open-consul/agent/consul/rpc.go
Jeff Mitchell e0068431f5 Chunking support (#6172)
* Initial chunk support

This uses the go-raft-middleware library to allow for chunked commits to the KV
2019-07-24 17:06:39 -04:00

554 lines
16 KiB
Go

package consul
import (
"crypto/tls"
"errors"
"fmt"
"io"
"net"
"strings"
"time"
"github.com/armon/go-metrics"
"github.com/hashicorp/consul/agent/consul/state"
"github.com/hashicorp/consul/agent/metadata"
"github.com/hashicorp/consul/agent/pool"
"github.com/hashicorp/consul/agent/structs"
"github.com/hashicorp/consul/lib"
memdb "github.com/hashicorp/go-memdb"
"github.com/hashicorp/go-raftchunking"
"github.com/hashicorp/memberlist"
msgpackrpc "github.com/hashicorp/net-rpc-msgpackrpc"
"github.com/hashicorp/raft"
"github.com/hashicorp/yamux"
)
const (
// maxQueryTime is used to bound the limit of a blocking query
maxQueryTime = 600 * time.Second
// defaultQueryTime is the amount of time we block waiting for a change
// if no time is specified. Previously we would wait the maxQueryTime.
defaultQueryTime = 300 * time.Second
// jitterFraction is a the limit to the amount of jitter we apply
// to a user specified MaxQueryTime. We divide the specified time by
// the fraction. So 16 == 6.25% limit of jitter. This same fraction
// is applied to the RPCHoldTimeout
jitterFraction = 16
// Warn if the Raft command is larger than this.
// If it's over 1MB something is probably being abusive.
raftWarnSize = 1024 * 1024
// enqueueLimit caps how long we will wait to enqueue
// a new Raft command. Something is probably wrong if this
// value is ever reached. However, it prevents us from blocking
// the requesting goroutine forever.
enqueueLimit = 30 * time.Second
)
var (
ErrChunkingResubmit = errors.New("please resubmit call for rechunking")
)
// listen is used to listen for incoming RPC connections
func (s *Server) listen(listener net.Listener) {
for {
// Accept a connection
conn, err := listener.Accept()
if err != nil {
if s.shutdown {
return
}
s.logger.Printf("[ERR] consul.rpc: failed to accept RPC conn: %v", err)
continue
}
go s.handleConn(conn, false)
metrics.IncrCounter([]string{"rpc", "accept_conn"}, 1)
}
}
// logConn is a wrapper around memberlist's LogConn so that we format references
// to "from" addresses in a consistent way. This is just a shorter name.
func logConn(conn net.Conn) string {
return memberlist.LogConn(conn)
}
// handleConn is used to determine if this is a Raft or
// Consul type RPC connection and invoke the correct handler
func (s *Server) handleConn(conn net.Conn, isTLS bool) {
// Read a single byte
buf := make([]byte, 1)
if _, err := conn.Read(buf); err != nil {
if err != io.EOF {
s.logger.Printf("[ERR] consul.rpc: failed to read byte: %v %s", err, logConn(conn))
}
conn.Close()
return
}
typ := pool.RPCType(buf[0])
// Enforce TLS if VerifyIncoming is set
if s.tlsConfigurator.VerifyIncomingRPC() && !isTLS && typ != pool.RPCTLS && typ != pool.RPCTLSInsecure {
s.logger.Printf("[WARN] consul.rpc: Non-TLS connection attempted with VerifyIncoming set %s", logConn(conn))
conn.Close()
return
}
// Switch on the byte
switch typ {
case pool.RPCConsul:
s.handleConsulConn(conn)
case pool.RPCRaft:
metrics.IncrCounter([]string{"rpc", "raft_handoff"}, 1)
s.raftLayer.Handoff(conn)
case pool.RPCTLS:
conn = tls.Server(conn, s.tlsConfigurator.IncomingRPCConfig())
s.handleConn(conn, true)
case pool.RPCMultiplexV2:
s.handleMultiplexV2(conn)
case pool.RPCSnapshot:
s.handleSnapshotConn(conn)
case pool.RPCTLSInsecure:
conn = tls.Server(conn, s.tlsConfigurator.IncomingInsecureRPCConfig())
s.handleInsecureConn(conn)
default:
if !s.handleEnterpriseRPCConn(typ, conn, isTLS) {
s.logger.Printf("[ERR] consul.rpc: unrecognized RPC byte: %v %s", typ, logConn(conn))
conn.Close()
}
}
}
// handleMultiplexV2 is used to multiplex a single incoming connection
// using the Yamux multiplexer
func (s *Server) handleMultiplexV2(conn net.Conn) {
defer conn.Close()
conf := yamux.DefaultConfig()
conf.LogOutput = s.config.LogOutput
server, _ := yamux.Server(conn, conf)
for {
