Serge Bazanski | 42e61c6 | 2021-03-18 15:07:18 +0100 | [diff] [blame] | 1 | // Copyright 2020 The Monogon Project Authors. |
| 2 | // |
| 3 | // SPDX-License-Identifier: Apache-2.0 |
| 4 | // |
| 5 | // Licensed under the Apache License, Version 2.0 (the "License"); |
| 6 | // you may not use this file except in compliance with the License. |
| 7 | // You may obtain a copy of the License at |
| 8 | // |
| 9 | // http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | // |
| 11 | // Unless required by applicable law or agreed to in writing, software |
| 12 | // distributed under the License is distributed on an "AS IS" BASIS, |
| 13 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | // See the License for the specific language governing permissions and |
| 15 | // limitations under the License. |
| 16 | |
Serge Bazanski | 37110c3 | 2023-03-01 13:57:27 +0000 | [diff] [blame] | 17 | // Package cluster implements low-level clustering logic, especially logic |
| 18 | // regarding to bootstrapping, registering into and joining a cluster. Its goal |
| 19 | // is to provide the rest of the node code with the following: |
| 20 | // - A mounted plaintext storage. |
| 21 | // - Node credentials/identity. |
| 22 | // - A locally running etcd server if the node is supposed to run one, and a |
| 23 | // client connection to that etcd cluster if so. |
| 24 | // - The state of the cluster as seen by the node, to enable code to respond to |
| 25 | // node lifecycle changes. |
Serge Bazanski | 42e61c6 | 2021-03-18 15:07:18 +0100 | [diff] [blame] | 26 | package cluster |
| 27 | |
| 28 | import ( |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 29 | "context" |
| 30 | "errors" |
Serge Bazanski | 42e61c6 | 2021-03-18 15:07:18 +0100 | [diff] [blame] | 31 | "fmt" |
Mateusz Zalega | 2930e99 | 2022-04-25 12:52:35 +0200 | [diff] [blame] | 32 | "strings" |
Serge Bazanski | 42e61c6 | 2021-03-18 15:07:18 +0100 | [diff] [blame] | 33 | |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 34 | "source.monogon.dev/metropolis/node/core/localstorage" |
| 35 | "source.monogon.dev/metropolis/node/core/network" |
Serge Bazanski | 6dff6d6 | 2022-01-28 18:15:14 +0100 | [diff] [blame] | 36 | "source.monogon.dev/metropolis/node/core/roleserve" |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 37 | "source.monogon.dev/metropolis/pkg/supervisor" |
| 38 | apb "source.monogon.dev/metropolis/proto/api" |
Mateusz Zalega | 2930e99 | 2022-04-25 12:52:35 +0200 | [diff] [blame] | 39 | cpb "source.monogon.dev/metropolis/proto/common" |
Serge Bazanski | 42e61c6 | 2021-03-18 15:07:18 +0100 | [diff] [blame] | 40 | ) |
| 41 | |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 42 | type Manager struct { |
| 43 | storageRoot *localstorage.Root |
| 44 | networkService *network.Service |
Serge Bazanski | 6dff6d6 | 2022-01-28 18:15:14 +0100 | [diff] [blame] | 45 | roleServer *roleserve.Service |
Lorenz Brun | 85ad26a | 2023-03-27 17:00:00 +0200 | [diff] [blame] | 46 | nodeParams *apb.NodeParameters |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 47 | |
Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 48 | oneway chan struct{} |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 49 | } |
| 50 | |
| 51 | // NewManager creates a new cluster Manager. The given localstorage Root must |
| 52 | // be places, but not yet started (and will be started as the Manager makes |
| 53 | // progress). The given network Service must already be running. |
Lorenz Brun | 85ad26a | 2023-03-27 17:00:00 +0200 | [diff] [blame] | 54 | func NewManager(storageRoot *localstorage.Root, networkService *network.Service, rs *roleserve.Service, nodeParams *apb.NodeParameters) *Manager { |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 55 | return &Manager{ |
| 56 | storageRoot: storageRoot, |
| 57 | networkService: networkService, |
Serge Bazanski | 6dff6d6 | 2022-01-28 18:15:14 +0100 | [diff] [blame] | 58 | roleServer: rs, |
Lorenz Brun | 85ad26a | 2023-03-27 17:00:00 +0200 | [diff] [blame] | 59 | nodeParams: nodeParams, |
Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 60 | oneway: make(chan struct{}), |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 61 | } |
| 62 | } |
| 63 | |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 64 | // Run is the runnable of the Manager, to be started using the Supervisor. It |
| 65 | // is one-shot, and should not be restarted. |
| 66 | func (m *Manager) Run(ctx context.Context) error { |
Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 67 | select { |
| 68 | case <-m.oneway: |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 69 | return fmt.Errorf("cannot restart cluster manager") |
Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 70 | default: |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 71 | } |
Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 72 | close(m.oneway) |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 73 | |
Lorenz Brun | 6c35e97 | 2021-12-14 03:08:23 +0100 | [diff] [blame] | 74 | configuration, err := m.storageRoot.ESP.Metropolis.SealedConfiguration.Unseal() |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 75 | if err == nil { |
| 76 | supervisor.Logger(ctx).Info("Sealed configuration present. attempting to join cluster") |
Mateusz Zalega | 2930e99 | 2022-04-25 12:52:35 +0200 | [diff] [blame] | 77 | |
| 78 | // Read Cluster Directory and unmarshal it. Since the node is already |
| 79 | // registered with the cluster, the directory won't be bootstrapped from |
| 80 | // Node Parameters. |
| 81 | cd, err := m.storageRoot.ESP.Metropolis.ClusterDirectory.Unmarshal() |
| 82 | if err != nil { |
| 83 | return fmt.Errorf("while reading cluster directory: %w", err) |
| 84 | } |
| 85 | return m.join(ctx, configuration, cd) |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 86 | } |
| 87 | |
| 88 | if !errors.Is(err, localstorage.ErrNoSealed) { |
| 89 | return fmt.Errorf("unexpected sealed config error: %w", err) |
| 90 | } |
| 91 | |
| 92 | supervisor.Logger(ctx).Info("No sealed configuration, looking for node parameters") |
| 93 | |
Lorenz Brun | 85ad26a | 2023-03-27 17:00:00 +0200 | [diff] [blame] | 94 | switch inner := m.nodeParams.Cluster.(type) { |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 95 | case *apb.NodeParameters_ClusterBootstrap_: |
Serge Bazanski | 5839e97 | 2021-11-16 15:46:19 +0100 | [diff] [blame] | 96 | err = m.bootstrap(ctx, inner.ClusterBootstrap) |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 97 | case *apb.NodeParameters_ClusterRegister_: |
Serge Bazanski | 5839e97 | 2021-11-16 15:46:19 +0100 | [diff] [blame] | 98 | err = m.register(ctx, inner.ClusterRegister) |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 99 | default: |
Serge Bazanski | 5839e97 | 2021-11-16 15:46:19 +0100 | [diff] [blame] | 100 | 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] | 101 | } |
Serge Bazanski | 5839e97 | 2021-11-16 15:46:19 +0100 | [diff] [blame] | 102 | |
| 103 | if err == nil { |
| 104 | supervisor.Logger(ctx).Info("Cluster enrolment done.") |
Serge Bazanski | fe5192d | 2023-03-16 11:33:56 +0100 | [diff] [blame] | 105 | return nil |
Serge Bazanski | 5839e97 | 2021-11-16 15:46:19 +0100 | [diff] [blame] | 106 | } |
| 107 | return err |
Serge Bazanski | a959cbd | 2021-06-17 15:56:51 +0200 | [diff] [blame] | 108 | } |
| 109 | |
Mateusz Zalega | 2930e99 | 2022-04-25 12:52:35 +0200 | [diff] [blame] | 110 | // logClusterDirectory verbosely logs the whole Cluster Directory passed to it. |
| 111 | func logClusterDirectory(ctx context.Context, cd *cpb.ClusterDirectory) { |
| 112 | for _, node := range cd.Nodes { |
Mateusz Zalega | 2930e99 | 2022-04-25 12:52:35 +0200 | [diff] [blame] | 113 | var addresses []string |
| 114 | for _, add := range node.Addresses { |
| 115 | addresses = append(addresses, add.Host) |
| 116 | } |
Mateusz Zalega | de82150 | 2022-04-29 16:37:17 +0200 | [diff] [blame] | 117 | supervisor.Logger(ctx).Infof(" Addresses: %s", strings.Join(addresses, ",")) |
Mateusz Zalega | 2930e99 | 2022-04-25 12:52:35 +0200 | [diff] [blame] | 118 | } |
| 119 | } |