open-consul/agent/consul/gateway_locator.go

450 lines
14 KiB
Go
Raw Normal View History

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
package consul
import (
"context"
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
"errors"
"math/rand"
"sort"
"sync"
"time"
"github.com/hashicorp/consul/agent/consul/state"
"github.com/hashicorp/consul/agent/structs"
"github.com/hashicorp/consul/api"
"github.com/hashicorp/consul/ipaddr"
2020-05-27 16:47:32 +00:00
"github.com/hashicorp/consul/lib/stringslice"
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
"github.com/hashicorp/consul/logging"
"github.com/hashicorp/go-hclog"
memdb "github.com/hashicorp/go-memdb"
)
// GatewayLocator assists in selecting an appropriate mesh gateway when wan
// federation via mesh gateways is enabled.
//
// This is exclusively used by the consul server itself when it needs to tunnel
// RPC or gossip through a mesh gateway to reach its ultimate destination.
//
// During secondary datacenter bootstrapping there is a phase where it is
// impossible for mesh gateways in the secondary datacenter to register
// themselves into the catalog to be discovered by the servers, so the servers
// maintain references for the mesh gateways in the primary in addition to its
// own local mesh gateways.
//
// After initial datacenter federation the primary mesh gateways are only used
// in extreme fallback situations (basically re-bootstrapping).
//
// For all other operations a consul server will ALWAYS contact a local mesh
// gateway to ultimately forward the request through a remote mesh gateway to
// reach its destination.
type GatewayLocator struct {
logger hclog.Logger
srv serverDelegate
datacenter string // THIS dc
primaryDatacenter string
// these ONLY contain ones that have the wanfed:1 meta
gatewaysLock sync.Mutex
primaryGateways []string // WAN addrs
localGateways []string // LAN addrs
// primaryMeshGatewayDiscoveredAddresses is the current fallback addresses
// for the mesh gateways in the primary datacenter.
primaryMeshGatewayDiscoveredAddresses []string
primaryMeshGatewayDiscoveredAddressesLock sync.Mutex
// This will be closed the FIRST time we get some gateways populated
primaryGatewaysReadyCh chan struct{}
primaryGatewaysReadyOnce sync.Once
agent: handle re-bootstrapping in a secondary datacenter when WAN federation via mesh gateways is configured (#7931) The main fix here is to always union the `primary-gateways` list with the list of mesh gateways in the primary returned from the replicated federation states list. This will allow any replicated (incorrect) state to be supplemented with user-configured (correct) state in the config file. Eventually the game of random selection whack-a-mole will pick a winning entry and re-replicate the latest federation states from the primary. If the user-configured state is actually the incorrect one, then the same eventual correct selection process will work in that case, too. The secondary fix is actually to finish making wanfed-via-mgws actually work as originally designed. Once a secondary datacenter has replicated federation states for the primary AND managed to stand up its own local mesh gateways then all of the RPCs from a secondary to the primary SHOULD go through two sets of mesh gateways to arrive in the consul servers in the primary (one hop for the secondary datacenter's mesh gateway, and one hop through the primary datacenter's mesh gateway). This was neglected in the initial implementation. While everything works, ideally we should treat communications that go around the mesh gateways as just provided for bootstrapping purposes. Now we heuristically use the success/failure history of the federation state replicator goroutine loop to determine if our current mesh gateway route is working as intended. If it is, we try using the local gateways, and if those don't work we fall back on trying the primary via the union of the replicated state and the go-discover configuration flags. This can be improved slightly in the future by possibly initializing the gateway choice to local on startup if we already have replicated state. This PR does not address that improvement. Fixes #7339
2020-05-27 16:31:10 +00:00
// these are a collection of measurements that factor into deciding if we
// should directly dial the primary's mesh gateways or if we should try to
// route through our local gateways (if they are up).
