl_enum_display_and_debug_using_to_str!(FrontendFailureCategory);
+impl_mysql_enum_coders!(FrontendFailureCategory);
+
+/// NB: Legacy API for older code.
+impl FrontendFailureCategory {
+  pub fn to_str(&self) -> &'static str {
+    match self {
+      Self::FaceNotDetected => "face_not_detected",
+      Self::KeepAliveElapsed => "keep_alive_elapsed",
+      Self::NotYetImplemented => "not_yet_implemented",
+      Self::RetryableWorkerError => "retryable_worker_error",
+    }
+  }
+
+  pub fn from_str(value: &str) -> Result<Self, String> {
+    match value {
+      "face_not_detected" => Ok(Self::FaceNotDetected),
+      "keep_alive_elapsed" => Ok(Self::KeepAliveElapsed),
+      "not_yet_implemented" => Ok(Self::NotYetImplemented),
+      "retryable_worker_error" => Ok(Self::RetryableWorkerError),
+      _ => Err(format!("invalid value: {:?}", value)),
+    }
+  }
+
+  pub fn all_variants() -> BTreeSet<Self> {
+    // NB: BTreeSet is sorted
+    // NB: BTreeSet::from() isn't const, but not worth using LazyStatic, etc.
+    BTreeSet::from([
+      Self::FaceNotDetected,
+      Self::KeepAliveElapsed,
+      Self::NotYetImplemented,
+      Self::RetryableWorkerError,
+    ])
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::by_table::generic_inference_jobs::frontend_failure_category::FrontendFailureCategory;
+  use crate::test_helpers::assert_serialization;
+
+  mod explicit_checks {
+    use super::*;
+
+    #[test]
+    fn test_serialization() {
+      assert_serialization(FrontendFailureCategory::FaceNotDetected, "face_not_detected");
+      assert_serialization(FrontendFailureCategory::KeepAliveElapsed, "keep_alive_elapsed");
+      assert_serialization(FrontendFailureCategory::NotYetImplemented, "not_yet_implemented");
+      assert_serialization(FrontendFailureCategory::RetryableWorkerError, "retryable_worker_error");
+    }
+
+    #[test]
+    fn to_str() {
+      assert_eq!(FrontendFailureCategory::FaceNotDetected.to_str(), "face_not_detected");
+      assert_eq!(FrontendFailureCategory::KeepAliveElapsed.to_str(), "keep_alive_elapsed");
+      assert_eq!(FrontendFailureCategory::NotYetImplemented.to_str(), "not_yet_implemented");
+      assert_eq!(FrontendFailureCategory::RetryableWorkerError.to_str(), "retryable_worker_error");
+    }
+
+    #[test]
+    fn from_str() {
+      assert_eq!(FrontendFailureCategory::from_str("face_not_detected").unwrap(), FrontendFailureCategory::FaceNotDetected);
+      assert_eq!(FrontendFailureCategory::from_str("keep_alive_elapsed").unwrap(), FrontendFailureCategory::KeepAliveElapsed);
+      assert_eq!(FrontendFailureCategory::from_str("not_yet_implemented").unwrap(), FrontendFailureCategory::NotYetImplemented);
+      assert_eq!(FrontendFailureCategory::from_str("retryable_worker_error").unwrap(), FrontendFailureCategory::RetryableWorkerError);
+    }
+
+    #[test]
+    fn all_variants() {
+      let mut variants = FrontendFailureCategory::all_variants();
+      assert_eq!(variants.len(), 4);
+      assert_eq!(variants.pop_first(), Some(FrontendFailureCategory::FaceNotDetected));
+      assert_eq!(variants.pop_first(), Some(FrontendFailureCategory::KeepAliveElapsed));
+      assert_eq!(variants.pop_first(), Some(FrontendFailureCategory::NotYetImplemented));
+      assert_eq!(variants.pop_first(), Some(FrontendFailureCategory::RetryableWorkerError));
+      assert_eq!(variants.pop_first(), None);
+    }
+  }
+
+  mod mechanical_checks {
+    use super::*;
+
+    #[test]
+    fn variant_length() {
+      use strum::IntoEnumIterator;
+      assert_eq!(FrontendFailureCategory::all_variants().len(), FrontendFailureCategory::iter().len());
+    }
+
+    #[test]
+    fn round_trip() {
+      for variant in FrontendFailureCategory::all_variants() {
+        assert_eq!(variant, FrontendFailureCategory::from_str(variant.to_str()).unwrap());
+        assert_eq!(variant, FrontendFailureCategory::from_str(&format!("{}", variant)).unwrap());
+        assert_eq!(variant, FrontendFailureCategory::from_str(&format!("{:?}", variant)).unwrap());
+      }
+    }
+  }
+}

crates/_types/enums/src/by_table/generic_inference_jobs/inference_category.rs

@@ -5,7 +5,6 @@ use strum::EnumCount;
 #[cfg(test)]
 use strum::EnumIter;
 
-
 /// Used in the `generic_inference_jobs` table in `VARCHAR(32)` field `inference_category`.
 ///
 /// Our "generic inference" pipeline supports a wide variety of ML models and other media.

crates/_types/enums/src/by_table/generic_inference_jobs/inference_input_source_token_type.rs

@@ -5,7 +5,6 @@ use strum::EnumCount;
 #[cfg(test)]
 use strum::EnumIter;
 
-
 /// Used in the `generic_inference_jobs` table in `VARCHAR(32)` field `maybe_input_source_token`.
 ///
 /// DO NOT CHANGE VALUES WITHOUT A MIGRATION STRATEGY.

crates/_types/enums/src/by_table/generic_inference_jobs/inference_model_type.rs

@@ -5,7 +5,6 @@ use strum::EnumCount;
 #[cfg(test)]
 use strum::EnumIter;
 
-
 /// Used in the `generic_inference_jobs` table in `VARCHAR(32)` field `maybe_model_type`.
 ///
 /// Our "generic inference" pipeline supports a wide variety of ML models and other media.
@@ -20,20 +19,18 @@ use strum::EnumIter;
 pub enum InferenceModelType {
   #[serde(rename = "rvc_v2")]
   RvcV2,
-
   // NB: sad_talker does use user-supplied models, so there is no "model token"
   #[serde(rename = "sad_talker")]
   SadTalker,
-
   #[serde(rename = "so_vits_svc")]
   SoVitsSvc,
-
   // TODO: Does this need to be "legacy_tacotron2" ?
   #[serde(rename = "tacotron2")]
   Tacotron2,
-
   #[serde(rename = "vits")]
   Vits,
+  #[serde(rename = "vall_e_x")]
+  VallEX,
 }
 
 // TODO(bt, 2022-12-21): This desperately needs MySQL integration tests!
@@ -49,6 +46,7 @@ impl InferenceModelType {
       Self::SoVitsSvc => "so_vits_svc",
       Self::Tacotron2 => "tacotron2",
       Self::Vits => "vits",
+      Self::VallEX => "vall_e_x"
     }
   }
 
