update workspace

cache sessions + purge cache on billing, edit profile, etc.

.env

@@ -14,10 +14,14 @@ MIGRATION_DIRECTORY=_sql/migrations/
 # See https://github.com/launchbadge/sqlx/tree/master/sqlx-cli#readme
 #SQLX_OFFLINE=true
 
-# We can't serve HTTPS locally without self-signed certificate nonsense.
-COOKIE_SECURE=false
-COOKIE_DOMAIN=.jungle.horse
-WEBSITE_HOMEPAGE_REDIRECT=http://jungle.horse
+# NB(bt, 2023-03-20):
+#  These are used in storyteller-web, image-generation-service, and websocket-gateway.
+#  I'm moving them to env configs that live under storyteller-web/.env.
+#  This will break the other services for now.
+# # We can't serve HTTPS locally without self-signed certificate nonsense.
+# COOKIE_SECURE=false
+# COOKIE_DOMAIN=.jungle.horse
+# WEBSITE_HOMEPAGE_REDIRECT=http://jungle.horse
 
 # Bucket config (non-secret)
 REGION_NAME=us-east1
@@ -71,14 +75,3 @@ TTS_HIFIGAN_SUPERRES_VOCODER_MODEL_FILENAME=hifigan_superres_twilight_33000.pt
 
 # Development key. Permissible to check in.
 NEWRELIC_API_KEY=c402be1c4c4216ec83d0cfc0851520cfFFFFNRAL
-
-# Stripe configs
-# All secrets are development and okay to check in
-STRIPE_CHECKOUT_SUCCESS_URL_PATH=/checkout_success
-STRIPE_CHECKOUT_CANCEL_URL_PATH=/checkout_cancel
-STRIPE_PORTAL_RETURN_URL_PATH=/portal_success
-STRIPE_PORTAL_DEFAULT_CONFIG_ID=bpc_1LyPPREU5se17MekYiViZF12
-# STRIPE_PORTAL_DEFAULT_CONFIG_ID=bpc_1M4jc3EU5se17MekmgdS4dDJ # Production (not a secret)
-STRIPE_PUBLISHABLE_KEY=pk_test_51KZvWXEU5se17Mekx6e0gw32ovDF78k2eeaGctLIs8oQVxN6DerEp1BDqh6gLMYKli4VsTZoVGHoxq5RmwKMlxtP00r59UEEad
-STRIPE_SECRET_KEY=sk_test_51KZvWXEU5se17Mek11ZdnWr4frnyw7tWOPuvl3pRzL290zvkT6KqcbGUl3fcvc3hA8oH6lWXz2kGWXXUlNcCuDxB003uVT9JX0
-STRIPE_SECRET_WEBHOOK_SIGNING_KEY=whsec_40b60bd75768052532b63dbeefa7a1b163e182c3e657fb27fa906ef49a2cd59f

.env-download-job

@@ -1,25 +0,0 @@
-# Environmental configurations specific to `download-job`.
-# These are defaulted for dev and are easy to overwrite in production with K8s environment variables.
-# !! DOES *NOT* CONTAIN SECRETS !!
-
-WEB_DOWNLOADER_COMMAND=./download_internet_file.py
-WEB_DOWNLOADER_MAYBE_VENV_SCRIPT=./python/bin/activate
-WEB_DOWNLOADER_MAYBE_DOCKER_IMAGE=d73f28ce3ff6
-
-SOFTVC_MODEL_CHECK_ROOT_DIRECTORY=/
-SOFTVC_MODEL_CHECK_MAYBE_VENV_COMMAND="source python/bin/activate"
-SOFTVC_MODEL_CHECK_MAYBE_DOCKER_IMAGE=a79106991ce3
-
-TACOTRON_MODEL_CHECK_ROOT_DIRECTORY=/models/tts
-TACOTRON_MODEL_CHECK_MAYBE_VENV_COMMAND="source python/bin/activate"
-TACOTRON_MODEL_CHECK_COMMAND=./vocodes_model_check_tacotron.py
-TACOTRON_MODEL_CHECK_MAYBE_DOCKER_IMAGE=5642d0fd7fc1
-
-HIFIGAN_SOFTVC_MODEL_CHECK_ROOT_DIRECTORY=/
-HIFIGAN_SOFTVC_MODEL_CHECK_MAYBE_VENV_COMMAND="source python/bin/activate"
-HIFIGAN_SOFTVC_MODEL_CHECK_MAYBE_DOCKER_IMAGE=a79106991ce3
-
-HIFIGAN_MODEL_CHECK_ROOT_DIRECTORY=/models/tts
-HIFIGAN_MODEL_CHECK_MAYBE_VENV_COMMAND="source python/bin/activate"
-HIFIGAN_MODEL_CHECK_COMMAND=./vocodes_model_check_hifigan.py
-HIFIGAN_MODEL_CHECK_MAYBE_DOCKER_IMAGE=5642d0fd7fc1

.github/workflows/docker-publish-cpu.yml

@@ -2,7 +2,7 @@
 # https://github.com/actions/starter-workflows/ci/docker-publish.yml
 # If using base images, don't forget to add READ access, eg. from a package settings page like:
 # https://github.com/orgs/storytold/packages/container/docker-base-images-rust-ssl/settings/actions_access
-name: Publish Docker Container to GHCR
+name: "[CPU] Build GHCR Docker artifact"
 on:
   push:
     branches:
@@ -64,6 +64,7 @@ jobs:
           username: ${{ github.actor }} # username: ${{ github.repository_owner }}
           password: ${{ secrets.GITHUB_TOKEN }}
           image_name: storytold/storyteller-rust
+          dockerfile: service_cpu.Dockerfile
           # NB: The `git_push_tag: true` option does not work as expected, so we put the sha in `image_tag`:
           # https://github.com/whoan/docker-build-with-cache-action/issues/72
           image_tag: latest,${{ github.sha }},${{ steps.vars.outputs.sha_short }}

.github/workflows/docker-publish-gpu.yml

@@ -0,0 +1,113 @@
+# Adapted from
+# https://github.com/actions/starter-workflows/ci/docker-publish.yml
+# If using base images, don't forget to add READ access, eg. from a package settings page like:
+# https://github.com/orgs/storytold/packages/container/docker-base-images-rust-ssl/settings/actions_access
+name: "[GPU] Build GHCR Docker artifact"
+on:
+  push:
+    branches:
+      # NB: Default-branch doesn't work, despite Github's documentation.
+      #- $default-branch
+      - master
+env:
+  IMAGE_NAME: storyteller-rust-gpu
+
+jobs:
+  # Push image to GitHub Packages.
+  # See also https://docs.docker.com/docker-hub/builds/
+  # See also https://depot.dev/blog/docker-layer-caching-in-github-actions
+  # See also https://evilmartians.com/chronicles/build-images-on-github-actions-with-docker-layer-caching
+  build-docker-image:
+    name: "Build Docker Image"
+    runs-on: ubuntu-22.04 # runs-on: ubuntu-latest
+    if: github.event_name == 'push'
+    permissions:
+      packages: write
+      contents: read
+
+    steps:
+      - uses: actions/checkout@v3
+
+      # - uses: docker/setup-buildx-action@v1
+      #   name: Set up Docker Buildx
+
+      - uses: docker/login-action@v1
+        name: Login to GitHub Container Registry
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }} # username: ${{ github.repository_owner }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Calculate short SHA
+        id: vars
+        # echo "sha_short=$(git rev-parse --short HEAD)" >> $GITHUB_OUTPUT
+        run: |
+          echo "sha_short=$(echo ${{ github.sha }} | cut -c1-12)" >> $GITHUB_OUTPUT
+
+      - name: Emit short SHA
+        run: |
+          echo "Full SHA: ${{ github.sha }}"
+          echo "Short SHA: ${{ steps.vars.outputs.sha_short }}"
+
+      #- uses: actions/cache@v2
+      #  name: Cache Docker layers
+      #  with:
+      #    path: /tmp/.buildx-cache
+      #    key: ${{ runner.os }}-buildx-${{ github.sha }}
+      #    restore-keys: |
+      #      ${{ runner.os }}-buildx-
+
+      # Docs for this action are at https://github.com/marketplace/actions/build-docker-images-using-cache
+      - uses: whoan/docker-build-with-cache-action@v6
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }} # username: ${{ github.repository_owner }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+          image_name: storytold/storyteller-rust-gpu
+          dockerfile: service_gpu.Dockerfile
+          # NB: The `git_push_tag: true` option does not work as expected, so we put the sha in `image_tag`:
+          # https://github.com/whoan/docker-build-with-cache-action/issues/72
+          image_tag: latest,${{ github.sha }},${{ steps.vars.outputs.sha_short }}
+          # NB: Inject the Git SHA into the container so the servers can emit their own version:
+          build_extra_args: "--build-arg=GIT_SHA=${{ github.sha }}"
+
+      # - name: Build image
+      #   run: |
+      #     SHORT_SHA=$(echo ${GITHUB_SHA} | cut -c1-12)
+      #     docker build . --file Dockerfile --tag $IMAGE_NAME --label "runnumber=${GITHUB_RUN_ID}" --build-arg GIT_SHA=${SHORT_SHA}
+
+      # - name: Push image to GitHub Container Registry
+      #   run: |
+      #     IMAGE_ID=ghcr.io/${{ github.repository_owner }}/$IMAGE_NAME
+
+      #     # Change all uppercase to lowercase
+      #     IMAGE_ID=$(echo $IMAGE_ID | tr '[A-Z]' '[a-z]')
+
+      #     # Strip git ref prefix from version
+      #     VERSION=$(echo "${{ github.ref }}" | sed -e 's,.*/\(.*\),\1,')
+
+      #     # Strip "v" prefix from tag name
+      #     [[ "${{ github.ref }}" == "refs/tags/"* ]] && VERSION=$(echo $VERSION | sed -e 's/^v//')
+
+      #     # Use Docker `latest` tag convention
+      #     [ "$VERSION" == "$default-branch" ] && VERSION=latest
+
+      #     echo IMAGE_ID=$IMAGE_ID
+      #     echo VERSION=$VERSION
+
+      #     docker tag $IMAGE_NAME $IMAGE_ID:$VERSION
+      #     docker push $IMAGE_ID:$VERSION
+
+      #     SHORT_SHA=$(echo ${GITHUB_SHA} | cut -c1-12)
+
+      #     docker tag $IMAGE_NAME $IMAGE_ID:$SHORT_SHA
+      #     docker push $IMAGE_ID:$SHORT_SHA
+
+      - name: Cleanup / Debug
+        # This job will run even on failures to help with debugging.
+        if: always()
+        run: |
+          echo "Report at final github action step:"
+          echo "Disk usage:"
+          pwd
+          du -hsc * | sort -hr

.gitignore

@@ -13,6 +13,10 @@ secrets.yaml
 _develop/tools/.secrets
 _develop/tools/db_secrets.toml
 twitch_secrets.toml
+crates/service/web/storyteller_web/config/storyteller-web.development-secrets.env
+
+# New env secrets (per-app)
+*-secrets.env
 
 # Rust build directory
 target/
@@ -49,3 +53,7 @@ models/tacotron2/python/
 # Stuff I want to keep around, but not commit
 scraps/
 
+# Windows build files
+# These should be built by the platform
+windows/sqlite/sqlite3.exp
+windows/sqlite/sqlite3.lib

.idea/dictionaries/bt.xml

@@ -8,6 +8,7 @@
       <w>crockford</w>
       <w>culted</w>
       <w>dalle</w>
+      <w>demuxer</w>
       <w>downvote</w>
       <w>fakeyou</w>
       <w>ffprobe</w>
@@ -16,16 +17,22 @@
       <w>greenlit</w>
       <w>hanashi</w>
       <w>hifigan</w>
+      <w>hparams</w>
+      <w>huggingface</w>
+      <w>pathbuf</w>
       <w>replayable</w>
       <w>retryably</w>
       <w>scrollback</w>
       <w>softvc</w>
+      <w>sql's</w>
       <w>symphonia</w>
       <w>tacotron</w>
       <w>techcrunch</w>
       <w>transcoded</w>
       <w>transcodings</w>
+      <w>unpermitted</w>
       <w>upserting</w>
+      <w>vits</w>
       <w>vocoded</w>
       <w>vocoder</w>
       <w>vocoders</w>

Cargo.toml

@@ -1,4 +1,5 @@
 [workspace]
+resolver="2"
 
 members = [
   # Types: common strongly-typed tokens, ids, payloads, etc.
@@ -15,26 +16,30 @@ members = [
 
   # 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/clients/cloud_storage",
-  "crates/lib/clients/fakeyou_client",
-  "crates/lib/clients/google_drive_common",
-  "crates/lib/clients/twitch_common",
   "crates/lib/concurrency",
-  "crates/lib/config", # TODO(bt, 2023-02-13): Deprecate or repurpose
-  "crates/lib/container_common", # TODO(bt, 2023-02-13): Pull apart and rehome pieces
   "crates/lib/crockford",
-  "crates/lib/database_queries", # TODO(bt2, 2023-02-13): Rename "mysql_queries", though this will not live beyond a single DB
   "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/errors",
-  "crates/lib/files",
-  "crates/lib/hashing",
-  "crates/lib/http_server_common",
+  "crates/lib/files/filesys",
+  "crates/lib/files/hashing",
+  "crates/lib/files/media",
+  "crates/lib/files/mimetypes",
   "crates/lib/idempotency",
   "crates/lib/jobs_common",
-  "crates/lib/media",
+  "crates/lib/mysql_queries",
   "crates/lib/redis_common",
+  "crates/lib/server_environment",
   "crates/lib/sqlite_queries",
   "crates/lib/sqlx_mysql_helpers",
   "crates/lib/subprocess_common",
@@ -45,6 +50,7 @@ members = [
 
