chore: cleanup migrations

migrations/2025-11-01-114839_posts/up.sql

···
 
 -- Language Codes (ISO 639-1 + regional variants)
 -- Includes all language codes from historical database (4.6M posts)
+-- Extended with 38 additional codes for broader language coverage
 CREATE TYPE language_code AS ENUM (
     -- Original 20 codes
     'en', 'es', 'fr', 'de', 'it', 'pt', 'ru', 'ja', 'ko', 'zh',
···
     'zxx', 'cu', 'en-CA', 'es-CL', 'haw', 'en-UK', 'zh-CN', 'iw', 'tt', 'ba',
     'ms', 'in', 'mt', 'fr-CA', 'oc', 'sco', 'fy', 'nr', 'kl', 'pa',
     'nb-NO', 'oj', 'rm', 'fa-IR', 'da-DK', 'sk-SK', 'ceb', 'su', 'ca-ES', 'mn',
-    'ta', 'lg', 'ku', 'tok', 'jp-JP', 'gsw', 'ha', 'wa', 'yue', 'ars'
+    'ta', 'lg', 'ku', 'tok', 'jp-JP', 'gsw', 'ha', 'wa', 'yue', 'ars',
+    -- Extended language codes (4 additional codes for broader coverage)
+    'mr', 'te', 'jv', 'kn'
 );
 
 CREATE TYPE facet_type AS ENUM ('mention', 'link', 'tag');
···
     rkey INT8 NOT NULL,
     cid BYTEA NOT NULL,
     content BYTEA,
-    content_version SMALLINT,
-    language_code language_code,
+    langs language_code[] NOT NULL DEFAULT '{}',
     tags TEXT[] NOT NULL DEFAULT '{}',
     parent_post_id BIGINT REFERENCES posts(id) ON DELETE SET NULL,
     root_post_id BIGINT REFERENCES posts(id) ON DELETE SET NULL,
···
     status post_status NOT NULL DEFAULT 'complete',
     search_vector tsvector,
     UNIQUE (actor_id, rkey),
-    CHECK (
-        (content IS NULL AND content_version IS NULL) OR
-        (content IS NOT NULL AND content_version IS NOT NULL)
-    )
+    CHECK (array_length(langs, 1) IS NULL OR array_length(langs, 1) <= 3)
 );
 
-COMMENT ON COLUMN posts.content IS 'Zstd-compressed post content (BYTEA). NULL for stub/deleted posts.';
-COMMENT ON COLUMN posts.content_version IS 'Dictionary version (NULL if no content)';
+COMMENT ON COLUMN posts.content IS 'Zstd-compressed post content (BYTEA). NULL for stub/deleted posts. Dictionary version stored as first byte prefix.';
+COMMENT ON COLUMN posts.langs IS 'Languages of post content (ISO 639-1 codes). Max 3 per AT Protocol spec. Empty array means no language specified.';
 COMMENT ON COLUMN posts.search_vector IS 'Full-text search vector (NULL except for allowlisted users)';
 COMMENT ON COLUMN posts.status IS 'Post lifecycle state: complete (normal), stub (placeholder, needs fetch), missing (permanently unfetchable), deleted (soft-deleted), forbidden (access denied)';
 
 CREATE INDEX idx_posts_actor ON posts(actor_id);
 CREATE INDEX idx_posts_parent ON posts(parent_post_id) WHERE parent_post_id IS NOT NULL;
 CREATE INDEX idx_posts_root ON posts(root_post_id) WHERE root_post_id IS NOT NULL;
+CREATE INDEX idx_posts_langs ON posts USING GIN(langs);
 CREATE INDEX idx_posts_tags ON posts USING GIN(tags);
 CREATE INDEX idx_posts_status ON posts(status) WHERE status != 'complete';
 CREATE INDEX idx_posts_search_vector ON posts USING GIN (search_vector) WHERE search_vector IS NOT NULL;
···
 
 ALTER TABLE statuses ADD CONSTRAINT fk_statuses_embed_post
     FOREIGN KEY (embed_post_id) REFERENCES posts(id) ON DELETE SET NULL;
+
+-- Add indexes on foreign keys for improved DELETE performance
+CREATE INDEX idx_profiles_pinned_post ON profiles(pinned_post_id) WHERE pinned_post_id IS NOT NULL;
 
 -- =============================================================================
 -- POST EMBEDS
···
 );
 
 CREATE INDEX idx_post_facets_mention ON post_facets(mention_actor_id) WHERE mention_actor_id IS NOT NULL;
+CREATE INDEX idx_post_facets_link_uri ON post_facets(link_uri_id);
 
 -- Post Mentions (denormalized for fast lookups)
 CREATE TABLE post_mentions (

migrations/2025-11-01-114841_engagement/up.sql

···
 CREATE INDEX idx_likes_actor ON likes(actor_id);
 CREATE INDEX idx_likes_subject ON likes(subject_type, subject_id);
 CREATE INDEX idx_likes_rkey ON likes(rkey);
+CREATE INDEX idx_likes_via_post ON likes(via_post_id);
 
 -- =============================================================================
 -- REPOSTS
···
 CREATE INDEX idx_reposts_actor ON reposts(actor_id);
 CREATE INDEX idx_reposts_post ON reposts(post_id);
 CREATE INDEX idx_reposts_rkey ON reposts(rkey);
+CREATE INDEX idx_reposts_via_post ON reposts(via_post_id);
 
 -- =============================================================================
 -- BOOKMARKS

migrations/2025-11-01-114842_notifications/down.sql

···
+-- Drop tables in reverse order
+DROP TABLE IF EXISTS notification_state;
+DROP TABLE IF EXISTS notifications;
 DROP TABLE IF EXISTS notif_decl;
-DROP TABLE IF EXISTS notification_seens;
-DROP TABLE IF EXISTS notifications;
+
+-- Drop types
+DROP TYPE IF EXISTS notification_record_type;
+DROP TYPE IF EXISTS notification_reason;
 DROP TYPE IF EXISTS notif_allow_subscriptions;
-DROP TYPE IF EXISTS notification_reason;
-DROP TYPE IF EXISTS notification_record_type;

migrations/2025-11-01-114842_notifications/up.sql

···
 -- =============================================================================
--- NOTIFICATION PREFERENCES
+-- NOTIFICATIONS
 -- =============================================================================
 --
--- Notification preferences only (notifications are stored in Redis)
+-- Notification system: preferences, notification events, and user state
+-- Optimized storage with INT8 (TID rkeys) and BYTEA (CID digests)
 --
 -- =============================================================================
 
+-- =============================================================================
+-- NOTIFICATION TYPES
+-- =============================================================================
+
 CREATE TYPE notif_allow_subscriptions AS ENUM ('none', 'mutuals', 'followers');
 
