Skip to content

Commit 71da325

Browse files
committed
refactor: migrate to run Corrosion service as managed container instead of systemd unit, automatically migrate data to v1
1 parent 9236aa3 commit 71da325

8 files changed

Lines changed: 599 additions & 185 deletions

File tree

‎internal/machine/cluster.go‎

Lines changed: 11 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -15,6 +15,7 @@ import (
1515
"github.com/psviderski/uncloud/internal/machine/api/pb"
1616
"github.com/psviderski/uncloud/internal/machine/caddyconfig"
1717
"github.com/psviderski/uncloud/internal/machine/constants"
18+
"github.com/psviderski/uncloud/internal/machine/corromigrate"
1819
"github.com/psviderski/uncloud/internal/machine/corroservice"
1920
"github.com/psviderski/uncloud/internal/machine/dns"
2021
"github.com/psviderski/uncloud/internal/machine/docker"
@@ -38,6 +39,9 @@ type clusterController struct {
3839

3940
server *grpc.Server
4041
corroService corroservice.Service
42+
// corrosionDir is the disk path that holds the Corrosion config and data.
43+
// TODO: remove in 0.22 assuming all pre 0.20 clusters upgraded their pre-v1 Corrosion.
44+
corrosionDir string
4145
dockerCtrl *docker.Controller
4246
// dockerReady is signalled when Docker is configured and ready for containers.
4347
dockerReady chan<- struct{}
@@ -60,6 +64,7 @@ func newClusterController(
6064
store *store.Store,
6165
server *grpc.Server,
6266
corroService corroservice.Service,
67+
corrosionDir string,
6368
dockerService *docker.Service,
6469
dockerReady chan<- struct{},
6570
clusterReady chan<- struct{},
@@ -82,6 +87,7 @@ func newClusterController(
8287
endpointChanges: endpointChanges,
8388
server: server,
8489
corroService: corroService,
90+
corrosionDir: corrosionDir,
8591
dockerCtrl: docker.NewController(state.ID, dockerService, store),
8692
dockerReady: dockerReady,
8793
clusterReady: clusterReady,
@@ -126,6 +132,11 @@ func (cc *clusterController) Run(ctx context.Context) error {
126132
slog.Info("Corrosion service started.")
127133
}
128134

135+
// Apply the seed to finish Corrosion migrations from 0.x to 2026.5.14 (upstream v1.0.0) if applicable.
136+
if err := corromigrate.ApplySeedIfPresent(ctx, cc.corrosionDir, cc.store); err != nil {
137+
return fmt.Errorf("apply corrosion migration seed: %w", err)
138+
}
139+
129140
errGroup, ctx := errgroup.WithContext(ctx)
130141

