consul/agent/structs/config_entry.go

1109 lines
30 KiB
Go
Raw Normal View History

2019-03-19 17:06:46 +00:00
package structs
import (
"fmt"
"strconv"
"strings"
connect: intentions are now managed as a new config entry kind "service-intentions" (#8834) - Upgrade the ConfigEntry.ListAll RPC to be kind-aware so that older copies of consul will not see new config entries it doesn't understand replicate down. - Add shim conversion code so that the old API/CLI method of interacting with intentions will continue to work so long as none of these are edited via config entry endpoints. Almost all of the read-only APIs will continue to function indefinitely. - Add new APIs that operate on individual intentions without IDs so that the UI doesn't need to implement CAS operations. - Add a new serf feature flag indicating support for intentions-as-config-entries. - The old line-item intentions way of interacting with the state store will transparently flip between the legacy memdb table and the config entry representations so that readers will never see a hiccup during migration where the results are incomplete. It uses a piece of system metadata to control the flip. - The primary datacenter will begin migrating intentions into config entries on startup once all servers in the datacenter are on a version of Consul with the intentions-as-config-entries feature flag. When it is complete the old state store representations will be cleared. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up. - The secondary datacenters continue to run the old intentions replicator until all servers in the secondary DC and primary DC support intentions-as-config-entries (via serf flag). Once this condition it met the old intentions replicator ceases. - The secondary datacenters replicate the new config entries as they are migrated in the primary. When they detect that the primary has zeroed it's old state store table it waits until all config entries up to that point are replicated and then zeroes its own copy of the old state store table. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up.
2020-10-06 18:24:05 +00:00
"time"
"github.com/hashicorp/go-msgpack/codec"
"github.com/hashicorp/go-multierror"
"github.com/mitchellh/hashstructure"
"github.com/mitchellh/mapstructure"
"github.com/hashicorp/consul/acl"
"github.com/hashicorp/consul/agent/cache"
"github.com/hashicorp/consul/lib"
"github.com/hashicorp/consul/lib/decode"
)
2019-03-19 17:06:46 +00:00
const (
ServiceDefaults string = "service-defaults"
ProxyDefaults string = "proxy-defaults"
ServiceRouter string = "service-router"
ServiceSplitter string = "service-splitter"
ServiceResolver string = "service-resolver"
IngressGateway string = "ingress-gateway"
TerminatingGateway string = "terminating-gateway"
connect: intentions are now managed as a new config entry kind "service-intentions" (#8834) - Upgrade the ConfigEntry.ListAll RPC to be kind-aware so that older copies of consul will not see new config entries it doesn't understand replicate down. - Add shim conversion code so that the old API/CLI method of interacting with intentions will continue to work so long as none of these are edited via config entry endpoints. Almost all of the read-only APIs will continue to function indefinitely. - Add new APIs that operate on individual intentions without IDs so that the UI doesn't need to implement CAS operations. - Add a new serf feature flag indicating support for intentions-as-config-entries. - The old line-item intentions way of interacting with the state store will transparently flip between the legacy memdb table and the config entry representations so that readers will never see a hiccup during migration where the results are incomplete. It uses a piece of system metadata to control the flip. - The primary datacenter will begin migrating intentions into config entries on startup once all servers in the datacenter are on a version of Consul with the intentions-as-config-entries feature flag. When it is complete the old state store representations will be cleared. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up. - The secondary datacenters continue to run the old intentions replicator until all servers in the secondary DC and primary DC support intentions-as-config-entries (via serf flag). Once this condition it met the old intentions replicator ceases. - The secondary datacenters replicate the new config entries as they are migrated in the primary. When they detect that the primary has zeroed it's old state store table it waits until all config entries up to that point are replicated and then zeroes its own copy of the old state store table. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up.