lastReplLock sync.Mutex
lastReplSuccess time.Time
lastReplFailure time.Time
lastReplSuccesses uint64
lastReplFailures uint64
useReplicationSignal bool // this should be set to true on the leader
agent: handle re-bootstrapping in a secondary datacenter when WAN federation via mesh gateways is configured (#7931) The main fix here is to always union the `primary-gateways` list with the list of mesh gateways in the primary returned from the replicated federation states list. This will allow any replicated (incorrect) state to be supplemented with user-configured (correct) state in the config file. Eventually the game of random selection whack-a-mole will pick a winning entry and re-replicate the latest federation states from the primary. If the user-configured state is actually the incorrect one, then the same eventual correct selection process will work in that case, too. The secondary fix is actually to finish making wanfed-via-mgws actually work as originally designed. Once a secondary datacenter has replicated federation states for the primary AND managed to stand up its own local mesh gateways then all of the RPCs from a secondary to the primary SHOULD go through two sets of mesh gateways to arrive in the consul servers in the primary (one hop for the secondary datacenter's mesh gateway, and one hop through the primary datacenter's mesh gateway). This was neglected in the initial implementation. While everything works, ideally we should treat communications that go around the mesh gateways as just provided for bootstrapping purposes. Now we heuristically use the success/failure history of the federation state replicator goroutine loop to determine if our current mesh gateway route is working as intended. If it is, we try using the local gateways, and if those don't work we fall back on trying the primary via the union of the replicated state and the go-discover configuration flags. This can be improved slightly in the future by possibly initializing the gateway choice to local on startup if we already have replicated state. This PR does not address that improvement. Fixes #7339
2020-05-27 16:31:10 +00:00
}
// SetLastFederationStateReplicationError is used to indicate if the federation
// state replication loop has succeeded (nil) or failed during the last
// execution.
//
// Rather than introduce a completely new mechanism to periodically probe that
// our chosen mesh-gateway configuration can reach the primary's servers (like
// a ping or status RPC) we cheat and use the federation state replicator
// goroutine's success or failure as a proxy.
func (g *GatewayLocator) SetLastFederationStateReplicationError(err error, fromReplication bool) {
if g == nil {
return
}
agent: handle re-bootstrapping in a secondary datacenter when WAN federation via mesh gateways is configured (#7931) The main fix here is to always union the `primary-gateways` list with the list of mesh gateways in the primary returned from the replicated federation states list. This will allow any replicated (incorrect) state to be supplemented with user-configured (correct) state in the config file. Eventually the game of random selection whack-a-mole will pick a winning entry and re-replicate the latest federation states from the primary. If the user-configured state is actually the incorrect one, then the same eventual correct selection process will work in that case, too. The secondary fix is actually to finish making wanfed-via-mgws actually work as originally designed. Once a secondary datacenter has replicated federation states for the primary AND managed to stand up its own local mesh gateways then all of the RPCs from a secondary to the primary SHOULD go through two sets of mesh gateways to arrive in the consul servers in the primary (one hop for the secondary datacenter's mesh gateway, and one hop through the primary datacenter's mesh gateway). This was neglected in the initial implementation. While everything works, ideally we should treat communications that go around the mesh gateways as just provided for bootstrapping purposes. Now we heuristically use the success/failure history of the federation state replicator goroutine loop to determine if our current mesh gateway route is working as intended. If it is, we try using the local gateways, and if those don't work we fall back on trying the primary via the union of the replicated state and the go-discover configuration flags. This can be improved slightly in the future by possibly initializing the gateway choice to local on startup if we already have replicated state. This PR does not address that improvement. Fixes #7339
2020-05-27 16:31:10 +00:00
g.lastReplLock.Lock()
defer g.lastReplLock.Unlock()
agent: handle re-bootstrapping in a secondary datacenter when WAN federation via mesh gateways is configured (#7931) The main fix here is to always union the `primary-gateways` list with the list of mesh gateways in the primary returned from the replicated federation states list. This will allow any replicated (incorrect) state to be supplemented with user-configured (correct) state in the config file. Eventually the game of random selection whack-a-mole will pick a winning entry and re-replicate the latest federation states from the primary. If the user-configured state is actually the incorrect one, then the same eventual correct selection process will work in that case, too. The secondary fix is actually to finish making wanfed-via-mgws actually work as originally designed. Once a secondary datacenter has replicated federation states for the primary AND managed to stand up its own local mesh gateways then all of the RPCs from a secondary to the primary SHOULD go through two sets of mesh gateways to arrive in the consul servers in the primary (one hop for the secondary datacenter's mesh gateway, and one hop through the primary datacenter's mesh gateway). This was neglected in the initial implementation. While everything works, ideally we should treat communications that go around the mesh gateways as just provided for bootstrapping purposes. Now we heuristically use the success/failure history of the federation state replicator goroutine loop to determine if our current mesh gateway route is working as intended. If it is, we try using the local gateways, and if those don't work we fall back on trying the primary via the union of the replicated state and the go-discover configuration flags. This can be improved slightly in the future by possibly initializing the gateway choice to local on startup if we already have replicated state. This PR does not address that improvement. Fixes #7339
2020-05-27 16:31:10 +00:00
oldChoice := g.dialPrimaryThroughLocalGateway()
if err == nil {
g.lastReplSuccess = time.Now().UTC()
g.lastReplSuccesses++
g.lastReplFailures = 0
if fromReplication {
// If we get info from replication, assume replication is operating.
