github.com/maier/nomad@v0.4.1-0.20161110003312-a9e3d0b8549d/nomad/config.go (about) 1 package nomad 2 3 import ( 4 "fmt" 5 "io" 6 "net" 7 "os" 8 "runtime" 9 "time" 10 11 "github.com/hashicorp/memberlist" 12 "github.com/hashicorp/nomad/helper/tlsutil" 13 "github.com/hashicorp/nomad/nomad/structs" 14 "github.com/hashicorp/nomad/nomad/structs/config" 15 "github.com/hashicorp/nomad/scheduler" 16 "github.com/hashicorp/raft" 17 "github.com/hashicorp/serf/serf" 18 ) 19 20 const ( 21 DefaultRegion = "global" 22 DefaultDC = "dc1" 23 DefaultSerfPort = 4648 24 ) 25 26 // These are the protocol versions that Nomad can understand 27 const ( 28 ProtocolVersionMin uint8 = 1 29 ProtocolVersionMax = 1 30 ) 31 32 // ProtocolVersionMap is the mapping of Nomad protocol versions 33 // to Serf protocol versions. We mask the Serf protocols using 34 // our own protocol version. 35 var protocolVersionMap map[uint8]uint8 36 37 func init() { 38 protocolVersionMap = map[uint8]uint8{ 39 1: 4, 40 } 41 } 42 43 var ( 44 DefaultRPCAddr = &net.TCPAddr{IP: net.ParseIP("127.0.0.1"), Port: 4647} 45 ) 46 47 // Config is used to parameterize the server 48 type Config struct { 49 // Bootstrap mode is used to bring up the first Nomad server. It is 50 // required so that it can elect a leader without any other nodes 51 // being present 52 Bootstrap bool 53 54 // BootstrapExpect mode is used to automatically bring up a 55 // collection of Nomad servers. This can be used to automatically 56 // bring up a collection of nodes. All operations on BootstrapExpect 57 // must be handled via `atomic.*Int32()` calls. 58 BootstrapExpect int32 59 60 // DataDir is the directory to store our state in 61 DataDir string 62 63 // DevMode is used for development purposes only and limits the 64 // use of persistence or state. 65 DevMode bool 66 67 // DevDisableBootstrap is used to disable bootstrap mode while 68 // in DevMode. This is largely used for testing. 69 DevDisableBootstrap bool 70 71 // LogOutput is the location to write logs to. If this is not set, 72 // logs will go to stderr. 73 LogOutput io.Writer 74 75 // ProtocolVersion is the protocol version to speak. This must be between 76 // ProtocolVersionMin and ProtocolVersionMax. 77 ProtocolVersion uint8 78 79 // RPCAddr is the RPC address used by Nomad. This should be reachable 80 // by the other servers and clients 81 RPCAddr *net.TCPAddr 82 83 // RPCAdvertise is the address that is advertised to other nodes for 84 // the RPC endpoint. This can differ from the RPC address, if for example 85 // the RPCAddr is unspecified "0.0.0.0:4646", but this address must be 86 // reachable 87 RPCAdvertise *net.TCPAddr 88 89 // RaftConfig is the configuration used for Raft in the local DC 90 RaftConfig *raft.Config 91 92 // RaftTimeout is applied to any network traffic for raft. Defaults to 10s. 93 RaftTimeout time.Duration 94 95 // RequireTLS ensures that all RPC traffic is protected with TLS 96 RequireTLS bool 97 98 // SerfConfig is the configuration for the serf cluster 99 SerfConfig *serf.Config 100 101 // Node name is the name we use to advertise. Defaults to hostname. 102 NodeName string 103 104 // Region is the region this Nomad server belongs to. 105 Region string 106 107 // Datacenter is the datacenter this Nomad server belongs to. 108 Datacenter string 109 110 // Build is a string that is gossiped around, and can be used to help 111 // operators track which versions are actively deployed 112 Build string 113 114 // NumSchedulers is the number of scheduler thread that are run. 115 // This can be as many as one per core, or zero to disable this server 116 // from doing any scheduling work. 117 NumSchedulers int 118 119 // EnabledSchedulers controls the set of sub-schedulers that are 120 // enabled for this server to handle. This will restrict the evaluations 121 // that the workers dequeue for processing. 122 EnabledSchedulers []string 123 124 // ReconcileInterval controls how often we reconcile the strongly 125 // consistent store with the Serf info. This is used to handle nodes 126 // that are force removed, as well as intermittent unavailability during 127 // leader election. 128 ReconcileInterval time.Duration 129 130 // EvalGCInterval is how often we dispatch a job to GC evaluations 131 EvalGCInterval time.Duration 132 133 // EvalGCThreshold is how "old" an evaluation must be to be eligible 134 // for GC. This gives users some time to debug a failed evaluation. 135 EvalGCThreshold time.Duration 136 137 // JobGCInterval is how often we dispatch a job to GC jobs that are 138 // available for garbage collection. 139 JobGCInterval time.Duration 140 141 // JobGCThreshold is how old a job must be before it eligible for GC. This gives 142 // the user time to inspect the job. 143 JobGCThreshold time.Duration 144 145 // NodeGCInterval is how often we dispatch a job to GC failed nodes. 146 NodeGCInterval time.Duration 147 148 // NodeGCThreshold is how "old" a nodemust be to be eligible 149 // for GC. This gives users some time to view and debug a failed nodes. 