2020-10-06 18:24:05 +00:00
ServiceIntentions string = "service-intentions"
MeshConfig string = "mesh"
ProxyConfigGlobal string = "global"
MeshConfigMesh string = "mesh"
DefaultServiceProtocol = "tcp"
2019-03-19 17:06:46 +00:00
)
connect: intentions are now managed as a new config entry kind "service-intentions" (#8834) - Upgrade the ConfigEntry.ListAll RPC to be kind-aware so that older copies of consul will not see new config entries it doesn't understand replicate down. - Add shim conversion code so that the old API/CLI method of interacting with intentions will continue to work so long as none of these are edited via config entry endpoints. Almost all of the read-only APIs will continue to function indefinitely. - Add new APIs that operate on individual intentions without IDs so that the UI doesn't need to implement CAS operations. - Add a new serf feature flag indicating support for intentions-as-config-entries. - The old line-item intentions way of interacting with the state store will transparently flip between the legacy memdb table and the config entry representations so that readers will never see a hiccup during migration where the results are incomplete. It uses a piece of system metadata to control the flip. - The primary datacenter will begin migrating intentions into config entries on startup once all servers in the datacenter are on a version of Consul with the intentions-as-config-entries feature flag. When it is complete the old state store representations will be cleared. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up. - The secondary datacenters continue to run the old intentions replicator until all servers in the secondary DC and primary DC support intentions-as-config-entries (via serf flag). Once this condition it met the old intentions replicator ceases. - The secondary datacenters replicate the new config entries as they are migrated in the primary. When they detect that the primary has zeroed it's old state store table it waits until all config entries up to that point are replicated and then zeroes its own copy of the old state store table. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up.
2020-10-06 18:24:05 +00:00
var AllConfigEntryKinds = []string{
ServiceDefaults,
ProxyDefaults,
ServiceRouter,
ServiceSplitter,
ServiceResolver,
IngressGateway,
TerminatingGateway,
ServiceIntentions,
MeshConfig,
connect: intentions are now managed as a new config entry kind "service-intentions" (#8834) - Upgrade the ConfigEntry.ListAll RPC to be kind-aware so that older copies of consul will not see new config entries it doesn't understand replicate down. - Add shim conversion code so that the old API/CLI method of interacting with intentions will continue to work so long as none of these are edited via config entry endpoints. Almost all of the read-only APIs will continue to function indefinitely. - Add new APIs that operate on individual intentions without IDs so that the UI doesn't need to implement CAS operations. - Add a new serf feature flag indicating support for intentions-as-config-entries. - The old line-item intentions way of interacting with the state store will transparently flip between the legacy memdb table and the config entry representations so that readers will never see a hiccup during migration where the results are incomplete. It uses a piece of system metadata to control the flip. - The primary datacenter will begin migrating intentions into config entries on startup once all servers in the datacenter are on a version of Consul with the intentions-as-config-entries feature flag. When it is complete the old state store representations will be cleared. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up. - The secondary datacenters continue to run the old intentions replicator until all servers in the secondary DC and primary DC support intentions-as-config-entries (via serf flag). Once this condition it met the old intentions replicator ceases. - The secondary datacenters replicate the new config entries as they are migrated in the primary. When they detect that the primary has zeroed it's old state store table it waits until all config entries up to that point are replicated and then zeroes its own copy of the old state store table. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up.
2020-10-06 18:24:05 +00:00
}
// ConfigEntry is the interface for centralized configuration stored in Raft.
// Currently only service-defaults and proxy-defaults are supported.
type ConfigEntry interface {
GetKind() string
2019-03-19 17:06:46 +00:00
GetName() string
// This is called in the RPC endpoint and can apply defaults or limits.
Normalize() error
2019-03-19 17:06:46 +00:00
Validate() error
2019-04-23 06:55:11 +00:00
// CanRead and CanWrite return whether or not the given Authorizer
// has permission to read or write to the config entry, respectively.
2019-04-23 06:55:11 +00:00
CanRead(acl.Authorizer) bool
CanWrite(acl.Authorizer) bool
GetMeta() map[string]string
GetEnterpriseMeta() *EnterpriseMeta
GetRaftIndex() *RaftIndex
2019-03-19 17:06:46 +00:00
}
connect: intentions are now managed as a new config entry kind "service-intentions" (#8834) - Upgrade the ConfigEntry.ListAll RPC to be kind-aware so that older copies of consul will not see new config entries it doesn't understand replicate down. - Add shim conversion code so that the old API/CLI method of interacting with intentions will continue to work so long as none of these are edited via config entry endpoints. Almost all of the read-only APIs will continue to function indefinitely. - Add new APIs that operate on individual intentions without IDs so that the UI doesn't need to implement CAS operations. - Add a new serf feature flag indicating support for intentions-as-config-entries. - The old line-item intentions way of interacting with the state store will transparently flip between the legacy memdb table and the config entry representations so that readers will never see a hiccup during migration where the results are incomplete. It uses a piece of system metadata to control the flip. - The primary datacenter will begin migrating intentions into config entries on startup once all servers in the datacenter are on a version of Consul with the intentions-as-config-entries feature flag. When it is complete the old state store representations will be cleared. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up. - The secondary datacenters continue to run the old intentions replicator until all servers in the secondary DC and primary DC support intentions-as-config-entries (via serf flag). Once this condition it met the old intentions replicator ceases. - The secondary datacenters replicate the new config entries as they are migrated in the primary. When they detect that the primary has zeroed it's old state store table it waits until all config entries up to that point are replicated and then zeroes its own copy of the old state store table. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up.