   # Programs: Desktop apps
   "crates/programs/aichatbot-sidecar",
+  "crates/programs/bevy-experiments",
 
   # Services: API servers, jobs, workers, etc.
   "crates/service/job/analytics_job",

_develop/tools/.gitignore

@@ -0,0 +1,6 @@
+# Secrets
+db_secrets.toml
+
+# Python Virtual Env
+python/
+

_develop/tools/find_101soundboards_accounts.py

@@ -0,0 +1,100 @@
+#!/usr/bin/env python
+
+# FIXME: Needs packages 'toml' and 'mysql-connector-python'. Put in requirements.txt.
+
+import time
+import toml
+import mysql.connector
+
+DB_SECRETS_FILE = 'db_secrets.toml'
+
+# TODO: Share db-related setup when more scripts are added.
+
+def read_database_secrets(secrets_file = DB_SECRETS_FILE):
+    secrets = None
+    with open(secrets_file, "r") as f:
+        secrets = toml.load(f)
+
+    required_keys = ['database', 'hostname', 'port', 'password', 'username']
+    for key in required_keys:
+        if key not in secrets:
+            raise Exception(f"Required key '#{key}' not present")
+
+    return secrets
+
+def connect_to_mysql():
+    secrets = read_database_secrets()
+    connection = mysql.connector.connect(
+        host = secrets['hostname'],
+        port = secrets['port'],
+        user = secrets['username'],
+        password = secrets['password'],
+        database = secrets['database'],
+    )
+    return connection
+
+connection = connect_to_mysql()
+cursor = connection.cursor()
+
+SAFE_ID = 76000000 # Reasonably completed on 2022-06-24
+LIMIT = 100000
+
+#def get_record_count():
+#    query = f"SELECT COUNT(*) FROM tts_inference_jobs"
+#    cursor.execute(query)
+#    result = cursor.fetchall()
+#    print(result)
+#    return result[0]
+#
+##total_count = get_record_count()
+
+#total_count = 75082057
+total_count = 21925602
+
+def get_user_tokens(cursor):
+    query = f"""
+		SELECT
+			r.maybe_creator_user_token, r.creator_ip_address
+		FROM
+			tts_results as r
+				LEFT OUTER JOIN
+			users as u
+			ON
+					r.maybe_creator_user_token = u.token
+		WHERE
+				r.raw_inference_text LIKE '%spidermant%'
+		AND r.created_at > ( CURDATE() - INTERVAL 1 HOUR )
+    """
+    cursor.execute(query)
+    results = cursor.fetchall()
+
+    user_tokens = [row[0] for row in results]
+    user_tokens = [token for token in user_tokens if token is not None]
+    user_tokens = set(user_tokens)
+
+    ip_addresses = [row[1] for row in results]
+    ip_addresses = [ip for ip in ip_addresses if ip is not None]
+    ip_addresses = set(ip_addresses)
+
+    return (user_tokens, ip_addresses)
+
+(user_tokens, ip_addresses) = get_user_tokens(cursor)
+
+print(user_tokens)
+print(len(user_tokens))
+print(len(ip_addresses))
+
+with open('101_user_tokens.txt', 'w') as f:
+    for user_token in user_tokens:
+        f.write(user_token)
+        f.write("\n")
+
+with open('101_ip_addresses.txt', 'w') as f:
+    for ip in ip_addresses:
+        f.write(ip)
+        f.write("\n")
+
+
+
+
+

_develop/tools/requirements.txt

@@ -0,0 +1,3 @@
+mysql-connector-python==8.0.32
+protobuf==3.20.3
+toml==0.10.2

_sql/actions/generic_inference_jobs.sql

@@ -0,0 +1,9 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Kill routed jobs
+update generic_inference_jobs set status = 'dead' where maybe_routing_tag IS NOT NULL limit 100;
+
+-- Get pending jobs
+select count(*) from generic_inference_jobs where status = 'pending';

_sql/actions/tts_inference_jobs.sql

@@ -0,0 +1,24 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Kill pending jobs
+update tts_inference_jobs set status = 'dead' where status = 'pending';
+
+-- Kill all low priority pending, waiting, and in-progress jobs
+update tts_inference_jobs
+set status = 'dead'
+where status IN ('pending', 'started', 'attempt_failed')
+  and priority_level IN (0, 1);
+
+-- Kill all pending jobs
+update tts_inference_jobs
+set status = 'dead'
+where status = 'pending';
+
+-- Kill all waiting and in-progress jobs
+update tts_inference_jobs
+set status = 'dead'
+where status IN ('pending', 'started', 'attempt_failed');
+
+

_sql/actions/tts_results.sql

@@ -0,0 +1,11 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Delete TTS results for a single model
+-- Do this in short batches so a lock isn't held for prohibitively long.
+update tts_results
+set mod_deleted_at = NOW()
+where model_token = 'TM:ztt5s1be5tq6'
+  and mod_deleted_at IS NULL
+    limit 5000;

_sql/actions/users.sql

@@ -0,0 +1,35 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Upgrade top TTS model uploaders to a loyalty plan
+update users
+set maybe_loyalty_program_key = 'fakeyou_contributor'
+where token IN (
+    select creator_user_token from
+        (
+            select count(*) as upload_count,
+                   creator_user_token
+            from tts_models
+            group by creator_user_token
+            having upload_count >= 10
+            order by upload_count desc
+        ) as t
+);
+
+-- Upgrade top VC model uploaders to a loyalty plan
+update users
+set maybe_loyalty_program_key = 'fakeyou_contributor'
+where token IN (
+    select creator_user_token from
+        (
+            select count(*) as upload_count,
+                   creator_user_token
+            from voice_conversion_models
+            group by creator_user_token
+            having upload_count >= 10
+            order by upload_count desc
+        ) as t
+);
+
+

_sql/actions/voice_conversion_models.sql

@@ -0,0 +1,55 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Remove temporary test models
+update voice_conversion_models
+set
+    mod_deleted_at = NOW(),
+    maybe_mod_comments = 'temporary test model'
+where token IN (
+    'vcm_jt9mdz9xjjcb',
+    'vcm_4xzkzdhs4kk5',
+    'vcm_502wkt2pca3y'
+);
+
+-- Remove broken models
+update voice_conversion_models
+set
+    mod_deleted_at = NOW(),
+    maybe_mod_comments = 'broken model'
+where token IN (
+    'vcm_982y9g6v3jrg',
+    'vcm_16rpyehf7f0a',
+    'vcm_8vjtz3dbnw17',
+    'vcm_tc9mwr60cjth',
+    'vcm_bshsyp01w9vg',
+    'vcm_31whj8apmqj7',
+    'vcm_84qa0an44p8p',
+    'vcm_9xx4v5qz5fbg',
+    'vcm_7zp2b2hs2at2',
+    'vcm_zhvk0mnnb7zp',
+    'vcm_1vc4scq9qp6m',
+    'vcm_sr1b54mwy0ym',
+    'vcm_qx96y8aasydk'
+);
+
+-- Remove duplicate models
+update voice_conversion_models
+set
+    mod_deleted_at = NOW(),
+    maybe_mod_comments = 'duplicate model'
+where token IN (
+    'vcm_6w21em1bs7q3',
+    'vcm_sbhvyj7926w4',
+    'vcm_anv469g89adz',
+    'vcm_xsx5yx585c7x',
+    'vcm_dpahzjjpkf18',
+    'vcm_y09dqe6zgwyk',
+    'vcm_xrp96dwgppv2'
+);
+
+
+
+
+

_sql/actions/w2l_inference_jobs.sql

@@ -0,0 +1,19 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Kill pending jobs
+update w2l_inference_jobs set status='dead' where status IN ('pending', 'started', 'attempt_failed');
+
+-- Unstick old jobs that might be stuck as "started"
+update w2l_template_upload_jobs
+set status='dead'
+where status='started'
+  and created_at < ( CURDATE() - INTERVAL 1 DAY );
+
+-- Kill pending jobs with tiktok URLs:
+-- These need to be blocked in the server and job directly.
+update w2l_template_upload_jobs
+set status='dead'
+where status='pending'
+  and download_url LIKE '%tiktok%';

_sql/analytics/generic_inference/generic_inference_jobs.sql

@@ -0,0 +1,80 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- See how long recent jobs are taking
+select
+    id,
+    maybe_input_source_token,
+    assigned_worker,
+    assigned_cluster,
+    success_execution_millis / 1000 / 60 as minutes
+from generic_inference_jobs
+where status != 'pending'
+and success_execution_millis IS NOT NULL
+order by id desc
+limit 50;
+
+
+-- 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
+--   sections where models / files get downloaded.
+select *
+from (
+         select
+             jobs.created_at,
+             jobs.first_started_at,
+             jobs.id,
+             jobs.status,
+             jobs.attempt_count as attempts,
+             trim(replace(replace(jobs.internal_debugging_failure_reason, "\n", ""), "\r", ""))
+                 as i_failure_reason,
+             -- u.username,
+             jobs.assigned_worker,
+             jobs.assigned_cluster,
+             jobs.maybe_input_source_token,
+             jobs.maybe_model_token,
+             m.media_source,
+             m.maybe_original_mime_type as mime_type,
+             TRUNCATE(jobs.success_execution_millis / 1000 / 60, 2) as execution_mins,
+             TRUNCATE(jobs.success_inference_execution_millis / 1000 / 60, 2) as inference_mins,
+             TRUNCATE((jobs.success_execution_millis - jobs.success_inference_execution_millis) / 1000 / 60, 2) as extra_mins,
+             TRUNCATE(m.original_duration_millis / 1000 / 60, 2) as input_mins,
+             jobs.success_execution_millis / m.original_duration_millis as ratio
+         from generic_inference_jobs AS jobs
+                  left join users AS u on
+                 u.token = jobs.maybe_creator_user_token
+                  left join media_uploads AS m on
+                 m.token = jobs.maybe_input_source_token
+         where
+             jobs.status != 'pending'
+        AND jobs.created_at > NOW() - INTERVAL 20 MINUTE
+         order by id desc
+             limit 5000
+     ) as t
+order by execution_mins desc;
+
+order by assigned_worker desc, first_started_at asc;
+
+order by first_started_at asc;
+
+order by assigned_worker desc, execution_mins desc;
+
+
+-- Get pending so-vits-svc jobs
+select count(*)
+from generic_inference_jobs
+where maybe_model_type = 'so_vits_svc'
+  and status='pending';
+
+
+-- Get jobs that have routing tags
+select
+    id,
+    maybe_input_source_token,
+    maybe_creator_user_token,
+    maybe_routing_tag,
+    status
+from generic_inference_jobs
+where maybe_routing_tag IS NOT NULL
+    limit 10;

_sql/analytics/generic_inference/media_uploads.sql

@@ -0,0 +1,50 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+select
+    mu.id,
+    mu.token,
+    u.username,
+    mu.media_source,
+    maybe_original_filename as filename,
+    original_file_size_bytes as size_bytes,
+    original_duration_millis as duration_millis,
+    maybe_original_mime_type as mime_type
+from media_uploads as mu
+    left join users AS u
+        on u.token = mu.maybe_creator_user_token
+order by id desc limit 10;
+
+-- Analyze all media
+select
+    mu.id,
+    mu.token,
+    u.username,
+    mu.media_source,
+    maybe_original_filename as filename,
+    original_file_size_bytes as size_bytes,
+    original_duration_millis as duration_millis,
+    maybe_original_mime_type as mime_type
+from media_uploads as mu
+         left join users AS u
+                   on u.token = mu.maybe_creator_user_token
+order by id desc
+    limit 40;
+
+-- Analyze broken media (no duration)
+select
+    mu.id,
+    mu.token,
+    u.username,
+    mu.media_source,
+    maybe_original_filename as filename,
+    original_file_size_bytes as size_bytes,
+    original_duration_millis as duration_millis,
+    maybe_original_mime_type as mime_type
+from media_uploads as mu
+    left join users AS u
+        on u.token = mu.maybe_creator_user_token
+where mu.original_duration_millis = 0
+order by id desc
+limit 40;

_sql/analytics/tts/tts_inference_jobs.sql

@@ -2,9 +2,6 @@
 -- noinspection SqlNoDataSourceInspectionForFile
 -- noinspection SqlResolveForFile
 
--- Kill pending jobs
-update tts_inference_jobs set status = 'dead' where status = 'pending';
-
 -- Total pending jobs
 select count(*)
 from tts_inference_jobs
@@ -32,17 +29,6 @@ group by model_token
 order by attempts desc
     limit 50;
 
-
--- Kill pending jobs
-update tts_inference_jobs
-set status = 'dead'
-where status = 'pending';
-
--- Kill all waiting and in-progress jobs
-update tts_inference_jobs
-set status = 'dead'
-where status IN ('pending', 'started', 'attempt_failed');
-
 -- Sample the pending inference text
 select creator_ip_address,
        model_token,

_sql/analytics/tts/tts_models.sql

@@ -12,9 +12,9 @@ select
     m.title,
     u.username,
     r.use_count,
+    m.created_at,
     m.user_deleted_at,
-    m.mod_deleted_at,
-    m.created_at
+    m.mod_deleted_at
 from (
          select model_token, count(*) as use_count
          from tts_results
@@ -63,7 +63,41 @@ from (
     join users as u
     on u.token = m.creator_user_token
 order by r.use_count desc
-    limit 25;
+    limit 100;
+
+
+--
+-- Top models binned by IP address.
+-- Should give us a look at demand independent of model quality / repeated use.
+--
+select
+    m.token,
+    m.title,
+    m.user_ratings_positive_count / m.user_ratings_total_count as rating,
+    m.text_pipeline_type,
+    m.ietf_language_tag,
+    u.username,
+    r.use_count,
+    m.created_at,
+    m.user_deleted_at,
+    m.mod_deleted_at
+from (
+        select model_token, count(*) as use_count
+        from (
+            select model_token, creator_ip_address
+            from tts_results
+            where created_at > (CURDATE() - INTERVAL 12 HOUR)
+            group by model_token, creator_ip_address
+        ) as x
+        group by model_token
+     ) as r
+         join tts_models as m
+              on m.token = r.model_token
+         join users as u
+              on u.token = m.creator_user_token
+         where m.ietf_language_tag IN ('en', 'en-US', 'en-AU', 'en-CA', 'en-GB')
+order by r.use_count desc
+    limit 100;
 
 --
 -- Top 20 ***NEW*** models to share in #announcements
@@ -73,12 +107,13 @@ select
     m.token,
     m.text_pipeline_type,
     m.ietf_language_tag,
+    m.creator_set_visibility,
     m.title,
     u.username,
     r.use_count,
+    m.created_at,
     m.user_deleted_at,
-    m.mod_deleted_at,
-    m.created_at
+    m.mod_deleted_at
 from (
          select model_token, count(*) as use_count
          from tts_results
@@ -90,7 +125,7 @@ from (
          join users as u
               on u.token = m.creator_user_token
 where m.created_at > ( CURDATE() - INTERVAL 31 DAY )
-  AND u.username IN ( 'vegito1089', 'justinjohn0306')
+  -- AND u.username IN ( 'vegito1089', 'justinjohn0306')
 order by r.use_count desc
     limit 100;
 
@@ -105,9 +140,9 @@ select
     m.title,
     u.username,
     r.use_count,
+    m.created_at,
     m.user_deleted_at,
-    m.mod_deleted_at,
-    m.created_at
+    m.mod_deleted_at
 from (
     select model_token, count(*) as use_count
     from tts_results
@@ -138,9 +173,9 @@ select
     m.ietf_language_tag,
     u.username,
     r.use_count,
+    m.created_at,
     m.user_deleted_at,
-    m.mod_deleted_at,
-    m.created_at
+    m.mod_deleted_at
 from (
          select model_token, count(*) as use_count
          from tts_results
@@ -164,9 +199,9 @@ select
     m.ietf_language_tag,
     u.username,
     r.use_count,
+    m.created_at,
     m.user_deleted_at,
-    m.mod_deleted_at,
-    m.created_at
+    m.mod_deleted_at
 from (
     select model_token, count(*) as use_count
     from tts_results
@@ -192,9 +227,9 @@ select
     m.ietf_language_tag,
     u.username,
     r.use_count,
+    m.created_at,
     m.user_deleted_at,
-    m.mod_deleted_at,
-    m.created_at
+    m.mod_deleted_at
 from (
     select model_token, count(*) as use_count
     from tts_results
@@ -224,8 +259,8 @@ select
     m.ietf_language_tag,
     u.username,
     r.use_count,
-    m.user_deleted_at,
-    m.created_at
+    m.created_at,
+    m.user_deleted_at
 from (
          select model_token, count(*) as use_count
          from tts_results
@@ -256,9 +291,9 @@ select
     m.ietf_language_tag,
     u.username,
     r.use_count,
+    m.created_at,
     m.user_deleted_at,
-    m.mod_deleted_at,
-    m.created_at
+    m.mod_deleted_at
 from (
          select model_token, count(*) as use_count
          from tts_results
@@ -300,3 +335,31 @@ where m.mod_deleted_at > ( CURDATE() - INTERVAL 14 DAY )
 order by r.use_count desc
     limit 500;
 
