open-consul/agent/consul/rpc.go

1100 lines
33 KiB
Go
Raw Normal View History

2013-12-07 00:54:33 +00:00
package consul
import (
"context"
"crypto/tls"
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
"encoding/binary"
"errors"
"fmt"
2013-12-11 22:57:40 +00:00
"io"
2013-12-07 00:54:33 +00:00
"net"
"strings"
"sync/atomic"
"time"
"github.com/armon/go-metrics"
"github.com/armon/go-metrics/prometheus"
connlimit "github.com/hashicorp/go-connlimit"
"github.com/hashicorp/go-hclog"
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"
"google.golang.org/grpc"
"github.com/hashicorp/consul/acl"
"github.com/hashicorp/consul/agent/consul/state"
"github.com/hashicorp/consul/agent/consul/wanfed"
"github.com/hashicorp/consul/agent/metadata"
"github.com/hashicorp/consul/agent/pool"
"github.com/hashicorp/consul/agent/structs"
"github.com/hashicorp/consul/lib"
"github.com/hashicorp/consul/logging"
2013-12-07 00:54:33 +00:00
)
var RPCCounters = []prometheus.CounterDefinition{
{
Name: []string{"rpc", "accept_conn"},
2020-11-16 19:02:11 +00:00
Help: "Increments when a server accepts an RPC connection.",
},
{
Name: []string{"rpc", "raft_handoff"},
2020-11-16 19:02:11 +00:00
Help: "Increments when a server accepts a Raft-related RPC connection.",
},
{
Name: []string{"rpc", "request_error"},
2020-11-16 19:02:11 +00:00
Help: "Increments when a server returns an error from an RPC request.",
},
{
Name: []string{"rpc", "request"},
2020-11-16 19:02:11 +00:00
Help: "Increments when a server receives a Consul-related RPC request.",
},
{
Name: []string{"rpc", "cross-dc"},
2020-11-16 19:02:11 +00:00
Help: "Increments when a server sends a (potentially blocking) cross datacenter RPC query.",
},
{
Name: []string{"rpc", "query"},
Help: "Increments when a server receives a read request, indicating the rate of new read queries.",
},
}
var RPCGauges = []prometheus.GaugeDefinition{
{
Name: []string{"rpc", "queries_blocking"},
2020-11-16 19:02:11 +00:00
Help: "Shows the current number of in-flight blocking queries the server is handling.",
},
}
var RPCSummaries = []prometheus.SummaryDefinition{
{
Name: []string{"rpc", "consistentRead"},
2020-11-16 19:02:11 +00:00
Help: "Measures the time spent confirming that a consistent read can be performed.",
},
}
const (
// 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")
// partitionUnsetter is used to describe requests values that can unset their
// EnterpriseMeta.Partition value.
type partitionUnsetter interface {
// UnsetPartition is used to strip a Partition value from the request before
// it is forwarded to a remote datacenter. By unsetting the value, the server
// that handles the request can decide which partition should be used (or do nothing).
// This ensures that servers that are Partition-enabled (pre-1.11, or non-Enterprise)
// don't inadvertently cause servers that are not Partition-enabled (<= 1.10 or non-Enterprise)
// to filter their responses by Partition. In other words, this ensures upgraded servers
// remain compatible with non-upgraded servers.
UnsetPartition()
}
func (s *Server) rpcLogger() hclog.Logger {
return s.loggers.Named(logging.RPC)
}
2013-12-07 00:54:33 +00:00
// listen is used to listen for incoming RPC connections
func (s *Server) listen(listener net.Listener) {
2013-12-07 00:54:33 +00:00
for {
// Accept a connection
conn, err := listener.Accept()
2013-12-07 00:54:33 +00:00
if err != nil {
if s.shutdown {
return
}
s.rpcLogger().Error("failed to accept RPC conn", "error", err)
2013-12-07 00:54:33 +00:00
continue
}
free, err := s.rpcConnLimiter.Accept(conn)
if err != nil {
2020-02-12 15:37:16 +00:00
s.rpcLogger().Error("rejecting RPC conn from because rpc_max_conns_per_client exceeded", "conn", logConn(conn))
conn.Close()
continue
}
// Wrap conn so it will be auto-freed from conn limiter when it closes.
