open-nomad/client/client.go

815 lines
21 KiB
Go
Raw Normal View History

2015-08-20 22:25:09 +00:00
package client
import (
2015-08-20 23:07:26 +00:00
"fmt"
"io/ioutil"
2015-08-20 22:25:09 +00:00
"log"
2015-08-20 23:07:26 +00:00
"net"
2015-08-20 22:25:09 +00:00
"os"
"path/filepath"
2015-08-20 23:07:26 +00:00
"strconv"
2015-08-20 22:25:09 +00:00
"sync"
2015-08-20 23:07:26 +00:00
"time"
"github.com/hashicorp/go-multierror"
"github.com/hashicorp/nomad/client/config"
2015-08-20 23:53:43 +00:00
"github.com/hashicorp/nomad/client/driver"
2015-08-20 23:41:29 +00:00
"github.com/hashicorp/nomad/client/fingerprint"
2015-08-20 23:07:26 +00:00
"github.com/hashicorp/nomad/nomad"
2015-08-20 23:41:29 +00:00
"github.com/hashicorp/nomad/nomad/structs"
2015-08-20 22:25:09 +00:00
)
2015-08-20 23:07:26 +00:00
const (
// clientRPCCache controls how long we keep an idle connection
// open to a server
clientRPCCache = 30 * time.Second
// clientMaxStreams controsl how many idle streams we keep
// open to a server
clientMaxStreams = 2
2015-08-21 00:49:04 +00:00
// registerRetryIntv is minimum interval on which we retry
// registration. We pick a value between this and 2x this.
2015-08-24 00:40:14 +00:00
registerRetryIntv = 15 * time.Second
2015-08-23 02:31:22 +00:00
// getAllocRetryIntv is minimum interval on which we retry
// to fetch allocations. We pick a value between this and 2x this.
getAllocRetryIntv = 30 * time.Second
2015-08-24 00:40:14 +00:00
// devModeRetryIntv is the retry interval used for development
devModeRetryIntv = time.Second
2015-08-31 00:19:20 +00:00
// stateSnapshotIntv is how often the client snapshots state
stateSnapshotIntv = 60 * time.Second
// registerErrGrace is the grace period where we don't log about
// register errors after start. This is to improve the user experience
// in dev mode where the leader isn't elected for a few seconds.
registerErrGrace = 10 * time.Second
// initialHeartbeatStagger is used to stagger the interval between
// starting and the intial heartbeat. After the intial heartbeat,
// we switch to using the TTL specified by the servers.
initialHeartbeatStagger = 10 * time.Second
2015-08-20 23:07:26 +00:00
)
2015-08-20 22:25:09 +00:00
// DefaultConfig returns the default configuration
func DefaultConfig() *config.Config {
return &config.Config{
2015-08-20 22:25:09 +00:00
LogOutput: os.Stderr,
Region: "global",
2015-08-20 22:25:09 +00:00
}
}
// Client is used to implement the client interaction with Nomad. Clients
// are expected to register as a schedulable node to the servers, and to
// run allocations as determined by the servers.
type Client struct {
config *config.Config
start time.Time
2015-08-20 23:07:26 +00:00
2015-08-20 22:25:09 +00:00
logger *log.Logger
2015-11-24 20:34:26 +00:00
consulService *ConsulService
2015-08-20 23:07:26 +00:00
lastServer net.Addr
lastRPCTime time.Time
lastServerLock sync.Mutex
servers []string
serverLock sync.RWMutex
2015-08-20 23:07:26 +00:00
connPool *nomad.ConnPool
2015-08-23 01:16:05 +00:00
lastHeartbeat time.Time
heartbeatTTL time.Duration
2015-08-23 21:54:52 +00:00
// allocs is the current set of allocations
allocs map[string]*AllocRunner
2015-08-23 22:06:47 +00:00
allocLock sync.RWMutex
2015-08-23 21:54:52 +00:00
2015-08-20 22:25:09 +00:00
shutdown bool
shutdownCh chan struct{}
shutdownLock sync.Mutex
}
// NewClient is used to create a new client from the given configuration
func NewClient(cfg *config.Config) (*Client, error) {
2015-08-20 22:25:09 +00:00
// Create a logger
logger := log.New(cfg.LogOutput, "", log.LstdFlags)
2015-08-20 22:25:09 +00:00
2015-08-20 23:41:29 +00:00
// Create the client
2015-08-20 22:25:09 +00:00
c := &Client{
config: cfg,
start: time.Now(),
connPool: nomad.NewPool(cfg.LogOutput, clientRPCCache, clientMaxStreams, nil),