@@ -59,6 +57,7 @@ impl InferenceModelType {
       "so_vits_svc" => Ok(Self::SoVitsSvc),
       "tacotron2" => Ok(Self::Tacotron2),
       "vits" => Ok(Self::Vits),
+      "vall_e_x" => Ok(Self::VallEX),
       _ => Err(format!("invalid value: {:?}", value)),
     }
   }
@@ -72,6 +71,7 @@ impl InferenceModelType {
       InferenceModelType::SoVitsSvc,
       InferenceModelType::Tacotron2,
       InferenceModelType::Vits,
+      InferenceModelType::VallEX
     ])
   }
 }
@@ -88,6 +88,7 @@ mod tests {
     assert_serialization(InferenceModelType::SoVitsSvc, "so_vits_svc");
     assert_serialization(InferenceModelType::Tacotron2, "tacotron2");
     assert_serialization(InferenceModelType::Vits, "vits");
+    assert_serialization(InferenceModelType::VallEX, "vall_e_x");
   }
 
   #[test]
@@ -97,6 +98,7 @@ mod tests {
     assert_eq!(InferenceModelType::SoVitsSvc.to_str(), "so_vits_svc");
     assert_eq!(InferenceModelType::Tacotron2.to_str(), "tacotron2");
     assert_eq!(InferenceModelType::Vits.to_str(), "vits");
+    assert_eq!(InferenceModelType::VallEX.to_str(), "vall_e_x");
   }
 
   #[test]
@@ -106,18 +108,20 @@ mod tests {
     assert_eq!(InferenceModelType::from_str("so_vits_svc").unwrap(), InferenceModelType::SoVitsSvc);
     assert_eq!(InferenceModelType::from_str("tacotron2").unwrap(), InferenceModelType::Tacotron2);
     assert_eq!(InferenceModelType::from_str("vits").unwrap(), InferenceModelType::Vits);
+    assert_eq!(InferenceModelType::from_str("vall_e_x").unwrap(), InferenceModelType::VallEX);
   }
 
   #[test]
   fn all_variants() {
     // Static check
     let mut variants = InferenceModelType::all_variants();
-    assert_eq!(variants.len(), 5);
+    assert_eq!(variants.len(), 6);
     assert_eq!(variants.pop_first(), Some(InferenceModelType::RvcV2));
     assert_eq!(variants.pop_first(), Some(InferenceModelType::SadTalker));
     assert_eq!(variants.pop_first(), Some(InferenceModelType::SoVitsSvc));
     assert_eq!(variants.pop_first(), Some(InferenceModelType::Tacotron2));
     assert_eq!(variants.pop_first(), Some(InferenceModelType::Vits));
+    assert_eq!(variants.pop_first(), Some(InferenceModelType::VallEX));
     assert_eq!(variants.pop_first(), None);
 
     // Generated check

crates/_types/enums/src/by_table/generic_inference_jobs/inference_result_type.rs

@@ -5,7 +5,6 @@ use strum::EnumCount;
 #[cfg(test)]
 use strum::EnumIter;
 
-
 /// Used in the `generic_inference_jobs` table in `VARCHAR(32)` field `on_success_result_entity_type`.
 ///
 /// Our "generic inference" pipeline supports a wide variety of output types.

crates/_types/enums/src/by_table/generic_inference_jobs/mod.rs

@@ -1,3 +1,4 @@
+pub mod frontend_failure_category;
 pub mod inference_category;
 pub mod inference_input_source_token_type;
 pub mod inference_model_type;

crates/_types/enums/src/by_table/generic_synthetic_ids/id_category.rs

@@ -5,7 +5,6 @@ use strum::EnumCount;
 #[cfg(test)]
 use strum::EnumIter;
 
-
 /// Used in the `generic_synthetic_ids` table in `VARCHAR(32)` field `id_category`.
 ///
 /// This lets us create synthetic increment IDs on a per-user, per-category basis.
@@ -21,6 +20,23 @@ pub enum IdCategory {
   /// Results from lipsync animations (which may live in the media_files table)
   #[serde(rename = "lipsync_animation")]
   LipsyncAnimation,
+
+  /// Results from voice conversion (which may live in the media_files table)
+  /// Applies for RVC and SVC
+  #[serde(rename = "voice_conversion")]
+  VoiceConversion,
+
+  /// Results from the zero shot tts
+  #[serde(rename = "zs_audio_tts")]
+  ZeroShotTTS,
+
+  /// Zs dataset which lives in the zs_voice_datasets table
+  #[serde(rename = "zs_dataset")]
+  ZsDataset,
+
+  /// Zs voice which lives in the zs_voices table
+  #[serde(rename = "zs_voice")]
+  ZsVoice,
 }
 
 // TODO(bt, 2022-12-21): This desperately needs MySQL integration tests!
@@ -33,6 +49,10 @@ impl IdCategory {
     match self {
       Self::MediaFile => "media_file",
       Self::LipsyncAnimation => "lipsync_animation",
+      Self::VoiceConversion => "voice_conversion",
+      Self::ZsDataset => "zs_dataset",
+      Self::ZsVoice => "zs_voice",
+      Self::ZeroShotTTS => "zs_audio_tts"
     }
   }
 
@@ -40,6 +60,10 @@ impl IdCategory {
     match value {
       "media_file" => Ok(Self::MediaFile),
       "lipsync_animation" => Ok(Self::LipsyncAnimation),
+      "voice_conversion" => Ok(Self::VoiceConversion),
+      "zs_dataset" => Ok(Self::ZsDataset),
+      "zs_voice" => Ok(Self::ZsVoice),
+      "zs_audio_tts" => Ok(Self::ZeroShotTTS),
       _ => Err(format!("invalid value: {:?}", value)),
     }
   }
@@ -50,6 +74,10 @@ impl IdCategory {
     BTreeSet::from([
       Self::MediaFile,
       Self::LipsyncAnimation,
+      Self::VoiceConversion,
+      Self::ZsDataset,
+      Self::ZsVoice,
+      Self::ZeroShotTTS
     ])
   }
 }
@@ -59,35 +87,73 @@ mod tests {
   use crate::by_table::generic_synthetic_ids::id_category::IdCategory;
   use crate::test_helpers::assert_serialization;
 
+  mod explicit_checks {
+    use super::*;
+
   #[test]
   fn test_serialization() {
     assert_serialization(IdCategory::MediaFile, "media_file");
     assert_serialization(IdCategory::LipsyncAnimation, "lipsync_animation");
+    assert_serialization(IdCategory::VoiceConversion, "voice_conversion");
+    assert_serialization(IdCategory::ZsDataset, "zs_dataset");
+    assert_serialization(IdCategory::ZsVoice, "zs_voice");
   }
 