+--
+-- Non-public models
+--
+select
+    m.token,
+    m.title,
+    m.text_pipeline_type,
+    m.ietf_language_tag,
+    m.creator_set_visibility,
+    u.username,
+    r.use_count,
+    m.created_at,
+    m.user_deleted_at,
+    m.mod_deleted_at
+from (
+         select model_token, count(*) as use_count
+         from tts_results
+         where created_at > ( CURDATE() - INTERVAL 30 DAY )
+         group by model_token
+     ) as r
+         left outer join tts_models as m
+              on m.token = r.model_token
+         join users as u
+              on u.token = m.creator_user_token
+where m.creator_set_visibility != 'public'
+order by r.use_count desc
+    limit 500;
+

_sql/analytics/tts/tts_results.sql

@@ -64,14 +64,6 @@ ORDER BY created_at DESC
 
 
 
--- Delete TTS results for a single model
--- Do this in short batches so a lock isn't held for prohibitively long.
-update tts_results
-set mod_deleted_at = NOW()
-where model_token = 'TM:ztt5s1be5tq6'
-and mod_deleted_at IS NULL
-limit 5000;
-
 --- Search for particular text by a particular model
 SELECT
     raw_inference_text,
@@ -83,6 +75,39 @@ WHERE
     model_token = 'TM:ztt5s1be5tq6'
     AND raw_inference_text LIKE '%fakeyou%'
 
+--- Search for 101soundboards.com using us
+SELECT
+    r.raw_inference_text,
+    r.maybe_creator_user_token,
+    u.username,
+    u.ip_address_creation,
+    r.creator_ip_address,
+    r.created_at
+FROM
+    tts_results as r
+LEFT OUTER JOIN
+    users as u
+ON
+    r.maybe_creator_user_token = u.token
+WHERE
+    r.raw_inference_text LIKE '%bananas%'
+    AND r.created_at > ( CURDATE() - INTERVAL 1 HOUR )
+
+
+--- Search for 101soundboards.com using us
+SELECT
+    r.maybe_creator_user_token, r.creator_ip_address
+FROM
+    tts_results as r
+        LEFT OUTER JOIN
+    users as u
+    ON
+            r.maybe_creator_user_token = u.token
+WHERE
+        r.raw_inference_text LIKE '%pickle%'
+  AND r.created_at > ( CURDATE() - INTERVAL 1 HOUR )
+
+
 ---
 -- WTF WHY IS THIS SLOW???
 SELECT

_sql/analytics/users/paying_user_retention.sql

@@ -13,6 +13,52 @@ select count(*) from user_subscriptions;
 -- NB: "active" status users are current, treating "past_due" as active too (for now)
 select count(*) from user_subscriptions where maybe_stripe_subscription_status NOT IN ("canceled", "incomplete", "incomplete_expired");
 
+-- ==================== TTS + VC + W2L ====================
+
+-- Make sure the time intervals are the same!
+SELECT
+    username,
+    count(*) as total_use_count
+FROM (
+   SELECT
+        u.username as username,
+        j.token as token
+    FROM users AS u
+    JOIN tts_inference_jobs AS j
+        ON u.token = j.maybe_creator_user_token
+        WHERE u.token IN (
+            SELECT DISTINCT user_token
+            FROM user_subscriptions
+        )
+        AND j.created_at > NOW() - INTERVAL 31 DAY
+   UNION
+        SELECT
+            u.username as username,
+            j.token as token
+        FROM users AS u
+            JOIN generic_inference_jobs AS j
+        ON u.token = j.maybe_creator_user_token
+        WHERE u.token IN (
+            SELECT DISTINCT user_token
+            FROM user_subscriptions
+            )
+          AND j.created_at > NOW() - INTERVAL 31 DAY
+    UNION
+        SELECT
+            u.username as username,
+            j.token as token
+        FROM users AS u
+             JOIN w2l_inference_jobs AS j
+        ON u.token = j.maybe_creator_user_token
+        WHERE u.token IN (
+            SELECT DISTINCT user_token
+            FROM user_subscriptions
+            )
+        AND j.created_at > NOW() - INTERVAL 31 DAY
+) as t
+GROUP BY username
+ORDER BY total_use_count desc
+
 -- ==================== TEXT TO SPEECH====================
 
 -- All paid subscribers generating results in the last week
@@ -23,7 +69,7 @@ on u.token = j.maybe_creator_user_token
 WHERE u.token IN (
     select distinct user_token
     from user_subscriptions
-    where maybe_stripe_subscription_status NOT IN ("canceled", "incomplete", "incomplete_expired")
+    -- where maybe_stripe_subscription_status NOT IN ("canceled", "incomplete", "incomplete_expired")
 )
 AND j.created_at > NOW() - interval 7 day
 GROUP BY u.username

_sql/analytics/users/user_top_contributors.sql

@@ -25,20 +25,26 @@ on u.token = m.creator_user_token
 where m.upload_count >= 5;
 
 
-
--- Upgrade top TTS model uploaders to a loyalty plan
-update users
-set maybe_loyalty_program_key = 'fakeyou_contributor'
-where token IN (
-  select creator_user_token from
-    (
-      select count(*) as upload_count,
-      creator_user_token
-      from tts_models
-      group by creator_user_token
-      having upload_count >= 10
-      order by upload_count desc
-  ) as t
-);
+-- Top VC model uploaders
+select
+    u.id,
+    u.token,
+    u.username,
+    u.created_at,
+    u.maybe_stripe_customer_id,
+    u.maybe_loyalty_program_key,
+    m.upload_count
+from users as u
+         join
+     (
+         select
+             count(*) as upload_count,
+             creator_user_token
+         from voice_conversion_models
+         group by creator_user_token
+         order by upload_count desc
+     ) as m
+     on u.token = m.creator_user_token
+where m.upload_count >= 5;
 
 

_sql/analytics/voice_conversion/voice_conversion_models.sql

@@ -0,0 +1,37 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Inventory all
+select
+    vc.id,
+    vc.token,
+    vc.model_type,
+    vc.title,
+    u.username,
+    vc.original_download_url,
+    vc.created_at,
+    vc.updated_at,
+    vc.user_deleted_at,
+    vc.mod_deleted_at,
+    vc.maybe_mod_comments
+from voice_conversion_models as vc
+left join users AS u on
+    u.token = vc.creator_user_token;
+
+-- Inventory non-deleted
+select
+    vc.id,
+    vc.token,
+    vc.model_type,
+    vc.title,
+    u.username,
+    vc.original_download_url,
+    vc.created_at,
+    vc.updated_at
+from voice_conversion_models as vc
+left join users AS u on
+    u.token = vc.creator_user_token
+where
+    vc.user_deleted_at IS NULL
+    and vc.mod_deleted_at IS NULL;

_sql/analytics/w2l/w2l_template_upload_jobs.sql

@@ -14,15 +14,3 @@ from w2l_template_upload_jobs
 where status='started'
 and created_at < ( CURDATE() - INTERVAL 1 DAY );
 
--- Unstick old jobs that might be stuck as "started"
-update w2l_template_upload_jobs
-set status='dead'
-where status='started'
-  and created_at < ( CURDATE() - INTERVAL 1 DAY );
-
--- Kill pending jobs with tiktok URLs:
--- These need to be blocked in the server and job directly.
-update w2l_template_upload_jobs
-set status='dead'
-where status='pending'
-  and download_url LIKE '%tiktok%';

_sql/migrations/2022-12-14-050849_generic_inference_jobs/up.sql

@@ -88,7 +88,7 @@ CREATE TABLE generic_inference_jobs (
   is_from_api_user BOOLEAN NOT NULL DEFAULT FALSE,
   is_for_twitch BOOLEAN NOT NULL DEFAULT FALSE,
 
-  -- If true, the request gets routed to a special "debug" worker.
+  -- NB: The meaning of this column has changed. See the "squashed" schema for docs/details.
   is_debug_request BOOLEAN NOT NULL DEFAULT FALSE,
 
   -- ========== JOB SYSTEM DETAILS ==========

_sql/migrations/2023-03-16-003630_comments/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+DROP TABLE comments;

_sql/migrations/2023-03-16-003630_comments/up.sql

@@ -0,0 +1,69 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Comments that can be attached to just about anything.
+-- This is a polymorphic table and can apply to a wide variety of entities.
+CREATE TABLE comments (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" (PUBLIC)
+  token VARCHAR(32) NOT NULL,
+
+  -- Idempotency token from client
+  -- This is so the frontend client doesn't submit duplicate items.
+  uuid_idempotency_token VARCHAR(36) NOT NULL,
+
+  -- User making the comment
+  user_token VARCHAR(32) NOT NULL,
+
+  -- The type of entity the comment applies to
+  entity_type VARCHAR(32) NOT NULL,
+
+  -- The token of the entity the comment applies to
+  entity_token VARCHAR(32) NOT NULL,
+
+  -- The comment in markdown
+  comment_markdown TEXT NOT NULL,
+
+  -- Generated HTML (not user-editable).
+  comment_rendered_html TEXT NOT NULL,
+
+  -- For abuse tracking.
+  -- Wide enough for IPv4/6
+  creator_ip_address VARCHAR(40) NOT NULL,
+  editor_ip_address VARCHAR(40) NOT NULL,
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- 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,
+
+  -- Technically, this conveys *user* edits, not things that update the record itself.
+  edited_at TIMESTAMP NULL,
+
+  -- If this is removed by a mod.
+  -- It completely disappears from the system.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+  -- The owner of the thing the comment is attached to can delete the comment too.
+  object_owner_deleted_at TIMESTAMP NULL,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+  UNIQUE KEY (uuid_idempotency_token),
+  KEY index_entity_type_entity_token (entity_type, entity_token),
+  KEY index_user_token (user_token),
+  KEY index_created_at (created_at),
+  KEY index_user_deleted_at (user_deleted_at),
+  KEY index_mod_deleted_at (mod_deleted_at),
+  KEY index_object_owner_deleted_at (object_owner_deleted_at)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations/2023-03-21-014742_alter_generic_inference_jobs_add_debug_columns/down.sql

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

_sql/migrations/2023-03-21-014742_alter_generic_inference_jobs_add_debug_columns/up.sql

@@ -0,0 +1,10 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN maybe_routing_tag VARCHAR(32) DEFAULT NULL
+    AFTER is_debug_request;
+
+ALTER TABLE generic_inference_jobs
+    ADD INDEX index_maybe_routing_tag (maybe_routing_tag);

_sql/migrations/2023-04-04-070629_alter_tts_models_add_vits/down.sql

@@ -0,0 +1,23 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- REVERT ADDING 'vits' MODEL TYPE
+
+ALTER TABLE tts_models
+MODIFY COLUMN tts_model_type ENUM(
+    'not-set',
+    'tacotron2',
+    'glowtts',
+    'glowtts-vocodes',
+    'talknet'
+) NOT NULL DEFAULT 'not-set';
+
+ALTER TABLE tts_model_upload_jobs
+MODIFY COLUMN tts_model_type ENUM(
+    'not-set',
+    'tacotron2',
+    'glowtts',
+    'glowtts-vocodes',
+    'talknet'
+) NOT NULL DEFAULT 'not-set';

_sql/migrations/2023-04-04-070629_alter_tts_models_add_vits/up.sql

@@ -0,0 +1,28 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Add 'vits' to the tts_model_type.
+
+ALTER TABLE tts_models
+MODIFY COLUMN tts_model_type ENUM(
+    'not-set',
+    'tacotron2',
+    'glowtts',
+    'glowtts-vocodes',
+    'talknet',
+    'vits'
+) NOT NULL DEFAULT 'not-set';
+
+-- Technically we won't be using the model upload jobs table anymore,
+-- but add 'vits' for consistency.
+
+ALTER TABLE tts_model_upload_jobs
+MODIFY COLUMN tts_model_type ENUM(
+    'not-set',
+    'tacotron2',
+    'glowtts',
+    'glowtts-vocodes',
+    'talknet',
+    'vits'
+) NOT NULL DEFAULT 'not-set';

_sql/migrations/2023-04-05-074546_alter_tts_models_index_model_type/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+DROP INDEX `index_tts_model_type` ON `tts_models`;

_sql/migrations/2023-04-05-074546_alter_tts_models_index_model_type/up.sql

@@ -0,0 +1,6 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE tts_models ADD INDEX index_tts_model_type (tts_model_type);
+

_sql/migrations/2023-04-09-020030_alter_inference_jobs_rename_inference_type/down.sql

@@ -0,0 +1,7 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs RENAME COLUMN inference_category TO inference_type;
+
+ALTER TABLE generic_inference_jobs RENAME INDEX index_inference_category TO index_inference_type;

_sql/migrations/2023-04-09-020030_alter_inference_jobs_rename_inference_type/up.sql

@@ -0,0 +1,7 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs RENAME COLUMN inference_type TO inference_category;
+
+ALTER TABLE generic_inference_jobs RENAME INDEX index_inference_type TO index_inference_category;

_sql/migrations/2023-04-09-020102_alter_inference_jobs_reorder_model_token/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs MODIFY maybe_model_token VARCHAR(32) DEFAULT NULL AFTER maybe_raw_inference_text;

_sql/migrations/2023-04-09-020102_alter_inference_jobs_reorder_model_token/up.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs MODIFY maybe_model_token VARCHAR(32) DEFAULT NULL AFTER inference_category;

_sql/migrations/2023-04-09-022655_alter_inference_jobs_add_model_type/down.sql

@@ -0,0 +1,8 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+DROP INDEX index_maybe_model_type_and_maybe_model_token ON generic_inference_jobs;
+DROP INDEX index_maybe_model_type ON generic_inference_jobs;
+
+ALTER TABLE generic_inference_jobs DROP COLUMN maybe_model_type;

_sql/migrations/2023-04-09-022655_alter_inference_jobs_add_model_type/up.sql

@@ -0,0 +1,10 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN maybe_model_type VARCHAR(32) DEFAULT NULL
+    AFTER inference_category;
+
+CREATE INDEX index_maybe_model_type_and_maybe_model_token ON generic_inference_jobs (maybe_model_type, maybe_model_token);
+CREATE INDEX index_maybe_model_type ON generic_inference_jobs (maybe_model_type);

_sql/migrations/2023-04-22-085636_voice_conversion_results_change_columns/down.sql