g.useReplicationSignal = true
}
agent: handle re-bootstrapping in a secondary datacenter when WAN federation via mesh gateways is configured (#7931) The main fix here is to always union the `primary-gateways` list with the list of mesh gateways in the primary returned from the replicated federation states list. This will allow any replicated (incorrect) state to be supplemented with user-configured (correct) state in the config file. Eventually the game of random selection whack-a-mole will pick a winning entry and re-replicate the latest federation states from the primary. If the user-configured state is actually the incorrect one, then the same eventual correct selection process will work in that case, too. The secondary fix is actually to finish making wanfed-via-mgws actually work as originally designed. Once a secondary datacenter has replicated federation states for the primary AND managed to stand up its own local mesh gateways then all of the RPCs from a secondary to the primary SHOULD go through two sets of mesh gateways to arrive in the consul servers in the primary (one hop for the secondary datacenter's mesh gateway, and one hop through the primary datacenter's mesh gateway). This was neglected in the initial implementation. While everything works, ideally we should treat communications that go around the mesh gateways as just provided for bootstrapping purposes. Now we heuristically use the success/failure history of the federation state replicator goroutine loop to determine if our current mesh gateway route is working as intended. If it is, we try using the local gateways, and if those don't work we fall back on trying the primary via the union of the replicated state and the go-discover configuration flags. This can be improved slightly in the future by possibly initializing the gateway choice to local on startup if we already have replicated state. This PR does not address that improvement. Fixes #7339
2020-05-27 16:31:10 +00:00
} else {
g.lastReplFailure = time.Now().UTC()
g.lastReplFailures++
g.lastReplSuccesses = 0
}
newChoice := g.dialPrimaryThroughLocalGateway()
if oldChoice != newChoice {
g.logPrimaryDialingMessage(newChoice)
}
}
func (g *GatewayLocator) SetUseReplicationSignal(newValue bool) {
if g == nil {
return
}
g.lastReplLock.Lock()
g.useReplicationSignal = newValue
g.lastReplLock.Unlock()
}
agent: handle re-bootstrapping in a secondary datacenter when WAN federation via mesh gateways is configured (#7931) The main fix here is to always union the `primary-gateways` list with the list of mesh gateways in the primary returned from the replicated federation states list. This will allow any replicated (incorrect) state to be supplemented with user-configured (correct) state in the config file. Eventually the game of random selection whack-a-mole will pick a winning entry and re-replicate the latest federation states from the primary. If the user-configured state is actually the incorrect one, then the same eventual correct selection process will work in that case, too. The secondary fix is actually to finish making wanfed-via-mgws actually work as originally designed. Once a secondary datacenter has replicated federation states for the primary AND managed to stand up its own local mesh gateways then all of the RPCs from a secondary to the primary SHOULD go through two sets of mesh gateways to arrive in the consul servers in the primary (one hop for the secondary datacenter's mesh gateway, and one hop through the primary datacenter's mesh gateway). This was neglected in the initial implementation. While everything works, ideally we should treat communications that go around the mesh gateways as just provided for bootstrapping purposes. Now we heuristically use the success/failure history of the federation state replicator goroutine loop to determine if our current mesh gateway route is working as intended. If it is, we try using the local gateways, and if those don't work we fall back on trying the primary via the union of the replicated state and the go-discover configuration flags. This can be improved slightly in the future by possibly initializing the gateway choice to local on startup if we already have replicated state. This PR does not address that improvement. Fixes #7339
2020-05-27 16:31:10 +00:00
func (g *GatewayLocator) logPrimaryDialingMessage(useLocal bool) {
if g.datacenter == g.primaryDatacenter {
// These messages are useless when the server is in the primary
// datacenter.
return
}
if useLocal {
g.logger.Info("will dial the primary datacenter using our local mesh gateways if possible")
} else {
g.logger.Info("will dial the primary datacenter through its mesh gateways")
}
}
// DialPrimaryThroughLocalGateway determines if we should dial the primary's
// mesh gateways directly or use our local mesh gateways (if they are up).