-  #[test]
-  fn to_str() {
-    assert_eq!(IdCategory::MediaFile.to_str(), "media_file");
-    assert_eq!(IdCategory::LipsyncAnimation.to_str(), "lipsync_animation");
+    #[test]
+    fn to_str() {
+      assert_eq!(IdCategory::MediaFile.to_str(), "media_file");
+      assert_eq!(IdCategory::LipsyncAnimation.to_str(), "lipsync_animation");
+      assert_eq!(IdCategory::VoiceConversion.to_str(), "voice_conversion");
+      assert_eq!(IdCategory::ZsDataset.to_str(), "zs_dataset");
+      assert_eq!(IdCategory::ZsVoice.to_str(), "zs_voice");
+    assert_eq!(IdCategory::ZeroShotTTS.to_str(),"zs_audio_tts");
+    }
+
+    #[test]
+    fn from_str() {
+      assert_eq!(IdCategory::from_str("media_file").unwrap(), IdCategory::MediaFile);
+      assert_eq!(IdCategory::from_str("lipsync_animation").unwrap(), IdCategory::LipsyncAnimation);
+      assert_eq!(IdCategory::from_str("voice_conversion").unwrap(), IdCategory::VoiceConversion);
+      assert_eq!(IdCategory::from_str("zs_dataset").unwrap(), IdCategory::ZsDataset);
+      assert_eq!(IdCategory::from_str("zs_voice").unwrap(), IdCategory::ZsVoice);
+    assert_eq!(IdCategory::from_str("zs_audio_tts").unwrap(), IdCategory::ZeroShotTTS);
+    }
+
+    #[test]
+    fn all_variants() {
+      // Static check
+      let mut variants = IdCategory::all_variants();
+      assert_eq!(variants.len(), 5);
+      assert_eq!(variants.pop_first(), Some(IdCategory::MediaFile));
+      assert_eq!(variants.pop_first(), Some(IdCategory::LipsyncAnimation));
+      assert_eq!(variants.pop_first(), Some(IdCategory::VoiceConversion));
+      assert_eq!(variants.pop_first(), Some(IdCategory::ZsDataset));
+      assert_eq!(variants.pop_first(), Some(IdCategory::ZsVoice));
+    assert_eq!(variants.pop_first(), Some(IdCategory::ZeroShotTTS));
+      assert_eq!(variants.pop_first(), None);
+
+      // Generated check
+      use strum::IntoEnumIterator;
+      assert_eq!(IdCategory::all_variants().len(), IdCategory::iter().len());
+    }
   }
 
-  #[test]
-  fn from_str() {
-    assert_eq!(IdCategory::from_str("media_file").unwrap(), IdCategory::MediaFile);
-    assert_eq!(IdCategory::from_str("lipsync_animation").unwrap(), IdCategory::LipsyncAnimation);
-  }
+  mod mechanical_checks {
+    use super::*;
 
-  #[test]
-  fn all_variants() {
-    // Static check
-    let mut variants = IdCategory::all_variants();
-    assert_eq!(variants.len(), 2);
-    assert_eq!(variants.pop_first(), Some(IdCategory::MediaFile));
-    assert_eq!(variants.pop_first(), Some(IdCategory::LipsyncAnimation));
-    assert_eq!(variants.pop_first(), None);
+    #[test]
+    fn variant_length() {
+      use strum::IntoEnumIterator;
+      assert_eq!(IdCategory::all_variants().len(), IdCategory::iter().len());
+    }
 
-    // Generated check
-    use strum::IntoEnumIterator;
-    assert_eq!(IdCategory::all_variants().len(), IdCategory::iter().len());
+    #[test]
+    fn round_trip() {
+      for variant in IdCategory::all_variants() {
+        assert_eq!(variant, IdCategory::from_str(variant.to_str()).unwrap());
+        assert_eq!(variant, IdCategory::from_str(&format!("{}", variant)).unwrap());
+        assert_eq!(variant, IdCategory::from_str(&format!("{:?}", variant)).unwrap());
+      }
+    }
   }
 }

crates/_types/enums/src/by_table/media_files/media_file_origin_category.rs

@@ -1,6 +1,6 @@
+use std::collections::BTreeSet;
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]
@@ -10,8 +10,8 @@ use strum::EnumIter;
 ///
 /// DO NOT CHANGE VALUES WITHOUT A MIGRATION STRATEGY.
 #[cfg_attr(test, derive(EnumIter, EnumCount))]
-#[derive(Clone, Copy, Eq, PartialEq, Deserialize, Serialize)]
-#[serde(rename_all = "lowercase")]
+#[derive(Clone, Copy, Eq, PartialEq, Hash, Ord, PartialOrd, Deserialize, Serialize)]
+#[serde(rename_all = "snake_case")]
 pub enum MediaFileOriginCategory {
   /// ML model inference output - uploaded models or zero shot.
   Inference,
@@ -19,8 +19,11 @@ pub enum MediaFileOriginCategory {
   /// Processed results - (we don't have these systems yet, but eg. trim, transcode, etc).
   Processed,
 
-  /// User uploaded files
+  /// User uploaded files (from their filesystem)
   Upload,
+
+  /// User uploaded files recorded directly from their device (browser, mobile), typically using device APIs.
+  DeviceApi,
 }
 
 // TODO(bt, 2022-12-21): This desperately needs MySQL integration tests!
@@ -34,6 +37,7 @@ impl MediaFileOriginCategory {
       Self::Inference => "inference",
       Self::Processed => "processed",
       Self::Upload => "upload",
+      Self::DeviceApi => "device_api",
     }
   }
 
@@ -42,9 +46,21 @@ impl MediaFileOriginCategory {
       "inference" => Ok(Self::Inference),
       "processed" => Ok(Self::Processed),
       "upload" => Ok(Self::Upload),
+      "device_api" => Ok(Self::DeviceApi),
       _ => Err(format!("invalid value: {:?}", value)),
     }
   }
+
+  pub fn all_variants() -> BTreeSet<Self> {
+    // NB: BTreeSet is sorted
+    // NB: BTreeSet::from() isn't const, but not worth using LazyStatic, etc.
+    BTreeSet::from([
+      Self::Inference,
+      Self::Processed,
+      Self::Upload,
+      Self::DeviceApi,
+    ])
+  }
 }
 
 #[cfg(test)]
@@ -52,25 +68,62 @@ mod tests {
   use crate::by_table::media_files::media_file_origin_category::MediaFileOriginCategory;
   use crate::test_helpers::assert_serialization;
 
-  #[test]
-  fn test_serialization() {
-    assert_serialization(MediaFileOriginCategory::Inference, "inference");
-    assert_serialization(MediaFileOriginCategory::Processed, "processed");
-    assert_serialization(MediaFileOriginCategory::Upload, "upload");
+  mod explicit_checks {
+    use super::*;
+
+    #[test]
+    fn test_serialization() {
+      assert_serialization(MediaFileOriginCategory::Inference, "inference");
+      assert_serialization(MediaFileOriginCategory::Processed, "processed");
+      assert_serialization(MediaFileOriginCategory::Upload, "upload");
+      assert_serialization(MediaFileOriginCategory::DeviceApi, "device_api");
+    }
+
+    #[test]
+    fn test_to_str() {
+      assert_eq!(MediaFileOriginCategory::Inference.to_str(), "inference");
+      assert_eq!(MediaFileOriginCategory::Processed.to_str(), "processed");
+      assert_eq!(MediaFileOriginCategory::Upload.to_str(), "upload");
+      assert_eq!(MediaFileOriginCategory::DeviceApi.to_str(), "device_api");
+    }
+
+    #[test]
+    fn test_from_str() {
+      assert_eq!(MediaFileOriginCategory::from_str("inference").unwrap(), MediaFileOriginCategory::Inference);
+      assert_eq!(MediaFileOriginCategory::from_str("processed").unwrap(), MediaFileOriginCategory::Processed);
+      assert_eq!(MediaFileOriginCategory::from_str("upload").unwrap(), MediaFileOriginCategory::Upload);
+      assert_eq!(MediaFileOriginCategory::from_str("device_api").unwrap(), MediaFileOriginCategory::DeviceApi);
+      assert!(MediaFileOriginCategory::from_str("foo").is_err());
+    }
+
+    #[test]
+    fn all_variants() {
+      let mut variants = MediaFileOriginCategory::all_variants();
+      assert_eq!(variants.len(), 4);
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginCategory::Inference));
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginCategory::Processed));
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginCategory::Upload));
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginCategory::DeviceApi));
+      assert_eq!(variants.pop_first(), None);
+    }
   }
 