conn = connlimit.Wrap(conn, free)
go s.handleConn(conn, false)
metrics.IncrCounter([]string{"rpc", "accept_conn"}, 1)
2013-12-07 00:54:33 +00:00
}
}
// 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) {
// Limit how long the client can hold the connection open before they send the
// magic byte (and authenticate when mTLS is enabled). If `isTLS == true` then
// this also enforces a timeout on how long it takes for the handshake to
// complete since tls.Conn.Read implicitly calls Handshake().
if s.config.RPCHandshakeTimeout > 0 {
conn.SetReadDeadline(time.Now().Add(s.config.RPCHandshakeTimeout))
}
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
if !isTLS && s.tlsConfigurator.MutualTLSCapable() {
// See if actually this is native TLS multiplexed onto the old
// "type-byte" system.
peekedConn, nativeTLS, err := pool.PeekForTLS(conn)
if err != nil {
if err != io.EOF {
s.rpcLogger().Error(
"failed to read first byte",
"conn", logConn(conn),
"error", err,
)
}
conn.Close()
return
}
if nativeTLS {
s.handleNativeTLS(peekedConn)
return
}
conn = peekedConn
}
// Read a single byte
buf := make([]byte, 1)
if _, err := conn.Read(buf); err != nil {
if err != io.EOF {
s.rpcLogger().Error("failed to read byte",
"conn", logConn(conn),
"error", err,
)
}
conn.Close()
return
}
typ := pool.RPCType(buf[0])
// Reset the deadline as we aren't sure what is expected next - it depends on
// the protocol.
if s.config.RPCHandshakeTimeout > 0 {
conn.SetReadDeadline(time.Time{})
}
// Enforce TLS if VerifyIncoming is set
if s.tlsConfigurator.VerifyIncomingRPC() && !isTLS && typ != pool.RPCTLS && typ != pool.RPCTLSInsecure {
s.rpcLogger().Warn("Non-TLS connection attempted with VerifyIncoming set", "conn", logConn(conn))
conn.Close()
return
}
// Switch on the byte
switch typ {
case pool.RPCConsul:
s.handleConsulConn(conn)
case pool.RPCRaft:
2021-08-26 22:04:32 +00:00
s.handleRaftRPC(conn)
case pool.RPCTLS:
// Don't allow malicious client to create TLS-in-TLS for ever.
if isTLS {
2020-02-12 15:37:16 +00:00
s.rpcLogger().Error("TLS connection attempting to establish inner TLS connection", "conn", logConn(conn))
conn.Close()
return
}
conn = tls.Server(conn, s.tlsConfigurator.IncomingRPCConfig())
s.handleConn(conn, true)
case pool.RPCMultiplexV2:
s.handleMultiplexV2(conn)
case pool.RPCSnapshot:
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
s.handleSnapshotConn(conn)
case pool.RPCTLSInsecure:
// Don't allow malicious client to create TLS-in-TLS for ever.
if isTLS {
2020-02-12 15:37:16 +00:00
s.rpcLogger().Error("TLS connection attempting to establish inner TLS connection", "conn", logConn(conn))
conn.Close()
return
}
conn = tls.Server(conn, s.tlsConfigurator.IncomingInsecureRPCConfig())
s.handleInsecureConn(conn)
case pool.RPCGRPC:
s.grpcHandler.Handle(conn)
default:
if !s.handleEnterpriseRPCConn(typ, conn, isTLS) {
s.rpcLogger().Error("unrecognized RPC byte",
"byte", typ,
"conn", logConn(conn),
)
conn.Close()
}
}
}
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
func (s *Server) handleNativeTLS(conn net.Conn) {
s.rpcLogger().Trace(
"detected actual TLS over RPC port",
"conn", logConn(conn),
)
tlscfg := s.tlsConfigurator.IncomingALPNRPCConfig(pool.RPCNextProtos)
tlsConn := tls.Server(conn, tlscfg)
// Force the handshake to conclude.