@@ -0,0 +1,26 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE voice_conversion_results DROP COLUMN public_bucket_hash;
+ALTER TABLE voice_conversion_results DROP COLUMN bucket_has_wav;
+ALTER TABLE voice_conversion_results DROP COLUMN bucket_has_mp3;
+ALTER TABLE voice_conversion_results DROP COLUMN bucket_has_mp4;
+ALTER TABLE voice_conversion_results DROP COLUMN bucket_has_webm;
+ALTER TABLE voice_conversion_results DROP COLUMN bucket_has_spectrogram;
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN public_bucket_directory_full_path VARCHAR(255) NOT NULL
+    AFTER maybe_creator_synthetic_id;
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN public_bucket_wav_audio_object_name VARCHAR(255) NOT NULL
+    AFTER public_bucket_directory_full_path;
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN public_bucket_spectrogram_object_name VARCHAR(255) NOT NULL
+    AFTER public_bucket_wav_audio_object_name;
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN public_bucket_video_object_name VARCHAR(255) NOT NULL
+    AFTER public_bucket_spectrogram_object_name;

_sql/migrations/2023-04-22-085636_voice_conversion_results_change_columns/up.sql

@@ -0,0 +1,39 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- These columns are no longer useful.
+-- It's better to store booleans for each type of file we store and calculate the paths.
+
+ALTER TABLE voice_conversion_results DROP COLUMN public_bucket_directory_full_path;
+ALTER TABLE voice_conversion_results DROP COLUMN public_bucket_wav_audio_object_name;
+ALTER TABLE voice_conversion_results DROP COLUMN public_bucket_spectrogram_object_name;
+ALTER TABLE voice_conversion_results DROP COLUMN public_bucket_video_object_name;
+
+-- The hash will contain the key information for where files live
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN public_bucket_hash VARCHAR(64) NOT NULL
+    AFTER maybe_creator_synthetic_id;
+
+-- The boolean flags will tell what we store.
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN bucket_has_wav BOOLEAN NOT NULL DEFAULT FALSE
+    AFTER public_bucket_hash;
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN bucket_has_mp3 BOOLEAN NOT NULL DEFAULT FALSE
+    AFTER bucket_has_wav;
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN bucket_has_mp4 BOOLEAN NOT NULL DEFAULT FALSE
+    AFTER bucket_has_mp3;
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN bucket_has_webm BOOLEAN NOT NULL DEFAULT FALSE
+    AFTER bucket_has_mp4;
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN bucket_has_spectrogram BOOLEAN NOT NULL DEFAULT FALSE
+    AFTER bucket_has_webm;

_sql/migrations/2023-04-23-100915_voice_conversion_results_alterations/down.sql

@@ -0,0 +1,15 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+DROP INDEX `fk_media_token_and_media_token_type` ON `voice_conversion_results`;
+
+ALTER TABLE voice_conversion_results DROP COLUMN media_token_type;
+ALTER TABLE voice_conversion_results DROP COLUMN generated_by_cluster;
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN maybe_vocoder_token VARCHAR(64) DEFAULT NULL
+    AFTER model_token;
+
+CREATE INDEX `fk_maybe_vocoder_token`
+    ON `voice_conversion_results` (`maybe_vocoder_token`);

_sql/migrations/2023-04-23-100915_voice_conversion_results_alterations/up.sql

@@ -0,0 +1,20 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- There isn't a vocoder, and even if so, it doesn't need to be referenced here.
+
+DROP INDEX `fk_maybe_vocoder_token` ON `voice_conversion_results`;
+
+ALTER TABLE voice_conversion_results DROP COLUMN maybe_vocoder_token;
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN media_token_type VARCHAR(32) NOT NULL
+    AFTER media_token;
+
+ALTER TABLE voice_conversion_results
+    ADD COLUMN generated_by_cluster VARCHAR(255) NOT NULL
+    AFTER generated_by_worker;
+
+CREATE INDEX `fk_media_token_and_media_token_type`
+    ON `voice_conversion_results` (`media_token`, `media_token_type`);

_sql/migrations/2023-04-23-110735_generic_inference_jobs_alterations/down.sql

@@ -0,0 +1,9 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+DROP INDEX `fk_maybe_input_source_token` ON `generic_inference_jobs`;
+DROP INDEX `fk_maybe_input_source_token_and_type` ON `generic_inference_jobs`;
+
+ALTER TABLE generic_inference_jobs DROP COLUMN maybe_input_source_token;
+ALTER TABLE generic_inference_jobs DROP COLUMN maybe_input_source_token_type;

_sql/migrations/2023-04-23-110735_generic_inference_jobs_alterations/up.sql

@@ -0,0 +1,17 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN maybe_input_source_token VARCHAR(32) DEFAULT NULL
+    AFTER maybe_model_token;
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN maybe_input_source_token_type VARCHAR(32) DEFAULT NULL
+    AFTER maybe_input_source_token;
+
+CREATE INDEX `fk_maybe_input_source_token`
+    ON `generic_inference_jobs` (`maybe_input_source_token`);
+
+CREATE INDEX `fk_maybe_input_source_token_and_type`
+    ON `generic_inference_jobs` (`maybe_input_source_token`, `maybe_input_source_token_type`);

_sql/migrations/2023-04-23-112622_media_uploads_alterations/down.sql

@@ -0,0 +1,10 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE media_uploads
+    ADD COLUMN public_bucket_directory_full_path VARCHAR(255) NOT NULL
+    AFTER public_bucket_directory_hash;
+
+ALTER TABLE media_uploads
+    RENAME COLUMN maybe_extra_file_modification_info TO extra_file_modification_info ;

_sql/migrations/2023-04-23-112622_media_uploads_alterations/up.sql

@@ -0,0 +1,8 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE media_uploads DROP COLUMN public_bucket_directory_full_path;
+
+ALTER TABLE media_uploads
+    RENAME COLUMN extra_file_modification_info TO maybe_extra_file_modification_info;

_sql/migrations/2023-04-25-075931_media_uploads_add_upload_source/down.sql

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

_sql/migrations/2023-04-25-075931_media_uploads_add_upload_source/up.sql

@@ -0,0 +1,7 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE media_uploads
+    ADD COLUMN media_source VARCHAR(16) NOT NULL
+    AFTER media_type;

_sql/migrations/2023-04-29-070009_alter_download_job_add_worker_info/down.sql

@@ -0,0 +1,8 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_download_jobs DROP COLUMN first_started_at;
+ALTER TABLE generic_download_jobs DROP COLUMN assigned_cluster;
+ALTER TABLE generic_download_jobs DROP COLUMN assigned_worker;
+ALTER TABLE generic_download_jobs DROP COLUMN successfully_completed_at;

_sql/migrations/2023-04-29-070009_alter_download_job_add_worker_info/up.sql

@@ -0,0 +1,19 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_download_jobs
+    ADD COLUMN assigned_worker VARCHAR(128) DEFAULT NULL
+    AFTER failure_reason;
+
+ALTER TABLE generic_download_jobs
+    ADD COLUMN assigned_cluster VARCHAR(128) DEFAULT NULL
+    AFTER assigned_worker;
+
+ALTER TABLE generic_download_jobs
+    ADD COLUMN first_started_at TIMESTAMP NULL
+    AFTER retry_at;
+
+ALTER TABLE generic_download_jobs
+    ADD COLUMN successfully_completed_at TIMESTAMP NULL
+    AFTER first_started_at;

_sql/migrations/2023-04-29-070934_alter_inference_job_add_worker_info/down.sql

@@ -0,0 +1,8 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs DROP COLUMN first_started_at;
+ALTER TABLE generic_inference_jobs DROP COLUMN assigned_cluster;
+ALTER TABLE generic_inference_jobs DROP COLUMN assigned_worker;
+ALTER TABLE generic_inference_jobs DROP COLUMN successfully_completed_at;

_sql/migrations/2023-04-29-070934_alter_inference_job_add_worker_info/up.sql

@@ -0,0 +1,19 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN assigned_worker VARCHAR(128) DEFAULT NULL
+    AFTER last_assigned_worker;
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN assigned_cluster VARCHAR(128) DEFAULT NULL
+    AFTER assigned_worker;
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN first_started_at TIMESTAMP NULL
+    AFTER retry_at;
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN successfully_completed_at TIMESTAMP NULL
+    AFTER first_started_at;

_sql/migrations/2023-04-29-093735_alter_jobs_add_success_execution_millis/down.sql

@@ -0,0 +1,6 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_download_jobs DROP COLUMN success_execution_millis;
+ALTER TABLE generic_inference_jobs DROP COLUMN success_execution_millis;

_sql/migrations/2023-04-29-093735_alter_jobs_add_success_execution_millis/up.sql

@@ -0,0 +1,11 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_download_jobs
+    ADD COLUMN success_execution_millis INT(10) UNSIGNED DEFAULT NULL
+    AFTER failure_reason;
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN success_execution_millis INT(10) UNSIGNED DEFAULT NULL
+    AFTER internal_debugging_failure_reason;

_sql/migrations/2023-05-01-105648_alter_jobs_add_inference_duration_millis/down.sql

@@ -0,0 +1,6 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_download_jobs DROP COLUMN success_inference_execution_millis;
+ALTER TABLE generic_inference_jobs DROP COLUMN success_inference_execution_millis;

_sql/migrations/2023-05-01-105648_alter_jobs_add_inference_duration_millis/up.sql

@@ -0,0 +1,11 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_download_jobs
+    ADD COLUMN success_inference_execution_millis INT(10) UNSIGNED DEFAULT NULL
+    AFTER success_execution_millis;
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN success_inference_execution_millis INT(10) UNSIGNED DEFAULT NULL
+    AFTER success_execution_millis;

_sql/migrations/2023-05-21-235956_alter_jobs_add_keepalive_flag/down.sql

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

_sql/migrations/2023-05-21-235956_alter_jobs_add_keepalive_flag/up.sql

@@ -0,0 +1,7 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+ALTER TABLE generic_inference_jobs
+    ADD COLUMN is_keepalive_required BOOLEAN NOT NULL DEFAULT FALSE
+    AFTER priority_level;

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

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

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

@@ -0,0 +1,51 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+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,
+
+  -- Username is a lookup key; display_name allows the user to add custom case.
+  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) NOT 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),
+  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/generic_download_jobs.sql

@@ -0,0 +1,137 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+CREATE TABLE generic_download_jobs (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" (PUBLIC)
+  -- This is so the in-progress results can be looked up by the UI.
+  token VARCHAR(32) NOT NULL,
+
+  -- Idempotency token
+  -- This is so the frontend client doesn't submit duplicate jobs.
+  uuid_idempotency_token VARCHAR(36) NOT NULL,
+
+  -- ========== SUCCESS CASE ==========
+
+  -- If the job completes successfully, this is the result token.
+  -- This is only populated on a successful result.
+  on_success_downloaded_entity_token VARCHAR(32) DEFAULT NULL,
+
+  -- The type of the object will vary based on the type of the upload,
+  -- and we may include heuristics that auto-detect types in the future
+  -- Example types: 'hifigan', 'hifigan_rocket_vc' (codename for softvc), 'tacotron2', ...
+  on_success_downloaded_entity_type VARCHAR(32) DEFAULT NULL,
+
+  -- ========== UPLOAD DETAILS ==========
+
+  -- A "well defined" type of upload
+  -- Examples: "hifigan", "tacotron2", etc.
+  download_type VARCHAR(32) NOT NULL,
+
+  -- Assume the same URL can be uploaded more than once (in case of errors or
+  -- people sharing models)
+
+  download_url VARCHAR(512) NOT NULL,
+
+  -- The user-supplied title of the upload
+  -- This might have different meanings for different upload types
+  -- The "name" of the voice model, which might be complicated.
+  -- If maybe_subject_token (etc.) is set, then it's authoritative instead.
+  title VARCHAR(255) NOT NULL,
+
+  -- ========== CREATOR DETAILS AND PREFERENCES ==========
+
+  -- Foreign key to user
+  creator_user_token VARCHAR(32) NOT 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',
+
+  -- ========== JOB SYSTEM DETAILS ==========
+
+  -- Jobs begin as "pending", then transition to other states.
+  --
+  --  * Pending = job is ready to go
+  --  * Started = job is running
+  --  * Complete_Success = job is done (success)
+  --  * Complete_Failure = job is done (failure)
+  --  * Attempt_Failed = job failed but may retry.
+  --  * Dead = job failed permanently.
+  --
+  -- Pending -> Started -> Complete_Success
+  --                    |-> Complete_Failure
+  --                    \-> Attempt_Failed -> Started -> { Complete, Failed, Dead }
+  status ENUM(
+    'pending',
+    'started',
+    'complete_success',
+    'complete_failure',
+    'attempt_failed',
+    'dead') NOT NULL DEFAULT 'pending',
+
+  -- We can track this against a "max_attempt_count"
+  attempt_count INT(3) NOT NULL DEFAULT 0,
+
+  -- If there is a failure, tell the user why.
+  failure_reason VARCHAR(512) 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
+  -- derived execution time analytics with it.
+  success_execution_millis INT(10) UNSIGNED DEFAULT NULL,
+
+  -- This is only the inference portion and does not include downloads, uploads,
+  -- or any subsequent processing.
+  success_inference_execution_millis INT(10) UNSIGNED DEFAULT NULL,
+
+  -- Worker hostname (linux hostname, k8s pod name)
+  -- Assigned when a worker picks up the job
+  -- Reassigned if the job fails and gets picked up again
+  assigned_worker VARCHAR(128) DEFAULT NULL,
+
+  -- Cluster name (k8s)
+  -- Assigned when a worker picks up the job
+  -- Reassigned if the job fails and gets picked up again
+  assigned_cluster VARCHAR(128) DEFAULT NULL,
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- Failed jobs will set a next attempt time.
+  -- Subsequent tries can increase the timeout.
+  -- Failures because of permissions require human intervention => [retry_at=null].
+  -- Failures because of invalid files are dead => [status=dead].
+  retry_at TIMESTAMP NULL,
+
+  -- Set when the job first starts executing
+  first_started_at TIMESTAMP NULL,
+
+  -- Set when the job is successfully completed
+  successfully_completed_at TIMESTAMP NULL,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+  UNIQUE KEY (uuid_idempotency_token),
+  KEY fk_on_download_type (download_type),
+  KEY fk_creator_user_token (creator_user_token),
+  KEY index_status (status),
+  KEY index_creator_ip_address (creator_ip_address)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/generic_inference_jobs.sql

