Commit Graph

1 Commits

Author SHA1 Message Date
shankar0123 de4f93b35e refactor(cmd/server): extract migration block to migrations.go (Phase 9, 8b — behavior-aware)
Closes the third file Sprint 8 deferred. Sprint 8a (commit 3f1344e8)
shipped the pure-mechanical relocation of wire.go (helpers + adapter
types). Sprint 8b crosses the behavior-change boundary: extracts an
inline block from main()'s body into a new function, which introduces
a new function call frame.

What moved
==========
  cmd/server/migrations.go (new, 209 lines incl. BSL header + Phase 9
                           doc-comment + 6 imports + 2 functions)

Two unexported helpers:
  - parseMigrateOnlyFlag() bool — hand-parses os.Args[1:] for the
    `--migrate-only` token. Six-line implementation; matches the
    pre-Sprint-8b inline behavior exactly (bare match, no value form,
    no env override). Hand-parsed (not flag.Parse) for the same
    reason the original was: keeps flag.Parse's global state out of
    package main so future imports stay clean.
  - runBootMigrations(cfg, db, logger, migrateOnly) bool — owns the
    Phase 4 DEPL-M1 migration-via-hook posture. Reads
    CERTCTL_MIGRATIONS_VIA_HOOK, gates RunMigrations + RunSeed,
    handles the --migrate-only early-exit signal, runs RunDemoSeed
    when CERTCTL_DEMO_SEED=true.
    Returns true ONLY when migrateOnly was set; caller (main)
    handles the clean exit via `return` so deferred cleanup runs.
    Returns false in every other case — caller continues normal boot.
    On any migration / seed error: os.Exit(1) inline (matches the
    pre-extraction shape; recovery is impossible at this boot stage).

main.go delta
=============
  - Lines 54-72 (the --migrate-only flag parse + its Phase 4
    doc-comment): replaced with a single call
    `migrateOnly := parseMigrateOnlyFlag()` plus a 6-line pointer
    to migrations.go.
  - Lines 178-259 (the migrations-via-hook + RunMigrations +
    RunSeed + --migrate-only early-exit + RunDemoSeed inline
    block): replaced with a single call
    `if exitAfterMigrations := runBootMigrations(cfg, db, logger,
    migrateOnly); exitAfterMigrations { return }` plus an 8-line
    pointer to migrations.go.
  - No imports needed adjusting in main.go — the moved code's
    imports (database/sql, strings) were ALSO used by the rest of
    main(); they stay. (Notably, this is unlike Sprint 8a, which
    surfaced 5 unused imports requiring removal.)

main.go LOC: 2347 → 2260  (-87 lines)

Behavior-change contract (the single intentional shift)
========================================================
Every error path inside runBootMigrations calls os.Exit(1) directly
— byte-for-byte equivalent to the original inline shape (same log
message, same exit code, same no-defer-run on fatal).

THE ONE BEHAVIOR CHANGE: the --migrate-only SUCCESS path now returns
to main() rather than calling os.Exit(0) inline. Observable effect:
the `defer db.Close()` registered at line 175 in main() now runs at
clean exit instead of being skipped.

Why this is strictly an improvement (not a regression):
  - The original os.Exit(0) skipped every registered defer. db.Close
    never ran; the OS reclaimed the socket when the process died.
  - The new `return` causes db.Close to run on the orderly main()
    teardown path. PostgreSQL connection released cleanly via the
    Go *sql.DB.Close() contract rather than mid-flight socket
    teardown.
  - Migrations + seed are SYNCHRONOUS — by the time runBootMigrations
    returns true, all SQL work has fsync'd or returned errors. There's
    no async work that db.Close could truncate.
  - The exit code stays 0 (Kubernetes Job lifecycle still reports
    success).
  - The exit log message ("--migrate-only: migrations + seed
    complete; exiting without starting server lifecycle") fires
    BEFORE the return, identical to the pre-extraction position.

If an operator's monitoring is wired to detect "did the --migrate-only
container clean-shutdown its DB connection or did it just die," they
will see the new behavior. Every other observable signal is identical.

Documented in migrations.go's doc-comment so the next maintainer
doesn't think the change was accidental.

Why this is a separate commit from Sprint 8a
============================================
Sprint 8a was pure mechanical relocation — function definitions
moved between sibling files in the same package, zero runtime
semantics changed. Sprint 8b introduces a new function call frame,
which has a non-zero (if small + documented + improvement-shaped)
behavior delta.

Splitting these into two commits means git bisect against a future
boot-time regression gets a clean answer:
  3f1344e8 ... wire.go        — could not have changed behavior
  <this>   ... migrations.go  — one specific documented shift, see
                                commit body + migrations.go header

Anyone tracing a boot-time issue knows EXACTLY which commit to scrutinize.

Verification (all clean):
  go build ./cmd/server/...               → clean (no unused imports)
  go vet ./cmd/server/...                 → clean
  gofmt -l cmd/server/                    → clean
  go test ./cmd/server/... -count=1 -short → ok (0.39s; main_test.go
                                              + the existing
                                              preflight_*_test.go +
                                              finalhandler_test.go +
                                              auth_*_test.go +
                                              tls_test.go all pass —
                                              including main_test.go
                                              which exercises the
                                              boot flow through the
                                              new call site)
  staticcheck ./cmd/server/...            → clean
  grep -nE 'migrateOnly|migrationsViaHook|RunMigrations|RunSeed|RunDemoSeed'
       cmd/server/main.go   → just the runBootMigrations call site +
                                the parseMigrateOnlyFlag call site;
                                the inline block is gone.

LOC delta:
  main.go:       2347 → 2260  (-87 lines: -18 from flag-parse
                                          extraction, -75 from
                                          migration-block extraction,
                                          +6 from new call-site +
                                          pointer comments)
  migrations.go: new, 209 lines (incl. ~95-line Phase 9 doc-comment +
                                BSL header + package decl + 6-line
                                import block)

Phase 9 Sprint 8 closure
========================
Sprint 8a (wire.go) + Sprint 8b (this commit) together close the
Phase 9 prompt's three-file split for cmd/server/main.go:

  cmd/server/main.go        2966 → 2260  (-706 lines, -23.8%)
  cmd/server/wire.go        new, 758 LOC
  cmd/server/migrations.go  new, 209 LOC

Cumulative Phase 9 (Sprints 1-8b):
  config.go:                 3403 → 1342 LOC (-60.6% across 7 sprints)
  cmd/server/main.go:        2966 → 2260 LOC (-23.8% across this
                                              sprint + Sprint 8a)
  Combined LOC reduction in the two largest backend files: -2,767

Next queued (Sprint 9): internal/service/acme.go (1965 LOC). Per
the operator's decision after Sprint 8 (Option B = sibling files
in the same package, no subpackage split): the cut will keep the
package name `service` and split into
internal/service/{acme,acme_orders,acme_authz,acme_challenges,
acme_nonces,acme_gc}.go. Zero import-path churn for callers.

Closes: cowork/certctl-architecture-diligence-audit.html#fix-ARCH-M2
        (partial — Sprint 8 fully closed at 9 of 12 effective splits)
2026-05-14 09:13:38 +00:00