logger: logger,
allocs: make(map[string]*AllocRunner),
shutdownCh: make(chan struct{}),
}
// Setup the Consul Service
if err := c.setupConsulService(); err != nil {
return nil, fmt.Errorf("failed to create the consul service: %v", err)
2015-08-20 22:25:09 +00:00
}
2015-08-20 23:41:29 +00:00
2015-09-12 18:47:44 +00:00
// Initialize the client
if err := c.init(); err != nil {
return nil, fmt.Errorf("failed intializing client: %v", err)
}
2015-08-20 23:41:29 +00:00
// Setup the node
if err := c.setupNode(); err != nil {
return nil, fmt.Errorf("node setup failed: %v", err)
}
// Fingerprint the node
if err := c.fingerprint(); err != nil {
return nil, fmt.Errorf("fingerprinting failed: %v", err)
}
2015-08-20 23:53:43 +00:00
// Scan for drivers
if err := c.setupDrivers(); err != nil {
return nil, fmt.Errorf("driver setup failed: %v", err)
}
2015-08-21 00:49:04 +00:00
// Set up the known servers list
c.SetServers(c.config.Servers)
2015-11-09 23:55:31 +00:00
// Restore the state
if err := c.restoreState(); err != nil {
return nil, fmt.Errorf("failed to restore state: %v", err)
}
2015-08-21 00:49:04 +00:00
// Start the client!
go c.run()
2015-11-24 20:34:26 +00:00
// Start the consul service
go c.consulService.SyncWithConsul()
2015-08-20 22:25:09 +00:00
return c, nil
2015-08-23 23:53:15 +00:00
}
func (c *Client) setupConsulService() error {
var consulService *ConsulService
var err error
addr := c.config.ReadDefault("consul.address", "127.0.0.1:8500")
token := c.config.Read("consul.token")
auth := c.config.Read("consul.auth")
enableSSL := c.config.ReadBoolDefault("consul.ssl", false)
verifySSL := c.config.ReadBoolDefault("consul.verifyssl", true)
if consulService, err = NewConsulService(c.logger, addr, token, auth, enableSSL, verifySSL); err != nil {
return err
}
c.consulService = consulService
return nil
}
2015-09-12 18:47:44 +00:00
// init is used to initialize the client and perform any setup
// needed before we begin starting its various components.
func (c *Client) init() error {
2015-09-24 21:29:53 +00:00
// Ensure the state dir exists if we have one
if c.config.StateDir != "" {
if err := os.MkdirAll(c.config.StateDir, 0700); err != nil {
return fmt.Errorf("failed creating state dir: %s", err)
}
2015-11-11 00:03:18 +00:00
} else {
// Othewise make a temp directory to use.
p, err := ioutil.TempDir("", "NomadClient")
if err != nil {
return fmt.Errorf("failed creating temporary directory for the StateDir: %v", err)
}
c.config.StateDir = p
2015-09-24 21:29:53 +00:00
}
2015-11-11 00:03:18 +00:00
c.logger.Printf("[INFO] client: using state directory %v", c.config.StateDir)
2015-09-24 21:29:53 +00:00
// Ensure the alloc dir exists if we have one
if c.config.AllocDir != "" {
if err := os.MkdirAll(c.config.AllocDir, 0700); err != nil {
return fmt.Errorf("failed creating alloc dir: %s", err)
}
2015-09-26 01:12:11 +00:00
} else {
// Othewise make a temp directory to use.
p, err := ioutil.TempDir("", "NomadClient")
if err != nil {
return fmt.Errorf("failed creating temporary directory for the AllocDir: %v", err)
}
c.config.AllocDir = p
2015-09-12 18:47:44 +00:00
}
2015-09-25 23:46:21 +00:00
c.logger.Printf("[INFO] client: using alloc directory %v", c.config.AllocDir)
2015-09-12 18:47:44 +00:00
return nil
}
2015-08-23 23:53:15 +00:00
// Leave is used to prepare the client to leave the cluster
func (c *Client) Leave() error {
// TODO
return nil
2015-08-20 22:25:09 +00:00
}
// Shutdown is used to tear down the client
func (c *Client) Shutdown() error {
2015-08-20 23:07:26 +00:00
c.logger.Printf("[INFO] client: shutting down")
2015-08-20 22:25:09 +00:00
c.shutdownLock.Lock()
defer c.shutdownLock.Unlock()
if c.shutdown {
return nil
}
2015-10-04 20:36:03 +00:00
// Destroy all the running allocations.
