func main() {
var host = flag.String("H", "", "host (or hosts) to act on")
var role = flag.String("R", "", "role (or roles) to act on")
var proxy = flag.String("P", "127.0.0.1", "agent to interact with")
var all = flag.Bool("A", false, "act on all live nodes")
var fserial = flag.Bool("s", false, "print output when commands finish")
var fbinary = flag.Bool("b", false, "force binary mode output")
var ftext = flag.Bool("l", false, "force line mode output")
var glock = flag.String("G", "", "global lock to claim")
var llock = flag.String("L", "", "local lock to claim")
var dzrhost = flag.String("doozer", "localhost:8046",
"host:port for doozer")
var jobs = flag.Int("j", 0, "jobs to run in parallel")
var tout = flag.Int("t", 20, "timeout (in seconds) for jobs")
var fnowarn = flag.Bool("w", false, "suppress error text output")
flag.Usage = clientUsage
flag.Parse()
timeout = *tout
if timeout < 0 {
log.Error("timeout must be 0 (no timeout) or positive")
os.Exit(1)
}
nowarn = *fnowarn
// The output selection modes. By default, we assume that if the output
// is from a single machine, it's binary and we don't touch it. However,
// if the user is running against multiple targets, we assume that the
// output is line-based and we prefix the hostname returning each line.
//
// Line based defaults to interleaved. The serial flag can be used to
// ask us to buffer each host's output and then dump it all at once when
// that host is done. This is more useful for doing a batch job and having
// easily read output later, but still running commands in parallel.
//
// Finally, the user can force binary mode on multi-host outputs, which
// makes us not touch the output.
serial = *fserial
binary = *fbinary && !*ftext
// Uses the nice golog package to handle logging arguments and flags
// so we don't have to worry about it.
log = logging.InitFromFlags()
safedoozer.SetLogger(log)
// Connect to our doozer host. We have to do this early because alias
// validation requires doozer.
dzr = safedoozer.Dial(*dzrhost)
defer dzr.Close()
// Do some simple argument validation.
args := expandAlias(flag.Args())
if len(args) == 0 {
flag.Usage()
os.Exit(1)
}
if !isValidCommand(args[0], args[1:]) {
os.Exit(1)
}
// Figure out localhost.
var err error // If we use := below, we shadow the global, which is bad.
hostname, err = os.Hostname()
if err != nil {
log.Error("failed getting hostname: %s", err)
os.Exit(1)
}
hostname = strings.Split(hostname, ".")[0]
if len(hostname) <= 0 {
log.Error("hostname is empty!")
os.Exit(1)
}
zmq_ctx, err = zmq.NewContext()
if err != nil {
log.Error("failed to init zmq: %s", err)
os.Exit(1)
}
defer zmq_ctx.Close()
// Connect to the proxy agent. This is the agent we will be having do all
// of the work for us. This is probably localhost.
psock = socketForIp(*proxy)
if psock == nil {
log.Error("unable to connect to proxy agent")
os.Exit(1)
}
defer (*psock).Close()
// Determine which nodes we will be addressing.
hosts := make(map[string]bool)
if *all {
for _, host := range nodes() {
hosts[host] = false
}
} else {
if *host != "" {
lhosts := strings.Split(*host, ",")
for _, host := range lhosts {
host = strings.TrimSpace(host)
if len(host) > 0 {
hosts[strings.TrimSpace(host)] = false
}
}
}
if *role != "" {
roles := strings.Split(*role, ",")
for _, role := range roles {
lhosts := convertRoleToHosts(strings.TrimSpace(role))
for _, host := range lhosts {
hosts[host] = false
}
}
}
if *host == "" && *role == "" {
// Both empty, default to this machine.
hosts[hostname] = false
}
}
// If no hosts, bail out.
if len(hosts) <= 0 {
log.Error("no hosts or roles specified")
os.Exit(1)
} else if len(hosts) == 1 {
// If we're targetting a single machine, we want to use binary mode by
// default unless the user explicitly set text mode.
binary = true && !*ftext
}
// If we have been told to get a global lock, let's try to get that now.