sub, err := server.Accept()
if err != nil {
if err != io.EOF {
s.logger.Printf("[ERR] consul.rpc: multiplex conn accept failed: %v %s", err, logConn(conn))
}
return
}
go s.handleConsulConn(sub)
}
}
// handleConsulConn is used to service a single Consul RPC connection
func (s *Server) handleConsulConn(conn net.Conn) {
defer conn.Close()
rpcCodec := msgpackrpc.NewServerCodec(conn)
for {
select {
case <-s.shutdownCh:
return
default:
}
if err := s.rpcServer.ServeRequest(rpcCodec); err != nil {
if err != io.EOF && !strings.Contains(err.Error(), "closed") {
s.logger.Printf("[ERR] consul.rpc: RPC error: %v %s", err, logConn(conn))
metrics.IncrCounter([]string{"rpc", "request_error"}, 1)
}
return
}
metrics.IncrCounter([]string{"rpc", "request"}, 1)
}
}
// handleInsecureConsulConn is used to service a single Consul INSECURERPC connection
func (s *Server) handleInsecureConn(conn net.Conn) {
defer conn.Close()
rpcCodec := msgpackrpc.NewServerCodec(conn)
for {
select {
case <-s.shutdownCh:
return
default:
}
if err := s.insecureRPCServer.ServeRequest(rpcCodec); err != nil {
if err != io.EOF && !strings.Contains(err.Error(), "closed") {
s.logger.Printf("[ERR] consul.rpc: INSECURERPC error: %v %s", err, logConn(conn))
metrics.IncrCounter([]string{"rpc", "request_error"}, 1)
}
return
}
metrics.IncrCounter([]string{"rpc", "request"}, 1)
}
}
// handleSnapshotConn is used to dispatch snapshot saves and restores, which
// stream so don't use the normal RPC mechanism.
func (s *Server) handleSnapshotConn(conn net.Conn) {
go func() {
defer conn.Close()
if err := s.handleSnapshotRequest(conn); err != nil {
s.logger.Printf("[ERR] consul.rpc: Snapshot RPC error: %v %s", err, logConn(conn))
}
}()
}
// canRetry returns true if the given situation is safe for a retry.
func canRetry(args interface{}, err error) bool {
// No leader errors are always safe to retry since no state could have
// been changed.
if structs.IsErrNoLeader(err) {
return true
}
// If we are chunking and it doesn't seem to have completed, try again
intErr, ok := args.(error)
if ok && strings.Contains(intErr.Error(), ErrChunkingResubmit.Error()) {
return true
}
// Reads are safe to retry for stream errors, such as if a server was
// being shut down.
info, ok := args.(structs.RPCInfo)
if ok && info.IsRead() && lib.IsErrEOF(err) {
return true
}
return false
}
// forward is used to forward to a remote DC or to forward to the local leader
// Returns a bool of if forwarding was performed, as well as any error
func (s *Server) forward(method string, info structs.RPCInfo, args interface{}, reply interface{}) (bool, error) {
var firstCheck time.Time
// Handle DC forwarding
dc := info.RequestDatacenter()
if dc != s.config.Datacenter {
err := s.forwardDC(method, dc, args, reply)
return true, err
}
// Check if we can allow a stale read, ensure our local DB is initialized
if info.IsRead() && info.AllowStaleRead() && !s.raft.LastContact().IsZero() {
return false, nil
}
CHECK_LEADER:
// Fail fast if we are in the process of leaving
select {
case <-s.leaveCh:
return true, structs.ErrNoLeader
default:
}
// Find the leader
isLeader, leader := s.getLeader()
// Handle the case we are the leader
if isLeader {
return false, nil
}
// Handle the case of a known leader
rpcErr := structs.ErrNoLeader
if leader != nil {
rpcErr = s.connPool.RPC(s.config.Datacenter, leader.Addr,
leader.Version, method, leader.UseTLS, args, reply)
if rpcErr != nil && canRetry(info, rpcErr) {
goto RETRY
}
return true, rpcErr
}
RETRY:
// Gate the request until there is a leader
if firstCheck.IsZero() {
firstCheck = time.Now()
}
if time.Since(firstCheck) < s.config.RPCHoldTimeout {
jitter := lib.RandomStagger(s.config.RPCHoldTimeout / jitterFraction)
select {
case <-time.After(jitter):
goto CHECK_LEADER
case <-s.leaveCh:
case <-s.shutdownCh:
}
}
// No leader found and hold time exceeded
return true, rpcErr
}
// getLeader returns if the current node is the leader, and if not then it
// returns the leader which is potentially nil if the cluster has not yet
// elected a leader.
