prole/demo/ci-migration/01_schema_setup.sql

-- 01_schema_setup.sql
-- Creates the demo schema, legacy table, OLTP workload table,
-- and all monitoring views for the CI migration danger demo.
--
-- Run as superuser (postgres) against knoe-db.
-- Safe to re-run: all objects use IF NOT EXISTS / CREATE OR REPLACE.

CREATE SCHEMA IF NOT EXISTS demo;

-- ── Legacy table (source of the backfill) ────────────────────────────────────
-- Represents the old non-partitioned ci_job_artifacts table on a 10 TB GitLab
-- DB. In our scaled demo we'll fill this with ~5 million rows (~1 GB).
CREATE TABLE IF NOT EXISTS demo.ci_job_artifacts_legacy (
    id          BIGSERIAL PRIMARY KEY,
    job_id      BIGINT    NOT NULL,
    project_id  BIGINT    NOT NULL,
    file_type   SMALLINT  NOT NULL DEFAULT 0,
    size_bytes  BIGINT,
    file_store  SMALLINT  NOT NULL DEFAULT 1,
    checksum    TEXT,
    created_at  TIMESTAMP NOT NULL,
    expire_at   TIMESTAMP,
    locked      BOOLEAN   DEFAULT FALSE
);

CREATE INDEX IF NOT EXISTS idx_legacy_created_at
    ON demo.ci_job_artifacts_legacy (created_at);

-- ── OLTP workload table (concurrent traffic during migration) ─────────────────
-- Simulates background CI pipeline activity during the migration window.
-- UPDATEs to this table generate dead tuples that autovacuum must clean.
-- When a long-running migration transaction holds an old snapshot, autovacuum
-- is blocked and dead tuples pile up — this is the MVCC bloat we'll observe.
CREATE TABLE IF NOT EXISTS demo.ci_build_status (
    id         BIGSERIAL PRIMARY KEY,
    job_id     BIGINT    NOT NULL,
    status     TEXT      NOT NULL DEFAULT 'pending',
    updated_at TIMESTAMP NOT NULL DEFAULT NOW()
);

-- ── Monitoring views ──────────────────────────────────────────────────────────

CREATE OR REPLACE VIEW demo.mvcc_bloat_monitor AS
SELECT
    relname                                                        AS table_name,
    n_live_tup                                                     AS live_rows,
    n_dead_tup                                                     AS dead_rows,
    ROUND(n_dead_tup::NUMERIC / NULLIF(n_live_tup, 0) * 100, 2)  AS dead_pct,
    last_autovacuum,
    last_autoanalyze,
    pg_size_pretty(pg_total_relation_size('demo.' || relname))    AS total_size,
    pg_size_pretty(pg_relation_size('demo.' || relname))          AS table_size
FROM pg_stat_user_tables
WHERE schemaname = 'demo'
ORDER BY n_dead_tup DESC;

-- Shows any transaction open longer than 5 seconds — the migration blocker.
CREATE OR REPLACE VIEW demo.long_running_tx AS
SELECT
    pid,
    now() - xact_start                                  AS tx_duration,
    now() - query_start                                 AS query_duration,
    state,
    wait_event_type,
    wait_event,
    left(query, 200)                                    AS query_snippet,
    backend_type,
    application_name
FROM pg_stat_activity
WHERE xact_start IS NOT NULL
  AND now() - xact_start > INTERVAL '5 seconds'
  AND pid <> pg_backend_pid()
ORDER BY tx_duration DESC;

-- WAL generation tracker — shows how fast WAL is growing.
-- Run before and after to calculate the delta.
CREATE OR REPLACE VIEW demo.wal_progress AS
SELECT
    pg_current_wal_lsn()                                                        AS current_lsn,
    pg_size_pretty(pg_wal_lsn_diff(pg_current_wal_lsn(), '0/0'::pg_lsn))      AS total_wal,
    (SELECT count(*) FROM pg_ls_waldir())                                       AS wal_segment_count,
    pg_size_pretty((SELECT sum(size) FROM pg_ls_waldir()))                      AS wal_dir_size;

-- Lock wait graph — shows who is blocking whom.
CREATE OR REPLACE VIEW demo.lock_waits AS
SELECT
    blocked_a.pid                            AS blocked_pid,
    blocking_a.pid                           AS blocking_pid,
    now() - blocked_a.query_start           AS wait_duration,
    left(blocked_a.query, 120)              AS blocked_query,
    left(blocking_a.query, 120)             AS blocking_query,
    blocked_l.locktype,
    blocked_l.relation::regclass            AS locked_relation
FROM pg_catalog.pg_locks         blocked_l
JOIN pg_catalog.pg_locks         blocking_l
    ON  blocking_l.locktype    = blocked_l.locktype
    AND blocking_l.relation    = blocked_l.relation
    AND blocking_l.granted
    AND NOT blocked_l.granted
JOIN pg_stat_activity blocked_a  ON blocked_a.pid  = blocked_l.pid
JOIN pg_stat_activity blocking_a ON blocking_a.pid = blocking_l.pid
ORDER BY wait_duration DESC;

-- Autovacuum activity — shows if autovacuum is currently running or stuck.
CREATE OR REPLACE VIEW demo.autovacuum_status AS
SELECT
    pid,
    now() - xact_start    AS running_for,
    query                 AS vacuum_query,
    wait_event_type,
    wait_event
FROM pg_stat_activity
WHERE query LIKE 'autovacuum:%'
ORDER BY xact_start;

-- Oldest transaction horizon — the horizon that blocks dead tuple cleanup.
-- When this is far in the past and dead tuples are climbing, we have MVCC bloat.
CREATE OR REPLACE VIEW demo.xmin_horizon AS
SELECT
    pid,
    backend_xmin                                                   AS xmin,
    age(backend_xmin)                                              AS xmin_age,
    now() - xact_start                                             AS tx_age,
    state,
    left(query, 120)                                               AS query_snippet
FROM pg_stat_activity
WHERE backend_xmin IS NOT NULL
ORDER BY age(backend_xmin) DESC;

\echo 'Schema and views created. Run 02_generate_data.sql next.'