blob: 890a30bbcecd9b7dbf145767a3afde3c6f08b101 [file] [log] [blame]
Tim Windelschmidt6d33a432025-02-04 14:34:25 +01001// Copyright The Monogon Project Authors.
Lorenz Brunae0d90d2019-09-05 17:53:56 +02002// SPDX-License-Identifier: Apache-2.0
Lorenz Brunae0d90d2019-09-05 17:53:56 +02003
4package main
5
6import (
Serge Bazanskicdb8c782020-02-17 12:34:02 +01007 "context"
Lorenz Brundd8c80e2019-10-07 16:19:49 +02008 "fmt"
Serge Bazanski3b098232023-03-16 17:57:02 +01009 "strings"
Lorenz Brunae0d90d2019-09-05 17:53:56 +020010
Lorenz Brunae0d90d2019-09-05 17:53:56 +020011 "golang.org/x/sys/unix"
Serge Bazanski99f47742021-08-04 20:21:42 +020012
Serge Bazanski3c5d0632024-09-12 10:49:12 +000013 "source.monogon.dev/go/logging"
Serge Bazanski31370b02021-01-07 16:31:14 +010014 "source.monogon.dev/metropolis/node/core/cluster"
Lorenz Brun6c454342023-06-01 12:23:38 +020015 "source.monogon.dev/metropolis/node/core/devmgr"
Serge Bazanski31370b02021-01-07 16:31:14 +010016 "source.monogon.dev/metropolis/node/core/localstorage"
17 "source.monogon.dev/metropolis/node/core/localstorage/declarative"
Serge Bazanskiefbde192024-07-31 14:53:20 +000018 "source.monogon.dev/metropolis/node/core/metrics"
Serge Bazanski31370b02021-01-07 16:31:14 +010019 "source.monogon.dev/metropolis/node/core/network"
Serge Bazanskif9edf522021-06-17 15:57:13 +020020 "source.monogon.dev/metropolis/node/core/roleserve"
Serge Bazanski58ddc092022-06-30 18:23:33 +020021 "source.monogon.dev/metropolis/node/core/rpc/resolver"
Serge Bazanski0d9e1252024-09-03 12:16:47 +020022 "source.monogon.dev/metropolis/node/core/tconsole"
Lorenz Brune306d782021-09-01 13:01:06 +020023 timesvc "source.monogon.dev/metropolis/node/core/time"
Lorenz Brun35fcf032023-06-29 04:15:58 +020024 "source.monogon.dev/metropolis/node/core/update"
Serge Bazanski8d64a3b2023-11-20 12:58:42 +010025 mversion "source.monogon.dev/metropolis/version"
Tim Windelschmidt7dac92b2024-12-16 02:51:04 +010026 "source.monogon.dev/osbase/bringup"
Tim Windelschmidt9f21f532024-05-07 15:14:20 +020027 "source.monogon.dev/osbase/logtree"
Jan Schär91bf1c82024-07-29 17:31:33 +020028 "source.monogon.dev/osbase/net/dns"
Tim Windelschmidt9f21f532024-05-07 15:14:20 +020029 "source.monogon.dev/osbase/supervisor"
Serge Bazanski154e6d92024-09-11 17:26:31 +020030 "source.monogon.dev/osbase/sysctl"
Tim Windelschmidt9f21f532024-05-07 15:14:20 +020031 "source.monogon.dev/osbase/tpm"
Serge Bazanski8d64a3b2023-11-20 12:58:42 +010032 "source.monogon.dev/version"
Lorenz Brunae0d90d2019-09-05 17:53:56 +020033)
34
35func main() {
Tim Windelschmidt7dac92b2024-12-16 02:51:04 +010036 bringup.Runnable(root).RunWith(bringup.Config{
37 Console: bringup.ConsoleConfig{
38 ShortenDictionary: logtree.MetropolisShortenDict,
39 Filter: consoleFilter,
Serge Bazanski3b098232023-03-16 17:57:02 +010040 },
Tim Windelschmidt7dac92b2024-12-16 02:51:04 +010041 Supervisor: bringup.SupervisorConfig{
42 Metrics: []supervisor.Metrics{
43 supervisor.NewMetricsPrometheus(metrics.CoreRegistry),
44 },
Lorenz Brunf0b22ff2023-05-02 16:04:20 +020045 },
Tim Windelschmidt7dac92b2024-12-16 02:51:04 +010046 })
Serge Bazanski57b43752020-07-13 19:17:48 +020047}
Serge Bazanski3b098232023-03-16 17:57:02 +010048
49// consoleFilter is used to filter out some uselessly verbose logs from the
50// console.
51//
52// This should be limited to external services, our internal services should
53// instead just have good logging by default.