if err := tlsConn.Handshake(); err != nil {
s.rpcLogger().Error(
"TLS handshake failed",
"conn", logConn(conn),
"error", err,
)
conn.Close()
return
}
// Reset the deadline as we aren't sure what is expected next - it depends on
// the protocol.
if s.config.RPCHandshakeTimeout > 0 {
conn.SetReadDeadline(time.Time{})
}
var (
cs = tlsConn.ConnectionState()
sni = cs.ServerName
nextProto = cs.NegotiatedProtocol
transport = s.memberlistTransportWAN
)
s.rpcLogger().Trace(
"accepted nativeTLS RPC",
"sni", sni,
"protocol", nextProto,
"conn", logConn(conn),
)
switch nextProto {
case pool.ALPN_RPCConsul:
s.handleConsulConn(tlsConn)
case pool.ALPN_RPCRaft:
2021-08-26 22:04:32 +00:00
s.handleRaftRPC(tlsConn)
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
case pool.ALPN_RPCMultiplexV2:
s.handleMultiplexV2(tlsConn)
case pool.ALPN_RPCSnapshot:
s.handleSnapshotConn(tlsConn)
case pool.ALPN_RPCGRPC:
s.grpcHandler.Handle(tlsConn)
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
case pool.ALPN_WANGossipPacket:
if err := s.handleALPN_WANGossipPacketStream(tlsConn); err != nil && err != io.EOF {
s.rpcLogger().Error(
"failed to ingest RPC",
"sni", sni,
"protocol", nextProto,
"conn", logConn(conn),
"error", err,
)
}
case pool.ALPN_WANGossipStream:
// No need to defer the conn.Close() here, the Ingest methods do that.
if err := transport.IngestStream(tlsConn); err != nil {
s.rpcLogger().Error(
"failed to ingest RPC",
"sni", sni,
"protocol", nextProto,
"conn", logConn(conn),
"error", err,
)
}
default:
if !s.handleEnterpriseNativeTLSConn(nextProto, conn) {
s.rpcLogger().Error(
"discarding RPC for unknown negotiated protocol",
"failed to ingest RPC",
"protocol", nextProto,
"conn", 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()
2014-05-28 23:32:10 +00:00
conf := yamux.DefaultConfig()
// override the default because LogOutput conflicts with Logger
conf.LogOutput = nil
// TODO: should this be created once and cached?
conf.Logger = s.logger.StandardLogger(&hclog.StandardLoggerOptions{InferLevels: true})
2014-05-28 23:32:10 +00:00
server, _ := yamux.Server(conn, conf)
for {
sub, err := server.Accept()
if err != nil {
if err != io.EOF {
s.rpcLogger().Error("multiplex conn accept failed",
"conn", logConn(conn),
"error", err,
)
}
return
}
// In the beginning only RPC was supposed to be multiplexed
// with yamux. In order to add the ability to multiplex network
// area connections, this workaround was added.
// This code peeks the first byte and checks if it is
// RPCGossip, in which case this is handled by enterprise code.
// Otherwise this connection is handled like before by the RPC
// handler.
// This wouldn't work if a normal RPC could start with
// RPCGossip(6). In messagepack a 6 encodes a positive fixint:
// https://github.com/msgpack/msgpack/blob/master/spec.md.
// None of the RPCs we are doing starts with that, usually it is
// a string for datacenter.