@@ -0,0 +1,215 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+CREATE TABLE generic_inference_jobs (
+
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" (PUBLIC)
+  -- This is so the in-progress results can be looked up by the UI.
+  token VARCHAR(32) NOT NULL,
+
+  -- Idempotency token from client
+  -- This is so the frontend client doesn't submit duplicate jobs.
+  uuid_idempotency_token VARCHAR(36) NOT NULL,
+
+  -- ========== INFERENCE DETAILS ==========
+
+  -- Broad category of inference
+  -- Examples (may not be up to date):
+  --  * text_to_speech
+  --  * voice_conversion
+  inference_category VARCHAR(32) NOT NULL,
+
+  -- Potential part of the composite foreign key to the primary model being used, if any.
+  -- This will normally live in `maybe_inference_args`, but in this case, it's useful for
+  -- running easy database analytical queries.
+  maybe_model_type VARCHAR(32) DEFAULT NULL,
+
+  -- Potential part of the composite foreign key to the primary model being used, if any.
+  -- This will normally live in `maybe_inference_args`, but in this case, it's useful for
+  -- running easy database analytical queries.
+  maybe_model_token VARCHAR(32) DEFAULT NULL,
+
+  -- If the inference job takes some other kind of entity, this will be the token
+  -- Not all inference jobs have record-like input sources.
+  maybe_input_source_token VARCHAR(32) DEFAULT NULL,
+
+  -- If the inference job takes some other kind of entity, this will be the type of the
+  -- token. For now, this is just `media_upload`.
+  -- Not all inference jobs have record-like input sources.
+  maybe_input_source_token_type VARCHAR(32) DEFAULT NULL,
+
+  -- Polymorphic arguments payload that depends on the type of inference job.
+  -- TEXT = 65,535 bytes (64 KiB), ~= 4 bytes per UTF-8 character, ~ 16383 characters.
+  maybe_inference_args TEXT DEFAULT NULL,
+
+  -- For text-based workloads, the raw, unprocessed user input.
+  -- TEXT = 65,535 bytes (64 KiB), ~= 4 bytes per UTF-8 character, ~ 16383 characters.
+  maybe_raw_inference_text TEXT DEFAULT NULL,
+
+  -- ========== SUCCESS CASE ==========
+
+  -- The type of the object will vary based on the type of the upload,
+  -- and we may include heuristics that auto-detect types in the future
+  on_success_result_entity_type VARCHAR(32) DEFAULT NULL,
+
+  -- If the job completes successfully, this is the result token.
+  -- This is only populated on a successful result.
+  on_success_result_entity_token VARCHAR(32) 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',
+
+  -- ========== PREMIUM FEATURES METADATA ==========
+
+  -- The maximum duration for generated audio or video in seconds.
+  -- Zero is implied to be the default value, which is typically 12 seconds.
+  -- A negative value implies "unlimited".
+  max_duration_seconds INTEGER NOT NULL DEFAULT 0,
+
+  -- ========== WORKLOAD SOURCE AND PRIORITY ==========
+
+  -- Priority *increases*, so a level of 2 will be higher than 1.
+  -- These are the level semantics currently:
+  --   - All jobs from anonymous FakeYou users have level 0.
+  --   - All jobs from logged in FakeYou users have level 1.
+  --   - All jobs from Twitch TTS (unpaid) have level 10 (ten).
+  --   - (There will be future levels for paid Twitch and social FakeYou rewards.)
+  priority_level SMALLINT UNSIGNED NOT NULL DEFAULT 0,
+
+  -- For non-premium workloads (or some other workloads), we may require that the
+  -- user remains on the website. This will help us kill jobs if the user leaves.
+  -- The state for this keepalive signal will be job-type dependent and may live
+  -- in Redis or some other system. Typically the frontend job status polling will
+  -- populate this keepalive signal (eg. in a Redis key with TTL).
+  is_keepalive_required BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- TODO: add is_for_storyteller_product
+  is_from_premium_user BOOLEAN NOT NULL DEFAULT FALSE,
+  is_from_api_user BOOLEAN NOT NULL DEFAULT FALSE,
+  is_for_twitch BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- ========== DEVELOPMENT AND DEBUGGING METADATA ==========
+
+  -- If true, the request gets a "debug" flag, which may do different
+  -- things depending on the type of work.
+  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.
+  maybe_routing_tag VARCHAR(32) DEFAULT NULL,
+
+  -- ========== JOB SYSTEM DETAILS ==========
+
+  -- Jobs begin as "pending", then transition to other states.
+  --
+  --  * Pending = job is ready to go
+  --  * Started = job is running
+  --  * Complete_Success = job is done (success)
+  --  * Complete_Failure = job is done (failure)
+  --  * Attempt_Failed = job failed but may retry.
+  --  * Dead = job failed permanently.
+  --
+  -- Pending -> Started -> Complete_Success
+  --                    |-> Complete_Failure
+  --                    \-> Attempt_Failed -> Started -> { Complete, Failed, Dead }
+  status ENUM(
+    'pending',
+    'started',
+    'complete_success',
+    'complete_failure',
+    'attempt_failed',
+    'dead') 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.
+  failure_reason VARCHAR(512) DEFAULT NULL,
+
+  -- Optional internal-only debugging information in the case of failure.
+  internal_debugging_failure_reason VARCHAR(512) 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
+  -- derived execution time analytics with it.
+  success_execution_millis INT(10) UNSIGNED DEFAULT NULL,
+
+  -- This is only the inference portion and does not include downloads, uploads,
+  -- or any subsequent processing.
+  success_inference_execution_millis INT(10) UNSIGNED DEFAULT NULL,
+
+  -- TODO: Remove once `assigned_worker` column is added
+  --   and existing queries for this column are removed.
+  -- The last worker (hostname or pod name) to touch the job, either in the case of success or failure.
+  last_assigned_worker VARCHAR(255) DEFAULT NULL,
+
+  -- Worker hostname (linux hostname, k8s pod name)
+  -- Assigned when a worker picks up the job
+  -- Reassigned if the job fails and gets picked up again
+  assigned_worker VARCHAR(128) DEFAULT NULL,
+
+  -- Cluster name (k8s)
+  -- Assigned when a worker picks up the job
+  -- Reassigned if the job fails and gets picked up again
+  assigned_cluster VARCHAR(128) DEFAULT NULL,
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- Failed jobs will set a next attempt time.
+  -- Subsequent tries can increase the timeout.
+  -- Failures because of permissions require human intervention => [retry_at=null].
+  -- Failures because of invalid files are dead => [status=dead].
+  retry_at TIMESTAMP NULL,
+
+  -- Set when the job first starts executing
+  first_started_at TIMESTAMP NULL,
+
+  -- Set when the job is successfully completed
+  successfully_completed_at TIMESTAMP NULL,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+  UNIQUE KEY (uuid_idempotency_token),
+  KEY index_inference_category (inference_category),
+  KEY index_maybe_model_type_and_maybe_model_token (maybe_model_type, maybe_model_token),
+  KEY index_maybe_model_type (maybe_model_type),
+  KEY fk_maybe_model_token (maybe_model_token),
+  KEY fk_maybe_input_source_token (maybe_input_source_token),
+  KEY fk_maybe_input_source_token_and_type (maybe_input_source_token, maybe_input_source_token_type),
+  KEY fk_on_success_result_entity_token (on_success_result_entity_token),
+  KEY fk_maybe_creator_user_token (maybe_creator_user_token),
+  KEY index_creator_ip_address (creator_ip_address),
+  KEY index_priority_level (priority_level),
+  KEY index_is_debug_request (is_debug_request),
+  KEY index_maybe_routing_tag (maybe_routing_tag),
+  KEY index_status (status)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/media_uploads.sql

@@ -0,0 +1,135 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+CREATE TABLE media_uploads (
+
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" (PUBLIC)
+  -- This is so the in-progress results can be looked up by the UI.
+  token VARCHAR(32) NOT NULL,
+
+  -- Idempotency token from client
+  -- This is so the frontend client doesn't submit duplicate jobs.
+  uuid_idempotency_token VARCHAR(36) NOT NULL,
+
+  -- ========== MEDIA DETAILS ==========
+
+  -- Type of media:
+  --   * 'audio' for wav, mp3, etc.
+  --   * 'image' for png, jpg, etc.
+  --   * 'video' for a variety of video types.
+  media_type VARCHAR(16) NOT NULL,
+
+  -- Where the media originated.
+  --   * 'file' for user file uploads
+  --   * 'device_api' for user device API sessions (eg. browser+microphone)
+  --   * 'unknown' for unknown source
+  media_source VARCHAR(16) NOT NULL,
+
+  -- The original filename of the media
+  maybe_original_filename VARCHAR(255) DEFAULT NULL,
+
+  -- The original (non-transcoded or truncated) file size
+  original_file_size_bytes INT(10) NOT NULL DEFAULT 0,
+
+  -- The original (non-transcoded or truncated) duration in millis.
+  original_duration_millis INT(10) NOT NULL DEFAULT 0,
+
+  -- The original file's mime type.
+  maybe_original_mime_type VARCHAR(32) DEFAULT NULL,
+
+  -- Audio encoder details
+  -- Only present if the file was audio or a video that had audio.
+  maybe_original_audio_encoding VARCHAR(32) DEFAULT NULL,
+
+  -- Video encoder details
+  -- Only present if the file was a video.
+  maybe_original_video_encoding VARCHAR(32) DEFAULT NULL,
+
+  -- For videos, the original frame width.
+  maybe_original_frame_width INT(5) DEFAULT NULL,
+
+  -- For videos, the original frame height.
+  maybe_original_frame_height INT(5) DEFAULT NULL,
+
+  -- Checksum of the original media
+  -- SHA1 hash [SHA2 = CHAR(64), SHA1 = CHAR(40), MD5 = CHAR(32)]
+  -- checksum_sha1 CHAR(40) NOT NULL,
+
+  -- Checksum of the original media
+  -- SHA2 hash [SHA2 = CHAR(64), SHA1 = CHAR(40), MD5 = CHAR(32)]
+  -- Note that SHA2 is a hash family (SHA-228, SHA-256, SHA-384, SHA-512, SHA-512/224, SHA-512/256, ...)
+  -- and produces different lengths output depending on the choice of algorithm.
+  checksum_sha2 CHAR(64) NOT NULL,
+
+  -- ========== UPLOAD, TRANSCODING, AND TRUNCATION DETAILS ==========
+
+  -- The hash for the bucket directory that contains the original upload
+  -- as well as any transcodings, downsamplings, etc.
+  public_bucket_directory_hash  VARCHAR(32) NOT NULL,
+
+  -- We'll likely transcode (and potentially truncate) most media given to us.
+  -- This will store a json-encoded struct that details the changes.
+  -- TEXT = 65,535 bytes (64 KiB), ~= 4 bytes per UTF-8 character, ~ 16383 characters.
+  maybe_extra_file_modification_info TEXT 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.
+  -- We should also build an audit table to supersede this.
+  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,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+  UNIQUE KEY (uuid_idempotency_token),
+  KEY index_media_type (media_type),
+  KEY index_checksum_sha2 (checksum_sha2),
+  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/tts_tables/tts_inference_jobs.sql

@@ -63,6 +63,7 @@ CREATE TABLE tts_inference_jobs (
   is_for_twitch BOOLEAN NOT NULL DEFAULT FALSE,
 
   -- If true, the request gets routed to a special "debug" worker.
+  -- NB: This has a *DIFFERENT* meaning on the generic_inference_jobs table.
   is_debug_request BOOLEAN NOT NULL DEFAULT FALSE,
 
   -- ========== PREMIUM FEATURES METADATA ==========

_sql/migrations_squashed/tts_tables/tts_model_upload_jobs.sql

@@ -37,7 +37,8 @@ CREATE TABLE tts_model_upload_jobs (
       'tacotron2',
       'glowtts',
       'glowtts-vocodes',
-      'talknet'
+      'talknet',
+      'vits'
   ) NOT NULL DEFAULT 'not-set',
 
   -- Can be linked to a well-known voice

_sql/migrations_squashed/tts_tables/tts_models.sql

@@ -19,7 +19,8 @@ CREATE TABLE tts_models (
       'tacotron2',
       'glowtts',
       'glowtts-vocodes',
-      'talknet'
+      'talknet',
+      'vits'
   ) NOT NULL DEFAULT 'not-set',
 
   -- The name of the type of pipeline to use, eg "legacy_fakeyou", "legacy_vocodes", "english_v1", etc.
@@ -225,6 +226,7 @@ CREATE TABLE tts_models (
   KEY index_private_bucket_hash (private_bucket_hash),
   KEY index_is_public_listing_approved (is_public_listing_approved),
   KEY index_is_locked_from_user_modification (is_locked_from_user_modification),
-  KEY index_is_locked_from_use (is_locked_from_use)
+  KEY index_is_locked_from_use (is_locked_from_use),
+  KEY index_tts_model_type (tts_model_type)
 
 ) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/voice_conversion_models.sql

@@ -0,0 +1,144 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+CREATE TABLE voice_conversion_models (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" (PUBLIC)
+  token VARCHAR(32) NOT NULL,
+
+  -- ========== MODEL FUNCTIONALITY ==========
+
+  -- The type of voice conversion model
+  -- eg. 'rocket_vc' (codename for 'softvc'), 'diffsvc'
+  model_type VARCHAR(32) NOT NULL,
+
+  -- If set, we'll use the vocoder defined.
+  -- If not set, some global default will be used.
+  maybe_vocoder_token VARCHAR(32) DEFAULT NULL,
+
+  -- ========== MODEL DESCRIPTION ==========
+
+  -- The "name" of the voice model, which might be complicated.
+  -- If maybe_subject_token (etc.) is set, then it's authoritative instead.
+  title VARCHAR(255) NOT NULL,
+
+  -- The description of the model in markdown.
+  description_markdown TEXT NOT NULL,
+
+  -- Generated HTML (not user-editable).
+  description_rendered_html TEXT NOT NULL,
+
+  -- The full IETF BCP47 language tag (eg. en, en-US, es-419, ja-JP, pt, etc.)
+  -- (Not that it matters apart from categorization, since voice conversion is universal.)
+  ietf_language_tag VARCHAR(64) NOT NULL DEFAULT 'en',
+
+  -- The IETF BCP47 language tag's primary language subtag (eg. "es-419" becomes "es")
+  -- (Not that it matters apart from categorization, since voice conversion is universal.)
+  ietf_primary_language_subtag VARCHAR(12) NOT NULL DEFAULT 'en',
+
+  -- ========== MODEL PLACEMENT ==========
+
+  -- Whether the voice features on the front of FakeYou
+  is_front_page_featured BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- ========== IMMUTABLE PROVENANCE AND METADATA ==========
+
+  -- Where the file was originally downloaded (if it was downloaded)
+  original_download_url VARCHAR(512) DEFAULT NULL,
+
+  -- The filename that was used at upload time (if available)
+  original_filename VARCHAR(255) DEFAULT NULL,
+
+  -- File characteristics
+  file_size_bytes INT(10) NOT NULL DEFAULT 0,
+
+  -- ========== CREATOR DETAILS ==========
+
+  -- The person that created the template.
+  creator_user_token VARCHAR(32) NOT NULL,
+
+  -- For abuse tracking.
+  -- Wide enough for IPv4/6.
+  -- This will pick up moderator IP addresses, but a future audit logs table
+  -- will capture all edits.
+  ip_address_creation VARCHAR(40) NOT NULL,
+  ip_address_last_update VARCHAR(40) NOT NULL,
+
+  -- The last user to edit the model. This could be the creator or a moderator.
+  -- Check the future audit logs table for more info.
+  maybe_last_update_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- (THIS MIGHT NOT BE USED)
+  -- NB: DO NOT SORT!
+  -- THIS MUST MATCH THE RESPECTIVE JOBS TABLE.
+  creator_set_visibility ENUM(
+    'public',
+    'hidden',
+    'private'
+  ) NOT NULL DEFAULT 'public',
+
+  -- ========== BUCKET STORAGE ==========
+
+  -- The pytorch model
+  -- For now, this will be a hash of the file contents.
+  -- NB: NOT UNIQUE! We can allow duplicate uploads.
+  private_bucket_hash CHAR(64) NOT NULL,
+
+  -- The "full url" version of the path
+  private_bucket_object_name VARCHAR(255) NOT NULL,
+
+  -- ========== RATING AND RANKING ==========
+
+  -- Calculated average, on a scale of 0-100
+  -- Null with zero ratings.
+  -- calculated_average_score INT(3) DEFAULT NULL,
+
+  -- Other metrics
+  -- calculated_total_ratings_submitted_count INT(10) NOT NULL DEFAULT 0,
+  -- calculated_total_uses_count BIGINT(10) NOT NULL DEFAULT 0,
+
+  -- ========== MODERATION DETAILS ==========
+
+  -- Mods may have to approve of vc models for them to show up in a public index.
+  -- Models can be used by the author (or other parties that know the URL) before
+  -- they're approved, but unapproved models won't show up in public indices.
+  -- This is independent of other visibility controls (is_mod_disabled, deleted_at,
+  -- and creator_set_visibility)
+  is_public_listing_approved BOOLEAN DEFAULT NULL,
+
+  -- Extremely popular models may be locked from deletion or modification by users.
+  is_locked_from_user_modification BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- If a moderator has comments.
+  maybe_mod_comments VARCHAR(255) DEFAULT NULL,
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- 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,
+
+  -- If this is removed by a mod.
+  -- It completely disappears from the system.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+  KEY index_model_type (model_type),
+  KEY fk_maybe_vocoder_token (maybe_vocoder_token),
+  KEY fk_creator_user_token (creator_user_token),
+  KEY index_creator_set_visibility (creator_set_visibility),
+  KEY index_private_bucket_hash (private_bucket_hash),
+  KEY index_is_public_listing_approved (is_public_listing_approved),
+  KEY index_is_locked_from_user_modification (is_locked_from_user_modification)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/voice_conversion_results.sql