-  #[test]
-  fn test_to_str() {
-    assert_eq!(MediaFileOriginCategory::Inference.to_str(), "inference");
-    assert_eq!(MediaFileOriginCategory::Processed.to_str(), "processed");
-    assert_eq!(MediaFileOriginCategory::Upload.to_str(), "upload");
-  }
+  mod mechanical_checks {
+    use super::*;
 
-  #[test]
-  fn test_from_str() {
-    assert_eq!(MediaFileOriginCategory::from_str("inference").unwrap(), MediaFileOriginCategory::Inference);
-    assert_eq!(MediaFileOriginCategory::from_str("processed").unwrap(), MediaFileOriginCategory::Processed);
-    assert_eq!(MediaFileOriginCategory::from_str("upload").unwrap(), MediaFileOriginCategory::Upload);
-    assert!(MediaFileOriginCategory::from_str("foo").is_err());
+    #[test]
+    fn variant_length() {
+      use strum::IntoEnumIterator;
+      assert_eq!(MediaFileOriginCategory::all_variants().len(), MediaFileOriginCategory::iter().len());
+    }
+
+    #[test]
+    fn round_trip() {
+      for variant in MediaFileOriginCategory::all_variants() {
+        assert_eq!(variant, MediaFileOriginCategory::from_str(variant.to_str()).unwrap());
+        assert_eq!(variant, MediaFileOriginCategory::from_str(&format!("{}", variant)).unwrap());
+        assert_eq!(variant, MediaFileOriginCategory::from_str(&format!("{:?}", variant)).unwrap());
+      }
+    }
   }
 }

crates/_types/enums/src/by_table/media_files/media_file_origin_model_type.rs

@@ -1,6 +1,7 @@
+use std::collections::BTreeSet;
+
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]
@@ -10,11 +11,21 @@ use strum::EnumIter;
 ///
 /// DO NOT CHANGE VALUES WITHOUT A MIGRATION STRATEGY.
 #[cfg_attr(test, derive(EnumIter, EnumCount))]
-#[derive(Clone, Copy, Eq, PartialEq, Deserialize, Serialize)]
+#[derive(Clone, Copy, Eq, PartialEq, Hash, Ord, PartialOrd, Deserialize, Serialize)]
 #[serde(rename_all = "snake_case")]
 pub enum MediaFileOriginModelType {
+  /// RVC (v2) voice conversion models
+  #[serde(rename = "rvc_v2")]
+  RvcV2,
+
   /// SadTalker -- v1, we may add another enum value for future versions
   SadTalker,
+
+  /// so-vits-svc voice conversion models
+  #[serde(rename = "so_vits_svc")]
+  SoVitsSvc,
+  #[serde(rename = "vall_e_x")]
+  VallEX,
 }
 
 // TODO(bt, 2022-12-21): This desperately needs MySQL integration tests!
@@ -25,16 +36,33 @@ impl_mysql_enum_coders!(MediaFileOriginModelType);
 impl MediaFileOriginModelType {
   pub fn to_str(&self) -> &'static str {
     match self {
+      Self::RvcV2  => "rvc_v2",
       Self::SadTalker => "sad_talker",
+      Self::SoVitsSvc => "so_vits_svc",
+      Self::VallEX => "vall_e_x"
     }
   }
 
   pub fn from_str(value: &str) -> Result<Self, String> {
     match value {
+      "rvc_v2" => Ok(Self::RvcV2),
       "sad_talker" => Ok(Self::SadTalker),
+      "so_vits_svc" => Ok(Self::SoVitsSvc),
+      "vall_e_x" => Ok(Self::VallEX),
       _ => Err(format!("invalid value: {:?}", value)),
     }
   }
+
+  pub fn all_variants() -> BTreeSet<Self> {
+    // NB: BTreeSet is sorted
+    // NB: BTreeSet::from() isn't const, but not worth using LazyStatic, etc.
+    BTreeSet::from([
+      Self::RvcV2,
+      Self::SadTalker,
+      Self::SoVitsSvc,
+      Self::VallEX
+    ])
+  }
 }
 
 #[cfg(test)]
@@ -42,19 +70,60 @@ mod tests {
   use crate::by_table::media_files::media_file_origin_model_type::MediaFileOriginModelType;
   use crate::test_helpers::assert_serialization;
 
+  mod explicit_checks {
+    use super::*;
+
   #[test]
   fn test_serialization() {
     assert_serialization(MediaFileOriginModelType::SadTalker, "sad_talker");
   }
 
-  #[test]
-  fn test_to_str() {
-    assert_eq!(MediaFileOriginModelType::SadTalker.to_str(), "sad_talker");
+    #[test]
+    fn test_to_str() {
+      assert_eq!(MediaFileOriginModelType::RvcV2.to_str(), "rvc_v2");
+      assert_eq!(MediaFileOriginModelType::SadTalker.to_str(), "sad_talker");
+      assert_eq!(MediaFileOriginModelType::SoVitsSvc.to_str(), "so_vits_svc");
+      assert_eq!(MediaFileOriginModelType::VallEX.to_str(), "vall_e_x");
   }
 
-  #[test]
-  fn test_from_str() {
-    assert_eq!(MediaFileOriginModelType::from_str("sad_talker").unwrap(), MediaFileOriginModelType::SadTalker);
+    #[test]
+    fn test_from_str() {
+      assert_eq!(MediaFileOriginModelType::from_str("rvc_v2").unwrap(), MediaFileOriginModelType::RvcV2);
+      assert_eq!(MediaFileOriginModelType::from_str("sad_talker").unwrap(), MediaFileOriginModelType::SadTalker);
+      assert_eq!(MediaFileOriginModelType::from_str("so_vits_svc").unwrap(), MediaFileOriginModelType::SoVitsSvc);
+      assert_eq!(MediaFileOriginModelType::from_str("vall_e_x").unwrap(), MediaFileOriginModelType::VallEX);
     assert!(MediaFileOriginModelType::from_str("foo").is_err());
+    }
+
+    #[test]
+    fn all_variants() {
+      let mut variants = MediaFileOriginModelType::all_variants();
+      assert_eq!(variants.len(), 4);
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginModelType::RvcV2));
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginModelType::SadTalker));
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginModelType::SoVitsSvc));
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginModelType::VallEX));
+      assert_eq!(variants.pop_first(), None);
+    }
+  }
+
+
+  mod mechanical_checks {
+    use super::*;
+
+    #[test]
+    fn variant_length() {
+      use strum::IntoEnumIterator;
+      assert_eq!(MediaFileOriginModelType::all_variants().len(), MediaFileOriginModelType::iter().len());
+    }
+
+    #[test]
+    fn round_trip() {
+      for variant in MediaFileOriginModelType::all_variants() {
+        assert_eq!(variant, MediaFileOriginModelType::from_str(variant.to_str()).unwrap());
+        assert_eq!(variant, MediaFileOriginModelType::from_str(&format!("{}", variant)).unwrap());
+        assert_eq!(variant, MediaFileOriginModelType::from_str(&format!("{:?}", variant)).unwrap());
+      }
+    }
   }
 }

crates/_types/enums/src/by_table/media_files/media_file_origin_product_category.rs