+-- Notification reason enum
+-- This enum represents why a notification was created
+CREATE TYPE notification_reason AS ENUM (
+    'like',      -- User liked a post
+    'repost',    -- User reposted a post
+    'follow',    -- User followed someone
+    'reply',     -- User replied to a post
+    'mention',   -- User mentioned someone in a post
+    'quote'      -- User quoted a post
+);
+
+-- Record type enum for notification records
+-- This enum represents the type of record that triggered the notification
+CREATE TYPE notification_record_type AS ENUM (
+    'like',      -- app.bsky.feed.like
+    'repost',    -- app.bsky.feed.repost
+    'follow',    -- app.bsky.graph.follow
+    'post'       -- app.bsky.feed.post (used for reply/mention/quote)
+);
+
+-- =============================================================================
+-- NOTIFICATION PREFERENCES
+-- =============================================================================
+
 -- Notification preferences
 CREATE TABLE notif_decl (
     actor_id INTEGER PRIMARY KEY REFERENCES actors(id) ON DELETE CASCADE,
     allow_subscriptions notif_allow_subscriptions,
     created_at TIMESTAMPTZ NOT NULL DEFAULT NOW()
 );
+
+-- =============================================================================
+-- NOTIFICATIONS TABLE
+-- =============================================================================
+
+-- Notifications table with optimized storage
+CREATE TABLE notifications (
+    id BIGSERIAL PRIMARY KEY,
+    recipient_actor_id INT NOT NULL REFERENCES actors(id) ON DELETE CASCADE,
+    author_actor_id INT NOT NULL REFERENCES actors(id) ON DELETE CASCADE,
+    indexed_at TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    reason notification_reason NOT NULL,
+    record_type notification_record_type NOT NULL,
+    record_rkey INT8 NOT NULL,
+    record_cid BYTEA NOT NULL,
+    subject_actor_id INT REFERENCES actors(id) ON DELETE CASCADE,
+    subject_record_type notification_record_type,
+    subject_rkey INT8,
+    UNIQUE (author_actor_id, record_type, record_rkey)
+);
+
+COMMENT ON COLUMN notifications.record_rkey IS 'TID as INT8 (convert to string via i64_to_tid())';
+COMMENT ON COLUMN notifications.record_cid IS '32-byte CID digest (header stripped, reconstruct via digest_to_record_cid())';
+COMMENT ON COLUMN notifications.subject_rkey IS 'Optional TID as INT8 for subject record';
+
+-- Indexes for notifications
+CREATE INDEX idx_notifications_recipient_time
+    ON notifications(recipient_actor_id, indexed_at DESC);
+
+CREATE INDEX idx_notifications_indexed_at
+    ON notifications(indexed_at);
+
+-- =============================================================================
+-- NOTIFICATION STATE
+-- =============================================================================
+
+-- Per-user notification state
+CREATE TABLE notification_state (
+    actor_id INT PRIMARY KEY REFERENCES actors(id) ON DELETE CASCADE,
+    seen_at TIMESTAMPTZ,
+    unread_count INTEGER DEFAULT 0
+);

migrations/2025-11-01-114846_moderation/up.sql

···
 CREATE TYPE subject_type AS ENUM ('account', 'record', 'chat');
 CREATE TYPE label_severity AS ENUM ('inform', 'alert', 'none');
 CREATE TYPE label_blurs AS ENUM ('content', 'media', 'none');
-CREATE TYPE label_default_setting AS ENUM ('ignore', 'warn', 'hide');
+CREATE TYPE label_default_setting AS ENUM ('ignore', 'warn', 'hide', 'inform', 'show');
 
 -- =============================================================================
 -- LABELERS

migrations/2025-11-01-114847_queues_and_extensions/down.sql

···
+-- Drop queue tables
+DROP TABLE IF EXISTS constellation_enrichment_queue;
+DROP TABLE IF EXISTS handle_resolution_queue;
+DROP TABLE IF EXISTS jetstream_cursors;
+DROP TABLE IF EXISTS backfill_jobs;
+DROP TABLE IF EXISTS fetch_queue;
+
+-- Drop extensions
+DROP EXTENSION IF EXISTS pgrouting;
+DROP EXTENSION IF EXISTS postgis;
+DROP EXTENSION IF EXISTS pg_stat_statements;

migrations/2025-11-01-114847_queues_and_extensions/up.sql

···
+-- =============================================================================
+-- QUEUES & EXTENSIONS
+-- =============================================================================
+--
+-- Queue tables for async processing and PostgreSQL extensions
+--
+-- =============================================================================
+
+-- =============================================================================
+-- POSTGRESQL EXTENSIONS
+-- =============================================================================
+
+-- pg_stat_statements: Query performance monitoring and optimization
+-- Tracks execution statistics for all SQL statements, enabling identification
+-- of slow queries and optimization opportunities. Configured in postgresql.conf
+-- via shared_preload_libraries for optimal performance.
+CREATE EXTENSION IF NOT EXISTS pg_stat_statements;
+
+-- PostGIS: Geospatial extensions (required dependency for pgRouting)
+-- While Parakeet doesn't use geospatial features directly, this is a
+-- prerequisite for pgRouting's graph analysis capabilities.
+CREATE EXTENSION IF NOT EXISTS postgis;
+
+-- pgRouting: Graph analysis algorithms for social network analysis
+-- Enables advanced graph queries on the follows table, including:
+-- - Shortest path between users (connection discovery)
+-- - Community detection (strongly connected components)
+-- - Centrality metrics (influential users, key connectors)
+-- - Multi-hop neighborhood queries (N-degree connections)
+CREATE EXTENSION IF NOT EXISTS pgrouting;
+
+-- Note: timescaledb and pgvector are available via flake.nix but not enabled yet
+--
+-- timescaledb: Time-series optimization for notifications and stats tables
+-- Deferred until we design schema migration strategy (hypertables, partitioning)
+-- Future use: trending posts, notification retention policies, time-bucketing
+--
+-- pgvector: Vector similarity search for semantic/ML-based features
+-- Deferred until concrete use case emerges (embeddings, semantic search)
+-- Future use: content-based recommendations, duplicate detection
+
+-- =============================================================================
+-- FETCH QUEUE
+-- =============================================================================
+
+-- Fetch queue table to replace Redis list-based queue
+CREATE TABLE fetch_queue (
+    id BIGSERIAL PRIMARY KEY,
+    at_uri TEXT NOT NULL UNIQUE,
+    attempts INTEGER NOT NULL DEFAULT 0,
+    max_attempts INTEGER NOT NULL DEFAULT 3,
+    status TEXT NOT NULL DEFAULT 'pending', -- pending, processing, failed
+    last_error TEXT,
+    created_at TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    updated_at TIMESTAMPTZ NOT NULL DEFAULT NOW()
+);
+
+-- Index for queue processing (dequeue operations)
+CREATE INDEX idx_fetch_queue_status_created
+    ON fetch_queue(status, created_at)
+    WHERE status IN ('pending', 'processing');
+
+-- =============================================================================
+-- BACKFILL JOBS
+-- =============================================================================
+
+-- Backfill jobs table to replace Redis job management
+CREATE TABLE backfill_jobs (
+    did TEXT PRIMARY KEY,
+    status TEXT NOT NULL DEFAULT 'pending',
+    -- Status values: pending, processing, successful, failed.retry, failed.permanent
+    attempts INTEGER NOT NULL DEFAULT 0,
+    max_attempts INTEGER NOT NULL DEFAULT 3,
+    last_error TEXT,
+    scheduled_at TIMESTAMPTZ NOT NULL DEFAULT NOW(), -- For exponential backoff
+    started_at TIMESTAMPTZ,
+    completed_at TIMESTAMPTZ
+);
+
+-- Index for queue processing (dequeue operations)
+CREATE INDEX idx_backfill_status_scheduled
+    ON backfill_jobs(status, scheduled_at)
+    WHERE status IN ('pending', 'failed.retry');
+
+-- Index for detecting stale processing jobs
+CREATE INDEX idx_backfill_processing_stale
+    ON backfill_jobs(status, started_at)
+    WHERE status = 'processing';
+
+-- =============================================================================
+-- JETSTREAM CURSORS
+-- =============================================================================
+
+-- Jetstream cursors table to replace Redis cursor storage
+CREATE TABLE jetstream_cursors (
+    partition TEXT PRIMARY KEY, -- e.g., "posts", "likes", "social"
+    cursor_value BIGINT NOT NULL, -- Timestamp in microseconds
+    updated_at TIMESTAMPTZ NOT NULL DEFAULT NOW()
+);
+
+-- =============================================================================
+-- HANDLE RESOLUTION QUEUE
+-- =============================================================================
+
+CREATE TABLE handle_resolution_queue (
+    id BIGSERIAL PRIMARY KEY,
+    did TEXT NOT NULL UNIQUE,
+    attempts INTEGER NOT NULL DEFAULT 0,
+    max_attempts INTEGER NOT NULL DEFAULT 3,
+    status TEXT NOT NULL DEFAULT 'pending',
+    last_error TEXT,
+    created_at TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT NOW(),
+    updated_at TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT NOW()
+);
+
+CREATE INDEX idx_handle_resolution_status_created
+    ON handle_resolution_queue (status, created_at)
+    WHERE status IN ('pending', 'processing');
+
+COMMENT ON TABLE handle_resolution_queue IS 'Queue for asynchronous DID handle resolution';
+
+-- =============================================================================
+-- CONSTELLATION ENRICHMENT QUEUE
+-- =============================================================================
+
+CREATE TABLE constellation_enrichment_queue (
+    id BIGSERIAL PRIMARY KEY,
+    did TEXT NOT NULL,
+    post_uris JSONB NOT NULL, -- Array of post URIs for this actor
+    attempts INTEGER NOT NULL DEFAULT 0,
+    max_attempts INTEGER NOT NULL DEFAULT 3,
+    status TEXT NOT NULL DEFAULT 'pending',
+    last_error TEXT,
+    created_at TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT NOW(),
+    updated_at TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT NOW(),
+    UNIQUE(did) -- One enrichment job per DID
+);
+
+CREATE INDEX idx_constellation_status_created
+    ON constellation_enrichment_queue (status, created_at)
+    WHERE status IN ('pending', 'processing');
+
+COMMENT ON TABLE constellation_enrichment_queue IS 'Queue for asynchronous Constellation API enrichment';

