refactor: migrate to run Corrosion service as managed container instead of systemd unit, automatically migrate data to v1

2026-08-26 19:13:34 +00:00 · 2026-05-25 17:15:44 +10:00
parent 9236aa341a
commit 71da32549a
8 changed files with 599 additions and 185 deletions
@@ -0,0 +1,282 @@
+// Package corromigrate handles the one-time migration of the on-disk Corrosion store from
+// systemd-managed v0.x to v1.0.0 running in a uncloudd-managed container.
+// TODO: remove in 0.22 assuming all pre 0.20 clusters upgraded their pre-v1 Corrosion.
+package corromigrate
+
+import (
+	"context"
+	"database/sql"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"log/slog"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"time"
+
+	"github.com/psviderski/uncloud/internal/fs"
+	"github.com/psviderski/uncloud/internal/machine/api/pb"
+	"github.com/psviderski/uncloud/internal/machine/store"
+	"google.golang.org/protobuf/encoding/protojson"
+	_ "modernc.org/sqlite"
+)
+
+// legacyUnitName is the systemd unit that ran Corrosion before uncloudd took over its lifecycle.
+// The unit file at /etc/systemd/system/uncloud-corrosion.service and the binary at
+// /usr/local/bin/uncloud-corrosion are left in place after migration because uncloudd's sandbox
+// (ProtectSystem=full) blocks their removal. Operators should delete them manually as documented
+// in the release notes.
+const legacyUnitName = "uncloud-corrosion.service"
+
+// Seed is the on-disk representation of the durable rows dumped from a v0.x Corrosion store,
+// to be re-applied to the fresh v1.0.0 store. Existence of the seed file is the signal that
+// migration has not yet fully completed.
+type Seed struct {
+	Cluster  []ClusterEntry `json:"cluster"`
+	Machines []MachineEntry `json:"machines"`
+}
+
+type ClusterEntry struct {
+	Key   string `json:"key"`
+	Value string `json:"value"`
+}
+
+type MachineEntry struct {
+	ID   string `json:"id"`
+	Info string `json:"info"`
+}
+
+func seedPath(dir string) string { return dir + ".seed-v1.json" }
+
+func backupPath(dir string) string {
+	return dir + ".backup-" + time.Now().UTC().Format("20060102.150405")
+}
+
+// MigrateIfNeeded stops the legacy systemd Corrosion unit and, when a v0.x store.db is detected, dumps its durable
+// rows to a seed file next to the data dir, then backs up the dir so the new version can start fresh.
+// Idempotent: if the seed file already exists, the dump step is skipped (a prior run already produced it).
+func MigrateIfNeeded(ctx context.Context, dir, owner string) error {
+	stopLegacyUnit()
+
+	seedFile := seedPath(dir)
+	if _, err := os.Stat(seedFile); err == nil {
+		slog.Info("Corrosion migrations seed file found, dump already produced by a prior run.",
+			"path", seedFile)
+		return nil
+	} else if !errors.Is(err, os.ErrNotExist) {
+		return fmt.Errorf("stat seed file '%s': %w", seedFile, err)
+	}
+
+	dbFile := filepath.Join(dir, "store.db")
+	if _, err := os.Stat(dbFile); errors.Is(err, os.ErrNotExist) {
+		return nil
+	} else if err != nil {
+		return fmt.Errorf("stat '%s': %w", dbFile, err)
+	}
+
+	isV0, err := isV0Store(ctx, dbFile)
+	if err != nil {
+		return fmt.Errorf("detect old store version: %w", err)
+	}
+	if !isV0 {
+		return nil
+	}
+
+	slog.Info("Migrating Corrosion store from v0.x to v1.0.0.", "db", dbFile)
+
+	dump, err := dumpOldStore(ctx, dbFile)
+	if err != nil {
+		return fmt.Errorf("dump old store: %w", err)
+	}
+	slog.Info("Corrosion store dumped.",
+		"cluster_rows", len(dump.Cluster), "machine_rows", len(dump.Machines))
+
+	if err = writeSeedAtomic(seedFile, dump, owner); err != nil {
+		return fmt.Errorf("write seed file: %w", err)
+	}
+	slog.Info("Corrosion migration seed file written.", "path", seedFile)
+
+	backup := backupPath(dir)
+	if err = os.Rename(dir, backup); err != nil {
+		return fmt.Errorf("backup corrosion dir: %w", err)
+	}
+	slog.Info("Old Corrosion dir backed up.", "from", dir, "to", backup)
+
+	return nil
+}
+
+// stopLegacyUnit stops and disables the legacy uncloud-corrosion systemd unit if present.
+// Best-effort: errors are swallowed because the unit may not exist (fresh install) or systemctl
+// may be unavailable (containerised hosts).
+func stopLegacyUnit() {
+	for _, cmd := range []string{"stop", "disable"} {
+		if err := exec.Command("systemctl", cmd, legacyUnitName).Run(); err != nil {
+			slog.Debug("systemctl on legacy corrosion unit failed (likely absent).",
+				"cmd", cmd, "unit", legacyUnitName, "err", err)
+		}
+	}
+}
+
+// isV0Store reports whether the SQLite database at dbPath is in the old v0.x Corrosion format.
+// v0.x has a __corro_bookkeeping table that was dropped in v1.0.0.
