feat(storage): add PostgreSQL storage-table schema definitions

[Kiran01bm]

Summary

Adds the PostgreSQL translation of SchemaBot's 13 embedded storage-table DDL files under pkg/schema/postgres/, exposed as schema.PostgresFS. This is the schema-definition slice of the dialect-split storage bootstrap; the PostgreSQL EnsureSchema bootstrapper and advisory locker follow in a separate PR.

What

• pkg/schema/postgres/*.sql: one file per storage table, mirroring pkg/schema/mysql/ table-for-table and column-for-column. Each file holds one CREATE TABLE plus its CREATE INDEX statements.
• schema.PostgresFS embed alongside MySQLFS.
• Unit tests: file-set parity with the MySQL directory, index parity (every MySQL index must have a postgres counterpart with the same table, ordered column list, and uniqueness — and vice versa), a lint for MySQL-only constructs (backticks, AUTO_INCREMENT, ON UPDATE CURRENT_TIMESTAMP, unsigned, tinyint, datetime, engine/charset/collation options), table-name/filename agreement, and index-name rules (table prefix, schema-wide uniqueness, 63-char limit).
• Integration test: executes every file against a real postgres:16 testcontainer, then diffs information_schema.columns against the TiDB-parsed MySQL files — column sets, nullability, the declared type mapping, and varchar widths must match exactly.
• New deps: testcontainers-go/modules/postgres (pinned to the repo's existing testcontainers v0.40.0) and pgx/v5 stdlib driver (test-only for now; the PostgreSQL store will use it next).

Why

The storage bootstrap dispatch already fails closed for non-MySQL dialects; this supplies the PostgreSQL schema definitions it will route to.

Why a separate DDL directory rather than sharing the MySQL files (or generating both from one source): the differences are grammatical, not just type names — backtick quoting, AUTO_INCREMENT, inline KEY clauses, ENGINE/CHARSET table options, and unsigned integer types are all syntax errors in PostgreSQL, while index names move from a per-table to a schema-wide namespace. The MySQL files are also format-pinned: EnsureSchema diffs them via Spirit's TiDB parser, which requires canonical SHOW CREATE TABLE form, so they cannot drift toward anything dialect-neutral. And several translations are deliberate engineering choices (jsonb over json, identity columns, application-side updated_at) that a mechanical transliteration layer would get wrong or bury in special cases. Behavioral equivalence between the two schemas is enforced where it belongs — the cross-dialect storagetest parity suite (#935) will validate every store against the same interface contract once both land, and this PR's structural tests (file-set, index, column-set/nullability/type/width parity against the TiDB-parsed MySQL files) catch schema drift directly.

Translation decisions worth review:

• updated_at: application-side stamping, no trigger. MySQL's ON UPDATE CURRENT_TIMESTAMP has no PostgreSQL equivalent. Rather than installing per-table triggers in the bootstrap, the PostgreSQL store will stamp updated_at in its UPDATE statements, and the cross-dialect parity suite enforces the behavior. Columns keep DEFAULT CURRENT_TIMESTAMP for inserts.
• Index names are table-prefixed (idx_applies_repo_pr, not idx_repo_pr): MySQL index names are per-table, PostgreSQL's share one schema-wide namespace, so the MySQL names (reused across tables) would collide. A few names are also normalized where the MySQL name embedded its table (idx_apply_control_request_status → idx_apply_control_requests_status); the index parity test pins table/columns/uniqueness so renames can't hide a structural change.
• Type mapping: identity PKs (GENERATED BY DEFAULT AS IDENTITY, allowing explicit ids like AUTO_INCREMENT), jsonb for json, boolean for tinyint(1) (the Go structs are bool), bigint for int unsigned (PostgreSQL has no unsigned types; integer would halve the value range), timestamp for datetime.
• Timestamp precision widens to PostgreSQL's default microseconds. MySQL datetime is whole-second (except the one deliberate datetime(6) lease-expiry column); bare PG timestamp is microsecond everywhere. This is deliberate — second-truncating with timestamp(0) would replicate a MySQL storage artifact rather than a requirement, and the store layer already treats time precision as dialect-specific. Consequence to keep in mind: written time.Time values round-trip second-truncated on MySQL but exact on PostgreSQL, so cross-dialect behavioral tests must not assert exact stored-time equality across dialects.
• Collation: the PG schema is case-sensitive; MySQL's utf8mb4_0900_ai_ci is case- and accent-insensitive. Every varchar comparison, and every varchar unique index, dedups case-insensitively on MySQL but byte-wise on PostgreSQL. Nothing in the store layer normalizes case today, so this stands as an open decision for the PostgreSQL store PR: citext, LOWER() expression indexes, or a documented app-side canonicalization guarantee (e.g. for repository names, which GitHub treats case-insensitively but case-preservingly).
• Identity columns accept explicit ids but do not advance the backing sequence past them. Unlike AUTO_INCREMENT, an explicit-id backfill must setval the sequence afterwards or later default inserts collide. No code inserts explicit ids today; noting it for the store/backfill PR.

Merge coordination: the file-set and index parity tests intentionally turn main red if a MySQL table or index lands without a postgres counterpart. Two open PRs interact: #867 adds mysql/check_refresh_requests.sql (and an index on checks), #648 changes applies/apply_operations indexes. Whichever merges second must add the postgres counterpart — that's the tests doing their job.

[Copilot]

Pull request overview

Adds a PostgreSQL dialect copy of SchemaBot’s embedded storage-table schema files and wires them into pkg/schema as an embedded FS, with tests to enforce cross-dialect parity and validate that the DDL executes against a real Postgres instance.