//
// Generally the system has three states:
//
// 1. Servers dial primary MGWs using fallback addresses from the agent config.
// 2. Servers dial primary MGWs using replicated federation state data.
// 3. Servers dial primary MGWs indirectly through local MGWs.
//
// After initial bootstrapping most communication should go through (3). If the
// local mesh gateways are not coming up for chicken/egg problems (mostly the
// kind that arise from secondary datacenter bootstrapping) then (2) is useful
// to solve the chicken/egg problem and get back to (3). In the worst case
// where we completely lost communication with the primary AND all of their old
// mesh gateway addresses are changed then we need to go all the way back to
// square one and re-bootstrap via (1).
//
// Since both (1) and (2) are meant to be temporary we simplify things and make
// the system only consider two overall configurations: (1+2, with the
// addresses being unioned) or (3).
//
// This method returns true if in state (3) and false if in state (1+2).
func (g *GatewayLocator) DialPrimaryThroughLocalGateway() bool {
if g.datacenter == g.primaryDatacenter {
return false // not important
}
g.lastReplLock.Lock()
defer g.lastReplLock.Unlock()
return g.dialPrimaryThroughLocalGateway()
}
const localFederationStateReplicatorFailuresBeforeDialingDirectly = 3
func (g *GatewayLocator) dialPrimaryThroughLocalGateway() bool {
if !g.useReplicationSignal {
// Followers should blindly assume these gateways work. The leader will
// try to bypass them and correct the replicated federation state info
// that the followers will eventually pick up on.
return true
}
agent: handle re-bootstrapping in a secondary datacenter when WAN federation via mesh gateways is configured (#7931) The main fix here is to always union the `primary-gateways` list with the list of mesh gateways in the primary returned from the replicated federation states list. This will allow any replicated (incorrect) state to be supplemented with user-configured (correct) state in the config file. Eventually the game of random selection whack-a-mole will pick a winning entry and re-replicate the latest federation states from the primary. If the user-configured state is actually the incorrect one, then the same eventual correct selection process will work in that case, too. The secondary fix is actually to finish making wanfed-via-mgws actually work as originally designed. Once a secondary datacenter has replicated federation states for the primary AND managed to stand up its own local mesh gateways then all of the RPCs from a secondary to the primary SHOULD go through two sets of mesh gateways to arrive in the consul servers in the primary (one hop for the secondary datacenter's mesh gateway, and one hop through the primary datacenter's mesh gateway). This was neglected in the initial implementation. While everything works, ideally we should treat communications that go around the mesh gateways as just provided for bootstrapping purposes. Now we heuristically use the success/failure history of the federation state replicator goroutine loop to determine if our current mesh gateway route is working as intended. If it is, we try using the local gateways, and if those don't work we fall back on trying the primary via the union of the replicated state and the go-discover configuration flags. This can be improved slightly in the future by possibly initializing the gateway choice to local on startup if we already have replicated state. This PR does not address that improvement. Fixes #7339
2020-05-27 16:31:10 +00:00
if g.lastReplSuccess.IsZero() && g.lastReplFailure.IsZero() {
return false // no data yet
}
if g.lastReplSuccess.After(g.lastReplFailure) {
return true // we have viable data
}
if g.lastReplFailures < localFederationStateReplicatorFailuresBeforeDialingDirectly {
return true // maybe it's just a little broken
}
return false
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
}
// PrimaryMeshGatewayAddressesReadyCh returns a channel that will be closed
// when federation state replication ships back at least one primary mesh
// gateway (not via fallback config).
func (g *GatewayLocator) PrimaryMeshGatewayAddressesReadyCh() <-chan struct{} {
return g.primaryGatewaysReadyCh
}
// PickGateway returns the address for a gateway suitable for reaching the
// provided datacenter.