2020-10-06 18:24:05 +00:00
// UpdatableConfigEntry is the optional interface implemented by a ConfigEntry
// if it wants more control over how the update part of upsert works
// differently than a straight create. By default without this implementation
// all upsert operations are replacements.
type UpdatableConfigEntry interface {
// UpdateOver is called from the state machine when an identically named
// config entry already exists. This lets the config entry optionally
// choose to use existing information from a config entry (such as
// CreateTime) to slightly adjust how the update actually happens.
UpdateOver(prev ConfigEntry) error
ConfigEntry
}
2019-03-19 17:06:46 +00:00
// ServiceConfiguration is the top-level struct for the configuration of a service
// across the entire cluster.
type ServiceConfigEntry struct {
Kind string
Name string
Protocol string
Mode ProxyMode `json:",omitempty"`
TransparentProxy TransparentProxyConfig `json:",omitempty" alias:"transparent_proxy"`
MeshGateway MeshGatewayConfig `json:",omitempty" alias:"mesh_gateway"`
Expose ExposeConfig `json:",omitempty"`
ExternalSNI string `json:",omitempty" alias:"external_sni"`
UpstreamConfig *UpstreamConfiguration `json:",omitempty" alias:"upstream_config"`
2019-03-19 17:06:46 +00:00
Meta map[string]string `json:",omitempty"`
EnterpriseMeta `hcl:",squash" mapstructure:",squash"`
2019-03-19 17:06:46 +00:00
RaftIndex
}
server: config entry replication now correctly uses namespaces in comparisons (#9024) Previously config entries sharing a kind & name but in different namespaces could occasionally cause "stuck states" in replication because the namespace fields were ignored during the differential comparison phase. Example: Two config entries written to the primary: kind=A,name=web,namespace=bar kind=A,name=web,namespace=foo Under the covers these both get saved to memdb, so they are sorted by all 3 components (kind,name,namespace) during natural iteration. This means that before the replication code does it's own incomplete sort, the underlying data IS sorted by namespace ascending (bar comes before foo). After one pass of replication the primary and secondary datacenters have the same set of config entries present. If "kind=A,name=web,namespace=bar" were to be deleted, then things get weird. Before replication the two sides look like: primary: [ kind=A,name=web,namespace=foo ] secondary: [ kind=A,name=web,namespace=bar kind=A,name=web,namespace=foo ] The differential comparison phase walks these two lists in sorted order and first compares "kind=A,name=web,namespace=foo" vs "kind=A,name=web,namespace=bar" and falsely determines they are the SAME and are thus cause an update of "kind=A,name=web,namespace=foo". Then it compares "<nothing>" with "kind=A,name=web,namespace=foo" and falsely determines that the latter should be DELETED. During reconciliation the deletes are processed before updates, and so for a brief moment in the secondary "kind=A,name=web,namespace=foo" is erroneously deleted and then immediately restored. Unfortunately after this replication phase the final state is identical to the initial state, so when it loops around again (rate limited) it repeats the same set of operations indefinitely.