if c.config.DevMode {
for _, ar := range c.allocs {
ar.Destroy()
<-ar.WaitCh()
}
}
2015-11-24 20:34:26 +00:00
// Stop the consul service
c.consulService.ShutDown()
2015-08-20 22:25:09 +00:00
c.shutdown = true
close(c.shutdownCh)
2015-08-21 00:49:04 +00:00
c.connPool.Shutdown()
return c.saveState()
2015-08-20 22:25:09 +00:00
}
2015-08-20 23:07:26 +00:00
// RPC is used to forward an RPC call to a nomad server, or fail if no servers
func (c *Client) RPC(method string, args interface{}, reply interface{}) error {
// Invoke the RPCHandle if it exists
if c.config.RPCHandler != nil {
return c.config.RPCHandler.RPC(method, args, reply)
}
// Pick a server to request from
addr, err := c.pickServer()
if err != nil {
return err
}
// Make the RPC request
err = c.connPool.RPC(c.config.Region, addr, 1, method, args, reply)
// Update the last server information
c.lastServerLock.Lock()
if err != nil {
c.lastServer = nil
c.lastRPCTime = time.Time{}
} else {
c.lastServer = addr
c.lastRPCTime = time.Now()
}
c.lastServerLock.Unlock()
return err
}
// pickServer is used to pick a target RPC server
func (c *Client) pickServer() (net.Addr, error) {
c.lastServerLock.Lock()
defer c.lastServerLock.Unlock()
// Check for a valid last-used server
if c.lastServer != nil && time.Now().Sub(c.lastRPCTime) < clientRPCCache {
return c.lastServer, nil
}
// Bail if we can't find any servers
servers := c.Servers()
if len(servers) == 0 {
2015-08-20 23:07:26 +00:00
return nil, fmt.Errorf("no known servers")
}
// Shuffle so we don't always use the same server
2015-08-20 23:07:26 +00:00
shuffleStrings(servers)
// Try to resolve each server
for i := 0; i < len(servers); i++ {
addr, err := net.ResolveTCPAddr("tcp", servers[i])
if err == nil {
c.lastServer = addr
c.lastRPCTime = time.Now()
return addr, nil
}
c.logger.Printf("[WARN] client: failed to resolve '%s': %s", servers[i], err)
2015-08-20 23:07:26 +00:00
}
// Bail if we reach this point
return nil, fmt.Errorf("failed to resolve any servers")
}
// Servers is used to return the current known servers list. When an agent
// is first started, this list comes directly from configuration files.
func (c *Client) Servers() []string {
c.serverLock.RLock()
defer c.serverLock.RUnlock()
return c.servers
}
// SetServers is used to modify the known servers list. This avoids forcing
// a config rollout + rolling restart and enables auto-join features. The
// full set of servers is passed to support adding and/or removing servers.