func (s *Server) getLeader() (bool, *metadata.Server) {
// Check if we are the leader
if s.IsLeader() {
return true, nil
}
// Get the leader
leader := s.raft.Leader()
if leader == "" {
return false, nil
}
// Lookup the server
server := s.serverLookup.Server(leader)
// Server could be nil
return false, server
}
// forwardDC is used to forward an RPC call to a remote DC, or fail if no servers
func (s *Server) forwardDC(method, dc string, args interface{}, reply interface{}) error {
manager, server, ok := s.router.FindRoute(dc)
if !ok {
s.logger.Printf("[WARN] consul.rpc: RPC request for DC %q, no path found", dc)
return structs.ErrNoDCPath
}
metrics.IncrCounterWithLabels([]string{"rpc", "cross-dc"}, 1,
[]metrics.Label{{Name: "datacenter", Value: dc}})
if err := s.connPool.RPC(dc, server.Addr, server.Version, method, server.UseTLS, args, reply); err != nil {
manager.NotifyFailedServer(server)
s.logger.Printf("[ERR] consul: RPC failed to server %s in DC %q: %v", server.Addr, dc, err)
return err
}
return nil
}
// globalRPC is used to forward an RPC request to one server in each datacenter.
// This will only error for RPC-related errors. Otherwise, application-level
// errors can be sent in the response objects.
func (s *Server) globalRPC(method string, args interface{},
reply structs.CompoundResponse) error {
// Make a new request into each datacenter
dcs := s.router.GetDatacenters()
replies, total := 0, len(dcs)
errorCh := make(chan error, total)
respCh := make(chan interface{}, total)
for _, dc := range dcs {
go func(dc string) {
rr := reply.New()
if err := s.forwardDC(method, dc, args, &rr); err != nil {
errorCh <- err
return
}
respCh <- rr
}(dc)
}
for replies < total {
select {
case err := <-errorCh:
return err
case rr := <-respCh:
reply.Add(rr)
replies++
}
}
return nil
}
// raftApply is used to encode a message, run it through raft, and return
// the FSM response along with any errors
func (s *Server) raftApply(t structs.MessageType, msg interface{}) (interface{}, error) {
buf, err := structs.Encode(t, msg)
if err != nil {
return nil, fmt.Errorf("Failed to encode request: %v", err)
}
// Warn if the command is very large
if n := len(buf); n > raftWarnSize {
s.logger.Printf("[WARN] consul: Attempting to apply large raft entry (%d bytes)", n)
}
var chunked bool
var future raft.ApplyFuture
switch {
case len(buf) <= raft.SuggestedMaxDataSize || t != structs.KVSRequestType:
future = s.raft.Apply(buf, enqueueLimit)
default:
chunked = true
future = raftchunking.ChunkingApply(buf, nil, enqueueLimit, s.raft.ApplyLog)
}
if err := future.Error(); err != nil {
return nil, err
}
resp := future.Response()
if chunked {
// In this case we didn't apply all chunks successfully, possibly due
// to a term change; resubmit
if resp == nil {
// This returns the error in the interface because the raft library
// returns errors from the FSM via the future, not via err from the
// apply function. Downstream client code expects to see any error
// from the FSM (as opposed to the apply itself) and decide whether
// it can retry in the future's response.
return ErrChunkingResubmit, nil
}
// We expect that this conversion should always work
chunkedSuccess, ok := resp.(raftchunking.ChunkingSuccess)
if !ok {
return nil, errors.New("unknown type of response back from chunking FSM")
}
// Return the inner wrapped response
return chunkedSuccess.Response, nil
}
return resp, nil
}
// queryFn is used to perform a query operation. If a re-query is needed, the
// passed-in watch set will be used to block for changes. The passed-in state
// store should be used (vs. calling fsm.State()) since the given state store
// will be correctly watched for changes if the state store is restored from
// a snapshot.