54func consoleFilter(p *logtree.LogEntry) bool {
55 if p.Raw != nil {
56 return false
57 }
58 if p.Leveled == nil {
59 return false
60 }
61 s := string(p.DN)
62 if strings.HasPrefix(s, "root.role.controlplane.launcher.consensus.etcd") {
Serge Bazanski3c5d0632024-09-12 10:49:12 +000063 return p.Leveled.Severity().AtLeast(logging.WARNING)
Serge Bazanski3b098232023-03-16 17:57:02 +010064 }
65 // TODO(q3k): turn off RPC traces instead
66 if strings.HasPrefix(s, "root.role.controlplane.launcher.curator.listener.rpc") {
67 return false
68 }
69 if strings.HasPrefix(s, "root.role.kubernetes.run.kubernetes.networked.kubelet") {
Serge Bazanski3c5d0632024-09-12 10:49:12 +000070 return p.Leveled.Severity().AtLeast(logging.WARNING)
Serge Bazanski3b098232023-03-16 17:57:02 +010071 }
72 if strings.HasPrefix(s, "root.role.kubernetes.run.kubernetes.networked.apiserver") {
Serge Bazanski3c5d0632024-09-12 10:49:12 +000073 return p.Leveled.Severity().AtLeast(logging.WARNING)
Serge Bazanski3b098232023-03-16 17:57:02 +010074 }
75 if strings.HasPrefix(s, "root.role.kubernetes.run.kubernetes.controller-manager") {
Serge Bazanski3c5d0632024-09-12 10:49:12 +000076 return p.Leveled.Severity().AtLeast(logging.WARNING)
Serge Bazanski3b098232023-03-16 17:57:02 +010077 }
78 if strings.HasPrefix(s, "root.role.kubernetes.run.kubernetes.scheduler") {
Serge Bazanski3c5d0632024-09-12 10:49:12 +000079 return p.Leveled.Severity().AtLeast(logging.WARNING)
Serge Bazanski3b098232023-03-16 17:57:02 +010080 }
Lorenz Brun6eb3fb32023-08-09 17:19:24 +020081 if strings.HasPrefix(s, "root.kernel") {
82 // Linux writes high-severity logs directly to the console anyways and
83 // its low-severity logs are too verbose.
84 return false
85 }
Serge Bazanski6b7731e2023-03-22 17:58:04 +010086 if strings.HasPrefix(s, "supervisor") {
Serge Bazanski3c5d0632024-09-12 10:49:12 +000087 return p.Leveled.Severity().AtLeast(logging.WARNING)
Serge Bazanski6b7731e2023-03-22 17:58:04 +010088 }
Serge Bazanski3b098232023-03-16 17:57:02 +010089 return true
90}
91
Tim Windelschmidt7dac92b2024-12-16 02:51:04 +010092// Function which performs core, one-way initialization of the node. This means
93// waiting for the network, starting the cluster manager, and then starting all
94// services related to the node's roles.
95func root(ctx context.Context) error {
96 logger := supervisor.Logger(ctx)
97
98 logger.Info("Starting Metropolis node init")
99 logger.Infof("Version: %s", version.Semver(mversion.Version))
100
101 // Linux kernel default is 4096 which is far too low. Raise it to 1M which
102 // is what gVisor suggests.
103 if err := unix.Setrlimit(unix.RLIMIT_NOFILE, &unix.Rlimit{Cur: 1048576, Max: 1048576}); err != nil {
104 logger.Fatalf("Failed to raise rlimits: %v", err)
105 }
106
107 haveTPM := true
108 if err := tpm.Initialize(logger); err != nil {
109 logger.Warningf("Failed to initialize TPM 2.0: %v", err)
110 haveTPM = false
111 }
112
113 metrics.CoreRegistry.MustRegister(dns.MetricsRegistry)
114 networkSvc := network.New(nil, []string{"hosts", "kubernetes"})
115 networkSvc.DHCPVendorClassID = "dev.monogon.metropolis.node.v1"
116 timeSvc := timesvc.New()
117 devmgrSvc := devmgr.New()
118
119 // This function initializes a headless Delve if this is a debug build or
120 // does nothing if it's not
121 initializeDebugger(networkSvc)
122
123 // Prepare local storage.
124 root := &localstorage.Root{}
125 if err := declarative.PlaceFS(root, "/"); err != nil {
126 panic(fmt.Errorf("when placing root FS: %w", err))
127 }
128
129 updateSvc := &update.Service{
130 Logger: supervisor.MustSubLogger(ctx, "update"),
131 }
132 // Make node-wide cluster resolver.