func (c *Client) SetServers(servers []string) {
c.serverLock.Lock()
defer c.serverLock.Unlock()
if servers == nil {
servers = make([]string, 0)
}
c.servers = servers
}
2015-08-20 23:07:26 +00:00
// Stats is used to return statistics for debugging and insight
// for various sub-systems
func (c *Client) Stats() map[string]map[string]string {
toString := func(v uint64) string {
return strconv.FormatUint(v, 10)
}
2015-08-31 00:24:12 +00:00
c.allocLock.RLock()
numAllocs := len(c.allocs)
c.allocLock.RUnlock()
2015-08-20 23:07:26 +00:00
stats := map[string]map[string]string{
2015-08-31 00:24:12 +00:00
"client": map[string]string{
"known_servers": toString(uint64(len(c.Servers()))),
2015-08-31 00:24:12 +00:00
"num_allocations": toString(uint64(numAllocs)),
2015-09-22 22:29:30 +00:00
"last_heartbeat": fmt.Sprintf("%v", time.Since(c.lastHeartbeat)),
"heartbeat_ttl": fmt.Sprintf("%v", c.heartbeatTTL),
2015-08-20 23:07:26 +00:00
},
"runtime": nomad.RuntimeStats(),
}
return stats
}
2015-08-20 23:41:29 +00:00
// Node returns the locally registered node
func (c *Client) Node() *structs.Node {
return c.config.Node
}
2015-08-23 21:12:26 +00:00
// restoreState is used to restore our state from the data dir
func (c *Client) restoreState() error {
if c.config.DevMode {
return nil
}
// Scan the directory
list, err := ioutil.ReadDir(filepath.Join(c.config.StateDir, "alloc"))
if err != nil && os.IsNotExist(err) {
return nil
} else if err != nil {
return fmt.Errorf("failed to list alloc state: %v", err)
}
// Load each alloc back
var mErr multierror.Error
for _, entry := range list {
id := entry.Name()
alloc := &structs.Allocation{ID: id}
2015-11-24 20:34:26 +00:00
ar := NewAllocRunner(c.logger, c.config, c.updateAllocStatus, alloc, c.consulService)
c.allocs[id] = ar
if err := ar.RestoreState(); err != nil {
2015-11-09 23:55:31 +00:00
c.logger.Printf("[ERR] client: failed to restore state for alloc %s: %v", id, err)
mErr.Errors = append(mErr.Errors, err)
} else {
go ar.Run()
}
}
return mErr.ErrorOrNil()
2015-08-23 21:12:26 +00:00
}
// saveState is used to snapshot our state into the data dir
func (c *Client) saveState() error {
if c.config.DevMode {
return nil
}
var mErr multierror.Error
c.allocLock.RLock()
defer c.allocLock.RUnlock()
for id, ar := range c.allocs {
if err := ar.SaveState(); err != nil {
c.logger.Printf("[ERR] client: failed to save state for alloc %s: %v",
id, err)
mErr.Errors = append(mErr.Errors, err)
}
}
return mErr.ErrorOrNil()
2015-08-23 21:12:26 +00:00
}
2015-09-22 17:31:47 +00:00
// nodeID restores a persistent unique ID or generates a new one
func (c *Client) nodeID() (string, error) {
// Do not persist in dev mode
if c.config.DevMode {
return structs.GenerateUUID(), nil
}
// Attempt to read existing ID
path := filepath.Join(c.config.StateDir, "client-id")
buf, err := ioutil.ReadFile(path)
if err != nil && !os.IsNotExist(err) {
return "", err
}
// Use existing ID if any
if len(buf) != 0 {
return string(buf), nil
}
// Generate new ID
id := structs.GenerateUUID()
// Persist the ID
if err := ioutil.WriteFile(path, []byte(id), 0700); err != nil {
return "", err
}
return id, nil
}
2015-08-20 23:41:29 +00:00
// setupNode is used to setup the initial node
func (c *Client) setupNode() error {
node := c.config.Node
if node == nil {
node = &structs.Node{}
c.config.Node = node
}
if node.Attributes == nil {
node.Attributes = make(map[string]string)
}
if node.Links == nil {
node.Links = make(map[string]string)
}
if node.Meta == nil {
node.Meta = make(map[string]string)
}
2015-08-21 00:49:04 +00:00
if node.Resources == nil {
node.Resources = &structs.Resources{}
}
if node.ID == "" {
2015-09-22 17:31:47 +00:00
id, err := c.nodeID()
if err != nil {
return fmt.Errorf("node ID setup failed: %v", err)
}
node.ID = id
2015-08-21 00:49:04 +00:00
}
if node.Datacenter == "" {
node.Datacenter = "dc1"
}
if node.Name == "" {
node.Name, _ = os.Hostname()
}
if node.Name == "" {
node.Name = node.ID
}
node.Status = structs.NodeStatusInit
2015-08-20 23:41:29 +00:00
return nil
}
// fingerprint is used to fingerprint the client and setup the node
func (c *Client) fingerprint() error {
whitelist := c.config.ReadStringListToMap("fingerprint.whitelist")
whitelistEnabled := len(whitelist) > 0
2015-08-20 23:41:29 +00:00
var applied []string
var skipped []string
2015-09-23 04:25:12 +00:00
for _, name := range fingerprint.BuiltinFingerprints {
// Skip modules that are not in the whitelist if it is enabled.
