mesh-catalog/modules/postgres/cmd/postgres-provider/retire_pg.go

package main

// What retired means in postgres (novox/hq ADR 0230).
//
// **Retired is the login locked, the database kept, and the role marked.** `ALTER ROLE … NOLOGIN` —
// the consumer's login can no longer connect, as itself, to anything — and its open sessions are
// ended. The database, its owner, its grants and every byte in it stay exactly as they were: CONNECT is
// not revoked and the database still allows connections, because the backup contribution dumps every
// database and a retired one is still worth backing up, and because re-enabling must give it back as it
// was. The mark is the role's comment, a JSON object the mesh owns:
//
//	{"mesh":"consumer","node":"ace"}                                        active
//	{"mesh":"consumer","node":"ace","retired":"2026-10-06T…Z","why":"…"}     retired
//
// Asked for again, the ordinary create sets LOGIN and the password again and writes the active mark.
// Deleted — only through `cleanup delete` — the database is dropped, then the role, unless it still owns
// another database (a set-aside copy), which is said and kept.
//
// **What the mesh made is recognised by its mark, or by its shape before the mark existed**: a role
// valid until 'infinity' (only the mesh's role statement says that) that owns a database of its own
// name. A role of any other shape is somebody else's and is never listed, retired or deleted. A database
// renamed aside — `<name>_deleted_<yyyymmdd>`, by postgres_retire_database or by hand — is listed as
// retired too, since that date, so a person sees it and can delete it.

import (
	"context"
	"encoding/json"
	"fmt"
	"regexp"
	"strconv"
	"time"
)

// KindSetAside is a database renamed aside, listed for deletion beside the retired consumers.
const KindSetAside = "set-aside-database"

// roleMark is the comment the mesh keeps on a consumer's role.
type roleMark struct {
	Mesh    string `json:"mesh"`
	Node    string `json:"node,omitempty"`
	Retired string `json:"retired,omitempty"`
	Why     string `json:"why,omitempty"`
}

func readMark(comment string) (roleMark, bool) {
	var m roleMark
	if comment == "" || json.Unmarshal([]byte(comment), &m) != nil || m.Mesh != KindConsumer {
		return roleMark{}, false
	}
	return m, true
}

// MarkStatement is the comment that marks a role as the mesh's consumer, active or retired.
func MarkStatement(role string, m roleMark) string {
	m.Mesh = KindConsumer
	b, _ := json.Marshal(m)
	return fmt.Sprintf("COMMENT ON ROLE %s IS %s", Ident(role), Literal(string(b)))
}

// MarkActive writes the active mark — after create, which has already set LOGIN.
func (c *Client) MarkActive(ctx context.Context, role, node string) error {
	_, err := c.Query(ctx, "", MarkStatement(role, roleMark{Node: node}))
	return err
}

// RetireRole locks the login, ends its sessions and marks it retired with when and why. Its database
// is not touched. A role that does not exist has nothing to retire.
func (c *Client) RetireRole(ctx context.Context, role, why string, at time.Time) error {
	r, err := c.Query(ctx, "", "SELECT coalesce(shobj_description(oid, 'pg_authid'), '') FROM pg_roles WHERE rolname = "+
		Literal(role))
	if err != nil {
		return err
	}
	if len(r.Rows) == 0 {
		return nil
	}
	prev, _ := readMark(cell(r.Rows[0], 0))
	if err := c.LockRole(ctx, role); err != nil {
		return err
	}
	_, err = c.Query(ctx, "", MarkStatement(role, roleMark{Node: prev.Node, Retired: at.UTC().Format(time.RFC3339), Why: why}))
	return err
}

var setAside = regexp.MustCompile(`_deleted_(\d{8})$`)

// InventoryStatement lists every role that may be the mesh's, with its login, mark, shape and the size
// of its own database.
const InventoryStatement = `SELECT r.rolname, r.rolcanlogin, coalesce(shobj_description(r.oid, 'pg_authid'), '') AS mark,
  coalesce(r.rolvaliduntil = 'infinity', false) AS mesh_shaped,
  (SELECT pg_database_size(d.datname) FROM pg_database d WHERE d.datname = r.rolname AND d.datdba = r.oid) AS size
FROM pg_roles r
WHERE r.rolname !~ '^pg_' AND NOT r.rolsuper AND r.rolname <> ` + "'" + Reader + "'" + `
ORDER BY r.rolname`