func (g *GatewayLocator) PickGateway(dc string) string {
item := g.pickGateway(dc == g.primaryDatacenter)
g.logger.Trace("picking gateway for transit", "gateway", item, "source_datacenter", g.datacenter, "dest_datacenter", dc)
return item
}
func (g *GatewayLocator) pickGateway(primary bool) string {
addrs := g.listGateways(primary)
return getRandomItem(addrs)
}
func (g *GatewayLocator) listGateways(primary bool) []string {
g.gatewaysLock.Lock()
defer g.gatewaysLock.Unlock()
var addrs []string
if primary {
agent: handle re-bootstrapping in a secondary datacenter when WAN federation via mesh gateways is configured (#7931) The main fix here is to always union the `primary-gateways` list with the list of mesh gateways in the primary returned from the replicated federation states list. This will allow any replicated (incorrect) state to be supplemented with user-configured (correct) state in the config file. Eventually the game of random selection whack-a-mole will pick a winning entry and re-replicate the latest federation states from the primary. If the user-configured state is actually the incorrect one, then the same eventual correct selection process will work in that case, too. The secondary fix is actually to finish making wanfed-via-mgws actually work as originally designed. Once a secondary datacenter has replicated federation states for the primary AND managed to stand up its own local mesh gateways then all of the RPCs from a secondary to the primary SHOULD go through two sets of mesh gateways to arrive in the consul servers in the primary (one hop for the secondary datacenter's mesh gateway, and one hop through the primary datacenter's mesh gateway). This was neglected in the initial implementation. While everything works, ideally we should treat communications that go around the mesh gateways as just provided for bootstrapping purposes. Now we heuristically use the success/failure history of the federation state replicator goroutine loop to determine if our current mesh gateway route is working as intended. If it is, we try using the local gateways, and if those don't work we fall back on trying the primary via the union of the replicated state and the go-discover configuration flags. This can be improved slightly in the future by possibly initializing the gateway choice to local on startup if we already have replicated state. This PR does not address that improvement. Fixes #7339
2020-05-27 16:31:10 +00:00
if g.datacenter == g.primaryDatacenter {
addrs = g.primaryGateways
} else if g.DialPrimaryThroughLocalGateway() && len(g.localGateways) > 0 {
addrs = g.localGateways
} else {
// Note calling StringSliceMergeSorted only works because both
// inputs are pre-sorted. If for some reason one of the lists has
// *duplicates* (which shouldn't happen) it's not great but it
// won't break anything other than biasing our eventual random
// choice a little bit.
2020-05-27 16:47:32 +00:00
addrs = stringslice.MergeSorted(g.primaryGateways, g.PrimaryGatewayFallbackAddresses())
agent: handle re-bootstrapping in a secondary datacenter when WAN federation via mesh gateways is configured (#7931) The main fix here is to always union the `primary-gateways` list with the list of mesh gateways in the primary returned from the replicated federation states list. This will allow any replicated (incorrect) state to be supplemented with user-configured (correct) state in the config file. Eventually the game of random selection whack-a-mole will pick a winning entry and re-replicate the latest federation states from the primary. If the user-configured state is actually the incorrect one, then the same eventual correct selection process will work in that case, too. The secondary fix is actually to finish making wanfed-via-mgws actually work as originally designed. Once a secondary datacenter has replicated federation states for the primary AND managed to stand up its own local mesh gateways then all of the RPCs from a secondary to the primary SHOULD go through two sets of mesh gateways to arrive in the consul servers in the primary (one hop for the secondary datacenter's mesh gateway, and one hop through the primary datacenter's mesh gateway). This was neglected in the initial implementation. While everything works, ideally we should treat communications that go around the mesh gateways as just provided for bootstrapping purposes. Now we heuristically use the success/failure history of the federation state replicator goroutine loop to determine if our current mesh gateway route is working as intended. If it is, we try using the local gateways, and if those don't work we fall back on trying the primary via the union of the replicated state and the go-discover configuration flags. This can be improved slightly in the future by possibly initializing the gateway choice to local on startup if we already have replicated state. This PR does not address that improvement. Fixes #7339
2020-05-27 16:31:10 +00:00
}
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
} else {
addrs = g.localGateways
}
return addrs
}
// RefreshPrimaryGatewayFallbackAddresses is used to update the list of current
// fallback addresses for locating mesh gateways in the primary datacenter.
func (g *GatewayLocator) RefreshPrimaryGatewayFallbackAddresses(addrs []string) {
sort.Strings(addrs)
g.primaryMeshGatewayDiscoveredAddressesLock.Lock()
defer g.primaryMeshGatewayDiscoveredAddressesLock.Unlock()
2020-05-27 16:47:32 +00:00
if !stringslice.Equal(addrs, g.primaryMeshGatewayDiscoveredAddresses) {
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
g.primaryMeshGatewayDiscoveredAddresses = addrs
g.logger.Info("updated fallback list of primary mesh gateways", "mesh_gateways", addrs)
}
}
// PrimaryGatewayFallbackAddresses returns the current set of discovered
// fallback addresses for the mesh gateways in the primary datacenter.