150 NodeGCThreshold time.Duration 151 152 // EvalNackTimeout controls how long we allow a sub-scheduler to 153 // work on an evaluation before we consider it failed and Nack it. 154 // This allows that evaluation to be handed to another sub-scheduler 155 // to work on. Defaults to 60 seconds. This should be long enough that 156 // no evaluation hits it unless the sub-scheduler has failed. 157 EvalNackTimeout time.Duration 158 159 // EvalDeliveryLimit is the limit of attempts we make to deliver and 160 // process an evaluation. This is used so that an eval that will never 161 // complete eventually fails out of the system. 162 EvalDeliveryLimit int 163 164 // MinHeartbeatTTL is the minimum time between heartbeats. 165 // This is used as a floor to prevent excessive updates. 166 MinHeartbeatTTL time.Duration 167 168 // MaxHeartbeatsPerSecond is the maximum target rate of heartbeats 169 // being processed per second. This allows the TTL to be increased 170 // to meet the target rate. 171 MaxHeartbeatsPerSecond float64 172 173 // HeartbeatGrace is the additional time given as a grace period 174 // beyond the TTL to account for network and processing delays 175 // as well as clock skew. 176 HeartbeatGrace time.Duration 177 178 // FailoverHeartbeatTTL is the TTL applied to heartbeats after 179 // a new leader is elected, since we no longer know the status 180 // of all the heartbeats. 181 FailoverHeartbeatTTL time.Duration 182 183 // ConsulConfig is this Agent's Consul configuration 184 ConsulConfig *config.ConsulConfig 185 186 // VaultConfig is this Agent's Vault configuration 187 VaultConfig *config.VaultConfig 188 189 // RPCHoldTimeout is how long an RPC can be "held" before it is errored. 190 // This is used to paper over a loss of leadership by instead holding RPCs, 191 // so that the caller experiences a slow response rather than an error. 192 // This period is meant to be long enough for a leader election to take 193 // place, and a small jitter is applied to avoid a thundering herd. 194 RPCHoldTimeout time.Duration 195 196 // TLSConfig holds various TLS related configurations 197 TLSConfig *config.TLSConfig 198 } 199 200 // CheckVersion is used to check if the ProtocolVersion is valid 201 func (c *Config) CheckVersion() error { 202 if c.ProtocolVersion < ProtocolVersionMin { 203 return fmt.Errorf("Protocol version '%d' too low. Must be in range: [%d, %d]", 204 c.ProtocolVersion, ProtocolVersionMin, ProtocolVersionMax) 205 } else if c.ProtocolVersion > ProtocolVersionMax { 206 return fmt.Errorf("Protocol version '%d' too high. Must be in range: [%d, %d]", 207 c.ProtocolVersion, ProtocolVersionMin, ProtocolVersionMax) 208 } 209 return nil 210 } 211 212 // DefaultConfig returns the default configuration 213 func DefaultConfig() *Config { 214 hostname, err := os.Hostname() 215 if err != nil { 216 panic(err) 217 } 218 219 c := &Config{ 220 Region: DefaultRegion, 221 Datacenter: DefaultDC, 222 NodeName: hostname, 223 ProtocolVersion: ProtocolVersionMax, 224 RaftConfig: raft.DefaultConfig(), 225 RaftTimeout: 10 * time.Second, 226 LogOutput: os.Stderr, 227 RPCAddr: DefaultRPCAddr, 228 SerfConfig: serf.DefaultConfig(), 229 NumSchedulers: 1, 230 ReconcileInterval: 60 * time.Second, 231 EvalGCInterval: 5 * time.Minute, 232 EvalGCThreshold: 1 * time.Hour, 233 JobGCInterval: 5 * time.Minute, 234 JobGCThreshold: 4 * time.Hour, 235 NodeGCInterval: 5 * time.Minute, 236 NodeGCThreshold: 24 * time.Hour, 237 EvalNackTimeout: 60 * time.Second, 238 EvalDeliveryLimit: 3, 239 MinHeartbeatTTL: 10 * time.Second, 240 MaxHeartbeatsPerSecond: 50.0, 241 HeartbeatGrace: 10 * time.Second, 242 FailoverHeartbeatTTL: 300 * time.Second, 243 ConsulConfig: config.DefaultConsulConfig(), 244 VaultConfig: config.DefaultVaultConfig(), 245 RPCHoldTimeout: 5 * time.Second, 246 TLSConfig: &config.TLSConfig{}, 247 } 248 249 // Enable all known schedulers by default 250 c.EnabledSchedulers = make([]string, 0, len(scheduler.BuiltinSchedulers)) 251 for name := range scheduler.BuiltinSchedulers { 252 c.EnabledSchedulers = append(c.EnabledSchedulers, name) 253 } 254 c.EnabledSchedulers = append(c.EnabledSchedulers, structs.JobTypeCore) 255 256 // Default the number of schedulers to match the coores 257 c.NumSchedulers = runtime.NumCPU() 258 259 // Increase our reap interval to 3 days instead of 24h. 260 c.SerfConfig.ReconnectTimeout = 3 * 24 * time.Hour 261 262 // Serf should use the WAN timing, since we are using it 263 // to communicate between DC's 264 c.SerfConfig.MemberlistConfig = memberlist.DefaultWANConfig() 265 c.SerfConfig.MemberlistConfig.BindPort = DefaultSerfPort 266 267 // Disable shutdown on removal 268 c.RaftConfig.ShutdownOnRemove = false 269 return c 270 } 271 272 // tlsConfig returns a TLSUtil Config based on the server configuration 273 func (c *Config) tlsConfig() *tlsutil.Config { 274 tlsConf := &tlsutil.Config{ 275 VerifyIncoming: true, 276 VerifyOutgoing: true, 277 VerifyServerHostname: c.TLSConfig.VerifyServerHostname, 278 CAFile: c.TLSConfig.CAFile, 279 CertFile: c.TLSConfig.CertFile, 280 KeyFile: c.TLSConfig.KeyFile, 281 } 282 return tlsConf 283 }