if _, ok := whitelist[name]; whitelistEnabled && !ok {
skipped = append(skipped, name)
continue
}
2015-08-20 23:41:29 +00:00
f, err := fingerprint.NewFingerprint(name, c.logger)
if err != nil {
return err
}
applies, err := f.Fingerprint(c.config, c.config.Node)
2015-08-20 23:41:29 +00:00
if err != nil {
return err
}
if applies {
applied = append(applied, name)
}
2015-11-05 21:41:41 +00:00
p, period := f.Periodic()
if p {
// TODO: If more periodic fingerprinters are added, then
// fingerprintPeriodic should be used to handle all the periodic
// fingerprinters by using a priority queue.
go c.fingerprintPeriodic(name, f, period)
}
2015-08-20 23:41:29 +00:00
}
c.logger.Printf("[DEBUG] client: applied fingerprints %v", applied)
if len(skipped) != 0 {
c.logger.Printf("[DEBUG] client: fingerprint modules skipped due to whitelist: %v", skipped)
}
2015-08-20 23:41:29 +00:00
return nil
}
2015-08-20 23:53:43 +00:00
// fingerprintPeriodic runs a fingerprinter at the specified duration.
2015-11-05 21:41:41 +00:00
func (c *Client) fingerprintPeriodic(name string, f fingerprint.Fingerprint, d time.Duration) {
c.logger.Printf("[DEBUG] client: periodically fingerprinting %v at duration %v", name, d)
for {
select {
case <-time.After(d):
if _, err := f.Fingerprint(c.config, c.config.Node); err != nil {
c.logger.Printf("[DEBUG] client: periodic fingerprinting for %v failed: %v", name, err)
2015-11-05 21:41:41 +00:00
}
case <-c.shutdownCh:
return
}
}
}
2015-08-20 23:53:43 +00:00
// setupDrivers is used to find the available drivers
func (c *Client) setupDrivers() error {
2015-11-20 00:39:21 +00:00
// Build the whitelist of drivers.
whitelist := c.config.ReadStringListToMap("driver.whitelist")
2015-11-20 00:39:21 +00:00
whitelistEnabled := len(whitelist) > 0
2015-08-20 23:53:43 +00:00
var avail []string
var skipped []string
driverCtx := driver.NewDriverContext("", c.config, c.config.Node, c.logger)
2015-08-20 23:53:43 +00:00
for name := range driver.BuiltinDrivers {
2015-11-20 00:39:21 +00:00
// Skip fingerprinting drivers that are not in the whitelist if it is
// enabled.
if _, ok := whitelist[name]; whitelistEnabled && !ok {
skipped = append(skipped, name)
2015-11-20 00:39:21 +00:00
continue
}
d, err := driver.NewDriver(name, driverCtx)
2015-08-20 23:53:43 +00:00
if err != nil {
return err
}
applies, err := d.Fingerprint(c.config, c.config.Node)
2015-08-20 23:53:43 +00:00
if err != nil {
return err
}
if applies {
avail = append(avail, name)
}
}
2015-11-20 00:39:21 +00:00
2015-08-20 23:53:43 +00:00
c.logger.Printf("[DEBUG] client: available drivers %v", avail)
2015-11-20 00:39:21 +00:00
if len(skipped) != 0 {
c.logger.Printf("[DEBUG] client: drivers skipped due to whitelist: %v", skipped)
2015-11-20 00:39:21 +00:00
}
2015-08-20 23:53:43 +00:00
return nil
}
2015-08-21 00:49:04 +00:00
2015-08-24 00:40:14 +00:00
// retryIntv calculates a retry interval value given the base
func (c *Client) retryIntv(base time.Duration) time.Duration {
if c.config.DevMode {
return devModeRetryIntv
}
return base + randomStagger(base)
}
2015-08-21 00:49:04 +00:00
// run is a long lived goroutine used to run the client
func (c *Client) run() {
// Register the client
for {
if err := c.registerNode(); err == nil {
break
}
select {
2015-08-24 00:40:14 +00:00
case <-time.After(c.retryIntv(registerRetryIntv)):
2015-08-21 00:49:04 +00:00
case <-c.shutdownCh:
return
}
}
// Setup the heartbeat timer, for the initial registration
// we want to do this quickly. We want to do it extra quickly
// in development mode.