+func isV0Store(ctx context.Context, dbPath string) (bool, error) {
+	db, err := sql.Open("sqlite", "file:"+dbPath+"?mode=ro")
+	if err != nil {
+		return false, fmt.Errorf("open db '%s': %w", dbPath, err)
+	}
+	defer db.Close()
+
+	var one int
+	err = db.QueryRowContext(ctx,
+		"SELECT 1 FROM sqlite_master WHERE type='table' AND name='__corro_bookkeeping' LIMIT 1",
+	).Scan(&one)
+	if errors.Is(err, sql.ErrNoRows) {
+		return false, nil
+	}
+	if err != nil {
+		return false, err
+	}
+	return true, nil
+}
+
+func dumpOldStore(ctx context.Context, dbPath string) (*Seed, error) {
+	db, err := sql.Open("sqlite", "file:"+dbPath+"?mode=ro")
+	if err != nil {
+		return nil, fmt.Errorf("open %q: %w", dbPath, err)
+	}
+	defer db.Close()
+
+	seed := &Seed{}
+
+	clusterRows, err := db.QueryContext(ctx, "SELECT key, value FROM cluster")
+	if err != nil {
+		return nil, fmt.Errorf("query cluster: %w", err)
+	}
+	for clusterRows.Next() {
+		var key, value string
+		if err = clusterRows.Scan(&key, &value); err != nil {
+			clusterRows.Close()
+			return nil, fmt.Errorf("scan cluster row: %w", err)
+		}
+		seed.Cluster = append(seed.Cluster, ClusterEntry{Key: key, Value: value})
+	}
+	if err = clusterRows.Err(); err != nil {
+		clusterRows.Close()
+		return nil, fmt.Errorf("iterate cluster rows: %w", err)
+	}
+	clusterRows.Close()
+
+	machineRows, err := db.QueryContext(ctx, "SELECT id, info FROM machines")
+	if err != nil {
+		return nil, fmt.Errorf("query machines: %w", err)
+	}
+	defer machineRows.Close()
+	for machineRows.Next() {
+		var id, info string
+		if err = machineRows.Scan(&id, &info); err != nil {
+			return nil, fmt.Errorf("scan machine row: %w", err)
+		}
+		seed.Machines = append(seed.Machines, MachineEntry{ID: id, Info: info})
+	}
+	if err = machineRows.Err(); err != nil {
+		return nil, fmt.Errorf("iterate machine rows: %w", err)
+	}
+	return seed, nil
+}
+
+func writeSeedAtomic(path string, seed *Seed, owner string) error {
+	tmp := path + ".tmp"
+	data, err := json.MarshalIndent(seed, "", "  ")
+	if err != nil {
+		return err
+	}
+
+	f, err := os.OpenFile(tmp, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0o600)
+	if err != nil {
+		return err
+	}
+	if _, err = f.Write(data); err != nil {
+		f.Close()
+		os.Remove(tmp)
+		return fmt.Errorf("write seed: %w", err)
+	}
+	if err = f.Sync(); err != nil {
+		f.Close()
+		os.Remove(tmp)
+		return fmt.Errorf("fsync seed: %w", err)
+	}
+	if err = f.Close(); err != nil {
+		os.Remove(tmp)
+		return fmt.Errorf("close seed: %w", err)
+	}
+
+	if owner != "" {
+		if chErr := fs.Chown(tmp, owner, owner); chErr != nil {
+			os.Remove(tmp)
+			return chErr
+		}
+	}
+	if err = os.Rename(tmp, path); err != nil {
+		return fmt.Errorf("rename seed: %w", err)
+	}
+
+	// Fsync the parent directory so the rename entry is durable.
+	dir, err := os.Open(filepath.Dir(path))
+	if err != nil {
+		return fmt.Errorf("open parent dir for fsync: %w", err)
+	}
+	defer dir.Close()
+	if err = dir.Sync(); err != nil {
+		return fmt.Errorf("fsync parent dir: %w", err)
+	}
+	return nil
+}
+
+// ApplySeedIfPresent re-applies the durable rows from the seed file into the running Corrosion store,
+// then deletes the seed file as the final completion marker. Idempotent: re-running on the same seed
+// is a no-op for cluster rows (INSERT OR REPLACE) and machines rows (skipped via GetMachine).
+func ApplySeedIfPresent(ctx context.Context, dir string, st *store.Store) error {
+	seedFile := seedPath(dir)
+	data, err := os.ReadFile(seedFile)
+	if errors.Is(err, os.ErrNotExist) {
+		return nil
+	}
+	if err != nil {
+		return fmt.Errorf("read seed file: %w", err)
+	}
+
+	var seed Seed
+	if err = json.Unmarshal(data, &seed); err != nil {
+		return fmt.Errorf("parse seed file: %w", err)
+	}
+	slog.Info("Applying Corrosion migration seed.",
+		"cluster_rows", len(seed.Cluster), "machine_rows", len(seed.Machines))
+
+	for _, e := range seed.Cluster {
+		if err = st.Put(ctx, e.Key, e.Value); err != nil {
+			return fmt.Errorf("seed cluster row '%s': %w", e.Key, err)
+		}
+	}
+
+	for _, e := range seed.Machines {
+		var m pb.MachineInfo
+		if err = protojson.Unmarshal([]byte(e.Info), &m); err != nil {
+			return fmt.Errorf("parse machine '%s' info: %w", e.ID, err)
+		}
+		if _, err = st.GetMachine(ctx, m.Id); err == nil {
+			continue
+		} else if !errors.Is(err, store.ErrMachineNotFound) {
+			return fmt.Errorf("check machine '%s': %w", m.Id, err)
+		}
+		if err = st.CreateMachine(ctx, &m); err != nil {
+			return fmt.Errorf("seed machine '%s': %w", m.Id, err)
+		}
+	}
+
+	if err = os.Remove(seedFile); err != nil {
+		return fmt.Errorf("delete seed file: %w", err)
+	}
+	slog.Info("Corrosion migration completed.")
+	return nil
+}