@@ -0,0 +1,121 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+CREATE TABLE voice_conversion_results (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" (PUBLIC)
+  token VARCHAR(32) NOT NULL,
+
+  -- ========== INFERENCE DETAILS ==========
+
+  -- The TTS model that was used
+  model_token VARCHAR(32) NOT NULL,
+
+ -- Which media token was used as input (can be audio or video), potentially
+ -- from a wide number of tables (eg. media_uploads, tts_results, etc.)
+  media_token VARCHAR(32) NOT NULL,
+
+  -- For now we just support the `media_uploads` table, but we may grow to
+  -- support media inputs from other areas of the system (eg. tts, video, etc.)
+  media_token_type VARCHAR(32) NOT NULL,
+
+  -- ========== CREATOR DETAILS ==========
+
+  -- The person that created the result
+  -- If the user wasn't logged in, this is null
+  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,
+
+  -- (THIS MIGHT NOT BE USED)
+  -- NB: DO NOT SORT!
+  -- THIS MUST MATCH THE RESPECTIVE JOBS TABLE.
+  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,
+
+  -- ========== BUCKET STORAGE ==========
+
+  -- The hash lets us reconstitute the bucket directory
+  -- location (which stores potentially several files)
+  public_bucket_hash VARCHAR(64) NOT NULL,
+
+  -- Whether we're storing a wav result file.
+  bucket_has_wav BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- Whether we're storing a mp3 result file.
+  bucket_has_mp3 BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- Whether we're storing a mp4 result file.
+  bucket_has_mp4 BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- Whether we're storing a webm result file.
+  bucket_has_webm BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- Whether we're storing a spectrogram json file.
+  bucket_has_spectrogram BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- ========== METADATA ==========
+
+  file_size_bytes INT(10) NOT NULL DEFAULT 0,
+  duration_millis INT(10) NOT NULL DEFAULT 0,
+
+  -- ========== INFERENCE METADATA, DC, ROUTING, DEBUGGING ==========
+
+  -- If true, the request was routed to a special "debug" worker.
+  is_debug_request BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- Whether we generated this result from our own data center (vs. cloud).
+  is_generated_on_prem BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- Worker hostname (linux hostname, k8s pod name)
+  generated_by_worker VARCHAR(255) DEFAULT NULL,
+
+  -- Cluster name (k8s)
+  generated_by_cluster VARCHAR(255) DEFAULT NULL,
+
+  -- ========== MODERATION DETAILS ==========
+
+  -- The last user (possibly moderator) that made changes.
+  -- See the future audit logs table for better info and accounting.
+  maybe_last_update_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,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+  KEY fk_model_token (model_token),
+  KEY fk_media_token (media_token),
+  KEY fk_media_token_and_media_token_type (media_token, media_token_type),
+  KEY fk_maybe_creator_user_token (maybe_creator_user_token),
+  KEY fk_maybe_creator_anonymous_visitor_token (maybe_creator_anonymous_visitor_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),
+  KEY index_created_at(created_at)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_todo/2023-03-16-121816_voice_survey_answers/down.sql

@@ -0,0 +1,5 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+DROP TABLE voice_survey_answers;

_sql/migrations_todo/2023-03-16-121816_voice_survey_answers/up.sql

@@ -0,0 +1,53 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Voice survey answers
+CREATE TABLE voice_survey_answers (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Effective "primary key" (PUBLIC)
+  token VARCHAR(32) NOT NULL,
+
+  -- Idempotency token from client
+  -- This is so the frontend client doesn't submit duplicate items.
+  uuid_idempotency_token VARCHAR(36) NOT NULL,
+
+  -- User making the comment
+  maybe_user_token VARCHAR(32) DEFAULT NULL,
+
+  -- The voice that is being requested
+  voice_name VARCHAR(128) NOT NULL,
+
+  -- If the person would pay for FakeYou premium
+  would_pay BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- How much the user would hypothetically pay per month
+  would_pay_amount DOUBLE DEFAULT NULL,
+
+  -- For abuse tracking.
+  -- Wide enough for IPv4/6
+  creator_ip_address VARCHAR(40) NOT NULL,
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- 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,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (token),
+  UNIQUE KEY (uuid_idempotency_token),
+  KEY index_maybe_user_token (maybe_user_token),
+  KEY index_voice_name (voice_name),
+  KEY index_would_pay (would_pay),
+  KEY index_would_pay_amount (would_pay_amount),
+  KEY index_created_at (created_at)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

codegen_database.sh

@@ -17,7 +17,7 @@ SQLITE_DATABASE_PATH=./runtime_data/database.db
 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/database_queries
+  pushd crates/lib/mysql_queries
   SQLX_OFFLINE=true cargo sqlx prepare --merged
   popd
 }

container_includes/endpoint_disablement/endpoint_exact_matches.txt

@@ -3,3 +3,6 @@
 
 # Just an example
 /exact/match/path
+
+# Broken Python script
+/tts/job/None

container_includes/ip_bans/2023-03-18_broken_python_script.txt

@@ -0,0 +1,6 @@
+# This broken Python script has been around since at least the DDoS time
+# But now it's actively spamming the logs at 100QPS on single nodes!
+# It's from a single IP address, which we'll block:
+
+# Unbanning for now as the code fix should resolve the issue.
+# 104.251.211.22

container_includes/ip_bans/2023-03-23_unblocked_cloudflare_ipv6_cidrs.txt

@@ -0,0 +1,46 @@
+# These IPv6 were blocked within Cloudflare. The blast radius seemed to be too big - it hit Paul and other users.
+
+# From Cloudflare list I created called "top_attacker_ips":
+# (These were created around March 9th - March 11th.)
+# 2400:8500:1302:775::/64
+# 2a01:4f8:c17:750::/64
+
+# From Cloudflare list I created called "attacker_ips_v2":
+# (These were created around March 9th.)
+# 2001:448a:2061:2fc7::/64
+# 2001:569:be9b:8c00::/64
+# 2001:818:ea7c:5500::/64
+# 2001:871:266:4f57::/64
+# 2001:9e8:2478:9c00::/64
+# 2001:bc8:47b0:1719::/64
+# 2003:e0:712:9e80::/64
+# 2401:4900:190a:9085::/64
+# 2401:4900:60e2:41ca::/64
+# 2405:201:580b:7020::/64
+# 2405:201:a42e:c014::/64
+# 2405:3800:811:bd79::/64
+# 2409:4089:aa84:92f5::/64
+# 2600:387:f:5831::/64
+# 2800:200:e200:1307::/64
+# 2800:200:e800:2e69::/64
+# 2800:200:f140:33f::/64
+# 2800:a4:1a10:6200::/64
+# 2800:af0:103b:4871::/64
+# 2800:e2:637f:fcc1::/64
+# 2804:431:c7d9:8aa0::/64
+# 2804:7efc:42a:3901::/64
+# 2806:2a0:43c:82cb::/64
+# 2806:2f0:72a0:fef9::/64
+# 2a01:4f8:c17:750::/64
+# 2a01:c23:8d83:9200::/64
+# 2a01:cb06:2f9:da00::/64
+# 2a02:2f08:1201:2c00::/64
+# 2a02:2f08:6917:eb00::/64
+# 2a02:2f0d:b819:cb00::/64
+# 2a02:8084:61a3:db00::/64
+# 2a03:b0c0:3:d0::/64
+# 2a0c:5a80:4604:2100::/64
+
+# From Cloudflare list I created called "attacker_ips":
+# (These were created around March 9th.)
+# (None)

container_includes/ip_bans/2023-04-25_ddos.txt

@@ -0,0 +1,28 @@
+# There's another ongoing DDoS, but it appears to be Python clients
+
+# Top client appears to be using Python/3.10 aiohttp/3.8.3
+# Second client appears to be using python-urllib3/1.26.12
+# /tts/job/None is the top URL (191M)
+# /tts/queue_length is the second most visited URL (36k)
+
+# /tts/job/None (191M requests, lots of IPs)
+104.251.211.22
+47.253.48.57
+81.56.107.8
+13.234.107.200
+104.247.114.86
+66.214.82.240
+
+# Unban on 2023-05-19:
+# Vegito1089: I know you are super busy but is it possible to unblock this IP address? 
+# This one paying customer refuses refunds and says he really wants to keep using our website. 
+# His IP got blocked during the second DDOS attack
+# Vegito1089: 38.25.15.183
+# 38.25.15.183
+
+# 2600:1700:4d40:4920:ac48:939e:b6c:9e7
+# 2800:bf0:10b:f82:1516:7cdc:974a:17e8
+
+# /tts/queue_length (14k requests, just one IP)
+173.81.41.29
+

container_includes/troll_bans/ip_address_troll_bans/2023-03-23_101soundboards.com_ipv4_addresses.txt

@@ -0,0 +1,72 @@
+# It looks like 101soundboards has fewer IP addresses they can use (relative to user accounts).
+# IPv4 addresses are the primary pool.
+
+104.143.54.63
+141.98.84.147
+141.98.84.173
+141.98.84.44
+141.98.84.57
+147.78.181.146
+147.78.183.60
+174.107.171.93
+176.119.141.136
+178.20.30.32
+185.102.112.120
+185.102.113.129
+185.102.113.94
+185.152.92.89
+185.61.216.111
+185.61.216.174
+185.61.221.200
+185.81.144.108
+185.81.144.90
+185.81.145.102
+185.81.145.168
+185.81.145.76
+185.88.100.90
+185.89.100.27
+185.96.37.100
+185.96.37.83
+193.202.82.158
+193.203.10.225
+193.203.11.59
+193.31.126.43
+194.156.124.101
+204.116.214.102
+212.119.44.13
+212.119.46.138
+212.119.46.69
+212.119.47.163
+212.119.47.178
+212.119.47.99
+212.60.20.209
+212.60.21.120
+213.108.0.114
+213.108.0.140
+213.108.0.157
+213.108.0.179
+213.108.0.235
+213.108.0.60
+213.108.3.105
+213.108.3.14
+45.10.164.163
+45.10.164.26
+45.10.164.30
+45.10.167.215
+45.138.100.15
+45.148.234.61
+45.66.208.149
+45.66.208.222
+5.181.169.143
+5.181.169.167
+5.181.169.220
+5.181.169.35
+5.181.169.83
+5.183.252.70
+72.219.85.206
+76.0.25.232
+77.220.192.46
+83.142.52.192
+83.142.54.125
+83.142.55.165
+98.143.187.29

container_includes/troll_bans/ip_address_troll_bans/2023-03-23_101soundboards.com_ipv6_addresses.txt

@@ -0,0 +1,9 @@
+# It looks like 101soundboards has fewer IP addresses they can use (relative to user accounts).
+# They have a few IPv6 addresses that are infrequently used.
+
+# TODO: Support IPv6 addresses and CIDRs
+# 2600:6c64:4c7f:f11f:4171:9854:beed:6dde
+# 2601:205:c380:6810:5860:82ac:f25c:81de
+# 2601:300:4400:3290:1864:2e36:fdab:dd9b
+# 2601:648:8300:4cc0:a9c7:d8bc:9614:c76d
+# 2603:9001:3f00:4e29:e4e9:56c0:ffc:f1d5

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

@@ -1,6 +1,7 @@
 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 : &'static str = "original_upload.bin";
 
@@ -37,6 +38,10 @@ impl MediaUploadOriginalFilePath {
     &self.full_object_path
   }
 
+  pub fn to_full_object_pathbuf(&self) -> PathBuf {
+    PathBuf::from(&self.full_object_path)
+  }
+
   pub fn get_directory(&self) -> &MediaUploadDirectory {
     &self.directory
   }
@@ -53,6 +58,7 @@ impl MediaUploadOriginalFilePath {
 #[cfg(test)]
 mod tests {
   use crate::public::media_uploads::original_file::MediaUploadOriginalFilePath;
+  use std::path::PathBuf;
 
   #[test]
   pub fn generate_new_entropy() {
@@ -67,6 +73,12 @@ mod tests {
     assert_eq!(file.get_full_object_path_str(), "/media_upload/a/b/c/d/e/abcdefghijk/original_upload.bin");
   }
 
+  #[test]
+  pub fn to_full_object_pathbuf() {
+    let file = MediaUploadOriginalFilePath::from_object_hash("abcdefghijk");
+    assert_eq!(file.to_full_object_pathbuf(), PathBuf::from("/media_upload/a/b/c/d/e/abcdefghijk/original_upload.bin"));
+  }
+
   #[test]
   pub fn get_full_object_path_str_short_name() {
     let file = MediaUploadOriginalFilePath::from_object_hash("foo");

crates/_types/buckets/src/public/mod.rs

@@ -1,2 +1,3 @@
 pub mod media_uploads;
-pub mod public_path;
+pub mod public_path;
+pub mod voice_conversion_results;