2020-10-23 18:41:54 +00:00
func (e *ServiceConfigEntry) Clone() *ServiceConfigEntry {
e2 := *e
e2.Expose = e.Expose.Clone()
e2.UpstreamConfig = e.UpstreamConfig.Clone()
server: config entry replication now correctly uses namespaces in comparisons (#9024) Previously config entries sharing a kind & name but in different namespaces could occasionally cause "stuck states" in replication because the namespace fields were ignored during the differential comparison phase. Example: Two config entries written to the primary: kind=A,name=web,namespace=bar kind=A,name=web,namespace=foo Under the covers these both get saved to memdb, so they are sorted by all 3 components (kind,name,namespace) during natural iteration. This means that before the replication code does it's own incomplete sort, the underlying data IS sorted by namespace ascending (bar comes before foo). After one pass of replication the primary and secondary datacenters have the same set of config entries present. If "kind=A,name=web,namespace=bar" were to be deleted, then things get weird. Before replication the two sides look like: primary: [ kind=A,name=web,namespace=foo ] secondary: [ kind=A,name=web,namespace=bar kind=A,name=web,namespace=foo ] The differential comparison phase walks these two lists in sorted order and first compares "kind=A,name=web,namespace=foo" vs "kind=A,name=web,namespace=bar" and falsely determines they are the SAME and are thus cause an update of "kind=A,name=web,namespace=foo". Then it compares "<nothing>" with "kind=A,name=web,namespace=foo" and falsely determines that the latter should be DELETED. During reconciliation the deletes are processed before updates, and so for a brief moment in the secondary "kind=A,name=web,namespace=foo" is erroneously deleted and then immediately restored. Unfortunately after this replication phase the final state is identical to the initial state, so when it loops around again (rate limited) it repeats the same set of operations indefinitely.
2020-10-23 18:41:54 +00:00
return &e2
}
func (e *ServiceConfigEntry) GetKind() string {
2019-03-19 17:06:46 +00:00
return ServiceDefaults
}
func (e *ServiceConfigEntry) GetName() string {
if e == nil {
return ""
}
return e.Name
}
func (e *ServiceConfigEntry) GetMeta() map[string]string {
if e == nil {
return nil
}
return e.Meta
}
func (e *ServiceConfigEntry) Normalize() error {
if e == nil {
return fmt.Errorf("config entry is nil")
}
e.Kind = ServiceDefaults
e.Protocol = strings.ToLower(e.Protocol)
e.EnterpriseMeta.Normalize()
var validationErr error
if e.UpstreamConfig != nil {
for _, override := range e.UpstreamConfig.Overrides {
err := override.NormalizeWithName(&e.EnterpriseMeta)
if err != nil {
validationErr = multierror.Append(validationErr, fmt.Errorf("error in upstream override for %s: %v", override.ServiceName(), err))
}
}
if e.UpstreamConfig.Defaults != nil {
err := e.UpstreamConfig.Defaults.NormalizeWithoutName()
if err != nil {
validationErr = multierror.Append(validationErr, fmt.Errorf("error in upstream defaults: %v", err))
}
}
}
return validationErr
}
func (e *ServiceConfigEntry) Validate() error {
if e.Name == "" {
return fmt.Errorf("Name is required")
}
if e.Name == WildcardSpecifier {
return fmt.Errorf("service-defaults name must be the name of a service, and not a wildcard")
}
validationErr := validateConfigEntryMeta(e.Meta)
if e.UpstreamConfig != nil {
for _, override := range e.UpstreamConfig.Overrides {
err := override.ValidateWithName()
if err != nil {
validationErr = multierror.Append(validationErr, fmt.Errorf("error in upstream override for %s: %v", override.ServiceName(), err))
}
}
if e.UpstreamConfig.Defaults != nil {
if err := e.UpstreamConfig.Defaults.ValidateWithoutName(); err != nil {
validationErr = multierror.Append(validationErr, fmt.Errorf("error in upstream defaults: %v", err))
}
}
}
return validationErr
}
func (e *ServiceConfigEntry) CanRead(authz acl.Authorizer) bool {
var authzContext acl.AuthorizerContext
e.FillAuthzContext(&authzContext)
return authz.ServiceRead(e.Name, &authzContext) == acl.Allow
}
func (e *ServiceConfigEntry) CanWrite(authz acl.Authorizer) bool {
var authzContext acl.AuthorizerContext
e.FillAuthzContext(&authzContext)
return authz.ServiceWrite(e.Name, &authzContext) == acl.Allow
}
func (e *ServiceConfigEntry) GetRaftIndex() *RaftIndex {
if e == nil {
return &RaftIndex{}
}
return &e.RaftIndex
}
func (e *ServiceConfigEntry) GetEnterpriseMeta() *EnterpriseMeta {
if e == nil {
return nil
}
return &e.EnterpriseMeta
}
type UpstreamConfiguration struct {
// Overrides is a slice of per-service configuration. The name field is
// required.
Overrides []*UpstreamConfig `json:",omitempty"`
// Defaults contains default configuration for all upstreams of a given
// service. The name field must be empty.