@@ -0,0 +1,138 @@
+use std::collections::BTreeSet;
+
+#[cfg(test)]
+use strum::EnumCount;
+#[cfg(test)]
+use strum::EnumIter;
+
+/// Used in the `media_files` table in `VARCHAR(16)` field `origin_product_category`.
+///
+/// This value indicates what product originally created the media file. (Not the ML model or
+/// user upload process.) This will let us scope media files to the product that generated them
+/// and filter them out of unrelated products if necessary (eg. a user probably doesn't want
+/// "Voice Designer" dataset samples in a video generation flow.)
+///
+/// DO NOT CHANGE VALUES WITHOUT A MIGRATION STRATEGY.
+#[cfg_attr(test, derive(EnumIter, EnumCount))]
+#[derive(Clone, Copy, Eq, PartialEq, Hash, Ord, PartialOrd, Deserialize, Serialize)]
+pub enum MediaFileOriginProductCategory {
+  /// Media files created by (or uploaded for) the Face Animator product.
+  /// The underlying model could be SadTalker, Wav2Lip, or some future model
+  #[serde(rename = "face_animator")]
+  FaceAnimator,
+
+  // TODO: This should be a temporary category until we migrate the DB to remove this default value
+  /// Unknown which product is attached to the file (generated the file, the file was
+  /// uploaded on behalf of, etc.)
+  #[serde(rename = "unknown")]
+  Unknown,
+
+  /// Voice conversion (either RVC or SVC)
+  #[serde(rename = "voice_conversion")]
+  VoiceConversion,
+
+  /// Media files created by (or uploaded for) the Zero Shot voice product.
+  #[serde(rename = "zs_voice")]
+  ZeroShotVoice,
+}
+
+// TODO(bt, 2022-12-21): This desperately needs MySQL integration tests!
+impl_enum_display_and_debug_using_to_str!(MediaFileOriginProductCategory);
+impl_mysql_enum_coders!(MediaFileOriginProductCategory);
+
+/// NB: Legacy API for older code.
+impl MediaFileOriginProductCategory {
+  pub fn to_str(&self) -> &'static str {
+    match self {
+      Self::FaceAnimator => "face_animator",
+      Self::Unknown => "unknown",
+      Self::VoiceConversion => "voice_conversion",
+      Self::ZeroShotVoice => "zs_voice",
+    }
+  }
+
+  pub fn from_str(value: &str) -> Result<Self, String> {
+    match value {
+      "face_animator" => Ok(Self::FaceAnimator),
+      "unknown" => Ok(Self::Unknown),
+      "voice_conversion" => Ok(Self::VoiceConversion),
+      "zs_voice" => Ok(Self::ZeroShotVoice),
+      _ => Err(format!("invalid value: {:?}", value)),
+    }
+  }
+
+  pub fn all_variants() -> BTreeSet<Self> {
+    // NB: BTreeSet is sorted
+    // NB: BTreeSet::from() isn't const, but not worth using LazyStatic, etc.
+    BTreeSet::from([
+      Self::FaceAnimator,
+      Self::Unknown,
+      Self::VoiceConversion,
+      Self::ZeroShotVoice,
+    ])
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::by_table::media_files::media_file_origin_product_category::MediaFileOriginProductCategory;
+  use crate::test_helpers::assert_serialization;
+
+  mod explicit_checks {
+    use super::*;
+
+    #[test]
+    fn test_serialization() {
+      assert_serialization(MediaFileOriginProductCategory::FaceAnimator, "face_animator");
+      assert_serialization(MediaFileOriginProductCategory::Unknown, "unknown");
+      assert_serialization(MediaFileOriginProductCategory::VoiceConversion, "voice_conversion");
+      assert_serialization(MediaFileOriginProductCategory::ZeroShotVoice, "zs_voice");
+    }
+
+    #[test]
+    fn to_str() {
+      assert_eq!(MediaFileOriginProductCategory::FaceAnimator.to_str(), "face_animator");
+      assert_eq!(MediaFileOriginProductCategory::Unknown.to_str(), "unknown");
+      assert_eq!(MediaFileOriginProductCategory::VoiceConversion.to_str(), "voice_conversion");
+      assert_eq!(MediaFileOriginProductCategory::ZeroShotVoice.to_str(), "zs_voice");
+    }
+
+    #[test]
+    fn from_str() {
+      assert_eq!(MediaFileOriginProductCategory::from_str("face_animator").unwrap(), MediaFileOriginProductCategory::FaceAnimator);
+      assert_eq!(MediaFileOriginProductCategory::from_str("unknown").unwrap(), MediaFileOriginProductCategory::Unknown);
+      assert_eq!(MediaFileOriginProductCategory::from_str("voice_conversion").unwrap(), MediaFileOriginProductCategory::VoiceConversion);
+      assert_eq!(MediaFileOriginProductCategory::from_str("zs_voice").unwrap(), MediaFileOriginProductCategory::ZeroShotVoice);
+    }
+
+    #[test]
+    fn all_variants() {
+      let mut variants = MediaFileOriginProductCategory::all_variants();
+      assert_eq!(variants.len(), 4);
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginProductCategory::FaceAnimator));
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginProductCategory::Unknown));
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginProductCategory::VoiceConversion));
+      assert_eq!(variants.pop_first(), Some(MediaFileOriginProductCategory::ZeroShotVoice));
+      assert_eq!(variants.pop_first(), None);
+    }
+  }
+
+  mod mechanical_checks {
+    use super::*;
+
+    #[test]
+    fn variant_length() {
+      use strum::IntoEnumIterator;
+      assert_eq!(MediaFileOriginProductCategory::all_variants().len(), MediaFileOriginProductCategory::iter().len());
+    }
+
+    #[test]
+    fn round_trip() {
+      for variant in MediaFileOriginProductCategory::all_variants() {
+        assert_eq!(variant, MediaFileOriginProductCategory::from_str(variant.to_str()).unwrap());
+        assert_eq!(variant, MediaFileOriginProductCategory::from_str(&format!("{}", variant)).unwrap());
+        assert_eq!(variant, MediaFileOriginProductCategory::from_str(&format!("{:?}", variant)).unwrap());
+      }
+    }
+  }
+}

crates/_types/enums/src/by_table/media_files/media_file_type.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/media_files/mod.rs