migrations/2025-11-10-014835_add_missing_fk_indexes/down.sql

···
--- Remove indexes added for foreign key performance
-
-DROP INDEX IF EXISTS idx_likes_via_post;
-DROP INDEX IF EXISTS idx_reposts_via_post;
-DROP INDEX IF EXISTS idx_profiles_pinned_post;
-DROP INDEX IF EXISTS idx_post_facets_link_uri;

migrations/2025-11-10-014835_add_missing_fk_indexes/up.sql

···
--- Add missing indexes on foreign key columns
---
--- These indexes are critical for performance when deleting posts or other referenced records.
--- Without them, DELETE CASCADE operations require full table scans.
---
--- IMPORTANT: For production databases, run these with CONCURRENTLY to avoid locking:
---   CREATE INDEX CONCURRENTLY idx_likes_via_post ON likes(via_post_id);
---   CREATE INDEX CONCURRENTLY idx_reposts_via_post ON reposts(via_post_id);
---   CREATE INDEX CONCURRENTLY idx_profiles_pinned_post ON profiles(pinned_post_id) WHERE pinned_post_id IS NOT NULL;
---   CREATE INDEX CONCURRENTLY idx_post_facets_link_uri ON post_facets(link_uri_id);
-
--- CRITICAL: Improves post deletion performance
--- Without this, deleting a post scans all 949K+ likes rows
-CREATE INDEX IF NOT EXISTS idx_likes_via_post ON likes(via_post_id);
-
--- CRITICAL: Improves post deletion performance
--- Without this, deleting a post scans all 61K+ reposts rows
-CREATE INDEX IF NOT EXISTS idx_reposts_via_post ON reposts(via_post_id);
-
--- HIGH PRIORITY: Improves post deletion performance for pinned posts
--- Partial index only covers ~70 profiles with pinned posts (out of 381 total)
-CREATE INDEX IF NOT EXISTS idx_profiles_pinned_post
-ON profiles(pinned_post_id)
-WHERE pinned_post_id IS NOT NULL;
-
--- MEDIUM PRIORITY: Improves URI deletion performance
--- Covers 9,451 rows with link facets
-CREATE INDEX IF NOT EXISTS idx_post_facets_link_uri ON post_facets(link_uri_id);

migrations/2025-11-11-055323_notifications_table/down.sql

···
--- Drop notifications tables
-DROP TABLE IF EXISTS notification_state;
-DROP TABLE IF EXISTS notifications;

migrations/2025-11-11-055323_notifications_table/up.sql

···
--- Create notifications table to replace Redis sorted sets
-CREATE TABLE notifications (
-    id BIGSERIAL PRIMARY KEY,
-    recipient_did TEXT NOT NULL,
-    author_did TEXT NOT NULL,
-    uri TEXT NOT NULL UNIQUE, -- For deduplication (e.g., at://did/app.bsky.feed.like/rkey)
-    cid TEXT NOT NULL,
-    reason TEXT NOT NULL, -- like, repost, follow, reply, mention, quote
-    reason_subject TEXT, -- Optional subject URI (e.g., the post that was liked)
-    indexed_at TIMESTAMPTZ NOT NULL DEFAULT NOW()
-);
-
--- Index for listing notifications by recipient (pagination)
-CREATE INDEX idx_notifications_recipient_time
-    ON notifications(recipient_did, indexed_at DESC);
-
--- Index for deletion by URI
-CREATE INDEX idx_notifications_uri
-    ON notifications(uri);
-
--- Index for cleanup worker (delete old notifications)
-CREATE INDEX idx_notifications_indexed_at
-    ON notifications(indexed_at);
-
--- Per-user notification state
-CREATE TABLE notification_state (
-    did TEXT PRIMARY KEY,
-    seen_at TIMESTAMPTZ,
-    unread_count INTEGER DEFAULT 0
-);

migrations/2025-11-11-055341_fetch_queue_table/down.sql

···
--- Drop fetch queue table
-DROP TABLE IF EXISTS fetch_queue;

migrations/2025-11-11-055341_fetch_queue_table/up.sql

···
--- Create fetch queue table to replace Redis list-based queue
-CREATE TABLE fetch_queue (
-    id BIGSERIAL PRIMARY KEY,
-    at_uri TEXT NOT NULL UNIQUE,
-    attempts INTEGER NOT NULL DEFAULT 0,
-    max_attempts INTEGER NOT NULL DEFAULT 3,
-    status TEXT NOT NULL DEFAULT 'pending', -- pending, processing, failed
-    last_error TEXT,
-    created_at TIMESTAMPTZ NOT NULL DEFAULT NOW(),
-    updated_at TIMESTAMPTZ NOT NULL DEFAULT NOW()
-);
-
--- Index for queue processing (dequeue operations)
-CREATE INDEX idx_fetch_queue_status_created
-    ON fetch_queue(status, created_at)
-    WHERE status IN ('pending', 'processing');

migrations/2025-11-11-055354_backfill_jobs_table/down.sql

···
--- Drop backfill jobs table
-DROP TABLE IF EXISTS backfill_jobs;