131142
// Start the WireGuard control loop before waiting for store sync. This ensures endpoint rotation happens
Lines changed: 282 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,282 @@
1+
// Package corromigrate handles the one-time migration of the on-disk Corrosion store from
2+
// systemd-managed v0.x to v1.0.0 running in a uncloudd-managed container.
3+
// TODO: remove in 0.22 assuming all pre 0.20 clusters upgraded their pre-v1 Corrosion.
4+
package corromigrate
5+
6+
import (
7+
"context"
8+
"database/sql"
9+
"encoding/json"
10+
"errors"
11+
"fmt"
12+
"log/slog"
13+
"os"
14+
"os/exec"
15+
"path/filepath"
16+
"time"
17+
18+
"github.com/psviderski/uncloud/internal/fs"
19+
"github.com/psviderski/uncloud/internal/machine/api/pb"
20+
"github.com/psviderski/uncloud/internal/machine/store"
21+
"google.golang.org/protobuf/encoding/protojson"
22+
_ "modernc.org/sqlite"
23+
)
24+
25+
// legacyUnitName is the systemd unit that ran Corrosion before uncloudd took over its lifecycle.
26+
// The unit file at /etc/systemd/system/uncloud-corrosion.service and the binary at
27+
// /usr/local/bin/uncloud-corrosion are left in place after migration because uncloudd's sandbox
28+
// (ProtectSystem=full) blocks their removal. Operators should delete them manually as documented
29+
// in the release notes.
30+
const legacyUnitName = "uncloud-corrosion.service"
31+
32+
// Seed is the on-disk representation of the durable rows dumped from a v0.x Corrosion store,
33+
// to be re-applied to the fresh v1.0.0 store. Existence of the seed file is the signal that
34+
// migration has not yet fully completed.
35+
type Seed struct {
36+
Cluster []ClusterEntry `json:"cluster"`
37+
Machines []MachineEntry `json:"machines"`
38+
}
39+
40+
type ClusterEntry struct {
41+
Key string `json:"key"`
42+
Value string `json:"value"`
43+
}
44+
45+
type MachineEntry struct {
46+
ID string `json:"id"`
47+
Info string `json:"info"`
48+
}
49+
50+
func seedPath(dir string) string { return dir + ".seed-v1.json" }
51+
52+
func backupPath(dir string) string {
53+
return dir + ".backup-" + time.Now().UTC().Format("20060102.150405")
54+
}
55+
56+
// MigrateIfNeeded stops the legacy systemd Corrosion unit and, when a v0.x store.db is detected, dumps its durable
57+
// rows to a seed file next to the data dir, then backs up the dir so the new version can start fresh.
58+
// Idempotent: if the seed file already exists, the dump step is skipped (a prior run already produced it).
59+
func MigrateIfNeeded(ctx context.Context, dir, owner string) error {
60+
stopLegacyUnit()
61+
62+
seedFile := seedPath(dir)
63+
if _, err := os.Stat(seedFile); err == nil {
64+
slog.Info("Corrosion migrations seed file found, dump already produced by a prior run.",
65+
"path", seedFile)
66+
return nil
67+
} else if !errors.Is(err, os.ErrNotExist) {
68+
return fmt.Errorf("stat seed file '%s': %w", seedFile, err)
69+
}
70+
71+
dbFile := filepath.Join(dir, "store.db")
72+
if _, err := os.Stat(dbFile); errors.Is(err, os.ErrNotExist) {
73+
return nil
74+
} else if err != nil {
75+
return fmt.Errorf("stat '%s': %w", dbFile, err)
76+
}
77+
78+
isV0, err := isV0Store(ctx, dbFile)
79+
if err != nil {
80+
return fmt.Errorf("detect old store version: %w", err)
81+
}
82+
if !isV0 {
83+
return nil
84+
}
85+
86+
slog.Info("Migrating Corrosion store from v0.x to v1.0.0.", "db", dbFile)
87+
88+
dump, err := dumpOldStore(ctx, dbFile)
89+
if err != nil {
90+
return fmt.Errorf("dump old store: %w", err)
91+
}
92+
slog.Info("Corrosion store dumped.",
93+
"cluster_rows", len(dump.Cluster), "machine_rows", len(dump.Machines))
94+
95+
if err = writeSeedAtomic(seedFile, dump, owner); err != nil {
96+
return fmt.Errorf("write seed file: %w", err)
97+
}
98+
slog.Info("Corrosion migration seed file written.", "path", seedFile)
99+
100+
backup := backupPath(dir)
101+
if err = os.Rename(dir, backup); err != nil {
102+
return fmt.Errorf("backup corrosion dir: %w", err)
103+
}
104+
slog.Info("Old Corrosion dir backed up.", "from", dir, "to", backup)
105+
106+
return nil
107+
}
108+
109+
// stopLegacyUnit stops and disables the legacy uncloud-corrosion systemd unit if present.
110+
// Best-effort: errors are swallowed because the unit may not exist (fresh install) or systemctl
111+
// may be unavailable (containerised hosts).
112+
func stopLegacyUnit() {
113+
for _, cmd := range []string{"stop", "disable"} {
114+
if err := exec.Command("systemctl", cmd, legacyUnitName).Run(); err != nil {
115+
slog.Debug("systemctl on legacy corrosion unit failed (likely absent).",
116+
"cmd", cmd, "unit", legacyUnitName, "err", err)
117+
}
118+
}
119+
}
120+
121+
// isV0Store reports whether the SQLite database at dbPath is in the old v0.x Corrosion format.
122+
// v0.x has a __corro_bookkeeping table that was dropped in v1.0.0.
123+
func isV0Store(ctx context.Context, dbPath string) (bool, error) {
124+
db, err := sql.Open("sqlite", "file:"+dbPath+"?mode=ro")
125+
if err != nil {
126+