@@ -1,3 +1,4 @@
 pub mod media_file_origin_category;
 pub mod media_file_origin_model_type;
+pub mod media_file_origin_product_category;
 pub mod media_file_type;

crates/_types/enums/src/by_table/media_uploads/media_upload_source.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/media_uploads/media_upload_type.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/model_categories/model_type.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/trending_model_analytics/model_type.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/trending_model_analytics/window_name.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/tts_models/tts_model_type.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/tts_render_tasks/tts_render_status.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]
@@ -62,8 +61,8 @@ impl TtsRenderStatus {
 
 #[cfg(test)]
 mod tests {
-  use crate::test_helpers::assert_serialization;
   use crate::by_table::tts_render_tasks::tts_render_status::TtsRenderStatus;
+  use crate::test_helpers::assert_serialization;
 
   mod serde {
     use super::*;

crates/_types/enums/src/by_table/user_ratings/entity_type.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/voice_conversion_models/voice_conversion_model_type.rs

@@ -5,7 +5,6 @@ use strum::EnumCount;
 #[cfg(test)]
 use strum::EnumIter;
 
-
 /// Used in the `voice_conversion_models` table in `VARCHAR(32)` field `model_type`.
 ///
 /// DO NOT CHANGE VALUES WITHOUT A MIGRATION STRATEGY.

crates/_types/enums/src/by_table/voice_conversion_results/voice_conversion_media_token_type.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/web_rendition_targets/rendition_status.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/by_table/web_scraping_targets/scraping_status.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/common/job_status.rs

@@ -0,0 +1,147 @@
+use std::collections::BTreeSet;
+
+#[cfg(test)]
+use strum::EnumCount;
+#[cfg(test)]
+use strum::EnumIter;
+
+/// This is used in nearly every job system as an actual MySQL ENUM value:
+///
+///  - tts_download_job
+///  - tts_inference_job
+///  - w2l_download_job
+///  - w2l_inference_job
+///  - generic_download_job
+///  - (NOT generic_inference_job, which uses JobStatusPlus)
+///
+/// See the documentation on the table for usage.
+///
+/// DO NOT CHANGE VALUES WITHOUT A MIGRATION STRATEGY.
+
+#[cfg_attr(test, derive(EnumIter, EnumCount))]
+#[derive(Clone, Copy, PartialEq, Eq, Hash, Ord, PartialOrd, Serialize, Deserialize, sqlx::Type)]
+#[serde(rename_all = "snake_case")]
+#[sqlx(rename_all = "snake_case")]
+pub enum JobStatus {
+  Pending,
+  Started,
+  CompleteSuccess,
+  CompleteFailure,
+  AttemptFailed,
+  Dead,
+}
+
+// TODO(bt, 2022-12-21): This desperately needs MySQL integration tests!
+impl_enum_display_and_debug_using_to_str!(JobStatus);
+
+// NB: We can derive `sqlx::Type` instead of using `impl_mysql_enum_coders`
+
+impl JobStatus {
+  pub fn to_str(&self) -> &'static str {
+    match self {
+      Self::Pending => "pending",
+      Self::Started => "started",
+      Self::CompleteSuccess => "complete_success",
+      Self::CompleteFailure => "complete_failure",
+      Self::AttemptFailed => "attempt_failed",
+      Self::Dead => "dead",
+    }
+  }
+
+  pub fn from_str(job_status: &str) -> Result<Self, String> {
+    match job_status {
+      "pending" => Ok(Self::Pending),
+      "started" => Ok(Self::Started),
+      "complete_success" => Ok(Self::CompleteSuccess),
+      "complete_failure" => Ok(Self::CompleteFailure),
+      "attempt_failed" => Ok(Self::AttemptFailed),
+      "dead" => Ok(Self::Dead),
+      _ => Err(format!("invalid job_status: {:?}", job_status)),
+    }
+  }
+
+  pub fn all_variants() -> BTreeSet<Self> {
+    // NB: BTreeSet is sorted
+    // NB: BTreeSet::from() isn't const, but not worth using LazyStatic, etc.
+    BTreeSet::from([
+      Self::Pending,
+      Self::Started,
+      Self::CompleteSuccess,
+      Self::CompleteFailure,
+      Self::AttemptFailed,
+      Self::Dead,
+    ])
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::common::job_status::JobStatus;
+  use crate::test_helpers::assert_serialization;
+
+  mod explicit_checks {
+    use super::*;
+
+    #[test]
+    fn test_serialization() {
+      assert_serialization(JobStatus::Pending, "pending");
+      assert_serialization(JobStatus::Started, "started");
+      assert_serialization(JobStatus::CompleteSuccess, "complete_success");
+      assert_serialization(JobStatus::CompleteFailure, "complete_failure");
+      assert_serialization(JobStatus::AttemptFailed, "attempt_failed");
+      assert_serialization(JobStatus::Dead, "dead");
+    }
+
+    #[test]
+    fn to_str() {
+      assert_eq!(JobStatus::Pending.to_str(), "pending");
+      assert_eq!(JobStatus::Started.to_str(), "started");
+      assert_eq!(JobStatus::CompleteSuccess.to_str(), "complete_success");
+      assert_eq!(JobStatus::CompleteFailure.to_str(), "complete_failure");
+      assert_eq!(JobStatus::AttemptFailed.to_str(), "attempt_failed");
+      assert_eq!(JobStatus::Dead.to_str(), "dead");
+    }
+
+    #[test]
+    fn from_str() {
+      assert_eq!(JobStatus::from_str("pending").unwrap(), JobStatus::Pending);
+      assert_eq!(JobStatus::from_str("started").unwrap(), JobStatus::Started);
+      assert_eq!(JobStatus::from_str("complete_success").unwrap(), JobStatus::CompleteSuccess);
+      assert_eq!(JobStatus::from_str("complete_failure").unwrap(), JobStatus::CompleteFailure);
+      assert_eq!(JobStatus::from_str("attempt_failed").unwrap(), JobStatus::AttemptFailed);
+      assert_eq!(JobStatus::from_str("dead").unwrap(), JobStatus::Dead);
+    }
+
+    #[test]
+    fn all_variants() {
+      let mut variants = JobStatus::all_variants();
+      assert_eq!(variants.len(), 6);
+      assert_eq!(variants.pop_first(), Some(JobStatus::Pending));
+      assert_eq!(variants.pop_first(), Some(JobStatus::Started));
+      assert_eq!(variants.pop_first(), Some(JobStatus::CompleteSuccess));
+      assert_eq!(variants.pop_first(), Some(JobStatus::CompleteFailure));
+      assert_eq!(variants.pop_first(), Some(JobStatus::AttemptFailed));
+      assert_eq!(variants.pop_first(), Some(JobStatus::Dead));
+      assert_eq!(variants.pop_first(), None);
+    }
+  }
+
+  mod mechanical_checks {
+    use super::*;
+
+    #[test]
+    fn variant_length() {
+      use strum::IntoEnumIterator;
+      assert_eq!(JobStatus::all_variants().len(), JobStatus::iter().len());
+    }
+
+    #[test]
+    fn round_trip() {
+      for variant in JobStatus::all_variants() {
+        assert_eq!(variant, JobStatus::from_str(variant.to_str()).unwrap());
+        assert_eq!(variant, JobStatus::from_str(&format!("{}", variant)).unwrap());
+        assert_eq!(variant, JobStatus::from_str(&format!("{:?}", variant)).unwrap());
+      }
+    }
+  }
+}