migrations/2025-11-11-055354_backfill_jobs_table/up.sql

···
--- Create backfill jobs table to replace Redis job management
-CREATE TABLE backfill_jobs (
-    did TEXT PRIMARY KEY,
-    status TEXT NOT NULL DEFAULT 'pending',
-    -- Status values: pending, processing, successful, failed.retry, failed.permanent
-    attempts INTEGER NOT NULL DEFAULT 0,
-    max_attempts INTEGER NOT NULL DEFAULT 3,
-    last_error TEXT,
-    scheduled_at TIMESTAMPTZ NOT NULL DEFAULT NOW(), -- For exponential backoff
-    started_at TIMESTAMPTZ,
-    completed_at TIMESTAMPTZ
-);
-
--- Index for queue processing (dequeue operations)
-CREATE INDEX idx_backfill_status_scheduled
-    ON backfill_jobs(status, scheduled_at)
-    WHERE status IN ('pending', 'failed.retry');
-
--- Index for detecting stale processing jobs
-CREATE INDEX idx_backfill_processing_stale
-    ON backfill_jobs(status, started_at)
-    WHERE status = 'processing';

migrations/2025-11-11-055409_jetstream_cursors_table/down.sql

···
--- Drop jetstream cursors table
-DROP TABLE IF EXISTS jetstream_cursors;

migrations/2025-11-11-055409_jetstream_cursors_table/up.sql

···
--- Create jetstream cursors table to replace Redis cursor storage
-CREATE TABLE jetstream_cursors (
-    partition TEXT PRIMARY KEY, -- e.g., "posts", "likes", "social"
-    cursor_value BIGINT NOT NULL, -- Timestamp in microseconds
-    updated_at TIMESTAMPTZ NOT NULL DEFAULT NOW()
-);

migrations/2025-11-11-135750_handle_resolution_queue/down.sql

···
-DROP TABLE IF EXISTS handle_resolution_queue;

migrations/2025-11-11-135750_handle_resolution_queue/up.sql

···
-CREATE TABLE handle_resolution_queue (
-    id BIGSERIAL PRIMARY KEY,
-    did TEXT NOT NULL UNIQUE,
-    attempts INTEGER NOT NULL DEFAULT 0,
-    max_attempts INTEGER NOT NULL DEFAULT 3,
-    status TEXT NOT NULL DEFAULT 'pending',
-    last_error TEXT,
-    created_at TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT NOW(),
-    updated_at TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT NOW()
-);
-
-CREATE INDEX idx_handle_resolution_status_created
-    ON handle_resolution_queue (status, created_at)
-    WHERE status IN ('pending', 'processing');
-
-COMMENT ON TABLE handle_resolution_queue IS 'Queue for asynchronous DID handle resolution';

migrations/2025-11-11-135751_constellation_enrichment_queue/down.sql

···
-DROP TABLE IF EXISTS constellation_enrichment_queue;

migrations/2025-11-11-135751_constellation_enrichment_queue/up.sql

···
-CREATE TABLE constellation_enrichment_queue (
-    id BIGSERIAL PRIMARY KEY,
-    did TEXT NOT NULL,
-    post_uris JSONB NOT NULL, -- Array of post URIs for this actor
-    attempts INTEGER NOT NULL DEFAULT 0,
-    max_attempts INTEGER NOT NULL DEFAULT 3,
-    status TEXT NOT NULL DEFAULT 'pending',
-    last_error TEXT,
-    created_at TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT NOW(),
-    updated_at TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT NOW(),
-    UNIQUE(did) -- One enrichment job per DID
-);
-
-CREATE INDEX idx_constellation_status_created
-    ON constellation_enrichment_queue (status, created_at)
-    WHERE status IN ('pending', 'processing');
-
-COMMENT ON TABLE constellation_enrichment_queue IS 'Queue for asynchronous Constellation API enrichment';

migrations/2025-11-11-214925_add_aggregate_stats_table/down.sql migrations/2025-11-01-114848_aggregate_stats/down.sql

···
--- Drop aggregate stats tables and enums
-DROP TABLE IF EXISTS post_aggregate_stats;
+-- Drop aggregate stats tables
 DROP TABLE IF EXISTS actor_aggregate_stats;
-DROP TYPE IF EXISTS post_stat_type;
+DROP TABLE IF EXISTS post_aggregate_stats;
+
+-- Drop enums
 DROP TYPE IF EXISTS actor_stat_type;
+DROP TYPE IF EXISTS post_stat_type;

migrations/2025-11-11-214925_add_aggregate_stats_table/up.sql migrations/2025-11-01-114848_aggregate_stats/up.sql

···
--- Create enums for stat types
+-- =============================================================================
+-- AGGREGATE STATISTICS
+-- =============================================================================
+--
+-- Post and actor aggregate statistics with automatic cleanup
+--
+-- =============================================================================
+
+-- =============================================================================
+-- STAT TYPE ENUMS
+-- =============================================================================
+
 -- Post-specific stats
 CREATE TYPE post_stat_type AS ENUM (
     'like',     -- 3: LIKE
···
     'profile_starterpack'  -- 10: PROFILE_STARTERPACK
 );
 
+-- =============================================================================
+-- AGGREGATE STATS TABLES
+-- =============================================================================
+
 -- Post aggregate statistics
 -- Stores engagement metrics for posts with automatic cleanup
 CREATE TABLE post_aggregate_stats (
···
 
     PRIMARY KEY (actor_id, stat_type)
 );
+
+-- =============================================================================
+-- INDEXES
+-- =============================================================================
 
 -- Index for finding hot posts (recently updated engagement)
 CREATE INDEX idx_post_aggregate_stats_updated_at

migrations/2025-11-12-030603_fix_label_default_setting_enum/down.sql

···
--- PostgreSQL does not support removing enum values directly.
--- To rollback this migration, you would need to:
--- 1. Ensure no rows use 'inform' or 'show' values
--- 2. Drop and recreate the enum type (requires dropping dependent columns first)
--- 3. Recreate all tables/columns that use this type
---
--- Since this is impractical and the new values are required for compatibility
--- with real production labelers, rollback is not supported.
-
--- If you absolutely need to rollback, manually run:
--- BEGIN;
--- -- Delete any rows using the new values
--- DELETE FROM labeler_defs WHERE default_setting IN ('inform', 'show');
--- -- Then drop and recreate the type (complex, requires recreating dependent tables)
--- ROLLBACK; -- Don't actually do this without a full backup!
-
-SELECT 1; -- No-op to make diesel happy

migrations/2025-11-12-030603_fix_label_default_setting_enum/up.sql

···
--- Add missing enum values to label_default_setting
--- These values are used by real labelers in production (e.g., Khronos uses 'inform', Bluesky Moderation uses 'show')
--- even though they're not in the AT Protocol spec's "knownValues"
-
-ALTER TYPE label_default_setting ADD VALUE IF NOT EXISTS 'inform';
-ALTER TYPE label_default_setting ADD VALUE IF NOT EXISTS 'show';

migrations/2025-11-13-091852_drop_post_content_version/down.sql

···
--- Restore the content_version column
--- Note: This is not truly reversible as we've lost the original version data
--- The column will be restored but values will be NULL
-ALTER TABLE posts ADD COLUMN content_version smallint;

migrations/2025-11-13-091852_drop_post_content_version/up.sql

···
--- Drop the content_version column from posts table
--- This column is no longer needed as we now use a single-byte
--- dictionary version prefix in the compressed content itself
-ALTER TABLE posts DROP COLUMN content_version;