if *glock != "" {
resp := proxyCommand("global_lock", *glock)
if resp != "locked" {
log.Error("failed to get global lock %s: %s", *glock, resp)
os.Exit(1)
}
defer func(lock string) {
resp := proxyCommand("global_unlock", *glock)
if resp != "unlocked" {
log.Error("failed to release global lock %s: %s", *glock, resp)
}
}(*glock)
}
// Same, local.
if *llock != "" {
resp := proxyCommand("local_lock", *llock)
if resp != "locked" {
log.Error("failed to get local lock %s: %s", *llock, resp)
os.Exit(1)
}
defer func(lock string) {
resp := proxyCommand("local_unlock", *llock)
if resp != "unlocked" {
log.Error("failed to release local lock %s: %s", *llock, resp)
}
}(*llock)
}
// There are some commands which are client-only and don't run against a
// given set of hosts. For these, just execute locally and then we're done.
switch args[0] {
case "roles":
cmdRoles()
case "hosts":
cmdHosts()
case "alias":
cmdAlias(args[1:])
}
// Queue up the jobs and then execute them.
for host, _ := range hosts {
queueJob(host, args)
}
runJobs(*jobs) // Returns when jobs are done.
}
func main() {
var myhost = flag.String("hostname", "", "this machine's hostname")
var myport = flag.Int("port", 7330, "port number to listen on")
var dzrhost = flag.String("doozer", "localhost:8046",
"host:port for doozer")
flag.Parse()
// Uses the nice golog package to handle logging arguments and flags
// so we don't have to worry about it.
log = logging.InitFromFlags()
safedoozer.SetLogger(log)
log.Info("starting up - gid %s", gid)
rand.Seed(int64(time.Now().Nanosecond())) // Not the best but ok?
dzr = safedoozer.Dial(*dzrhost)
defer dzr.Close()
defer removeGlobalLocks()
// see config.go for this
initializeConfig()
var err error // If we use := below, we shadow the global, which is bad.
zmq_ctx, err = zmq.NewContext()
if err != nil {
log.Fatal("failed to init zmq: %s", err)
}
defer zmq_ctx.Close()
myinfo, err := getSelfInfo()
if err != nil {
log.Fatal("Failed getting local information: %s", err)
}
if *myhost == "" {
_, ok := myinfo["hostname"]
if !ok {
log.Fatal("getSelfInfo() did not return a hostname")
}
hostname = myinfo["hostname"]
} else {
log.Warn("user requested hostname override (command line argument)")
hostname = *myhost
}
// Global, easy to access variable since we use it everywhere. This is safe
// because it's read-only, too -- well, by definition. It isn't really. But
// if it changes in maintainInfo then we exit.
log.Info("client starting with hostname %s", hostname)
// Now we have enough information to see if anybody else is claiming to be
// this particular node. If so, we want to wait a bit and see if they
// update again. If they are updating, then it is assumed they are running
// okay, and we shouldn't start up again.
lock := "/s/nlock/" + hostname
rev := dzr.Stat(lock, nil)
// If the lock is claimed, attempt to wait and see if the remote seems
// to still be alive.
if rev > 0 {
log.Warn("node lock is claimed, waiting to see if it's lively")
time.Sleep(3 * time.Second)
nrev := dzr.Stat(lock, nil)
if nrev > rev {
log.Fatal("lock is lively, we can't continue!")
}
}
// Safe to claim the node lock. Let's get it.
log.Info("attempting to get the node lock")
nrev := dzr.Set(lock, rev, fmt.Sprintf("%d", time.Now().Unix()))
if nrev <= rev {
log.Fatal("failed to obtain the lock")
}
log.Info("lock successfully obtained! we are lively.")
go maintainLock(lock, nrev)
// Now we want to reset the gid map, asserting that we are now the living
// agent for this host.
lock = "/s/gid/" + hostname
rev = dzr.Stat(lock, nil)
dzr.Set(lock, rev, gid)
// There are certain miscellaneous tasks that need to get done somewhere,
// so we use global locks to make sure that somebody is doing them. We
// don't really care who.
go manageGlobalFunc(5, "gf.expire-hosts", gfExpireHosts)
go manageGlobalFunc(5, "gf.expire-glocks", gfExpireGlobalLocks)
// Personal maintenance here.
go maintainInfo(&myinfo)
go maintainStanzas()
runAgent(*myport) // Returns when dead.
}