return false, fmt.Errorf("open db '%s': %w", dbPath, err)
127+
}
128+
defer db.Close()
129+
130+
var one int
131+
err = db.QueryRowContext(ctx,
132+
"SELECT 1 FROM sqlite_master WHERE type='table' AND name='__corro_bookkeeping' LIMIT 1",
133+
).Scan(&one)
134+
if errors.Is(err, sql.ErrNoRows) {
135+
return false, nil
136+
}
137+
if err != nil {
138+
return false, err
139+
}
140+
return true, nil
141+
}
142+
143+
func dumpOldStore(ctx context.Context, dbPath string) (*Seed, error) {
144+
db, err := sql.Open("sqlite", "file:"+dbPath+"?mode=ro")
145+
if err != nil {
146+
return nil, fmt.Errorf("open %q: %w", dbPath, err)
147+
}
148+
defer db.Close()
149+
150+
seed := &Seed{}
151+
152+
clusterRows, err := db.QueryContext(ctx, "SELECT key, value FROM cluster")
153+
if err != nil {
154+
return nil, fmt.Errorf("query cluster: %w", err)
155+
}
156+
for clusterRows.Next() {
157+
var key, value string
158+
if err = clusterRows.Scan(&key, &value); err != nil {
159+
clusterRows.Close()
160+
return nil, fmt.Errorf("scan cluster row: %w", err)
161+
}
162+
seed.Cluster = append(seed.Cluster, ClusterEntry{Key: key, Value: value})
163+
}
164+
if err = clusterRows.Err(); err != nil {
165+
clusterRows.Close()
166+
return nil, fmt.Errorf("iterate cluster rows: %w", err)
167+
}
168+
clusterRows.Close()
169+
170+
machineRows, err := db.QueryContext(ctx, "SELECT id, info FROM machines")
171+
if err != nil {
172+
return nil, fmt.Errorf("query machines: %w", err)
173+
}
174+
defer machineRows.Close()
175+
for machineRows.Next() {
176+
var id, info string
177+
if err = machineRows.Scan(&id, &info); err != nil {
178+
return nil, fmt.Errorf("scan machine row: %w", err)
179+
}
180+
seed.Machines = append(seed.Machines, MachineEntry{ID: id, Info: info})
181+
}
182+
if err = machineRows.Err(); err != nil {
183+
return nil, fmt.Errorf("iterate machine rows: %w", err)
184+
}
185+
return seed, nil
186+
}
187+
188+
func writeSeedAtomic(path string, seed *Seed, owner string) error {
189+
tmp := path + ".tmp"
190+
data, err := json.MarshalIndent(seed, "", " ")
191+
if err != nil {
192+
return err
193+
}
194+
195+
f, err := os.OpenFile(tmp, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0o600)
196+
if err != nil {
197+
return err
198+
}
199+
if _, err = f.Write(data); err != nil {
200+
f.Close()
201+
os.Remove(tmp)
202+
return fmt.Errorf("write seed: %w", err)
203+
}
204+
if err = f.Sync(); err != nil {
205+
f.Close()
206+
os.Remove(tmp)
207+
return fmt.Errorf("fsync seed: %w", err)
208+
}
209+
if err = f.Close(); err != nil {
210+
os.Remove(tmp)
211+
return fmt.Errorf("close seed: %w", err)
212+
}
213+
214+
if owner != "" {
215+
if chErr := fs.Chown(tmp, owner, owner); chErr != nil {
216+
os.Remove(tmp)
217+
return chErr
218+
}
219+
}
220+
if err = os.Rename(tmp, path); err != nil {
221+
return fmt.Errorf("rename seed: %w", err)
222+
}
223+
224+
// Fsync the parent directory so the rename entry is durable.
225+
dir, err := os.Open(filepath.Dir(path))
226+
if err != nil {
227+
return fmt.Errorf("open parent dir for fsync: %w", err)
228+
}
229+
defer dir.Close()
230+
if err = dir.Sync(); err != nil {
231+
return fmt.Errorf("fsync parent dir: %w", err)
232+
}
233+
return nil
234+
}
235+
236+
// ApplySeedIfPresent re-applies the durable rows from the seed file into the running Corrosion store,
237+
// then deletes the seed file as the final completion marker. Idempotent: re-running on the same seed
238+
// is a no-op for cluster rows (INSERT OR REPLACE) and machines rows (skipped via GetMachine).
239+
func ApplySeedIfPresent(ctx context.Context, dir string, st *store.Store) error {
240+
seedFile := seedPath(dir)
241+
data, err := os.ReadFile(seedFile)
242+
if errors.Is(err, os.ErrNotExist) {
243+
return nil
244+
}
245+
if err != nil {
246+
return fmt.Errorf("read seed file: %w", err)
247+
}
248+
249+
var seed Seed
250+
if err = json.Unmarshal(data, &seed); err != nil {
251+
return fmt.Errorf("parse seed file: %w", err)
252+
}
253+
slog.Info("Applying Corrosion migration seed.",
254+
"cluster_rows", len(seed.Cluster), "machine_rows", len(seed.Machines))
255+
256+
for _, e := range seed.Cluster {
257+
if err = st.Put(ctx, e.Key, e.Value); err != nil {
258+
return fmt.Errorf("seed cluster row '%s': %w", e.Key, err)
259+
}
260+
}
261+
262+
for _, e := range seed.Machines {
263+
var m pb.MachineInfo
264+
if err = protojson.Unmarshal([]byte(e.Info), &m); err != nil {
265+
return fmt.Errorf("parse machine '%s' info: %w", e.ID, err)
266+
}
267+
if _, err = st.GetMachine(ctx, m.Id); err == nil {
268+
continue
269+
} else if !errors.Is(err, store.ErrMachineNotFound) {
270+
return fmt.Errorf("check machine '%s': %w", m.Id, err)
271+
}
272+
if err = st.CreateMachine(ctx, &m); err != nil {
273+
return fmt.Errorf("seed machine '%s': %w", m.Id, err)
274+
}
275+
}
276+
277+
if err = os.Remove(seedFile); err != nil {
278+
return fmt.Errorf("delete seed file: %w", err)
279+
}
280+
slog.Info("Corrosion migration completed.")
281+
return nil
282+
}

0 commit comments

Comments
 (0)