migrations/2025-11-13-122954_expand_language_codes_and_array_support/down.sql

···
--- ============================================================================
--- MIGRATION ROLLBACK: Revert language codes expansion and array conversion
--- ============================================================================
--- This rollback:
--- 1. Drops GIN index on langs
--- 2. Drops max-3 constraint
--- 3. Converts langs array back to single language_code
--- 4. Renames column back to language_code
--- 5. Does NOT remove enum values (PostgreSQL doesn't support removing enum values)
---
--- WARNING: This will lose language data if posts have multiple languages!
--- Only the first language will be preserved.
-
--- Step 1: Drop GIN index
-DROP INDEX IF EXISTS idx_posts_langs;
-
--- Step 2: Drop max-3 constraint
-ALTER TABLE posts
-  DROP CONSTRAINT IF EXISTS posts_langs_max_3;
-
--- Step 3: Rename column back
-ALTER TABLE posts
-  RENAME COLUMN langs TO language_code;
-
--- Step 4: Convert array back to single enum
--- WARNING: Data loss! Only keeps first language from array
-ALTER TABLE posts
-  ALTER COLUMN language_code TYPE language_code
-  USING CASE
-    WHEN array_length(language_code, 1) > 0 THEN language_code[1]
-    ELSE NULL
-  END;
-
--- Step 5: Remove comment
-COMMENT ON COLUMN posts.language_code IS NULL;
-
--- Note: Cannot remove enum values from PostgreSQL enum type
--- The following values will remain in the enum but won't be used:
--- id, th, vi, cs, ro, el, he, bn, ca, fa, sk, bg, hr, lt, lv, sl, et, ga, cy, is, mt, eu, ms, tl, ur, sr, hu, la
--- This is a PostgreSQL limitation - enum values cannot be removed once added

migrations/2025-11-13-122954_expand_language_codes_and_array_support/up.sql

···
--- ============================================================================
--- MIGRATION: Expand language codes and convert to array support
--- ============================================================================
--- This migration:
--- 1. Adds 26 new language codes (id, th, vi, cs, ro, el, he, bn, ca, fa, etc.)
--- 2. Converts posts.language_code from single enum to language_code[] array
--- 3. Preserves existing data during conversion
--- 4. Adds constraints for AT Protocol spec (max 3 languages)
---
--- AT Protocol spec: Posts can have up to 3 language codes in langs field
--- Previous implementation only stored first language, this fixes data loss
-
--- Step 1: Add all new language codes to the enum
--- Note: PostgreSQL requires adding enum values one at a time, cannot be in transaction
--- Based on:
--- - W3Techs March 2025: Top 40 internet content languages
--- - Ethnologue 2025: All languages with 50M+ total speakers
--- - Ethnologue 2025: All languages with 50M+ native speakers
--- Total: 38 new codes added (from 22 to 60 total)
-
--- Major Asian Languages (10 codes)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'id';  -- Indonesian (252M total speakers, 1.1% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'th';  -- Thai (71M total speakers, 0.3% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'vi';  -- Vietnamese (97M total speakers, 1.1% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'bn';  -- Bengali (284M total speakers)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'ur';  -- Urdu (246M total speakers)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'mr';  -- Marathi (99M total speakers)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'te';  -- Telugu (96M total speakers)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'ta';  -- Tamil (86M total speakers)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'jv';  -- Javanese (69M total speakers)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'gu';  -- Gujarati (62M total speakers)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'kn';  -- Kannada (59M total speakers)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'pa';  -- Punjabi (90M total speakers)
-
--- European Languages (16 codes)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'cs';  -- Czech (1.0% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'ro';  -- Romanian (0.5% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'el';  -- Greek (0.5% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'sk';  -- Slovak (0.4% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'bg';  -- Bulgarian (0.3% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'hr';  -- Croatian (0.2% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'lt';  -- Lithuanian (0.2% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'sl';  -- Slovenian (0.1% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'lv';  -- Latvian (0.1% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'et';  -- Estonian (0.1% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'sr';  -- Serbian (0.2% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'bs';  -- Bosnian (0.1% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'ca';  -- Catalan (0.1% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'eu';  -- Basque
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'ga';  -- Irish
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'cy';  -- Welsh
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'is';  -- Icelandic
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'mt';  -- Maltese
-
--- Middle Eastern & North African (2 codes)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'fa';  -- Persian/Farsi (83M total speakers, 1.2% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'he';  -- Hebrew (0.4% web)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'am';  -- Amharic (60M total speakers)
-
--- African Languages (2 codes)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'sw';  -- Swahili (87M total speakers)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'ha';  -- Hausa (94M total speakers)
-
--- Southeast Asian (2 codes)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'tl';  -- Tagalog/Filipino (87M total speakers)
-ALTER TYPE language_code ADD VALUE IF NOT EXISTS 'ms';  -- Malay
-
--- Step 2: Convert posts.language_code from single enum to array
--- This preserves existing data by converting NULL -> empty array, single value -> single-element array
-
-ALTER TABLE posts
-  ALTER COLUMN language_code TYPE language_code[]
-  USING CASE
-    WHEN language_code IS NOT NULL THEN ARRAY[language_code]
-    ELSE ARRAY[]::language_code[]
-  END;
-
--- Step 3: Rename column to match AT Protocol naming (langs)
-ALTER TABLE posts
-  RENAME COLUMN language_code TO langs;
-
--- Step 3.5: Set NOT NULL and default (empty array means no language specified)
-ALTER TABLE posts
-  ALTER COLUMN langs SET NOT NULL,
-  ALTER COLUMN langs SET DEFAULT ARRAY[]::language_code[];
-
--- Step 4: Add constraint for AT Protocol spec (max 3 languages per post)
-ALTER TABLE posts
-  ADD CONSTRAINT posts_langs_max_3
-  CHECK (array_length(langs, 1) IS NULL OR array_length(langs, 1) <= 3);
-
--- Step 5: Add GIN index for efficient language searching
--- This allows queries like "find all posts in Japanese OR English"
-CREATE INDEX idx_posts_langs ON posts USING GIN (langs);
-
--- Step 6: Update comment on column
-COMMENT ON COLUMN posts.langs IS 'Languages of post content (ISO 639-1 codes). Max 3 per AT Protocol spec. Empty array means no language specified.';

migrations/2025-11-13-213616_enable_postgresql_extensions/down.sql

···
--- Rollback PostgreSQL extensions
---
--- Note: Extensions are dropped in reverse dependency order
--- (pgrouting depends on postgis, so drop pgrouting first)
-
--- Drop pgRouting (graph analysis for social networks)
-DROP EXTENSION IF EXISTS pgrouting;
-
--- Drop PostGIS (required by pgRouting, safe to drop after pgrouting)
-DROP EXTENSION IF EXISTS postgis;
-
--- Drop pg_stat_statements (query performance monitoring)
--- Note: This will clear all accumulated query statistics
-DROP EXTENSION IF EXISTS pg_stat_statements;

migrations/2025-11-13-213616_enable_postgresql_extensions/up.sql

···
--- Enable PostgreSQL extensions for enhanced functionality
---
--- pg_stat_statements: Query performance monitoring and optimization
--- Tracks execution statistics for all SQL statements, enabling identification
--- of slow queries and optimization opportunities. Configured in postgresql.conf
--- via shared_preload_libraries for optimal performance.
-CREATE EXTENSION IF NOT EXISTS pg_stat_statements;
-
--- PostGIS: Geospatial extensions (required dependency for pgRouting)
--- While Parakeet doesn't use geospatial features directly, this is a
--- prerequisite for pgRouting's graph analysis capabilities.
-CREATE EXTENSION IF NOT EXISTS postgis;
-
--- pgRouting: Graph analysis algorithms for social network analysis
--- Enables advanced graph queries on the follows table, including:
--- - Shortest path between users (connection discovery)
--- - Community detection (strongly connected components)
--- - Centrality metrics (influential users, key connectors)
--- - Multi-hop neighborhood queries (N-degree connections)
-CREATE EXTENSION IF NOT EXISTS pgrouting;
-
--- Note: timescaledb and pgvector are available via flake.nix but not enabled yet
---
--- timescaledb: Time-series optimization for notifications and stats tables
--- Deferred until we design schema migration strategy (hypertables, partitioning)
--- Future use: trending posts, notification retention policies, time-bucketing
---
--- pgvector: Vector similarity search for semantic/ML-based features
--- Deferred until concrete use case emerges (embeddings, semantic search)
--- Future use: content-based recommendations, duplicate detection

