| Tim Windelschmidt | 6d33a43 | 2025-02-04 14:34:25 +0100 | [diff] [blame^] | 1 | // Copyright The Monogon Project Authors. |
| Serge Bazanski | 42e61c6 | 2021-03-18 15:07:18 +0100 | [diff] [blame] | 2 | // SPDX-License-Identifier: Apache-2.0 |
| Serge Bazanski | 42e61c6 | 2021-03-18 15:07:18 +0100 | [diff] [blame] | 3 | |
| Serge Bazanski | 37110c3 | 2023-03-01 13:57:27 +0000 | [diff] [blame] | 4 | // Package cluster implements low-level clustering logic, especially logic |
| 5 | // regarding to bootstrapping, registering into and joining a cluster. Its goal |
| 6 | // is to provide the rest of the node code with the following: |
| 7 | // - A mounted plaintext storage. |
| 8 | // - Node credentials/identity. |
| 9 | // - A locally running etcd server if the node is supposed to run one, and a |
| 10 | // client connection to that etcd cluster if so. |
| 11 | // - The state of the cluster as seen by the node, to enable code to respond to |
| 12 | // node lifecycle changes. |
| Serge Bazanski | 42e61c6 | 2021-03-18 15:07:18 +0100 | [diff] [blame] | 13 | package cluster |
| 14 | |
| 15 | import ( |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 16 | "context" |
| 17 | "errors" |
| Serge Bazanski | 42e61c6 | 2021-03-18 15:07:18 +0100 | [diff] [blame] | 18 | "fmt" |
| 19 | |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 20 | "source.monogon.dev/metropolis/node/core/localstorage" |
| 21 | "source.monogon.dev/metropolis/node/core/network" |
| Serge Bazanski | 6dff6d6 | 2022-01-28 18:15:14 +0100 | [diff] [blame] | 22 | "source.monogon.dev/metropolis/node/core/roleserve" |
| Lorenz Brun | 35fcf03 | 2023-06-29 04:15:58 +0200 | [diff] [blame] | 23 | "source.monogon.dev/metropolis/node/core/update" |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 24 | apb "source.monogon.dev/metropolis/proto/api" |
| Mateusz Zalega | 2930e99 | 2022-04-25 12:52:35 +0200 | [diff] [blame] | 25 | cpb "source.monogon.dev/metropolis/proto/common" |
| Tim Windelschmidt | 9f21f53 | 2024-05-07 15:14:20 +0200 | [diff] [blame] | 26 | "source.monogon.dev/osbase/supervisor" |
| Serge Bazanski | 42e61c6 | 2021-03-18 15:07:18 +0100 | [diff] [blame] | 27 | ) |
| 28 | |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 29 | type Manager struct { |
| 30 | storageRoot *localstorage.Root |
| 31 | networkService *network.Service |
| Serge Bazanski | 6dff6d6 | 2022-01-28 18:15:14 +0100 | [diff] [blame] | 32 | roleServer *roleserve.Service |
| Lorenz Brun | 35fcf03 | 2023-06-29 04:15:58 +0200 | [diff] [blame] | 33 | updateService *update.Service |
| Lorenz Brun | 85ad26a | 2023-03-27 17:00:00 +0200 | [diff] [blame] | 34 | nodeParams *apb.NodeParameters |
| Serge Bazanski | 5df62ba | 2023-03-22 17:56:46 +0100 | [diff] [blame] | 35 | haveTPM bool |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 36 | |
| Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 37 | oneway chan struct{} |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 38 | } |
| 39 | |
| 40 | // NewManager creates a new cluster Manager. The given localstorage Root must |
| 41 | // be places, but not yet started (and will be started as the Manager makes |
| 42 | // progress). The given network Service must already be running. |
| Lorenz Brun | 35fcf03 | 2023-06-29 04:15:58 +0200 | [diff] [blame] | 43 | func NewManager(storageRoot *localstorage.Root, networkService *network.Service, rs *roleserve.Service, updateService *update.Service, nodeParams *apb.NodeParameters, haveTPM bool) *Manager { |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 44 | return &Manager{ |
| 45 | storageRoot: storageRoot, |
| 46 | networkService: networkService, |
| Serge Bazanski | 6dff6d6 | 2022-01-28 18:15:14 +0100 | [diff] [blame] | 47 | roleServer: rs, |
| Lorenz Brun | 35fcf03 | 2023-06-29 04:15:58 +0200 | [diff] [blame] | 48 | updateService: updateService, |
| Lorenz Brun | 85ad26a | 2023-03-27 17:00:00 +0200 | [diff] [blame] | 49 | nodeParams: nodeParams, |
| Serge Bazanski | 5df62ba | 2023-03-22 17:56:46 +0100 | [diff] [blame] | 50 | haveTPM: haveTPM, |
| Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 51 | oneway: make(chan struct{}), |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 52 | } |
| 53 | } |
| 54 | |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 55 | // Run is the runnable of the Manager, to be started using the Supervisor. It |