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

@@ -0,0 +1,63 @@
+use crate::public::public_path::PublicPath;
+use crate::util::hashed_directory_path_long_string::hashed_directory_path_long_string;
+
+// TODO: Generate these from a macro.
+
+// TODO: Use a central path registry for quick reference
+const DIRECTORY: &'static str = "/voice_conversion";
+
+/// Directory for voice conversion results.
+/// Each result gets its own directory so that we can store the result
+/// alongside metadata and re-processed derivatives.
+#[derive(Clone)]
+pub struct VoiceConversionResultDirectory {
+  object_hash: String,
+  directory: String,
+}
+
+impl PublicPath for VoiceConversionResultDirectory {}
+
+impl VoiceConversionResultDirectory {
+
+  pub fn from_public_bucket_directory_hash(object_hash: &str) -> Self {
+    // TODO: Path construction could be cleaner.
+    let middle = hashed_directory_path_long_string(object_hash);
+    let directory = format!("{}/{}{}", DIRECTORY, middle, object_hash);
+    Self {
+      object_hash: object_hash.to_string(),
+      directory,
+    }
+  }
+
+  pub fn get_directory_path_str(&self) -> &str {
+    &self.directory
+  }
+
+  pub fn get_object_hash(&self) -> &str {
+    &self.object_hash
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::public::voice_conversion_results::directory::VoiceConversionResultDirectory;
+
+  #[test]
+  pub fn get_directory_path_str() {
+    let directory = VoiceConversionResultDirectory::from_public_bucket_directory_hash("abcdefghijk");
+    assert_eq!(directory.get_directory_path_str(), "/voice_conversion/a/b/c/d/e/abcdefghijk");
+  }
+
+  #[test]
+  pub fn get_directory_path_str_short_name() {
+    let directory = VoiceConversionResultDirectory::from_public_bucket_directory_hash("foo");
+    assert_eq!(directory.get_directory_path_str(), "/voice_conversion/f/o/foo");
+  }
+
+  #[test]
+  pub fn get_object_hash() {
+    let hash = "abcdefghijk";
+    let directory = VoiceConversionResultDirectory::from_public_bucket_directory_hash(hash);
+    assert_eq!(directory.get_object_hash(), hash);
+  }
+}

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

@@ -0,0 +1,2 @@
+pub mod directory;
+pub mod original_file;

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

@@ -0,0 +1,105 @@
+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 : &'static str = "fakeyou_";
+const ORIGINAL_FILE_SUFFIX : &'static str = ".wav";
+
+// TODO: Generate these from a macro.
+
+/// The original user upload file.
+/// It may have derivative files (down samples, crops, etc.) that live alongside it.
+#[derive(Clone)]
+pub struct VoiceConversionResultOriginalFilePath {
+  directory: VoiceConversionResultDirectory,
+  basename: String,
+  full_object_path: String,
+}
+
+impl PublicPath for VoiceConversionResultOriginalFilePath {}
+
+impl VoiceConversionResultOriginalFilePath {
+
+  pub fn generate_new() -> Self {
+    let entropy = crockford_entropy_lower(32);
+    Self::from_object_hash(&entropy)
+  }
+
+  pub fn from_object_hash(hash: &str) -> Self {
+    // TODO: Path construction could be cleaner.
+    let directory = VoiceConversionResultDirectory::from_public_bucket_directory_hash(hash);
+    let basename = format!("{}{}{}", ORIGINAL_FILE_PREFIX, hash, ORIGINAL_FILE_SUFFIX);
+    let full_object_path = format!("{}/{}{}{}", directory.get_directory_path_str(), ORIGINAL_FILE_PREFIX, hash, ORIGINAL_FILE_SUFFIX);
+    Self {
+      directory,
+      basename,
+      full_object_path,
+    }
+  }
+
+  pub fn get_full_object_path_str(&self) -> &str {
+    &self.full_object_path
+  }
+
+  pub fn to_full_object_pathbuf(&self) -> PathBuf {
+    PathBuf::from(&self.full_object_path)
+  }
+
+  pub fn get_directory(&self) -> &VoiceConversionResultDirectory {
+    &self.directory
+  }
+
+  pub fn get_object_hash(&self) -> &str {
+    &self.directory.get_object_hash()
+  }
+
+  pub fn get_basename(&self) -> &str {
+    &self.basename
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::public::voice_conversion_results::original_file::VoiceConversionResultOriginalFilePath;
+  use std::path::PathBuf;
+
+  #[test]
+  pub fn generate_new_entropy() {
+    let file = VoiceConversionResultOriginalFilePath::generate_new();
+    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 = VoiceConversionResultOriginalFilePath::from_object_hash("abcdefghijk");
+    assert_eq!(file.get_full_object_path_str(), "/voice_conversion/a/b/c/d/e/abcdefghijk/fakeyou_abcdefghijk.wav");
+  }
+
+  #[test]
+  pub fn to_full_object_pathbuf() {
+    let file = VoiceConversionResultOriginalFilePath::from_object_hash("abcdefghijk");
+    assert_eq!(file.to_full_object_pathbuf(), PathBuf::from("/voice_conversion/a/b/c/d/e/abcdefghijk/fakeyou_abcdefghijk.wav"));
+  }
+
+  #[test]
+  pub fn get_full_object_path_str_short_name() {
+    let file = VoiceConversionResultOriginalFilePath::from_object_hash("foo");
+    assert_eq!(file.get_full_object_path_str(), "/voice_conversion/f/o/foo/fakeyou_foo.wav");
+  }
+
+  #[test]
+  pub fn get_full_object_path_str_starts_with_directory() {
+    let file = VoiceConversionResultOriginalFilePath::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 get_object_hash() {
+    let hash = "abcdefghijk";
+    let file = VoiceConversionResultOriginalFilePath::from_object_hash(hash);
+    assert_eq!(file.get_object_hash(), hash);
+  }
+}

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

@@ -0,0 +1,84 @@
+use serde::Deserialize;
+use serde::Serialize;
+
+#[cfg(test)]
+use strum::EnumCount;
+#[cfg(test)]
+use strum::EnumIter;
+
+/// Used in the `audit_logs` table in a `VARCHAR(32)` field named `entity_action`.
+#[cfg_attr(test, derive(EnumIter, EnumCount))]
+#[derive(Clone, Copy, Eq, PartialEq, Hash, Deserialize, Serialize)]
+pub enum AuditLogEntityAction {
+  /// Create action
+  #[serde(rename = "create")]
+  Create,
+
+  /// Edit action
+  #[serde(rename = "edit")]
+  Edit,
+
+  /// Delete action
+  #[serde(rename = "delete")]
+  Delete,
+}
+
+// TODO(bt, 2023-01-17): This desperately needs MySQL integration tests!
+impl_enum_display_and_debug_using_to_str!(AuditLogEntityAction);
+impl_mysql_enum_coders!(AuditLogEntityAction);
+
+/// NB: Legacy API for older code.
+impl AuditLogEntityAction {
+  pub fn to_str(&self) -> &'static str {
+    match self {
+      Self::Create => "create",
+      Self::Edit => "edit",
+      Self::Delete => "delete",
+    }
+  }
+
+  pub fn from_str(value: &str) -> Result<Self, String> {
+    match value {
+      "create" => Ok(Self::Create),
+      "edit" => Ok(Self::Edit),
+      "delete" => Ok(Self::Delete),
+      _ => Err(format!("invalid value: {:?}", value)),
+    }
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::by_table::audit_logs::audit_log_entity_action::AuditLogEntityAction;
+  use crate::test_helpers::assert_serialization;
+
+  mod serde {
+    use super::*;
+
+    #[test]
+    fn test_serialization() {
+      assert_serialization(AuditLogEntityAction::Create, "create");
+      assert_serialization(AuditLogEntityAction::Edit, "edit");
+      assert_serialization(AuditLogEntityAction::Delete, "delete");
+    }
+  }
+
+  mod impl_methods {
+    use super::*;
+
+    #[test]
+    fn test_to_str() {
+      assert_eq!(AuditLogEntityAction::Create.to_str(), "create");
+      assert_eq!(AuditLogEntityAction::Edit.to_str(), "edit");
+      assert_eq!(AuditLogEntityAction::Delete.to_str(), "delete");
+    }
+
+    #[test]
+    fn test_from_str() {
+      assert_eq!(AuditLogEntityAction::from_str("create").unwrap(), AuditLogEntityAction::Create);
+      assert_eq!(AuditLogEntityAction::from_str("edit").unwrap(), AuditLogEntityAction::Edit);
+      assert_eq!(AuditLogEntityAction::from_str("delete").unwrap(), AuditLogEntityAction::Delete);
+      assert!(AuditLogEntityAction::from_str("foo").is_err());
+    }
+  }
+}

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

@@ -0,0 +1,102 @@
+use serde::Deserialize;
+use serde::Serialize;
+
+#[cfg(test)]
+use strum::EnumCount;
+#[cfg(test)]
+use strum::EnumIter;
+
+/// Used in the `audit_logs` table in a `VARCHAR(32)` field named `entity_type`.
+#[cfg_attr(test, derive(EnumIter, EnumCount))]
+#[derive(Clone, Copy, Eq, PartialEq, Hash, Deserialize, Serialize)]
+pub enum AuditLogEntityType {
+  /// Comment system
+  #[serde(rename = "comment")]
+  Comment,
+
+  /// TTS model (architecture does not matter)
+  #[serde(rename = "tts_model")]
+  TtsModel,
+
+  /// TTS result (architecture does not matter)
+  #[serde(rename = "tts_result")]
+  TtsResult,
+
+  /// W2L template
+  #[serde(rename = "w2l_template")]
+  W2lTemplate,
+
+  /// W2L result
+  #[serde(rename = "w2l_result")]
+  W2lResult,
+}
+
+// TODO(bt, 2023-01-17): This desperately needs MySQL integration tests!
+impl_enum_display_and_debug_using_to_str!(AuditLogEntityType);
+impl_mysql_enum_coders!(AuditLogEntityType);
+
+/// NB: Legacy API for older code.
+impl AuditLogEntityType {
+  pub fn to_str(&self) -> &'static str {
+    match self {
+      Self::Comment => "comment",
+      Self::TtsModel => "tts_model",
+      Self::TtsResult => "tts_result",
+      Self::W2lTemplate => "w2l_template",
+      Self::W2lResult => "w2l_result",
+    }
+  }
+
+  pub fn from_str(value: &str) -> Result<Self, String> {
+    match value {
+      "comment" => Ok(Self::Comment),
+      "tts_model" => Ok(Self::TtsModel),
+      "tts_result" => Ok(Self::TtsResult),
+      "w2l_template" => Ok(Self::W2lTemplate),
+      "w2l_result" => Ok(Self::W2lResult),
+      _ => Err(format!("invalid value: {:?}", value)),
+    }
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::by_table::audit_logs::audit_log_entity_type::AuditLogEntityType;
+  use crate::test_helpers::assert_serialization;
+
+  mod serde {
+    use super::*;
+
+    #[test]
+    fn test_serialization() {
+      assert_serialization(AuditLogEntityType::Comment, "comment");
+      assert_serialization(AuditLogEntityType::TtsModel, "tts_model");
+      assert_serialization(AuditLogEntityType::TtsResult, "tts_result");
+      assert_serialization(AuditLogEntityType::W2lTemplate, "w2l_template");
+      assert_serialization(AuditLogEntityType::W2lResult, "w2l_result");
+    }
+  }
+
+  mod impl_methods {
+    use super::*;
+
+    #[test]
+    fn test_to_str() {
+      assert_eq!(AuditLogEntityType::Comment.to_str(), "comment");
+      assert_eq!(AuditLogEntityType::TtsModel.to_str(), "tts_model");
+      assert_eq!(AuditLogEntityType::TtsResult.to_str(), "tts_result");
+      assert_eq!(AuditLogEntityType::W2lTemplate.to_str(), "w2l_template");
+      assert_eq!(AuditLogEntityType::W2lResult.to_str(), "w2l_result");
+    }
+
+    #[test]
+    fn test_from_str() {
+      assert_eq!(AuditLogEntityType::from_str("comment").unwrap(), AuditLogEntityType::Comment);
+      assert_eq!(AuditLogEntityType::from_str("tts_model").unwrap(), AuditLogEntityType::TtsModel);
+      assert_eq!(AuditLogEntityType::from_str("tts_result").unwrap(), AuditLogEntityType::TtsResult);
+      assert_eq!(AuditLogEntityType::from_str("w2l_template").unwrap(), AuditLogEntityType::W2lTemplate);
+      assert_eq!(AuditLogEntityType::from_str("w2l_result").unwrap(), AuditLogEntityType::W2lResult);
+      assert!(AuditLogEntityType::from_str("foo").is_err());
+    }
+  }
+}

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

@@ -0,0 +1,2 @@
+pub mod audit_log_entity_action;
+pub mod audit_log_entity_type;

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

@@ -0,0 +1,102 @@
+use serde::Deserialize;
+use serde::Serialize;
+
+#[cfg(test)]
+use strum::EnumCount;
+#[cfg(test)]
+use strum::EnumIter;
+
+/// Used in the `comments` table in a `VARCHAR(32)` field named `entity_type`.
+#[cfg_attr(test, derive(EnumIter, EnumCount))]
+#[derive(Clone, Copy, Eq, PartialEq, Hash, Deserialize, Serialize)]
+pub enum CommentEntityType {
+  /// User
+  #[serde(rename = "user")]
+  User,
+
+  /// TTS model (architecture does not matter)
+  #[serde(rename = "tts_model")]
+  TtsModel,
+
+  /// TTS result (architecture does not matter)
+  #[serde(rename = "tts_result")]
+  TtsResult,
+
+  /// W2L template
+  #[serde(rename = "w2l_template")]
+  W2lTemplate,
+
+  /// W2L result
+  #[serde(rename = "w2l_result")]
+  W2lResult,
+}
+
+// TODO(bt, 2023-01-17): This desperately needs MySQL integration tests!
+impl_enum_display_and_debug_using_to_str!(CommentEntityType);
+impl_mysql_enum_coders!(CommentEntityType);
+
+/// NB: Legacy API for older code.
+impl CommentEntityType {
+  pub fn to_str(&self) -> &'static str {
+    match self {
+      Self::User => "user",
+      Self::TtsModel => "tts_model",
+      Self::TtsResult => "tts_result",
+      Self::W2lTemplate => "w2l_template",
+      Self::W2lResult => "w2l_result",
+    }
+  }
+
+  pub fn from_str(value: &str) -> Result<Self, String> {
+    match value {
+      "user" => Ok(Self::User),
+      "tts_model" => Ok(Self::TtsModel),
+      "tts_result" => Ok(Self::TtsResult),
+      "w2l_template" => Ok(Self::W2lTemplate),
+      "w2l_result" => Ok(Self::W2lResult),
+      _ => Err(format!("invalid value: {:?}", value)),
+    }
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::by_table::comments::comment_entity_type::CommentEntityType;
+  use crate::test_helpers::assert_serialization;
+
+  mod serde {
+    use super::*;
+
+    #[test]
+    fn test_serialization() {
+      assert_serialization(CommentEntityType::User, "user");
+      assert_serialization(CommentEntityType::TtsModel, "tts_model");
+      assert_serialization(CommentEntityType::TtsResult, "tts_result");
+      assert_serialization(CommentEntityType::W2lTemplate, "w2l_template");
+      assert_serialization(CommentEntityType::W2lResult, "w2l_result");
+    }
+  }
+
+  mod impl_methods {
+    use super::*;
+
+    #[test]
+    fn test_to_str() {
+      assert_eq!(CommentEntityType::User.to_str(), "user");
+      assert_eq!(CommentEntityType::TtsModel.to_str(), "tts_model");
+      assert_eq!(CommentEntityType::TtsResult.to_str(), "tts_result");
+      assert_eq!(CommentEntityType::W2lTemplate.to_str(), "w2l_template");
+      assert_eq!(CommentEntityType::W2lResult.to_str(), "w2l_result");
+    }
+
+    #[test]
+    fn test_from_str() {
+      assert_eq!(CommentEntityType::from_str("user").unwrap(), CommentEntityType::User);
+      assert_eq!(CommentEntityType::from_str("tts_model").unwrap(), CommentEntityType::TtsModel);
+      assert_eq!(CommentEntityType::from_str("tts_result").unwrap(), CommentEntityType::TtsResult);
+      assert_eq!(CommentEntityType::from_str("w2l_template").unwrap(), CommentEntityType::W2lTemplate);
+      assert_eq!(CommentEntityType::from_str("w2l_result").unwrap(), CommentEntityType::W2lResult);
+      assert!(CommentEntityType::from_str("foo").is_err());
+    }
+  }
+}