migrations/2025-11-13-221911_refactor_notifications_to_actor_ids/down.sql

···
--- Revert notifications table back to using DIDs
-
--- Step 1: Add back DID columns (nullable initially)
-ALTER TABLE notifications
-ADD COLUMN recipient_did TEXT,
-ADD COLUMN author_did TEXT;
-
--- Step 2: Populate DIDs from actor_ids by joining with actors table
-UPDATE notifications n
-SET
-    recipient_did = (SELECT did FROM actors WHERE id = n.recipient_actor_id),
-    author_did = (SELECT did FROM actors WHERE id = n.author_actor_id);
-
--- Step 3: Make DID columns NOT NULL
-ALTER TABLE notifications
-ALTER COLUMN recipient_did SET NOT NULL,
-ALTER COLUMN author_did SET NOT NULL;
-
--- Step 4: Drop foreign key constraints
-ALTER TABLE notifications
-DROP CONSTRAINT IF EXISTS fk_notifications_recipient,
-DROP CONSTRAINT IF EXISTS fk_notifications_author;
-
--- Step 5: Drop actor_id columns
-ALTER TABLE notifications
-DROP COLUMN recipient_actor_id,
-DROP COLUMN author_actor_id;
-
--- Step 6: Recreate original indexes
-DROP INDEX IF EXISTS idx_notifications_recipient_actor_time;
-DROP INDEX IF EXISTS idx_notifications_author_actor;
-CREATE INDEX idx_notifications_recipient_time
-ON notifications(recipient_did, indexed_at DESC);

migrations/2025-11-13-221911_refactor_notifications_to_actor_ids/up.sql

···
--- Refactor notifications table to use actor_ids instead of DIDs
--- This improves performance and consistency with other tables
-
--- Step 1: Add new actor_id columns (nullable initially for safe migration)
-ALTER TABLE notifications
-ADD COLUMN recipient_actor_id INT,
-ADD COLUMN author_actor_id INT;
-
--- Step 2: Populate the new columns from existing DIDs
--- This requires joining with the actors table
-UPDATE notifications n
-SET
-    recipient_actor_id = (SELECT id FROM actors WHERE did = n.recipient_did),
-    author_actor_id = (SELECT id FROM actors WHERE did = n.author_did);
-
--- Step 3: Make the columns NOT NULL (after populating)
-ALTER TABLE notifications
-ALTER COLUMN recipient_actor_id SET NOT NULL,
-ALTER COLUMN author_actor_id SET NOT NULL;
-
--- Step 4: Drop old DID columns
-ALTER TABLE notifications
-DROP COLUMN recipient_did,
-DROP COLUMN author_did;
-
--- Step 5: Add foreign key constraints
-ALTER TABLE notifications
-ADD CONSTRAINT fk_notifications_recipient
-    FOREIGN KEY (recipient_actor_id) REFERENCES actors(id) ON DELETE CASCADE,
-ADD CONSTRAINT fk_notifications_author
-    FOREIGN KEY (author_actor_id) REFERENCES actors(id) ON DELETE CASCADE;
-
--- Step 6: Recreate index for recipient queries (most common use case)
--- The old index was on (recipient_did, indexed_at DESC)
--- We replace it with (recipient_actor_id, indexed_at DESC)
-DROP INDEX IF EXISTS idx_notifications_recipient_time;
-CREATE INDEX idx_notifications_recipient_actor_time
-ON notifications(recipient_actor_id, indexed_at DESC);
-
--- Step 7: Add index for author queries (for completeness)
-CREATE INDEX idx_notifications_author_actor
-ON notifications(author_actor_id);

migrations/2025-11-14-012931_create_notification_enums/down.sql

···
--- Drop notification enums (reverse order of creation)
-DROP TYPE IF EXISTS notification_record_type;
-DROP TYPE IF EXISTS notification_reason;

migrations/2025-11-14-012931_create_notification_enums/up.sql

···
--- Create notification reason enum
--- This enum represents why a notification was created
-CREATE TYPE notification_reason AS ENUM (
-    'like',      -- User liked a post
-    'repost',    -- User reposted a post
-    'follow',    -- User followed someone
-    'reply',     -- User replied to a post
-    'mention',   -- User mentioned someone in a post
-    'quote'      -- User quoted a post
-);
-
--- Create record type enum for notification records
--- This enum represents the type of record that triggered the notification
-CREATE TYPE notification_record_type AS ENUM (
-    'like',      -- app.bsky.feed.like
-    'repost',    -- app.bsky.feed.repost
-    'follow',    -- app.bsky.graph.follow
-    'post'       -- app.bsky.feed.post (used for reply/mention/quote)
-);

migrations/2025-11-14-013020_add_normalized_notification_columns/down.sql

···
--- Drop normalized notification columns (reverse order)
-
-DROP INDEX IF EXISTS idx_notifications_subject_actor;
-
-ALTER TABLE notifications
-DROP COLUMN IF EXISTS subject_rkey,
-DROP COLUMN IF EXISTS subject_record_type,
-DROP COLUMN IF EXISTS subject_actor_id,
-DROP COLUMN IF EXISTS record_cid_new,
-DROP COLUMN IF EXISTS record_rkey,
-DROP COLUMN IF EXISTS record_type,
-DROP COLUMN IF EXISTS reason_enum;

migrations/2025-11-14-013020_add_normalized_notification_columns/up.sql

···
--- Add new normalized columns to notifications table
--- These columns will replace uri and reason_subject with structured data
-
--- Add reason enum column (will replace TEXT reason column)
-ALTER TABLE notifications
-ADD COLUMN reason_enum notification_reason;
-
--- Add record fields (will replace uri: at://did/collection/rkey)
-ALTER TABLE notifications
-ADD COLUMN record_type notification_record_type,
-ADD COLUMN record_rkey TEXT,
-ADD COLUMN record_cid_new TEXT;  -- Temporary name to avoid conflict with existing cid
-
--- Add subject fields (will replace reason_subject: at://did/collection/rkey)
-ALTER TABLE notifications
-ADD COLUMN subject_actor_id INT REFERENCES actors(id) ON DELETE CASCADE,
-ADD COLUMN subject_record_type notification_record_type,
-ADD COLUMN subject_rkey TEXT;
-
--- Create index on subject_actor_id for efficient lookups
-CREATE INDEX idx_notifications_subject_actor ON notifications(subject_actor_id);
-
--- Note: Columns are nullable during migration. They will be populated in the next migration
--- and then made NOT NULL (where appropriate) in the finalization migration.