var heartbeat <-chan time.Time
if c.config.DevMode {
heartbeat = time.After(0)
} else {
heartbeat = time.After(randomStagger(initialHeartbeatStagger))
}
2015-08-23 01:16:05 +00:00
2015-08-23 02:31:22 +00:00
// Watch for changes in allocations
allocUpdates := make(chan []*structs.Allocation, 1)
go c.watchAllocations(allocUpdates)
2015-08-21 00:49:04 +00:00
2015-08-31 00:19:20 +00:00
// Create a snapshot timer
snapshot := time.After(stateSnapshotIntv)
2015-08-23 01:16:05 +00:00
// Periodically update our status and wait for termination
2015-08-24 00:40:14 +00:00
for {
select {
2015-08-31 00:19:20 +00:00
case <-snapshot:
snapshot = time.After(stateSnapshotIntv)
if err := c.saveState(); err != nil {
c.logger.Printf("[ERR] client: failed to save state: %v", err)
}
2015-08-24 00:40:14 +00:00
case allocs := <-allocUpdates:
c.runAllocs(allocs)
case <-heartbeat:
if err := c.updateNodeStatus(); err != nil {
heartbeat = time.After(c.retryIntv(registerRetryIntv))
} else {
heartbeat = time.After(c.heartbeatTTL)
}
2015-08-23 02:31:22 +00:00
2015-08-24 00:40:14 +00:00
case <-c.shutdownCh:
return
}
2015-08-21 00:49:04 +00:00
}
}
// registerNode is used to register the node or update the registration
func (c *Client) registerNode() error {
node := c.Node()
req := structs.NodeRegisterRequest{
Node: node,
WriteRequest: structs.WriteRequest{Region: c.config.Region},
}
var resp structs.NodeUpdateResponse
err := c.RPC("Node.Register", &req, &resp)
2015-08-21 00:49:04 +00:00
if err != nil {
if time.Since(c.start) > registerErrGrace {
c.logger.Printf("[ERR] client: failed to register node: %v", err)
}
2015-08-21 00:49:04 +00:00
return err
}
c.logger.Printf("[DEBUG] client: node registration complete")
if len(resp.EvalIDs) != 0 {
c.logger.Printf("[DEBUG] client: %d evaluations triggered by node registration", len(resp.EvalIDs))
}
2015-08-23 01:16:05 +00:00
c.lastHeartbeat = time.Now()
c.heartbeatTTL = resp.HeartbeatTTL
return nil
}
// updateNodeStatus is used to heartbeat and update the status of the node
func (c *Client) updateNodeStatus() error {
node := c.Node()
req := structs.NodeUpdateStatusRequest{
NodeID: node.ID,
Status: structs.NodeStatusReady,
WriteRequest: structs.WriteRequest{Region: c.config.Region},
}
var resp structs.NodeUpdateResponse
err := c.RPC("Node.UpdateStatus", &req, &resp)
2015-08-23 01:16:05 +00:00
if err != nil {
c.logger.Printf("[ERR] client: failed to update status: %v", err)
return err
}
if len(resp.EvalIDs) != 0 {
c.logger.Printf("[DEBUG] client: %d evaluations triggered by node update", len(resp.EvalIDs))
}
if resp.Index != 0 {
2015-08-24 00:40:14 +00:00
c.logger.Printf("[DEBUG] client: state updated to %s", req.Status)
2015-08-23 01:16:05 +00:00
}
c.lastHeartbeat = time.Now()
c.heartbeatTTL = resp.HeartbeatTTL
2015-08-21 00:49:04 +00:00
return nil
}
2015-08-23 02:31:22 +00:00
2015-08-29 21:22:24 +00:00
// updateAllocStatus is used to update the status of an allocation
func (c *Client) updateAllocStatus(alloc *structs.Allocation) error {
args := structs.AllocUpdateRequest{
Alloc: []*structs.Allocation{alloc},
WriteRequest: structs.WriteRequest{Region: c.config.Region},
}
var resp structs.GenericResponse
err := c.RPC("Node.UpdateAlloc", &args, &resp)
2015-08-29 21:22:24 +00:00
if err != nil {
c.logger.Printf("[ERR] client: failed to update allocation: %v", err)
return err
}
return nil
}
2015-08-23 02:31:22 +00:00
// watchAllocations is used to scan for updates to allocations
func (c *Client) watchAllocations(allocUpdates chan []*structs.Allocation) {
req := structs.NodeSpecificRequest{
NodeID: c.Node().ID,
QueryOptions: structs.QueryOptions{
2015-08-24 00:40:14 +00:00