func (g *GatewayLocator) PrimaryGatewayFallbackAddresses() []string {
g.primaryMeshGatewayDiscoveredAddressesLock.Lock()
defer g.primaryMeshGatewayDiscoveredAddressesLock.Unlock()
out := make([]string, len(g.primaryMeshGatewayDiscoveredAddresses))
copy(out, g.primaryMeshGatewayDiscoveredAddresses)
return out
}
func getRandomItem(items []string) string {
switch len(items) {
case 0:
return ""
case 1:
return items[0]
default:
idx := int(rand.Int31n(int32(len(items))))
return items[idx]
}
}
type serverDelegate interface {
blockingQuery(queryOpts structs.QueryOptionsCompat, queryMeta structs.QueryMetaCompat, fn queryFn) error
IsLeader() bool
LeaderLastContact() time.Time
}
func NewGatewayLocator(
logger hclog.Logger,
srv serverDelegate,
datacenter string,
primaryDatacenter string,
) *GatewayLocator {
agent: handle re-bootstrapping in a secondary datacenter when WAN federation via mesh gateways is configured (#7931) The main fix here is to always union the `primary-gateways` list with the list of mesh gateways in the primary returned from the replicated federation states list. This will allow any replicated (incorrect) state to be supplemented with user-configured (correct) state in the config file. Eventually the game of random selection whack-a-mole will pick a winning entry and re-replicate the latest federation states from the primary. If the user-configured state is actually the incorrect one, then the same eventual correct selection process will work in that case, too. The secondary fix is actually to finish making wanfed-via-mgws actually work as originally designed. Once a secondary datacenter has replicated federation states for the primary AND managed to stand up its own local mesh gateways then all of the RPCs from a secondary to the primary SHOULD go through two sets of mesh gateways to arrive in the consul servers in the primary (one hop for the secondary datacenter's mesh gateway, and one hop through the primary datacenter's mesh gateway). This was neglected in the initial implementation. While everything works, ideally we should treat communications that go around the mesh gateways as just provided for bootstrapping purposes. Now we heuristically use the success/failure history of the federation state replicator goroutine loop to determine if our current mesh gateway route is working as intended. If it is, we try using the local gateways, and if those don't work we fall back on trying the primary via the union of the replicated state and the go-discover configuration flags. This can be improved slightly in the future by possibly initializing the gateway choice to local on startup if we already have replicated state. This PR does not address that improvement. Fixes #7339
2020-05-27 16:31:10 +00:00
g := &GatewayLocator{
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
logger: logger.Named(logging.GatewayLocator),
srv: srv,
datacenter: datacenter,
primaryDatacenter: primaryDatacenter,
primaryGatewaysReadyCh: make(chan struct{}),
}
agent: handle re-bootstrapping in a secondary datacenter when WAN federation via mesh gateways is configured (#7931) The main fix here is to always union the `primary-gateways` list with the list of mesh gateways in the primary returned from the replicated federation states list. This will allow any replicated (incorrect) state to be supplemented with user-configured (correct) state in the config file. Eventually the game of random selection whack-a-mole will pick a winning entry and re-replicate the latest federation states from the primary. If the user-configured state is actually the incorrect one, then the same eventual correct selection process will work in that case, too. The secondary fix is actually to finish making wanfed-via-mgws actually work as originally designed. Once a secondary datacenter has replicated federation states for the primary AND managed to stand up its own local mesh gateways then all of the RPCs from a secondary to the primary SHOULD go through two sets of mesh gateways to arrive in the consul servers in the primary (one hop for the secondary datacenter's mesh gateway, and one hop through the primary datacenter's mesh gateway). This was neglected in the initial implementation. While everything works, ideally we should treat communications that go around the mesh gateways as just provided for bootstrapping purposes. Now we heuristically use the success/failure history of the federation state replicator goroutine loop to determine if our current mesh gateway route is working as intended. If it is, we try using the local gateways, and if those don't work we fall back on trying the primary via the union of the replicated state and the go-discover configuration flags. This can be improved slightly in the future by possibly initializing the gateway choice to local on startup if we already have replicated state. This PR does not address that improvement. Fixes #7339
2020-05-27 16:31:10 +00:00
g.logPrimaryDialingMessage(g.DialPrimaryThroughLocalGateway())
return g
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
}
var errGatewayLocalStateNotInitialized = errors.New("local state not initialized")
func (g *GatewayLocator) Run(ctx context.Context) {
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
var lastFetchIndex uint64
retryLoopBackoff(ctx, func() error {
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
idx, err := g.runOnce(lastFetchIndex)
if err != nil {
return err
}
lastFetchIndex = idx
return nil
}, func(err error) {
if !errors.Is(err, errGatewayLocalStateNotInitialized) {
g.logger.Error("error tracking primary and local mesh gateways", "error", err)
}
})
}
func (g *GatewayLocator) runOnce(lastFetchIndex uint64) (uint64, error) {
if err := g.checkLocalStateIsReady(); err != nil {
return 0, err
}
// NOTE: we can't do RPC here because we won't have a token so we'll just
// mostly assume that our FSM is caught up enough to answer locally. If
// this has drifted it's no different than a cache that drifts or an
// inconsistent read.