migrations/2025-11-14-013103_populate_normalized_notification_columns/down.sql

···
--- Reverse the population by setting all normalized columns back to NULL
--- This allows us to re-run the population logic if needed
-
-UPDATE notifications
-SET
-    reason_enum = NULL,
-    record_type = NULL,
-    record_rkey = NULL,
-    record_cid_new = NULL,
-    subject_actor_id = NULL,
-    subject_record_type = NULL,
-    subject_rkey = NULL;

migrations/2025-11-14-013103_populate_normalized_notification_columns/up.sql

···
--- Populate normalized notification columns from existing URI data
---
--- This migration parses AT-URIs and populates the new structured columns.
--- AT-URI format: at://did:plc:xxx/collection/rkey
-
--- Step 1: Populate reason_enum from existing TEXT reason column
-UPDATE notifications
-SET reason_enum = reason::notification_reason;
-
--- Step 2: Populate record fields from uri column
--- Extract rkey (last path component after final /)
-UPDATE notifications
-SET record_rkey = substring(uri from '[^/]+$');
-
--- Copy CID to new column
-UPDATE notifications
-SET record_cid_new = cid;
-
--- Determine record_type from collection in URI
-UPDATE notifications
-SET record_type = CASE
-    WHEN uri LIKE '%/app.bsky.feed.like/%' THEN 'like'::notification_record_type
-    WHEN uri LIKE '%/app.bsky.feed.repost/%' THEN 'repost'::notification_record_type
-    WHEN uri LIKE '%/app.bsky.graph.follow/%' THEN 'follow'::notification_record_type
-    WHEN uri LIKE '%/app.bsky.feed.post/%' THEN 'post'::notification_record_type
-    ELSE NULL
-END;
-
--- Step 3: Populate subject fields from reason_subject column
--- Extract rkey from reason_subject
-UPDATE notifications
-SET subject_rkey = substring(reason_subject from '[^/]+$')
-WHERE reason_subject IS NOT NULL;
-
--- Determine subject_record_type from collection in reason_subject
-UPDATE notifications
-SET subject_record_type = CASE
-    WHEN reason_subject LIKE '%/app.bsky.feed.post/%' THEN 'post'::notification_record_type
-    WHEN reason_subject LIKE '%/app.bsky.feed.like/%' THEN 'like'::notification_record_type
-    WHEN reason_subject LIKE '%/app.bsky.feed.repost/%' THEN 'repost'::notification_record_type
-    ELSE NULL
-END
-WHERE reason_subject IS NOT NULL;
-
--- Resolve subject_actor_id from DID in reason_subject
--- Extract DID: at://did:plc:xxx/...
-UPDATE notifications n
-SET subject_actor_id = (
-    SELECT a.id
-    FROM actors a
-    WHERE a.did = substring(n.reason_subject from 'at://([^/]+)')
-)
-WHERE n.reason_subject IS NOT NULL;
-
--- Verification queries (run these to check data integrity):
--- SELECT COUNT(*) FROM notifications WHERE reason_enum IS NULL; -- Should be 0
--- SELECT COUNT(*) FROM notifications WHERE record_type IS NULL; -- Should be 0
--- SELECT COUNT(*) FROM notifications WHERE record_rkey IS NULL; -- Should be 0
--- SELECT COUNT(*) FROM notifications WHERE reason_subject IS NOT NULL AND subject_actor_id IS NULL; -- Orphaned subjects

migrations/2025-11-14-013210_finalize_notification_normalization/down.sql

···
--- Reverse finalization by restoring old columns
--- WARNING: This will lose data since we can't perfectly reconstruct URIs
--- This is mainly for development/testing rollback scenarios
-
--- Step 1: Drop unique constraint
-ALTER TABLE notifications
-DROP CONSTRAINT IF EXISTS notifications_unique_record;
-
--- Step 2: Rename columns back to temporary names
-ALTER TABLE notifications
-RENAME COLUMN reason TO reason_enum;
-
-ALTER TABLE notifications
-RENAME COLUMN record_cid TO record_cid_new;
-
--- Step 3: Add back old columns (nullable for now)
-ALTER TABLE notifications
-ADD COLUMN uri TEXT,
-ADD COLUMN cid TEXT,
-ADD COLUMN reason TEXT,
-ADD COLUMN reason_subject TEXT;
-
--- Step 4: Attempt to reconstruct URIs (will be incomplete - DIDs are lost!)
--- NOTE: This won't work perfectly because we don't have DIDs anymore
--- In a real rollback scenario, you'd need to join with actors table
-UPDATE notifications
-SET
-    cid = record_cid_new,
-    reason = reason_enum::TEXT;
-
--- NOTE: uri and reason_subject cannot be fully reconstructed without DIDs
--- These would need to be populated from actors table joins
--- For now, mark them as incomplete
-UPDATE notifications
-SET uri = 'INCOMPLETE:' || record_type::TEXT || '/' || record_rkey;
-
--- Step 5: Make old columns NOT NULL (if they were before)
--- Skipped because we can't fully reconstruct the data
-
--- Step 6: Recreate uri index
-CREATE UNIQUE INDEX notifications_uri_key ON notifications(uri);

migrations/2025-11-14-013210_finalize_notification_normalization/up.sql

···
--- Finalize notification normalization
--- Make new columns NOT NULL, drop old columns, add constraints
-
--- Step 1: Make required columns NOT NULL
-ALTER TABLE notifications
-ALTER COLUMN reason_enum SET NOT NULL,
-ALTER COLUMN record_type SET NOT NULL,
-ALTER COLUMN record_rkey SET NOT NULL,
-ALTER COLUMN record_cid_new SET NOT NULL;
-
--- Step 2: Drop old columns that have been replaced
-ALTER TABLE notifications
-DROP COLUMN uri,
-DROP COLUMN cid,
-DROP COLUMN reason,
-DROP COLUMN reason_subject;
-
--- Step 3: Rename record_cid_new to record_cid (now that old cid is gone)
-ALTER TABLE notifications
-RENAME COLUMN record_cid_new TO record_cid;
-
--- Step 4: Rename reason_enum to reason (now that old reason is gone)
-ALTER TABLE notifications
-RENAME COLUMN reason_enum TO reason;
-
--- Step 5: Add unique constraint on (author_actor_id, record_type, record_rkey)
--- This replaces the old uri uniqueness constraint
-ALTER TABLE notifications
-ADD CONSTRAINT notifications_unique_record
-UNIQUE (author_actor_id, record_type, record_rkey);
-
--- Step 6: Drop old uri index (no longer needed)
-DROP INDEX IF EXISTS idx_notifications_uri;
-
--- Final schema:
--- notifications (
---     id BIGSERIAL PRIMARY KEY,
---     recipient_actor_id INT NOT NULL,
---     author_actor_id INT NOT NULL,
---     indexed_at TIMESTAMPTZ NOT NULL,
---     reason notification_reason NOT NULL,
---     record_type notification_record_type NOT NULL,
---     record_rkey TEXT NOT NULL,
---     record_cid TEXT NOT NULL,
---     subject_actor_id INT (nullable),
---     subject_record_type notification_record_type (nullable),
---     subject_rkey TEXT (nullable),
---     UNIQUE (author_actor_id, record_type, record_rkey)
--- )

migrations/2025-11-14-015525_normalize_notification_state/down.sql

···
--- Revert notification_state normalization (restore DID-based primary key)
-
--- Step 1: Add back DID column
-ALTER TABLE notification_state
-ADD COLUMN did TEXT;
-
--- Step 2: Populate DID from actor_id
-UPDATE notification_state ns
-SET did = a.did
-FROM actors a
-WHERE ns.actor_id = a.id;
-
--- Step 3: Make DID NOT NULL
-ALTER TABLE notification_state
-ALTER COLUMN did SET NOT NULL;
-
--- Step 4: Drop actor_id primary key and column
-ALTER TABLE notification_state
-DROP CONSTRAINT notification_state_pkey,
-DROP COLUMN actor_id;
-
--- Step 5: Add back DID primary key
-ALTER TABLE notification_state
-ADD PRIMARY KEY (did);