crates/_types/enums/src/common/job_status_plus.rs

@@ -0,0 +1,160 @@
+use std::collections::BTreeSet;
+
+#[cfg(test)]
+use strum::EnumCount;
+#[cfg(test)]
+use strum::EnumIter;
+
+/// This is used in newer jobs (that add additional enum states)
+///
+///  - generic_inference_job
+///  - (no other jobs yet)
+///
+/// See the documentation on the table for usage.
+///
+/// DO NOT CHANGE VALUES WITHOUT A MIGRATION STRATEGY.
+
+#[cfg_attr(test, derive(EnumIter, EnumCount))]
+#[derive(Clone, Copy, PartialEq, Eq, Hash, Ord, PartialOrd, Serialize, Deserialize, sqlx::Type)]
+#[serde(rename_all = "snake_case")]
+#[sqlx(rename_all = "snake_case")]
+pub enum JobStatusPlus {
+  Pending,
+  Started,
+  CompleteSuccess,
+  CompleteFailure,
+  AttemptFailed,
+  Dead,
+  CancelledByUser,
+  CancelledBySystem,
+}
+
+// TODO(bt, 2022-12-21): This desperately needs MySQL integration tests!
+impl_enum_display_and_debug_using_to_str!(JobStatusPlus);
+
+// NB: We can derive `sqlx::Type` instead of using `impl_mysql_enum_coders`
+
+impl JobStatusPlus {
+  pub fn to_str(&self) -> &'static str {
+    match self {
+      Self::Pending => "pending",
+      Self::Started => "started",
+      Self::CompleteSuccess => "complete_success",
+      Self::CompleteFailure => "complete_failure",
+      Self::AttemptFailed => "attempt_failed",
+      Self::Dead => "dead",
+      Self::CancelledByUser => "cancelled_by_user",
+      Self::CancelledBySystem => "cancelled_by_system",
+    }
+  }
+
+  pub fn from_str(job_status: &str) -> Result<Self, String> {
+    match job_status {
+      "pending" => Ok(Self::Pending),
+      "started" => Ok(Self::Started),
+      "complete_success" => Ok(Self::CompleteSuccess),
+      "complete_failure" => Ok(Self::CompleteFailure),
+      "attempt_failed" => Ok(Self::AttemptFailed),
+      "dead" => Ok(Self::Dead),
+      "cancelled_by_user" => Ok(Self::CancelledByUser),
+      "cancelled_by_system" => Ok(Self::CancelledBySystem),
+      _ => Err(format!("invalid job_status: {:?}", job_status)),
+    }
+  }
+
+  pub fn all_variants() -> BTreeSet<Self> {
+    // NB: BTreeSet is sorted
+    // NB: BTreeSet::from() isn't const, but not worth using LazyStatic, etc.
+    BTreeSet::from([
+      Self::Pending,
+      Self::Started,
+      Self::CompleteSuccess,
+      Self::CompleteFailure,
+      Self::AttemptFailed,
+      Self::Dead,
+      Self::CancelledByUser,
+      Self::CancelledBySystem,
+    ])
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::common::job_status_plus::JobStatusPlus;
+  use crate::test_helpers::assert_serialization;
+
+  mod explicit_checks {
+    use super::*;
+
+    #[test]
+    fn test_serialization() {
+      assert_serialization(JobStatusPlus::Pending, "pending");
+      assert_serialization(JobStatusPlus::Started, "started");
+      assert_serialization(JobStatusPlus::CompleteSuccess, "complete_success");
+      assert_serialization(JobStatusPlus::CompleteFailure, "complete_failure");
+      assert_serialization(JobStatusPlus::AttemptFailed, "attempt_failed");
+      assert_serialization(JobStatusPlus::Dead, "dead");
+      assert_serialization(JobStatusPlus::CancelledByUser, "cancelled_by_user");
+      assert_serialization(JobStatusPlus::CancelledBySystem, "cancelled_by_system");
+    }
+
+    #[test]
+    fn to_str() {
+      assert_eq!(JobStatusPlus::Pending.to_str(), "pending");
+      assert_eq!(JobStatusPlus::Started.to_str(), "started");
+      assert_eq!(JobStatusPlus::CompleteSuccess.to_str(), "complete_success");
+      assert_eq!(JobStatusPlus::CompleteFailure.to_str(), "complete_failure");
+      assert_eq!(JobStatusPlus::AttemptFailed.to_str(), "attempt_failed");
+      assert_eq!(JobStatusPlus::Dead.to_str(), "dead");
+      assert_eq!(JobStatusPlus::CancelledByUser.to_str(), "cancelled_by_user");
+      assert_eq!(JobStatusPlus::CancelledBySystem.to_str(), "cancelled_by_system");
+    }
+
+    #[test]
+    fn from_str() {
+      assert_eq!(JobStatusPlus::from_str("pending").unwrap(), JobStatusPlus::Pending);
+      assert_eq!(JobStatusPlus::from_str("started").unwrap(), JobStatusPlus::Started);
+      assert_eq!(JobStatusPlus::from_str("complete_success").unwrap(), JobStatusPlus::CompleteSuccess);
+      assert_eq!(JobStatusPlus::from_str("complete_failure").unwrap(), JobStatusPlus::CompleteFailure);
+      assert_eq!(JobStatusPlus::from_str("attempt_failed").unwrap(), JobStatusPlus::AttemptFailed);
+      assert_eq!(JobStatusPlus::from_str("dead").unwrap(), JobStatusPlus::Dead);
+      assert_eq!(JobStatusPlus::from_str("cancelled_by_user").unwrap(), JobStatusPlus::CancelledByUser);
+      assert_eq!(JobStatusPlus::from_str("cancelled_by_system").unwrap(), JobStatusPlus::CancelledBySystem);
+    }
+
+    #[test]
+    fn all_variants() {
+      let mut variants = JobStatusPlus::all_variants();
+      assert_eq!(variants.len(), 8);
+      assert_eq!(variants.pop_first(), Some(JobStatusPlus::Pending));
+      assert_eq!(variants.pop_first(), Some(JobStatusPlus::Started));
+      assert_eq!(variants.pop_first(), Some(JobStatusPlus::CompleteSuccess));
+      assert_eq!(variants.pop_first(), Some(JobStatusPlus::CompleteFailure));
+      assert_eq!(variants.pop_first(), Some(JobStatusPlus::AttemptFailed));
+      assert_eq!(variants.pop_first(), Some(JobStatusPlus::Dead));
+      assert_eq!(variants.pop_first(), Some(JobStatusPlus::CancelledByUser));
+      assert_eq!(variants.pop_first(), Some(JobStatusPlus::CancelledBySystem));
+      assert_eq!(variants.pop_first(), None);
+    }
+  }
+
+  mod mechanical_checks {
+    use super::*;
+
+    #[test]
+    fn variant_length() {
+      use strum::IntoEnumIterator;
+      assert_eq!(JobStatusPlus::all_variants().len(), JobStatusPlus::iter().len());
+    }
+
+    #[test]
+    fn round_trip() {
+      for variant in JobStatusPlus::all_variants() {
+        // Test to_str(), from_str(), Display, and Debug.
+        assert_eq!(variant, JobStatusPlus::from_str(variant.to_str()).unwrap());
+        assert_eq!(variant, JobStatusPlus::from_str(&format!("{}", variant)).unwrap());
+        assert_eq!(variant, JobStatusPlus::from_str(&format!("{:?}", variant)).unwrap());
+      }
+    }
+  }
+}