// SetAsideStatement lists the databases renamed aside.
const SetAsideStatement = `SELECT datname, pg_database_size(datname) FROM pg_database WHERE datname ~ '_deleted_[0-9]{8}$' ORDER BY datname`

// Inventory is what this server holds that the mesh made.
func (c *Client) Inventory(ctx context.Context) (Inventory, error) {
	inv := Inventory{Active: []string{}, Retired: []Retired{}}
	r, err := c.Query(ctx, "", InventoryStatement)
	if err != nil {
		return inv, err
	}
	for _, row := range r.Rows {
		name, login, comment, shaped, size := cell(row, 0), cell(row, 1) == "t", cell(row, 2), cell(row, 3) == "t", cell(row, 4)
		m, marked := readMark(comment)
		if !marked && !(shaped && size != "") {
			continue // not the mesh's
		}
		if login && m.Retired == "" {
			inv.Active = append(inv.Active, name)
			continue
		}
		at, _ := time.Parse(time.RFC3339, m.Retired)
		inv.Retired = append(inv.Retired, Retired{Consumer: name, Node: m.Node, RetiredAt: at, Why: m.Why,
			SizeBytes: sizeOf(size), Kind: KindConsumer})
	}
	r, err = c.Query(ctx, "", SetAsideStatement)
	if err != nil {
		return inv, err
	}
	for _, row := range r.Rows {
		name := cell(row, 0)
		m := setAside.FindStringSubmatch(name)
		if m == nil {
			continue
		}
		at, _ := time.Parse("20060102", m[1])
		inv.Retired = append(inv.Retired, Retired{Consumer: name, RetiredAt: at, SizeBytes: sizeOf(cell(row, 1)),
			Why: "renamed aside — by postgres_retire_database, or by hand", Kind: KindSetAside})
	}
	return inv, nil
}

// DeleteRetired drops what a retired entry names: a consumer's database and then its role, or one
// database set aside. Answers the bytes freed.
func (c *Client) DeleteRetired(ctx context.Context, r Retired) (int64, error) {
	if r.Kind == KindSetAside {
		if !setAside.MatchString(r.Consumer) {
			return 0, fmt.Errorf("%s is not a database set aside", r.Consumer)
		}
		return c.dropDatabase(ctx, r.Consumer)
	}
	// Only a retired one: the login must be locked here and now, whatever the caller believed.
	row, err := c.Query(ctx, "", "SELECT rolcanlogin FROM pg_roles WHERE rolname = "+Literal(r.Consumer))
	if err != nil {
		return 0, err
	}
	if len(row.Rows) > 0 && cell(row.Rows[0], 0) == "t" {
		return 0, fmt.Errorf("%s can log in — it is active, not retired, and is not deleted", r.Consumer)
	}
	freed, err := c.dropDatabase(ctx, r.Consumer)
	if err != nil {
		return freed, err
	}
	if len(row.Rows) == 0 {
		return freed, nil
	}
	owns, err := c.Query(ctx, "", "SELECT datname FROM pg_database WHERE datdba = (SELECT oid FROM pg_roles WHERE rolname = "+
		Literal(r.Consumer)+")")
	if err != nil {
		return freed, err
	}
	if len(owns.Rows) > 0 {
		return freed, fmt.Errorf("%s's database is dropped; the role is kept because it still owns %s — delete that first",
			r.Consumer, cell(owns.Rows[0], 0))
	}
	_, err = c.Query(ctx, "", fmt.Sprintf("DROP ROLE %s", Ident(r.Consumer)))
	return freed, err
}

func (c *Client) dropDatabase(ctx context.Context, database string) (int64, error) {
	r, err := c.Query(ctx, "", "SELECT pg_database_size(datname) FROM pg_database WHERE datname = "+Literal(database))
	if err != nil || len(r.Rows) == 0 {
		return 0, err
	}
	freed := sizeOf(cell(r.Rows[0], 0))
	if _, err := c.Query(ctx, "", "SELECT pg_terminate_backend(pid) FROM pg_stat_activity WHERE datname = "+
		Literal(database)+" AND pid <> pg_backend_pid()"); err != nil {
		return 0, err
	}
	if _, err := c.Query(ctx, "", fmt.Sprintf("DROP DATABASE %s", Ident(database))); err != nil {
		return 0, err
	}
	return freed, nil
}

func cell(row []*string, i int) string {
	if i < len(row) && row[i] != nil {
		return *row[i]
	}
	return ""
}

func sizeOf(s string) int64 {
	n, err := strconv.ParseInt(s, 10, 64)
	if err != nil {
		return -1
	}
	return n
}