| 56 | // is one-shot, and should not be restarted. |
| 57 | func (m *Manager) Run(ctx context.Context) error { |
| Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 58 | select { |
| 59 | case <-m.oneway: |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 60 | return fmt.Errorf("cannot restart cluster manager") |
| Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 61 | default: |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 62 | } |
| Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 63 | close(m.oneway) |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 64 | |
| Serge Bazanski | e4a4ce1 | 2023-03-22 18:29:54 +0100 | [diff] [blame] | 65 | // Try sealed configuration first. |
| Tim Windelschmidt | a10d0cb | 2025-01-13 14:44:15 +0100 | [diff] [blame] | 66 | configuration, err := m.storageRoot.ESP.Metropolis.SealedConfiguration.Unseal(cpb.NodeTPMUsage_NODE_TPM_USAGE_PRESENT_AND_USED) |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 67 | if err == nil { |
| 68 | supervisor.Logger(ctx).Info("Sealed configuration present. attempting to join cluster") |
| Mateusz Zalega | 2930e99 | 2022-04-25 12:52:35 +0200 | [diff] [blame] | 69 | |
| 70 | // Read Cluster Directory and unmarshal it. Since the node is already |
| 71 | // registered with the cluster, the directory won't be bootstrapped from |
| 72 | // Node Parameters. |
| 73 | cd, err := m.storageRoot.ESP.Metropolis.ClusterDirectory.Unmarshal() |
| 74 | if err != nil { |
| 75 | return fmt.Errorf("while reading cluster directory: %w", err) |
| 76 | } |
| Serge Bazanski | e4a4ce1 | 2023-03-22 18:29:54 +0100 | [diff] [blame] | 77 | return m.join(ctx, configuration, cd, true) |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 78 | } |
| 79 | |
| Serge Bazanski | 98054a1 | 2023-06-14 18:16:21 +0200 | [diff] [blame] | 80 | if !errors.Is(err, localstorage.ErrNoSealed) && !errors.Is(err, localstorage.ErrSealedCorrupted) { |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 81 | return fmt.Errorf("unexpected sealed config error: %w", err) |
| 82 | } |
| 83 | |
| Tim Windelschmidt | a10d0cb | 2025-01-13 14:44:15 +0100 | [diff] [blame] | 84 | configuration, err = m.storageRoot.ESP.Metropolis.SealedConfiguration.Unseal(cpb.NodeTPMUsage_NODE_TPM_USAGE_NOT_PRESENT) |
| Serge Bazanski | e4a4ce1 | 2023-03-22 18:29:54 +0100 | [diff] [blame] | 85 | if err == nil { |
| 86 | supervisor.Logger(ctx).Info("Non-sealed configuration present. attempting to join cluster") |
| 87 | |
| 88 | // Read Cluster Directory and unmarshal it. Since the node is already |
| 89 | // registered with the cluster, the directory won't be bootstrapped from |
| 90 | // Node Parameters. |
| 91 | cd, err := m.storageRoot.ESP.Metropolis.ClusterDirectory.Unmarshal() |
| 92 | if err != nil { |
| 93 | return fmt.Errorf("while reading cluster directory: %w", err) |
| 94 | } |
| 95 | return m.join(ctx, configuration, cd, false) |
| 96 | } |
| 97 | |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 98 | supervisor.Logger(ctx).Info("No sealed configuration, looking for node parameters") |
| 99 | |
| Lorenz Brun | 85ad26a | 2023-03-27 17:00:00 +0200 | [diff] [blame] | 100 | switch inner := m.nodeParams.Cluster.(type) { |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 101 | case *apb.NodeParameters_ClusterBootstrap_: |
| Serge Bazanski | 5839e97 | 2021-11-16 15:46:19 +0100 | [diff] [blame] | 102 | err = m.bootstrap(ctx, inner.ClusterBootstrap) |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 103 | case *apb.NodeParameters_ClusterRegister_: |
| Serge Bazanski | 5839e97 | 2021-11-16 15:46:19 +0100 | [diff] [blame] | 104 | err = m.register(ctx, inner.ClusterRegister) |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 105 | default: |
| Serge Bazanski | 5839e97 | 2021-11-16 15:46:19 +0100 | [diff] [blame] | 106 | err = fmt.Errorf("node parameters misconfigured: neither cluster_bootstrap nor cluster_register set") |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 107 | } |
| Serge Bazanski | 5839e97 | 2021-11-16 15:46:19 +0100 | [diff] [blame] | 108 | |
| 109 | if err == nil { |
| 110 | supervisor.Logger(ctx).Info("Cluster enrolment done.") |
| Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 111 | return nil |
| Serge Bazanski | 5839e97 | 2021-11-16 15:46:19 +0100 | [diff] [blame] | 112 | } |
| 113 | return err |
| Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 114 | } |