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

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

crates/_types/enums/src/by_table/generic_download_jobs/generic_download_type.rs

@@ -22,6 +22,12 @@ pub enum GenericDownloadType {
   #[sqlx(rename = "hifigan_rocket_vc")]
   HifiGanRocketVc,
 
+  /// NB: Note - this is hifigan for SoVitsSvc
+  /// Some work will need to be done to unify this with other hifigan types.
+  #[serde(rename = "hifigan_so_vits_svc")]
+  #[sqlx(rename = "hifigan_so_vits_svc")]
+  HifiGanSoVitsSvc,
+
   //#[serde(rename = "melgan_vocodes")]
   //#[sqlx(rename = "melgan_vocodes")]
   //MelGanVocodes,
@@ -32,10 +38,20 @@ pub enum GenericDownloadType {
   #[sqlx(rename = "rocket_vc")]
   RocketVc,
 
+  /// so-vits-svc voice conversion models
+  #[serde(rename = "so_vits_svc")]
+  #[sqlx(rename = "so_vits_svc")]
+  SoVitsSvc,
+
   /// Tacotron TTS models.
   #[serde(rename = "tacotron2")]
   #[sqlx(rename = "tacotron2")]
   Tacotron2,
+
+  /// VITS TTS models.
+  #[serde(rename = "vits")]
+  #[sqlx(rename = "vits")]
+  Vits,
 }
 
 /// NB: Legacy API for older code.
@@ -44,8 +60,11 @@ impl GenericDownloadType {
     match self {
       Self::HifiGan => "hifigan",
       Self::HifiGanRocketVc => "hifigan_rocket_vc",
+      Self::HifiGanSoVitsSvc => "hifigan_so_vits_svc",
       Self::RocketVc => "rocket_vc",
+      Self::SoVitsSvc => "so_vits_svc",
       Self::Tacotron2 => "tacotron2",
+      Self::Vits => "vits",
     }
   }
 
@@ -53,8 +72,11 @@ impl GenericDownloadType {
     match value {
       "hifigan" => Ok(Self::HifiGan),
       "hifigan_rocket_vc" => Ok(Self::HifiGanRocketVc),
+      "hifigan_so_vits_svc" => Ok(Self::HifiGanSoVitsSvc),
       "rocket_vc" => Ok(Self::RocketVc),
+      "so_vits_svc" => Ok(Self::SoVitsSvc),
       "tacotron2" => Ok(Self::Tacotron2),
+      "vits" => Ok(Self::Vits),
       _ => Err(format!("invalid value: {:?}", value)),
     }
   }
@@ -62,30 +84,39 @@ impl GenericDownloadType {
 
 #[cfg(test)]
 mod tests {
-  use crate::workers::generic_download_type::GenericDownloadType;
+  use crate::by_table::generic_download_jobs::generic_download_type::GenericDownloadType;
   use crate::test_helpers::assert_serialization;
 
   #[test]
   fn test_serialization() {
     assert_serialization(GenericDownloadType::HifiGan, "hifigan");
     assert_serialization(GenericDownloadType::HifiGanRocketVc, "hifigan_rocket_vc");
+    assert_serialization(GenericDownloadType::HifiGanSoVitsSvc, "hifigan_so_vits_svc");
     assert_serialization(GenericDownloadType::RocketVc, "rocket_vc");
+    assert_serialization(GenericDownloadType::SoVitsSvc, "so_vits_svc");
     assert_serialization(GenericDownloadType::Tacotron2, "tacotron2");
+    assert_serialization(GenericDownloadType::Vits, "vits");
   }
 
   #[test]
   fn to_str() {
     assert_eq!(GenericDownloadType::HifiGan.to_str(), "hifigan");
     assert_eq!(GenericDownloadType::HifiGanRocketVc.to_str(), "hifigan_rocket_vc");
+    assert_eq!(GenericDownloadType::HifiGanSoVitsSvc.to_str(), "hifigan_so_vits_svc");
     assert_eq!(GenericDownloadType::RocketVc.to_str(), "rocket_vc");
+    assert_eq!(GenericDownloadType::SoVitsSvc.to_str(), "so_vits_svc");
     assert_eq!(GenericDownloadType::Tacotron2.to_str(), "tacotron2");
+    assert_eq!(GenericDownloadType::Vits.to_str(), "vits");
   }
 
   #[test]
   fn from_str() {
     assert_eq!(GenericDownloadType::from_str("hifigan").unwrap(), GenericDownloadType::HifiGan);
     assert_eq!(GenericDownloadType::from_str("hifigan_rocket_vc").unwrap(), GenericDownloadType::HifiGanRocketVc);
+    assert_eq!(GenericDownloadType::from_str("hifigan_so_vits_svc").unwrap(), GenericDownloadType::HifiGanSoVitsSvc);
     assert_eq!(GenericDownloadType::from_str("rocket_vc").unwrap(), GenericDownloadType::RocketVc);
+    assert_eq!(GenericDownloadType::from_str("so_vits_svc").unwrap(), GenericDownloadType::SoVitsSvc);
     assert_eq!(GenericDownloadType::from_str("tacotron2").unwrap(), GenericDownloadType::Tacotron2);
+    assert_eq!(GenericDownloadType::from_str("vits").unwrap(), GenericDownloadType::Vits);
   }
 }

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

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

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

@@ -0,0 +1,93 @@
+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 `inference_category`.
+///
+/// Our "generic inference" pipeline supports a wide variety of ML models and other media.
+/// Each "category" of inference is identified by the following enum variants.
+/// These types are present in the HTTP API and database columns as serialized here.
+///
+/// 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 InferenceCategory {
+  #[serde(rename = "text_to_speech")]
+  TextToSpeech,
+
+  #[serde(rename = "voice_conversion")]
+  VoiceConversion,
+}
+
+// TODO(bt, 2022-12-21): This desperately needs MySQL integration tests!
+impl_enum_display_and_debug_using_to_str!(InferenceCategory);
+impl_mysql_enum_coders!(InferenceCategory);
+
+/// NB: Legacy API for older code.
+impl InferenceCategory {
+  pub fn to_str(&self) -> &'static str {
+    match self {
+      Self::TextToSpeech => "text_to_speech",
+      Self::VoiceConversion => "voice_conversion",
+    }
+  }
+
+  pub fn from_str(value: &str) -> Result<Self, String> {
+    match value {
+      "text_to_speech" => Ok(Self::TextToSpeech),
+      "voice_conversion" => Ok(Self::VoiceConversion),
+      _ => 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::TextToSpeech,
+      Self::VoiceConversion,
+    ])
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::by_table::generic_inference_jobs::inference_category::InferenceCategory;
+  use crate::test_helpers::assert_serialization;
+
+  #[test]
+  fn test_serialization() {
+    assert_serialization(InferenceCategory::TextToSpeech, "text_to_speech");
+    assert_serialization(InferenceCategory::VoiceConversion, "voice_conversion");
+  }
+
+  #[test]
+  fn to_str() {
+    assert_eq!(InferenceCategory::TextToSpeech.to_str(), "text_to_speech");
+    assert_eq!(InferenceCategory::VoiceConversion.to_str(), "voice_conversion");
+  }
+
+  #[test]
+  fn from_str() {
+    assert_eq!(InferenceCategory::from_str("text_to_speech").unwrap(), InferenceCategory::TextToSpeech);
+    assert_eq!(InferenceCategory::from_str("voice_conversion").unwrap(), InferenceCategory::VoiceConversion);
+  }
+
+  #[test]
+  fn all_variants() {
+    // Static check
+    let mut variants = InferenceCategory::all_variants();
+    assert_eq!(variants.len(), 2);
+    assert_eq!(variants.pop_first(), Some(InferenceCategory::TextToSpeech));
+    assert_eq!(variants.pop_first(), Some(InferenceCategory::VoiceConversion));
+    assert_eq!(variants.pop_first(), None);
+
+    // Generated check
+    use strum::IntoEnumIterator;
+    assert_eq!(InferenceCategory::all_variants().len(), InferenceCategory::iter().len());
+  }
+}

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

@@ -0,0 +1,79 @@
+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 `maybe_input_source_token`.
+///
+/// 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 InferenceInputSourceTokenType {
+  #[serde(rename = "media_upload")]
+  MediaUpload,
+}
+
+// TODO(bt, 2022-12-21): This desperately needs MySQL integration tests!
+impl_enum_display_and_debug_using_to_str!(InferenceInputSourceTokenType);
+impl_mysql_enum_coders!(InferenceInputSourceTokenType);
+
+/// NB: Legacy API for older code.
+impl InferenceInputSourceTokenType {
+  pub fn to_str(&self) -> &'static str {
+    match self {
+      Self::MediaUpload => "media_upload",
+    }
+  }
+
+  pub fn from_str(value: &str) -> Result<Self, String> {
+    match value {
+      "media_upload" => Ok(Self::MediaUpload),
+      _ => Err(format!("invalid value: {:?}", value)),
+    }
+  }
+
+  pub fn all_variants() -> BTreeSet<InferenceInputSourceTokenType> {
+    // NB: BTreeSet is sorted
+    // NB: BTreeSet::from() isn't const, but not worth using LazyStatic, etc.
+    BTreeSet::from([
+      InferenceInputSourceTokenType::MediaUpload,
+    ])
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::by_table::generic_inference_jobs::inference_input_source_token_type::InferenceInputSourceTokenType;
+  use crate::test_helpers::assert_serialization;
+
+  #[test]
+  fn test_serialization() {
+    assert_serialization(InferenceInputSourceTokenType::MediaUpload, "media_upload");
+  }
+
+  #[test]
+  fn to_str() {
+    assert_eq!(InferenceInputSourceTokenType::MediaUpload.to_str(), "media_upload");
+  }
+
+  #[test]
+  fn from_str() {
+    assert_eq!(InferenceInputSourceTokenType::from_str("media_upload").unwrap(), InferenceInputSourceTokenType::MediaUpload);
+  }
+
+  #[test]
+  fn all_variants() {
+    // Static check
+    let mut variants = InferenceInputSourceTokenType::all_variants();
+    assert_eq!(variants.len(), 1);
+    assert_eq!(variants.pop_first(), Some(InferenceInputSourceTokenType::MediaUpload));
+    assert_eq!(variants.pop_first(), None);
+
+    // Generated check
+    use strum::IntoEnumIterator;
+    assert_eq!(InferenceInputSourceTokenType::all_variants().len(), InferenceInputSourceTokenType::iter().len());
+  }
+}

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

@@ -0,0 +1,106 @@
+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 `maybe_model_type`.
+///
+/// Our "generic inference" pipeline supports a wide variety of ML models and other media.
+/// Each inference "model type" identified by the following enum variants, though some pipelines
+/// may use multiple models or no model (and may report NULL).
+///
+/// These types are present in the HTTP API and database columns as serialized here.
+///
+/// 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 InferenceModelType {
+  #[serde(rename = "so_vits_svc")]
+  SoVitsSvc,
+
+  // TODO: Does this need to be "legacy_tacotron2" ?
+  #[serde(rename = "tacotron2")]
+  Tacotron2,
+
+  #[serde(rename = "vits")]
+  Vits,
+}
+
+// TODO(bt, 2022-12-21): This desperately needs MySQL integration tests!
+impl_enum_display_and_debug_using_to_str!(InferenceModelType);
+impl_mysql_enum_coders!(InferenceModelType);
+
+/// NB: Legacy API for older code.
+impl InferenceModelType {
+  pub fn to_str(&self) -> &'static str {
+    match self {
+      Self::SoVitsSvc => "so_vits_svc",
+      Self::Tacotron2 => "tacotron2",
+      Self::Vits => "vits",
+    }
+  }
+
+  pub fn from_str(value: &str) -> Result<Self, String> {
+    match value {
+      "so_vits_svc" => Ok(Self::SoVitsSvc),
+      "tacotron2" => Ok(Self::Tacotron2),
+      "vits" => Ok(Self::Vits),
+      _ => Err(format!("invalid value: {:?}", value)),
+    }
+  }
+
+  pub fn all_variants() -> BTreeSet<InferenceModelType> {
+    // NB: BTreeSet is sorted
+    // NB: BTreeSet::from() isn't const, but not worth using LazyStatic, etc.
+    BTreeSet::from([
+      InferenceModelType::SoVitsSvc,
+      InferenceModelType::Tacotron2,
+      InferenceModelType::Vits,
+    ])
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::by_table::generic_inference_jobs::inference_model_type::InferenceModelType;
+  use crate::test_helpers::assert_serialization;
+
+  #[test]
+  fn test_serialization() {
+    assert_serialization(InferenceModelType::SoVitsSvc, "so_vits_svc");
+    assert_serialization(InferenceModelType::Tacotron2, "tacotron2");
+    assert_serialization(InferenceModelType::Vits, "vits");
+  }
+
+  #[test]
+  fn to_str() {
+    assert_eq!(InferenceModelType::SoVitsSvc.to_str(), "so_vits_svc");
+    assert_eq!(InferenceModelType::Tacotron2.to_str(), "tacotron2");
+    assert_eq!(InferenceModelType::Vits.to_str(), "vits");
+  }
+
+  #[test]
+  fn from_str() {
+    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);
+  }
+
+  #[test]
+  fn all_variants() {
+    // Static check
+    let mut variants = InferenceModelType::all_variants();
+    assert_eq!(variants.len(), 3);
+    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(), None);
+
+    // Generated check
+    use strum::IntoEnumIterator;
+    assert_eq!(InferenceModelType::all_variants().len(), InferenceModelType::iter().len());
+  }
+}