peeked, first, err := pool.PeekFirstByte(sub)
if err != nil {
s.rpcLogger().Error("Problem peeking connection", "conn", logConn(sub), "err", err)
sub.Close()
return
}
sub = peeked
switch first {
case byte(pool.RPCGossip):
buf := make([]byte, 1)
sub.Read(buf)
go func() {
if !s.handleEnterpriseRPCConn(pool.RPCGossip, sub, false) {
s.rpcLogger().Error("unrecognized RPC byte",
"byte", pool.RPCGossip,
"conn", logConn(conn),
)
sub.Close()
}
}()
default:
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.NewCodecFromHandle(true, true, conn, structs.MsgpackHandle)
for {
select {
case <-s.shutdownCh:
return
default:
}
2013-12-07 00:54:33 +00:00
if err := s.rpcServer.ServeRequest(rpcCodec); err != nil {
if err != io.EOF && !strings.Contains(err.Error(), "closed") {
s.rpcLogger().Error("RPC error",
"conn", logConn(conn),
"error", err,
)
metrics.IncrCounter([]string{"rpc", "request_error"}, 1)
2013-12-11 22:57:40 +00:00
}
2013-12-07 00:54:33 +00:00
return
}
metrics.IncrCounter([]string{"rpc", "request"}, 1)
2013-12-07 00:54:33 +00:00
}
}
// handleInsecureConsulConn is used to service a single Consul INSECURERPC connection
func (s *Server) handleInsecureConn(conn net.Conn) {
defer conn.Close()
rpcCodec := msgpackrpc.NewCodecFromHandle(true, true, conn, structs.MsgpackHandle)
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.rpcLogger().Error("INSECURERPC error",
"conn", logConn(conn),
"error", err,
)
metrics.IncrCounter([]string{"rpc", "request_error"}, 1)
}
return
}
metrics.IncrCounter([]string{"rpc", "request"}, 1)
}
}
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
// 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.rpcLogger().Error("Snapshot RPC error",
"conn", logConn(conn),
"error", err,
)
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
}
}()
}
2021-08-26 22:04:32 +00:00
func (s *Server) handleRaftRPC(conn net.Conn) {
if tlsConn, ok := conn.(*tls.Conn); ok {
err := s.tlsConfigurator.AuthorizeServerConn(s.config.Datacenter, tlsConn)
if err != nil {
s.rpcLogger().Warn(err.Error(), "from", conn.RemoteAddr(), "operation", "raft RPC")
conn.Close()
return
}
}
metrics.IncrCounter([]string{"rpc", "raft_handoff"}, 1)
s.raftLayer.Handoff(conn)
}
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
func (s *Server) handleALPN_WANGossipPacketStream(conn net.Conn) error {
defer conn.Close()
transport := s.memberlistTransportWAN
for {
select {
case <-s.shutdownCh:
return nil
default:
}
// Note: if we need to change this format to have additional header
// information we can just negotiate a different ALPN protocol instead
// of needing any sort of version field here.
prefixLen, err := readUint32(conn, wanfed.GossipPacketMaxIdleTime)
if err != nil {
return err
}
// Avoid a memory exhaustion DOS vector here by capping how large this
// packet can be to something reasonable.
if prefixLen > wanfed.GossipPacketMaxByteSize {
return fmt.Errorf("gossip packet size %d exceeds threshold of %d", prefixLen, wanfed.GossipPacketMaxByteSize)
}
lc := &limitedConn{
Conn: conn,
lr: io.LimitReader(conn, int64(prefixLen)),
}
if err := transport.IngestPacket(lc, conn.RemoteAddr(), time.Now(), false); err != nil {
return err
}
}
}
func readUint32(conn net.Conn, timeout time.Duration) (uint32, error) {
// Since requests are framed we can easily just set a deadline on
// reading that frame and then disable it for the rest of the body.
if err := conn.SetReadDeadline(time.Now().Add(timeout)); err != nil {
return 0, err
}
var v uint32
if err := binary.Read(conn, binary.BigEndian, &v); err != nil {
return 0, err
}
if err := conn.SetReadDeadline(time.Time{}); err != nil {
return 0, err
}
return v, nil
}
type limitedConn struct {
net.Conn
lr io.Reader
}
func (c *limitedConn) Read(b []byte) (n int, err error) {
return c.lr.Read(b)
}
// canRetry returns true if the request and error indicate that a retry is safe.
