179b2d8fbd
- sshmanager/deploy.go: initialize DeployResult with empty slices instead of nil to prevent null serialization in JSON - Machines.tsx: use ?? [] fallback for messages and errors arrays in deploy keys modal to handle null/undefined gracefully
5.6 KiB
5.6 KiB
Revertir RunRemote a approach SSH-a-source sin envío base64
Estado
Estamos en plan mode (READ-ONLY). Los cambios de código Go están bloqueados. Este documento describe los cambios necesarios para proceder cuando salgamos de plan mode.
Contexto
Las claves SSH ya están pre-instaladas en cada máquina:
| Máquina | Clave instalada | Pública autorizada en |
|---|---|---|
| Baby NAS (10.5.1.20) | /var/lib/syncserver/ssh/keys/qnap.key (OpenSSH nativo, 387 bytes) |
qnap.key.pub ya en Qnap /mnt/HDA_ROOT/.config/ssh/authorized_keys |
| Qnap (10.5.0.144) | n/a (server) | baby-nas.key.pub ya en Baby NAS /root/.ssh/authorized_keys |
| Baby NAS known_hosts | /var/lib/syncserver/ssh/known_hosts (3 host keys de Qnap en formato hashed) |
— |
Test crítico exitoso:
ssh -i /var/lib/syncserver/ssh/keys/baby-nas.key root@10.5.1.20 \
"ssh -i /var/lib/syncserver/ssh/keys/qnap.key -o StrictHostKeyChecking=yes \
admin@10.5.0.144 echo QNAP-FROM-BABY"
→ QNAP-FROM-BABY (exit 0)
Cambios pendientes
1. internal/syncengine/rsync_runner.go
Eliminar la función buildWrapperScript (líneas 86-101 del archivo actual).
Reemplazar la función RunRemote actual con la versión SSH-a-source:
func (r *RsyncRunner) RunRemote(ctx context.Context, pair *SyncPairConfig, src *MachineKeys, dst *MachineKeys, destKey string, onLine func(stream string, line string)) (*RsyncResult, error) {
if src.Port == 0 {
src.Port = 22
}
if src.PrivKey == "" {
src.PrivKey = filepath.Join(r.sshDir, "id_ed25519")
}
if destKey == "" {
destKey = filepath.Join(r.sshDir, "id_ed25519")
}
destUserHost := fmt.Sprintf("%s@%s", dst.SSHUser, dst.Host)
args := r.buildArgs(pair)
innerSSH := fmt.Sprintf("ssh -i %s -o StrictHostKeyChecking=accept-new -o UserKnownHostsFile=%s",
destKey, filepath.Join(r.sshDir, "known_hosts"))
rsyncFlags := strings.Join(args[:len(args)-2], " ")
sourcePath := args[len(args)-2]
destPath := args[len(args)-1]
remoteCmd := fmt.Sprintf("rsync %s -e %q %s %s",
rsyncFlags, innerSSH, sourcePath, destUserHost+":"+destPath)
sshArgs := []string{
"-i", src.PrivKey,
"-o", "StrictHostKeyChecking=accept-new",
"-o", "UserKnownHostsFile=" + filepath.Join(r.sshDir, "known_hosts"),
"-p", fmt.Sprintf("%d", src.Port),
fmt.Sprintf("%s@%s", src.SSHUser, src.Host),
}
sshArgs = append(sshArgs, remoteCmd)
cmd := exec.CommandContext(ctx, "ssh", sshArgs...)
return r.runCmd(ctx, cmd, onLine)
}
Modificar el struct MachineKeys para agregar SSHUser:
type MachineKeys struct {
Host string
Port int
User string
SSHUser string
PrivKey string
}
2. internal/syncengine/engine.go
Actualizar la llamada a RunRemote (líneas 263-283 actuales) para usar la nueva firma con destPrivKeyPath como argumento separado:
if isRemoteToRemote(srcMachine, dstMachine) {
destPrivKeyPath, err := e.resolveSSHKey(dstMachine)
if err != nil {
slog.Warn("failed to resolve destination SSH key, using server key", "error", err)
destPrivKeyPath = ""
}
result, err = runner.RunRemote(jobCtx, cfg,
&syncengine.MachineKeys{
Host: srcMachine.Host,
Port: srcMachine.Port,
User: srcMachine.SSHUser,
SSHUser: srcMachine.SSHUser,
PrivKey: privKey,
},
&syncengine.MachineKeys{
Host: dstMachine.Host,
Port: dstMachine.Port,
User: dstMachine.SSHUser,
SSHUser: dstMachine.SSHUser,
PrivKey: destPrivKeyPath,
},
destPrivKeyPath,
onLine)
}
3. Build y deploy
go build ./cmd/server
make clean && make package
./scripts/bump-version.sh # bump a 1.0.28
git add -f cmd/server/main.go Makefile internal/syncengine/
git commit -m "fix: RunRemote uses SSH-to-source, dest key pre-installed on source"
git push
# Deploy manual:
scp dist/syncserver_1.0.28_amd64.deb root@10.5.1.30:/root/move-data-nas/dist/
ssh root@10.5.1.30 "cd /root/move-data-nas && dpkg -i dist/syncserver_1.0.28_amd64.deb"
4. Test
Trigger job 23 desde UI o via API. Verificar en DB:
ssh root@10.5.1.30 'sqlite3 /var/lib/syncserver/app.db \
"SELECT id, status, error_code, substr(error_message,1,200) FROM jobs WHERE id=24;"'
Esperado: status=success, error_code=NULL, error_message=NULL.
Por qué este approach funciona
Antes:
- LXC SSH a Baby NAS
- Baby NAS ejecuta
echo 'BASE64' | base64 -d > /tmp/syncserver-dest-key && rsync ... - El base64 round-trip generaba archivo corrupto o vacío
- Baby NAS OpenSSH 8.9 no podía parsear el formato PKCS8
Ahora:
- LXC SSH a Baby NAS (usa baby-nas.key)
- Baby NAS ejecuta
rsync -e ssh -i /var/lib/syncserver/ssh/keys/qnap.key ... - qnap.key YA está en Baby NAS en formato OpenSSH nativo (parseable por 8.9)
- Host key de Qnap YA está en known_hosts de Baby NAS (no prompt)
- Pública de qnap.key YA está en authorized_keys de Qnap (auth pasa)
Comportamiento esperado
job started job_id=24 pair=Peliculas
ssh -i baby-nas.key root@10.5.1.20 'rsync -aP -e "ssh -i qnap.key -o StrictHostKeyChecking=accept-new -o UserKnownHostsFile=..." /mnt/storage/multimedia/peliculas admin@10.5.0.144:/share/media/peliculas'
sending incremental file list
...
sent X bytes received Y bytes Z bytes/sec
total size is T speedup is S
job completed job_id=24 pair=Peliculas
Si falla
| Error | Causa | Fix |
|---|---|---|
| Load key invalid format | clave corrupta en Baby NAS | re-copiar qnap.key |
| Permission denied (publickey) | pública no está en Qnap | re-agregar a authorized_keys |
| Host key verification failed | known_hosts desincronizado | re-correr ssh-keyscan |
| Connection timed out | problema L2/L3 real (no era el caso) | investigar red |