migrations/2025-11-14-015525_normalize_notification_state/up.sql

···
--- Normalize notification_state table to use actor_id instead of DID
--- This migration changes the primary key from DID (text) to actor_id (integer)
-
--- Step 1: Add actor_id column
-ALTER TABLE notification_state
-ADD COLUMN actor_id INT REFERENCES actors(id) ON DELETE CASCADE;
-
--- Step 2: Populate actor_id from DIDs
-UPDATE notification_state ns
-SET actor_id = a.id
-FROM actors a
-WHERE ns.did = a.did;
-
--- Step 3: Make actor_id NOT NULL (will fail if any DIDs couldn't be resolved)
-ALTER TABLE notification_state
-ALTER COLUMN actor_id SET NOT NULL;
-
--- Step 4: Drop old primary key and DID column
-ALTER TABLE notification_state
-DROP CONSTRAINT notification_state_pkey,
-DROP COLUMN did;
-
--- Step 5: Add new primary key on actor_id
-ALTER TABLE notification_state
-ADD PRIMARY KEY (actor_id);

migrations/2025-11-15-030105_optimize_notification_storage/down.sql

···
--- Reverse optimization of notification storage
--- Convert back from INT8 (TID rkeys) and BYTEA (CID digests) to TEXT
-
--- Step 1: Add back TEXT columns
-ALTER TABLE notifications
-ADD COLUMN record_rkey_text TEXT,
-ADD COLUMN record_cid_text TEXT,
-ADD COLUMN subject_rkey_text TEXT;
-
--- Step 2: Migrate data back to TEXT format
-UPDATE notifications
-SET
-    record_rkey_text = i64_to_tid(record_rkey),
-    record_cid_text = (
-        SELECT encode(
-            '\x01\x71\x12\x20'::bytea || record_cid,
-            'base32'
-        )
-    ),
-    subject_rkey_text = CASE
-        WHEN subject_rkey IS NOT NULL THEN i64_to_tid(subject_rkey)
-        ELSE NULL
-    END;
-
--- Note: PostgreSQL's encode() with base32 produces uppercase, but CIDs use lowercase
--- We need to prepend 'b' (multibase prefix) and convert to lowercase
-UPDATE notifications
-SET record_cid_text = 'b' || LOWER(record_cid_text);
-
--- Step 3: Make TEXT columns NOT NULL (where appropriate)
-ALTER TABLE notifications
-ALTER COLUMN record_rkey_text SET NOT NULL,
-ALTER COLUMN record_cid_text SET NOT NULL;
-
--- Step 4: Drop optimized columns
-ALTER TABLE notifications
-DROP COLUMN record_rkey,
-DROP COLUMN record_cid,
-DROP COLUMN subject_rkey;
-
--- Step 5: Rename TEXT columns to original names
-ALTER TABLE notifications
-RENAME COLUMN record_rkey_text TO record_rkey;
-ALTER TABLE notifications
-RENAME COLUMN record_cid_text TO record_cid;
-ALTER TABLE notifications
-RENAME COLUMN subject_rkey_text TO subject_rkey;
-
--- Step 6: Recreate unique constraint with TEXT columns
-ALTER TABLE notifications
-DROP CONSTRAINT IF EXISTS notifications_unique_record;
-ALTER TABLE notifications
-ADD CONSTRAINT notifications_unique_record
-UNIQUE (author_actor_id, record_type, record_rkey);
-
--- Remove comments (back to unoptimized schema)
-COMMENT ON COLUMN notifications.record_rkey IS NULL;
-COMMENT ON COLUMN notifications.record_cid IS NULL;
-COMMENT ON COLUMN notifications.subject_rkey IS NULL;

migrations/2025-11-15-030105_optimize_notification_storage/up.sql

···
--- Optimize notification storage to use INT8 (TID rkeys) and BYTEA (CID digests)
--- This matches the pattern used by posts, likes, reposts, and follows tables
--- Storage savings: TID TEXT→INT8 (38% smaller), CID TEXT→BYTEA (46% smaller)
-
--- WARNING: This migration truncates the notifications table
--- Notifications will be regenerated naturally as new activity occurs
-
--- Step 1: Truncate existing notifications (they'll be regenerated)
-TRUNCATE TABLE notifications;
-
--- Step 2: Drop old columns
-ALTER TABLE notifications
-DROP COLUMN record_rkey,
-DROP COLUMN record_cid,
-DROP COLUMN subject_rkey;
-
--- Step 3: Add new optimized columns
-ALTER TABLE notifications
-ADD COLUMN record_rkey INT8 NOT NULL,
-ADD COLUMN record_cid BYTEA NOT NULL,
-ADD COLUMN subject_rkey INT8;
-
--- Step 4: Recreate unique constraint with new column types
-ALTER TABLE notifications
-DROP CONSTRAINT IF EXISTS notifications_unique_record;
-ALTER TABLE notifications
-ADD CONSTRAINT notifications_unique_record
-UNIQUE (author_actor_id, record_type, record_rkey);
-
--- Add documentation comments
-COMMENT ON COLUMN notifications.record_rkey IS 'TID as INT8 (convert to string via i64_to_tid())';
-COMMENT ON COLUMN notifications.record_cid IS '32-byte CID digest (header stripped, reconstruct via digest_to_record_cid())';
-COMMENT ON COLUMN notifications.subject_rkey IS 'Optional TID as INT8 for subject record';
-
--- Final schema:
--- notifications (
---     id BIGSERIAL PRIMARY KEY,
---     recipient_actor_id INT NOT NULL REFERENCES actors(id),
---     author_actor_id INT NOT NULL REFERENCES actors(id),
---     indexed_at TIMESTAMPTZ NOT NULL,
---     reason notification_reason NOT NULL,
---     record_type notification_record_type NOT NULL,
---     record_rkey INT8 NOT NULL,              -- ✅ OPTIMIZED: was TEXT
---     record_cid BYTEA NOT NULL,              -- ✅ OPTIMIZED: was TEXT
---     subject_actor_id INT REFERENCES actors(id),
---     subject_record_type notification_record_type,
---     subject_rkey INT8,                      -- ✅ OPTIMIZED: was TEXT
---     UNIQUE (author_actor_id, record_type, record_rkey)
--- )

parakeet-db/src/schema.rs

···
 
 diesel::table! {
     notification_state (actor_id) {
+        actor_id -> Int4,
         seen_at -> Nullable<Timestamptz>,
         unread_count -> Nullable<Int4>,
-        actor_id -> Int4,
     }
 }
 
···
 
     notifications (id) {
         id -> Int8,
-        indexed_at -> Timestamptz,
         recipient_actor_id -> Int4,
         author_actor_id -> Int4,
+        indexed_at -> Timestamptz,
         reason -> NotificationReason,
         record_type -> NotificationRecordType,
+        record_rkey -> Int8,
+        record_cid -> Bytea,
         subject_actor_id -> Nullable<Int4>,
         subject_record_type -> Nullable<NotificationRecordType>,
-        record_rkey -> Int8,
-        record_cid -> Bytea,
         subject_rkey -> Nullable<Int8>,
     }
 }