func canRetry(info structs.RPCInfo, err error, start time.Time, config *Config) bool {
if info != nil && info.HasTimedOut(start, config.RPCHoldTimeout, config.MaxQueryTime, config.DefaultQueryTime) {
// RPCInfo timeout may include extra time for MaxQueryTime
return false
} else if info == nil && time.Since(start) > config.RPCHoldTimeout {
// When not RPCInfo, timeout is only RPCHoldTimeout
return false
}
// 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.
if err != nil && strings.Contains(err.Error(), ErrChunkingResubmit.Error()) {
return true
}
// Reads are safe to retry for stream errors, such as if a server was
// being shut down.
return info != nil && info.IsRead() && lib.IsErrEOF(err)
}
// ForwardRPC is used to potentially forward an RPC request to a remote DC or
// to the local leader depending upon the request.
//
// Returns a bool of if forwarding was performed, as well as any error. If
// false is returned (with no error) it is assumed that the current server
// should handle the request.
func (s *Server) ForwardRPC(method string, info structs.RPCInfo, reply interface{}) (bool, error) {
forwardToDC := func(dc string) error {
return s.forwardDC(method, dc, info, reply)
}
forwardToLeader := func(leader *metadata.Server) error {
return s.connPool.RPC(s.config.Datacenter, leader.ShortName, leader.Addr,
method, info, reply)
}
return s.forwardRPC(info, forwardToDC, forwardToLeader)
}
// ForwardGRPC is used to potentially forward an RPC request to a remote DC or
// to the local leader depending upon the request.
//
// Returns a bool of if forwarding was performed, as well as any error. If
// false is returned (with no error) it is assumed that the current server
// should handle the request.
func (s *Server) ForwardGRPC(connPool GRPCClientConner, info structs.RPCInfo, f func(*grpc.ClientConn) error) (handled bool, err error) {
forwardToDC := func(dc string) error {
conn, err := connPool.ClientConn(dc)
if err != nil {
return err
}
return f(conn)
}
forwardToLeader := func(leader *metadata.Server) error {
conn, err := connPool.ClientConnLeader()
if err != nil {
return err
}
return f(conn)
}
return s.forwardRPC(info, forwardToDC, forwardToLeader)
}
// forwardRPC is used to potentially forward an RPC request to a remote DC or
// to the local leader depending upon the request.
//
// If info.RequestDatacenter() does not match the local datacenter, then the
// request will be forwarded to the DC using forwardToDC.
//
// Stale read requests will be handled locally if the current node has an
// initialized raft database, otherwise requests will be forwarded to the local
// leader using forwardToLeader.
//
// Returns a bool of if forwarding was performed, as well as any error. If
// false is returned (with no error) it is assumed that the current server
// should handle the request.
func (s *Server) forwardRPC(
info structs.RPCInfo,
forwardToDC func(dc string) error,
forwardToLeader func(leader *metadata.Server) error,
) (handled bool, err error) {
// Forward the request to the requested datacenter.
if handled, err := s.forwardRequestToOtherDatacenter(info, forwardToDC); handled || err != nil {
return handled, err
}
// See if we should let this server handle the read request without
// shipping the request to the leader.
if s.canServeReadRequest(info) {
return false, nil
}
return s.forwardRequestToLeader(info, forwardToLeader)
}
// forwardRequestToOtherDatacenter is an implementation detail of forwardRPC.
// See the comment for forwardRPC for more details.
func (s *Server) forwardRequestToOtherDatacenter(info structs.RPCInfo, forwardToDC func(dc string) error) (handled bool, err error) {
// Handle DC forwarding
dc := info.RequestDatacenter()
if dc == "" {
dc = s.config.Datacenter
}
if dc != s.config.Datacenter {
// Local tokens only work within the current datacenter. Check to see
// if we are attempting to forward one to a remote datacenter and strip
// it, falling back on the anonymous token on the other end.
if token := info.TokenSecret(); token != "" {
done, ident, err := s.ResolveIdentityFromToken(token)
if done {
if err != nil && !acl.IsErrNotFound(err) {
return false, err
}
if ident != nil && ident.IsLocal() {
// Strip it from the request.
info.SetTokenSecret("")
defer info.SetTokenSecret(token)
}
}
}
// In order to interoperate with servers that can interpret Partition, but
// may not handle it correctly (eg. 1.10 servers), we need to unset the value.