133 res := resolver.New(ctx, resolver.WithLogger(supervisor.MustSubLogger(ctx, "resolver")))
134
135 // Start storage and network - we need this to get anything else done.
136 if err := root.Start(ctx, updateSvc); err != nil {
137 return fmt.Errorf("cannot start root FS: %w", err)
138 }
139
140 localNodeParams, err := getLocalNodeParams(ctx, root)
141 if err != nil {
142 return fmt.Errorf("cannot get local node parameters: %w", err)
143 }
144
145 if localNodeParams.NetworkConfig != nil {
146 networkSvc.StaticConfig = localNodeParams.NetworkConfig
147 if err := root.ESP.Metropolis.NetworkConfiguration.Marshal(localNodeParams.NetworkConfig); err != nil {
148 logger.Errorf("Error writing back network_config from NodeParameters: %v", err)
149 }
150 }
151 if networkSvc.StaticConfig == nil {
152 staticConfig, err := root.ESP.Metropolis.NetworkConfiguration.Unmarshal()
153 if err == nil {
154 networkSvc.StaticConfig = staticConfig
155 } else {
156 logger.Errorf("Unable to load static config, proceeding without it: %v", err)
157 }
158 }
159
160 if err := supervisor.Run(ctx, "devmgr", devmgrSvc.Run); err != nil {
161 return fmt.Errorf("when starting devmgr: %w", err)
162 }
163 if err := supervisor.Run(ctx, "network", networkSvc.Run); err != nil {
164 return fmt.Errorf("when starting network: %w", err)
165 }
166 if err := supervisor.Run(ctx, "time", timeSvc.Run); err != nil {
167 return fmt.Errorf("when starting time: %w", err)
168 }
169 if err := supervisor.Run(ctx, "sysctl", nodeSysctls); err != nil {
170 return fmt.Errorf("when applying sysctls: %w", err)
171 }
172
173 // The kernel does of course not run in this runnable, only the log pipe
174 // runs in it.
175 if err := supervisor.Run(ctx, "kernel", func(ctx context.Context) error {
176 return logtree.KmsgPipe(ctx, supervisor.Logger(ctx))
177 }); err != nil {
178 return fmt.Errorf("when starting kernel log pipe: %w", err)
179 }
180
181 // Start the role service. The role service connects to the curator and runs
182 // all node-specific role code (eg. Kubernetes services).
183 logger.Infof("Starting role service...")
184 rs := roleserve.New(roleserve.Config{
185 StorageRoot: root,
186 Network: networkSvc,
187 Resolver: res,
188 LogTree: supervisor.LogTree(ctx),
189 Update: updateSvc,
190 })
191 if err := supervisor.Run(ctx, "role", rs.Run); err != nil {
192 return fmt.Errorf("failed to start role service: %w", err)
193 }
194
195 if err := runDebugService(ctx, rs, supervisor.LogTree(ctx), root); err != nil {
196 return fmt.Errorf("when starting debug service: %w", err)
197 }
198
199 // Initialize interactive consoles.
200 interactiveConsoles := []string{"/dev/tty0"}
201 for _, c := range interactiveConsoles {
202 console, err := tconsole.New(tconsole.TerminalLinux, c, supervisor.LogTree(ctx), &networkSvc.Status, &rs.LocalRoles, &rs.CuratorConnection)
203 if err != nil {
204 logger.Infof("Failed to initialize interactive console at %s: %v", c, err)
205 } else {
206 logger.Infof("Started interactive console at %s", c)
207 supervisor.Run(ctx, "console-"+c, console.Run)
208 }
209 }
210
211 // Now that we have consoles, set console logging level to 1 (KERNEL_EMERG,
212 // minimum possible). This prevents the TUI console from being polluted by
213 // random printks.
214 opts := sysctl.Options{
215 "kernel.printk": "1",
216 }
217 if err := opts.Apply(); err != nil {
218 logger.Errorf("Failed to configure printk logging: %v", err)
219 }
220
221 nodeParams, err := getNodeParams(ctx, root)
222 if err != nil {
223 return fmt.Errorf("cannot get node parameters: %w", err)
224 }
225
226 // Start cluster manager. This kicks off cluster membership machinery,
227 // which will either start a new cluster, enroll into one or join one.
228 m := cluster.NewManager(root, networkSvc, rs, updateSvc, nodeParams, haveTPM)
229 if err := supervisor.Run(ctx, "cluster-manager", m.Run); err != nil {
230 return fmt.Errorf("when starting cluster manager: %w", err)
231 }
232
233 supervisor.Signal(ctx, supervisor.SignalHealthy)
234 supervisor.Signal(ctx, supervisor.SignalDone)
235 return nil
Serge Bazanski3b098232023-03-16 17:57:02 +0100236}