type queryFn func(memdb.WatchSet, *state.Store) error
// blockingQuery is used to process a potentially blocking query operation.
func (s *Server) blockingQuery(queryOpts *structs.QueryOptions, queryMeta *structs.QueryMeta,
fn queryFn) error {
var timeout *time.Timer
// Fast path right to the non-blocking query.
if queryOpts.MinQueryIndex == 0 {
goto RUN_QUERY
}
// Restrict the max query time, and ensure there is always one.
if queryOpts.MaxQueryTime > maxQueryTime {
queryOpts.MaxQueryTime = maxQueryTime
} else if queryOpts.MaxQueryTime <= 0 {
queryOpts.MaxQueryTime = defaultQueryTime
}
// Apply a small amount of jitter to the request.
queryOpts.MaxQueryTime += lib.RandomStagger(queryOpts.MaxQueryTime / jitterFraction)
// Setup a query timeout.
timeout = time.NewTimer(queryOpts.MaxQueryTime)
defer timeout.Stop()
RUN_QUERY:
// Update the query metadata.
s.setQueryMeta(queryMeta)
// If the read must be consistent we verify that we are still the leader.
if queryOpts.RequireConsistent {
if err := s.consistentRead(); err != nil {
return err
}
}
// Run the query.
metrics.IncrCounter([]string{"rpc", "query"}, 1)
// Operate on a consistent set of state. This makes sure that the
// abandon channel goes with the state that the caller is using to
// build watches.
state := s.fsm.State()
// We can skip all watch tracking if this isn't a blocking query.
var ws memdb.WatchSet
if queryOpts.MinQueryIndex > 0 {
ws = memdb.NewWatchSet()
// This channel will be closed if a snapshot is restored and the
// whole state store is abandoned.
ws.Add(state.AbandonCh())
}
// Block up to the timeout if we didn't see anything fresh.
err := fn(ws, state)
// Note we check queryOpts.MinQueryIndex is greater than zero to determine if
// blocking was requested by client, NOT meta.Index since the state function
// might return zero if something is not initialized and care wasn't taken to
// handle that special case (in practice this happened a lot so fixing it
// systematically here beats trying to remember to add zero checks in every
// state method). We also need to ensure that unless there is an error, we
// return an index > 0 otherwise the client will never block and burn CPU and
// requests.
if err == nil && queryMeta.Index < 1 {
queryMeta.Index = 1
}
if err == nil && queryOpts.MinQueryIndex > 0 && queryMeta.Index <= queryOpts.MinQueryIndex {
if expired := ws.Watch(timeout.C); !expired {
// If a restore may have woken us up then bail out from
// the query immediately. This is slightly race-ey since
// this might have been interrupted for other reasons,
// but it's OK to kick it back to the caller in either
// case.
select {
case <-state.AbandonCh():
default:
goto RUN_QUERY
}
}
}
return err
}
// setQueryMeta is used to populate the QueryMeta data for an RPC call
func (s *Server) setQueryMeta(m *structs.QueryMeta) {
if s.IsLeader() {
m.LastContact = 0
m.KnownLeader = true
} else {
m.LastContact = time.Since(s.raft.LastContact())
m.KnownLeader = (s.raft.Leader() != "")
}
}
// consistentRead is used to ensure we do not perform a stale
// read. This is done by verifying leadership before the read.
func (s *Server) consistentRead() error {
defer metrics.MeasureSince([]string{"rpc", "consistentRead"}, time.Now())
future := s.raft.VerifyLeader()
if err := future.Error(); err != nil {
return err //fail fast if leader verification fails
}
// poll consistent read readiness, wait for up to RPCHoldTimeout milliseconds
if s.isReadyForConsistentReads() {
return nil
}
jitter := lib.RandomStagger(s.config.RPCHoldTimeout / jitterFraction)
deadline := time.Now().Add(s.config.RPCHoldTimeout)
for time.Now().Before(deadline) {
select {
case <-time.After(jitter):
// Drop through and check before we loop again.
case <-s.shutdownCh:
return fmt.Errorf("shutdown waiting for leader")
}
if s.isReadyForConsistentReads() {
return nil
}
}
return structs.ErrNotReadyForConsistentReads
}