// Unsetting the Partition ensures that the server that handles the request
// uses its Partition, or an empty value (aka doing nothing).
// For requests that are not Partition-aware, this is a no-op.
if v, ok := info.(partitionUnsetter); ok {
v.UnsetPartition()
}
return true, forwardToDC(dc)
}
return false, nil
}
// canServeReadRequest determines if the request is a stale read request and
// the current node can safely process that request.
func (s *Server) canServeReadRequest(info structs.RPCInfo) bool {
// Check if we can allow a stale read, ensure our local DB is initialized
return info.IsRead() && info.AllowStaleRead() && !s.raft.LastContact().IsZero()
}
2014-04-19 00:26:59 +00:00
// forwardRequestToLeader is an implementation detail of forwardRPC.
// See the comment for forwardRPC for more details.
func (s *Server) forwardRequestToLeader(info structs.RPCInfo, forwardToLeader func(leader *metadata.Server) error) (handled bool, err error) {
firstCheck := time.Now()
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, rpcErr := s.getLeader()
// Handle the case we are the leader
if isLeader {
return false, nil
}
// Handle the case of a known leader
if leader != nil {
rpcErr = forwardToLeader(leader)
if rpcErr == nil {
return true, nil
}
}
if retry := canRetry(info, rpcErr, firstCheck, s.config); retry {
// Gate the request until there is a leader
jitter := lib.RandomStagger(s.config.RPCHoldTimeout / structs.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
}
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
// 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. In the case of not having a leader elected yet
// then a NoClusterLeader error gets returned. In the case of Raft having
// a leader but out internal tracking failing to find the leader we
// return a LeaderNotTracked error. Therefore if the err is nil AND
// the bool is false then the Server will be non-nil
func (s *Server) getLeader() (bool, *metadata.Server, error) {
// Check if we are the leader
if s.IsLeader() {
return true, nil, nil
}
// Get the leader
leader := s.raft.Leader()
2015-05-08 18:35:12 +00:00
if leader == "" {
return false, nil, structs.ErrNoLeader
}
// Lookup the server
2017-08-30 17:31:36 +00:00
server := s.serverLookup.Server(leader)
// if server is nil this indicates that while we have a Raft leader
// something has caused that node to be considered unhealthy which
// cascades into its removal from the serverLookup struct. In this case
// we should not report no cluster leader but instead report a different
// error so as not to confuse our users as to the what the root cause of
// an issue might be.
if server == nil {
s.logger.Warn("Raft has a leader but other tracking of the node would indicate that the node is unhealthy or does not exist. The network may be misconfigured.", "leader", leader)
return false, nil, structs.ErrLeaderNotTracked
}
return false, server, nil
}
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
// 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)
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
if !ok {
if s.router.HasDatacenter(dc) {
s.rpcLogger().Warn("RPC request to DC is currently failing as no server can be reached", "datacenter", dc)
return structs.ErrDCNotAvailable
}
s.rpcLogger().Warn("RPC request for DC is currently failing as no path was found",
"datacenter", dc,
"method", method,
)
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
return structs.ErrNoDCPath
}
2013-12-12 00:33:19 +00:00
metrics.IncrCounterWithLabels([]string{"rpc", "cross-dc"}, 1,
[]metrics.Label{{Name: "datacenter", Value: dc}})
if err := s.connPool.RPC(dc, server.ShortName, server.Addr, method, args, reply); err != nil {
manager.NotifyFailedServer(server)
s.rpcLogger().Error("RPC failed to server in DC",
"server", server.Addr,
"datacenter", dc,
"method", method,
"error", err,
)
return err
}
return nil
}
// keyringRPCs is used to forward an RPC request to a server in each dc. This
// will only error for RPC-related errors. Otherwise, application-level errors
// can be sent in the response objects.