Region: c.config.Region,
AllowStale: true,
2015-08-23 02:31:22 +00:00
},
}
var resp structs.NodeAllocsResponse
for {
// Get the allocations, blocking for updates
resp = structs.NodeAllocsResponse{}
err := c.RPC("Node.GetAllocs", &req, &resp)
2015-08-23 02:31:22 +00:00
if err != nil {
c.logger.Printf("[ERR] client: failed to query for node allocations: %v", err)
2015-08-24 00:40:14 +00:00
retry := c.retryIntv(getAllocRetryIntv)
2015-08-23 02:31:22 +00:00
select {
case <-time.After(retry):
continue
case <-c.shutdownCh:
return
}
}
// Check for shutdown
select {
case <-c.shutdownCh:
return
default:
}
// Check for updates
2015-08-24 00:40:14 +00:00
if resp.Index <= req.MinQueryIndex {
2015-08-23 02:31:22 +00:00
continue
}
2015-08-24 00:40:14 +00:00
req.MinQueryIndex = resp.Index
c.logger.Printf("[DEBUG] client: updated allocations at index %d (%d allocs)", resp.Index, len(resp.Allocs))
2015-08-23 02:31:22 +00:00
// Push the updates
select {
case allocUpdates <- resp.Allocs:
case <-c.shutdownCh:
return
}
}
}
// runAllocs is invoked when we get an updated set of allocations
2015-08-23 21:54:52 +00:00
func (c *Client) runAllocs(updated []*structs.Allocation) {
// Get the existing allocs
2015-08-23 22:06:47 +00:00
c.allocLock.RLock()
2015-08-29 21:33:30 +00:00
exist := make([]*structs.Allocation, 0, len(c.allocs))
for _, ar := range c.allocs {
exist = append(exist, ar.Alloc())
2015-08-23 21:54:52 +00:00
}
2015-08-23 22:06:47 +00:00
c.allocLock.RUnlock()
2015-08-23 21:54:52 +00:00
// Diff the existing and updated allocations
diff := diffAllocs(exist, updated)
c.logger.Printf("[DEBUG] client: %#v", diff)
// Remove the old allocations
for _, remove := range diff.removed {
if err := c.removeAlloc(remove); err != nil {
c.logger.Printf("[ERR] client: failed to remove alloc '%s': %v",
remove.ID, err)
}
}
// Update the existing allocations
for _, update := range diff.updated {
if err := c.updateAlloc(update.exist, update.updated); err != nil {
c.logger.Printf("[ERR] client: failed to update alloc '%s': %v",
update.exist.ID, err)
}
}
// Start the new allocations
for _, add := range diff.added {
if err := c.addAlloc(add); err != nil {
c.logger.Printf("[ERR] client: failed to add alloc '%s': %v",
add.ID, err)
}
}
// Persist our state
if err := c.saveState(); err != nil {
c.logger.Printf("[ERR] client: failed to save state: %v", err)
}
}
// removeAlloc is invoked when we should remove an allocation
func (c *Client) removeAlloc(alloc *structs.Allocation) error {
2015-08-29 21:33:30 +00:00
c.allocLock.Lock()
defer c.allocLock.Unlock()
ar, ok := c.allocs[alloc.ID]
2015-08-23 22:06:47 +00:00
if !ok {
c.logger.Printf("[WARN] client: missing context for alloc '%s'", alloc.ID)
return nil
}
ar.Destroy()
2015-08-29 21:33:30 +00:00
delete(c.allocs, alloc.ID)
2015-08-23 21:54:52 +00:00
return nil
}
// updateAlloc is invoked when we should update an allocation
func (c *Client) updateAlloc(exist, update *structs.Allocation) error {
2015-08-23 22:06:47 +00:00
c.allocLock.RLock()
defer c.allocLock.RUnlock()
ar, ok := c.allocs[exist.ID]
2015-08-23 22:06:47 +00:00
if !ok {
c.logger.Printf("[WARN] client: missing context for alloc '%s'", exist.ID)
return nil
}
ar.Update(update)
2015-08-23 21:54:52 +00:00
return nil
}
// addAlloc is invoked when we should add an allocation
func (c *Client) addAlloc(alloc *structs.Allocation) error {
2015-08-23 22:06:47 +00:00
c.allocLock.Lock()
defer c.allocLock.Unlock()
2015-11-24 20:34:26 +00:00
ar := NewAllocRunner(c.logger, c.config, c.updateAllocStatus, alloc, c.consulService)
c.allocs[alloc.ID] = ar
go ar.Run()
2015-08-23 21:54:52 +00:00
return nil
2015-08-23 02:31:22 +00:00
}