crates/_types/enums/src/common/mod.rs

@@ -1,3 +1,5 @@
+pub mod job_status;
+pub mod job_status_plus;
 pub mod sqlite;
 pub mod visibility;
 pub mod vocoder_type;

crates/_types/enums/src/common/sqlite/awaitable_job_status.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/common/sqlite/chatbot_job_status.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/common/sqlite/skip_reason.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]

crates/_types/enums/src/common/sqlite/web_content_type.rs

@@ -1,6 +1,5 @@
 use serde::Deserialize;
 use serde::Serialize;
-
 #[cfg(test)]
 use strum::EnumCount;
 #[cfg(test)]
@@ -83,8 +82,8 @@ impl WebContentType {
 
 #[cfg(test)]
 mod tests {
-  use crate::test_helpers::assert_serialization;
   use crate::common::sqlite::web_content_type::WebContentType;
+  use crate::test_helpers::assert_serialization;
 
   mod serde {
     use super::*;

crates/_types/enums/src/common/visibility.rs

@@ -144,6 +144,7 @@ mod tests {
 
   mod serde_serialization {
     use crate::common::visibility::Visibility;
+
     use super::CompositeType;
 
     #[test]
@@ -166,6 +167,7 @@ mod tests {
 
   mod serde_deserialization {
     use crate::common::visibility::Visibility;
+
     use super::CompositeType;
 
     #[test]

crates/_types/reusable_types/Cargo.toml

@@ -18,7 +18,7 @@ path = "src/lib.rs"
 # (none)
 
 # External
-anyhow = "1.0.40"
+anyhow = "1.0.75"
 chrono = { version = "0.4.19", features = ["serde"] }
 
 # Serialization

crates/_types/tokens/src/files/mod.rs

@@ -1 +0,0 @@
-pub mod media_upload;

crates/_types/tokens/src/jobs/mod.rs

@@ -1,2 +0,0 @@
-pub mod download;
-pub mod inference;

crates/_types/tokens/src/lib.rs

@@ -1,8 +1,20 @@
 //! tokens
 //!
-//! The purpose of this library is to have a strongly-typed primary/foreign key system.
-//! Every Database or Redis key will have a type here.
-//! Well known keys will have short identifiers (eg. "user" is prefixed with "U:")
+//! The purpose of this library is to provide a strongly-typed primary/foreign key system with
+//! well-known identifier prefixes to aid in manual database debugging.
+//!
+//! 1) All table foreign keys have a strongly-typed (non-stringly typed) wrapper type,
+//!    eg. `MediaFileToken` for the `token` field in the `media_files` table or `UserToken` for
+//!    the `token` field in the `users` table.
+//!
+//! 2) All such keys should have a well-defined constant prefix. Though tokens should be 100%
+//!    opaque to the user and the code, having well-defined prefixes can greatly aid when
+//!    debugging systems. Every table's primary key should have a unique prefix that identifies
+//!    that key, eg. `comment_` for the `comments` table or `m_` for the `media_files` table.
+//!
+//! This library does not directly introduce MySQL dependencies, so these strongly-typed token
+//! types can be used in other crates, eg. for defining HTTP APIs and returning strongly-typed
+//! tokens to the user.
 //!
 
 // Never allow these
@@ -39,8 +51,4 @@ mod macros;
 
 pub (crate) mod prefixes;
 
-pub mod files;
-pub mod jobs;
 pub mod tokens;
-pub mod users;
-pub mod voice_conversion;

crates/_types/tokens/src/prefixes.rs

@@ -13,6 +13,8 @@ use strum::EnumIter;
 #[derive(Debug, Copy, Clone, Eq, PartialEq)]
 #[cfg_attr(test, derive(EnumIter, EnumCount))]
 pub(crate) enum EntityType {
+  /// AVTs are not stored as primary keys in any table, but an index in many tables.
+  AnonymousVisitorTracking,
   AuditLog,
   Comment,
   DownloadJob,
@@ -24,15 +26,20 @@ pub(crate) enum EntityType {
   TtsModel,
   TtsRenderTask, // NB: aichatbot / sqlite
   User,
+  UserDeprecatedDoNotUse, // NB: Users prior to 2023-10-24 used this prefix.
   VoiceConversionModel,
   VoiceConversionResult,
   W2lTemplate,
+  ZsVoice,
+  ZsVoiceDataset,
+  ZsVoiceDatasetSample,
 }
 
 impl EntityType {
 
   pub fn prefix(self) -> &'static str {
     match self {
+      Self::AnonymousVisitorTracking => "avt_",
       Self::AuditLog => "audit_",
       Self::Comment => "comment_",
       Self::DownloadJob => "jdown_", // NB: Was "JGUP:"
@@ -43,21 +50,27 @@ impl EntityType {
       Self::NewsStory => "news_story_",
       Self::TtsModel => "TM:", // NB: Old-style prefix, do not use for future tokens.
       Self::TtsRenderTask => "tts_task_",
-      Self::User => "U:", // NB: Old-style prefix, do not use for future tokens.
+      Self::User => "user_",
+      Self::UserDeprecatedDoNotUse => "U:", // NB: Users prior to 2023-10-24 used this prefix.
       Self::VoiceConversionModel => "vcm_",
       Self::VoiceConversionResult => "vcr_",
       Self::W2lTemplate => "WT:", // NB: Old-style prefix, do not use for future tokens.
+      Self::ZsVoice => "zsv_",
+      Self::ZsVoiceDataset => "zsd_",
+      Self::ZsVoiceDatasetSample => "zss_",
     }
   }
 }
 
 #[cfg(test)]
 mod tests {
-  use crate::prefixes::EntityType;
   use std::collections::HashSet;
+
   use strum::EnumCount;
   use strum::IntoEnumIterator;
 
+  use crate::prefixes::EntityType;
+
   #[test]
   fn test_all_prefixes_are_unique() {
     let entities = EntityType::iter()

crates/_types/tokens/src/tokens/anonymous_visitor_tracking.rs

@@ -0,0 +1,16 @@
+use std::fmt::Debug;
+
+use serde::Deserialize;
+use serde::Serialize;
+
+use crate::prefixes::EntityType;
+
+/// A rare instance that is *not* a primary key.
+/// These are generated as cookies to track users for a better logged out experience and for analytics.
+/// We use these as indices into several columns (ML result types, uploads, etc.)
+#[derive(Clone, PartialEq, Eq, PartialOrd, Ord, Hash, sqlx::Type, Debug, Serialize, Deserialize)]
+#[sqlx(transparent)]
+pub struct AnonymousVisitorTrackingToken(pub String);
+
+impl_string_token!(AnonymousVisitorTrackingToken);
+impl_crockford_generator!(AnonymousVisitorTrackingToken, 32usize, EntityType::AnonymousVisitorTracking, CrockfordLower);