func (s *Server) keyringRPCs(method string, args interface{}, dcs []string) (*structs.KeyringResponses, error) {
errorCh := make(chan error, len(dcs))
respCh := make(chan *structs.KeyringResponses, len(dcs))
for _, dc := range dcs {
go func(dc string) {
rr := &structs.KeyringResponses{}
if err := s.forwardDC(method, dc, args, &rr); err != nil {
errorCh <- err
return
}
respCh <- rr
}(dc)
}
responses := &structs.KeyringResponses{}
for i := 0; i < len(dcs); i++ {
select {
case err := <-errorCh:
return nil, err
case rr := <-respCh:
responses.Add(rr)
}
}
return responses, nil
}
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
type raftEncoder func(structs.MessageType, interface{}) ([]byte, error)
// raftApplyMsgpack encodes the msg using msgpack and calls raft.Apply. See
// raftApplyWithEncoder.
// Deprecated: use raftApplyMsgpack
2013-12-19 20:03:57 +00:00
func (s *Server) raftApply(t structs.MessageType, msg interface{}) (interface{}, error) {
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
return s.raftApplyMsgpack(t, msg)
}
// raftApplyMsgpack encodes the msg using msgpack and calls raft.Apply. See
// raftApplyWithEncoder.
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
func (s *Server) raftApplyMsgpack(t structs.MessageType, msg interface{}) (interface{}, error) {
return s.raftApplyWithEncoder(t, msg, structs.Encode)
}
// raftApplyProtobuf encodes the msg using protobuf and calls raft.Apply. See
// raftApplyWithEncoder.
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
func (s *Server) raftApplyProtobuf(t structs.MessageType, msg interface{}) (interface{}, error) {
return s.raftApplyWithEncoder(t, msg, structs.EncodeProtoInterface)
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
}
// raftApplyWithEncoder encodes a message, and then calls raft.Apply with the
// encoded message. Returns the FSM response along with any errors. If the
// FSM.Apply response is an error it will be returned as the error return
// value with a nil response.
func (s *Server) raftApplyWithEncoder(
t structs.MessageType,
msg interface{},
encoder raftEncoder,
) (response interface{}, err error) {
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
if encoder == nil {
return nil, fmt.Errorf("Failed to encode request: nil encoder")
}
buf, err := encoder(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.rpcLogger().Warn("Attempting to apply large raft entry", "size_in_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 {
return nil, ErrChunkingResubmit
}
// 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")
}
resp = chunkedSuccess.Response
}
if err, ok := resp.(error); ok {
return nil, err
}
return resp, nil
}
// queryFn is used to perform a query operation. If a re-query is needed, the
2017-01-25 17:45:25 +00:00
// 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.
2017-04-21 00:46:29 +00:00
type queryFn func(memdb.WatchSet, *state.Store) error
// blockingQuery is used to process a potentially blocking query operation.
func (s *Server) blockingQuery(queryOpts structs.QueryOptionsCompat, queryMeta structs.QueryMetaCompat, fn queryFn) error {
var cancel func()
var ctx context.Context = &lib.StopChannelContext{StopCh: s.shutdownCh}
var queriesBlocking uint64
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
var queryTimeout time.Duration
// Instrument all queries run
metrics.IncrCounter([]string{"rpc", "query"}, 1)
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
minQueryIndex := queryOpts.GetMinQueryIndex()
// Fast path right to the non-blocking query.
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
if minQueryIndex == 0 {
goto RUN_QUERY
}
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
queryTimeout = queryOpts.GetMaxQueryTime()
// Restrict the max query time, and ensure there is always one.
if queryTimeout > s.config.MaxQueryTime {
queryTimeout = s.config.MaxQueryTime
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
} else if queryTimeout <= 0 {
queryTimeout = s.config.DefaultQueryTime
}
// Apply a small amount of jitter to the request.
queryTimeout += lib.RandomStagger(queryTimeout / structs.JitterFraction)
// wrap the base context with a deadline
ctx, cancel = context.WithDeadline(ctx, time.Now().Add(queryTimeout))
defer cancel()
// instrument blockingQueries
// atomic inc our server's count of in-flight blockingQueries and store the new value
queriesBlocking = atomic.AddUint64(&s.queriesBlocking, 1)
// atomic dec when we return from blockingQuery()
defer atomic.AddUint64(&s.queriesBlocking, ^uint64(0))
// set the gauge directly to the new value of s.blockingQueries
metrics.SetGauge([]string{"rpc", "queries_blocking"}, float32(queriesBlocking))
RUN_QUERY:
// Setup blocking loop
// Validate
// If the read must be consistent we verify that we are still the leader.