Changes:

• Added pkg/schema/postgres/*.sql storage-table DDL translations and embedded them as schema.PostgresFS.
• Added unit tests to enforce file-set parity, table/index naming rules, and to lint against MySQL-only syntax in Postgres files.
• Added an integration test that executes all Postgres DDL files in a Postgres testcontainer and diffs column names + nullability against the TiDB-parsed MySQL schema.

Reviewed changes

Copilot reviewed 17 out of 18 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
pkg/schema/schema.go	Embeds Postgres schema files alongside MySQL via PostgresFS.
pkg/schema/postgres/applies.sql	Postgres DDL for applies storage table and indexes.
pkg/schema/postgres/apply_comments.sql	Postgres DDL for apply_comments storage table and indexes.
pkg/schema/postgres/apply_control_requests.sql	Postgres DDL for apply_control_requests storage table and indexes.
pkg/schema/postgres/apply_logs.sql	Postgres DDL for apply_logs storage table and indexes.
pkg/schema/postgres/apply_operations.sql	Postgres DDL for apply_operations storage table and indexes.
pkg/schema/postgres/apply_target_locks.sql	Postgres DDL for apply_target_locks storage table and uniqueness/index rules.
pkg/schema/postgres/checks.sql	Postgres DDL for checks storage table and indexes.
pkg/schema/postgres/locks.sql	Postgres DDL for locks storage table and indexes.
pkg/schema/postgres/plan_comments.sql	Postgres DDL for plan_comments storage table and indexes.
pkg/schema/postgres/plans.sql	Postgres DDL for plans storage table and indexes.
pkg/schema/postgres/settings.sql	Postgres DDL for settings storage table and indexes.
pkg/schema/postgres/tasks.sql	Postgres DDL for tasks storage table and indexes.
pkg/schema/postgres/webhook_events.sql	Postgres DDL for webhook_events storage table and indexes.
pkg/schema/postgres_test.go	Unit tests enforcing MySQL/Postgres schema file parity and Postgres naming/syntax rules.
pkg/schema/postgres_integration_test.go	Integration test executing Postgres DDL and comparing column sets/nullability to MySQL (TiDB-parsed).
go.mod	Adds dependencies for pgx stdlib driver and testcontainers Postgres module.
go.sum	Adds checksums for newly introduced Postgres-related dependencies.

Suppressed comments (1)

pkg/schema/postgres_integration_test.go:98

• The rows close error is discarded. If the driver reports an error on Close (e.g., protocol/stream issues), it’s useful for the test to surface it instead of silently ignoring it.

	defer func() { _ = rows.Close() }()

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[aparajon]

🤖 Adversarial correctness review, requested by Armand and performed by his agent. Reviewed at head 3d04ec7.

Verdict: clean — no fix-before-merge findings. The translations are faithful, the parity tests fail closed, and the two dimensions the tests don't cover (primary keys and column defaults) I audited by hand across all 13 tables and found zero divergences.

The core risk in a hand-translated schema is a silent semantic gap the structural tests don't reach. The tests cover a lot mechanically — file-set parity both directions, index shape parity both directions (with a parse-count cross-check so an index the regex can't read fails loudly instead of escaping the lint), and a real-server integration test that reads back information_schema after the DDL runs, so it asserts what PostgreSQL actually created. What they don't cover, I checked manually: every one of the 13 files declares PRIMARY KEY (id), and a mechanical column-by-column defaults comparison found only the expected translations (identity columns are implicitly NOT NULL; tinyint(1) DEFAULT '1'/'0' → boolean DEFAULT TRUE/FALSE). The datetime(6) lease-expiry column maps into PostgreSQL's default microsecond timestamp with no precision loss, and the unique index on nullable idempotency_key has the same multiple-NULLs semantics in both dialects.

Follow-up, not blocking
PK parity is untested	A future edit could drop PRIMARY KEY (id) from a postgres file and every test would still pass. The integration test could assert it cheaply — information_schema.table_constraints on the PG side against the TiDB-parsed PK on the MySQL side.
Default parity is untested	Same shape: information_schema.columns.column_default is already one column away from what the test reads today. All 13 tables match at this head, but drift would be invisible.
updated_at enforcement lands later	The app-side stamping decision is sound, but nothing enforces it yet — the cross-dialect parity suite has no updated_at-advances assertion today. The PostgreSQL store PR should bring one.

Verified solid: every forbidden-construct lint token is genuinely absent from the postgres files and the lint's policy bans (charset/collate) are documented with their rationale; the index-name rules (table prefix, schema-wide uniqueness, 63-char identifier limit) match PostgreSQL's actual namespace semantics; the type mapping is declared exactly once (expectedPostgresColumnType) and its choices are right — bigint for unsigned ints rather than range-halving integer, jsonb, identity PKs that accept explicit ids; the integration test asserts column sets in both directions so an extra postgres column fails too; both Copilot threads are already resolved correctly (CloseAndLog applied; engine_migration_id kept because it mirrors an existing production column and a one-sided rename would fail the parity tests this PR adds); the branch is 0 behind main; CI is fully green; and locally at 3d04ec7 the build, the unit parity tests with -race, and the postgres:16 integration test all pass. The merge-coordination note in the PR body (#867/#648 must add postgres counterparts if they land second) is accurate — that's the file-set and index parity tests doing their job.

This review was generated by Claude Code (claude-fable-5).

[aparajon]

🤖 Approving on Armand's behalf after the adversarial correctness review above (no fix-before-merge findings). This stamp was left by Claude Code (claude-fable-5).