queryOpts := &structs.QueryOptions{
MinQueryIndex: lastFetchIndex,
RequireConsistent: false,
}
var (
results []*structs.FederationState
queryMeta structs.QueryMeta
)
err := g.srv.blockingQuery(
queryOpts,
&queryMeta,
func(ws memdb.WatchSet, state *state.Store) error {
// Get the existing stored version of this config that has replicated down.
// We could phone home to get this but that would incur extra WAN traffic
// when we already have enough information locally to figure it out
// (assuming that our replicator is still functioning).
idx, all, err := state.FederationStateList(ws)
if err != nil {
return err
}
queryMeta.Index = idx
results = all
return nil
})
if err != nil {
return 0, err
}
g.updateFromState(results)
return queryMeta.Index, nil
}
// checkLocalStateIsReady is inlined a bit from (*Server).ForwardRPC(). We need to
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
// wait until our own state machine is safe to read from.
func (g *GatewayLocator) checkLocalStateIsReady() error {
// Check if we can allow a stale read, ensure our local DB is initialized
if !g.srv.LeaderLastContact().IsZero() {
return nil // the raft leader talked to us
}
if g.srv.IsLeader() {
return nil // we are the leader
}
return errGatewayLocalStateNotInitialized
}
func (g *GatewayLocator) updateFromState(results []*structs.FederationState) {
var (
local structs.CheckServiceNodes
primary structs.CheckServiceNodes
)
for _, config := range results {
retained := retainGateways(config.MeshGateways)
if config.Datacenter == g.datacenter {
local = retained
}
// NOT else-if because conditionals are not mutually exclusive
if config.Datacenter == g.primaryDatacenter {
primary = retained
}
}
primaryAddrs := renderGatewayAddrs(primary, true)
localAddrs := renderGatewayAddrs(local, false)
g.gatewaysLock.Lock()
defer g.gatewaysLock.Unlock()
changed := false
primaryReady := false
2020-05-27 16:47:32 +00:00
if !stringslice.Equal(g.primaryGateways, primaryAddrs) {
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
g.primaryGateways = primaryAddrs
primaryReady = len(g.primaryGateways) > 0
changed = true
}
2020-05-27 16:47:32 +00:00
if !stringslice.Equal(g.localGateways, localAddrs) {
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
g.localGateways = localAddrs
changed = true
}
if changed {
g.logger.Info(
"new cached locations of mesh gateways",
"primary", primaryAddrs,
"local", localAddrs,
)
}
if primaryReady {
g.primaryGatewaysReadyOnce.Do(func() {
close(g.primaryGatewaysReadyCh)
})
}
}
func retainGateways(full structs.CheckServiceNodes) structs.CheckServiceNodes {
out := make([]structs.CheckServiceNode, 0, len(full))
for _, csn := range full {
if csn.Service.Meta[structs.MetaWANFederationKey] != "1" {
continue
}
// only keep healthy ones
ok := true
for _, chk := range csn.Checks {
if chk.Status == api.HealthCritical {
ok = false
}
}
if ok {
out = append(out, csn)
}
}
return out
}
func renderGatewayAddrs(gateways structs.CheckServiceNodes, wan bool) []string {
out := make([]string, 0, len(gateways))
for _, csn := range gateways {
addr, port := csn.BestAddress(wan)
completeAddr := ipaddr.FormatAddressPort(addr, port)
out = append(out, completeAddr)
}
sort.Strings(out)
return out
}