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
if queryOpts.GetRequireConsistent() {
if err := s.consistentRead(); err != nil {
return err
}
}
// Run query
// 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
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
if 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())
}
// Execute the queryFn
err := fn(ws, state)
// Update the query metadata.
s.setQueryMeta(queryMeta, queryOpts.GetToken())
// 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.
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
if err == nil && queryMeta.GetIndex() < 1 {
queryMeta.SetIndex(1)
}
// block up to the timeout if we don't see anything fresh.
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
if err == nil && minQueryIndex > 0 && queryMeta.GetIndex() <= minQueryIndex {
if err := ws.WatchCtx(ctx); err == nil {
// a non-nil error only occurs when the context is cancelled
// 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:
// loop back and look for an update again
goto RUN_QUERY
}
}
}
return err
}
// setQueryMeta is used to populate the QueryMeta data for an RPC call
//
// Note: This method must be called *after* filtering query results with ACLs.
func (s *Server) setQueryMeta(m structs.QueryMetaCompat, token string) {
if s.IsLeader() {
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
m.SetLastContact(0)
m.SetKnownLeader(true)
} else {
Add support for implementing new requests with protobufs instea… (#6502) * Add build system support for protobuf generation This is done generically so that we don’t have to keep updating the makefile to add another proto generation. Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date. If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild> Providing the PROTOFILES var will override the default behavior of finding all the .proto files. * Start adding types to the agent/proto package These will be needed for some other work and are by no means comprehensive. * Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store. * Use protobuf marshalling of raft requests instead of msgpack for protoc generated types. This does not change any encoding of existing types. * Removed structs package automatically encoding with protobuf marshalling Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf` * Run update-vendor to fixup modules.txt Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical. * Rename some things and implement the structs.RPCInfo interface bits agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth. * Use the right encoding function. * Renamed agent/proto package to agent/agentpb to prevent package name conflicts * Update modules.txt to fix ordering * Change blockingQuery to take in interfaces for the query options and meta * Add %T to error output. * Add/Update some comments
2019-09-20 18:37:22 +00:00
m.SetLastContact(time.Since(s.raft.LastContact()))
m.SetKnownLeader(s.raft.Leader() != "")
}
maskResultsFilteredByACLs(token, m)
}
// 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 / structs.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
}
// maskResultsFilteredByACLs blanks out the ResultsFilteredByACLs flag if the
// request is unauthenticated, to limit information leaking.
//
// Endpoints that support bexpr filtering could be used in combination with
// this flag/header to discover the existence of resources to which the user
// does not have access, therefore we only expose it when the user presents
// a valid ACL token. This doesn't completely remove the risk (by nature the
// purpose of this flag is to let the user know there are resources they can
// not access) but it prevents completely unauthenticated users from doing so.
//
// Notes:
//
// * The definition of "unauthenticated" here is incomplete, as it doesn't
// account for the fact that operators can modify the anonymous token with
// custom policies, or set namespace default policies. As these scenarios
// are less common and this flag is a best-effort UX improvement, we think
// the trade-off for reduced complexity is acceptable.
//
// * This method assumes that the given token has already been validated (and
// will only check whether it is blank or not). It's a safe assumption because
// ResultsFilteredByACLs is only set to try when applying the already-resolved
// token's policies.
func maskResultsFilteredByACLs(token string, meta structs.QueryMetaCompat) {
if token == "" {
meta.SetResultsFilteredByACLs(false)
}
}