Defaults *UpstreamConfig `json:",omitempty"`
}
func (c *UpstreamConfiguration) Clone() *UpstreamConfiguration {
if c == nil {
return nil
}
var c2 UpstreamConfiguration
if len(c.Overrides) > 0 {
c2.Overrides = make([]*UpstreamConfig, 0, len(c.Overrides))
for _, o := range c.Overrides {
dup := o.Clone()
c2.Overrides = append(c2.Overrides, &dup)
}
}
if c.Defaults != nil {
def2 := c.Defaults.Clone()
c2.Defaults = &def2
}
return &c2
2019-03-19 17:06:46 +00:00
}
// ProxyConfigEntry is the top-level struct for global proxy configuration defaults.
type ProxyConfigEntry struct {
Kind string
Name string
Config map[string]interface{}
Mode ProxyMode `json:",omitempty"`
TransparentProxy TransparentProxyConfig `json:",omitempty" alias:"transparent_proxy"`
MeshGateway MeshGatewayConfig `json:",omitempty" alias:"mesh_gateway"`
Expose ExposeConfig `json:",omitempty"`
Meta map[string]string `json:",omitempty"`
EnterpriseMeta `hcl:",squash" mapstructure:",squash"`
RaftIndex
2019-03-19 17:06:46 +00:00
}
func (e *ProxyConfigEntry) GetKind() string {
2019-03-19 17:06:46 +00:00
return ProxyDefaults
}
func (e *ProxyConfigEntry) GetName() string {
if e == nil {
return ""
}
return e.Name
}
func (e *ProxyConfigEntry) GetMeta() map[string]string {
if e == nil {
return nil
}
return e.Meta
}
func (e *ProxyConfigEntry) Normalize() error {
if e == nil {
return fmt.Errorf("config entry is nil")
}
e.Kind = ProxyDefaults
e.Name = ProxyConfigGlobal
e.EnterpriseMeta.Normalize()
return nil
}
func (e *ProxyConfigEntry) Validate() error {
if e == nil {
return fmt.Errorf("config entry is nil")
}
if e.Name != ProxyConfigGlobal {
return fmt.Errorf("invalid name (%q), only %q is supported", e.Name, ProxyConfigGlobal)
}
if err := validateConfigEntryMeta(e.Meta); err != nil {
return err
}
return e.validateEnterpriseMeta()
}
func (e *ProxyConfigEntry) CanRead(authz acl.Authorizer) bool {
return true
}
func (e *ProxyConfigEntry) CanWrite(authz acl.Authorizer) bool {
var authzContext acl.AuthorizerContext
e.FillAuthzContext(&authzContext)
return authz.OperatorWrite(&authzContext) == acl.Allow
}
func (e *ProxyConfigEntry) GetRaftIndex() *RaftIndex {
if e == nil {
return &RaftIndex{}
}
return &e.RaftIndex
}
func (e *ProxyConfigEntry) GetEnterpriseMeta() *EnterpriseMeta {
if e == nil {
return nil
}
return &e.EnterpriseMeta
}
func (e *ProxyConfigEntry) MarshalBinary() (data []byte, err error) {
// We mainly want to implement the BinaryMarshaller interface so that
// we can fixup some msgpack types to coerce them into JSON compatible
// values. No special encoding needs to be done - we just simply msgpack
// encode the struct which requires a type alias to prevent recursively
// calling this function.
type alias ProxyConfigEntry
a := alias(*e)
// bs will grow if needed but allocate enough to avoid reallocation in common
// case.
bs := make([]byte, 128)
enc := codec.NewEncoderBytes(&bs, MsgpackHandle)
err = enc.Encode(a)
if err != nil {
return nil, err
}
return bs, nil
}
func (e *ProxyConfigEntry) UnmarshalBinary(data []byte) error {
// The goal here is to add a post-decoding operation to
// decoding of a ProxyConfigEntry. The cleanest way I could
// find to do so was to implement the BinaryMarshaller interface
// and use a type alias to do the original round of decoding,
// followed by a MapWalk of the Config to coerce everything
// into JSON compatible types.
type alias ProxyConfigEntry
var a alias
dec := codec.NewDecoderBytes(data, MsgpackHandle)
if err := dec.Decode(&a); err != nil {
return err
}
*e = ProxyConfigEntry(a)
config, err := lib.MapWalk(e.Config)
if err != nil {
return err
}
e.Config = config
return nil
}
// DecodeConfigEntry can be used to decode a ConfigEntry from a raw map value.
// Currently its used in the HTTP API to decode ConfigEntry structs coming from
// JSON. Unlike some of our custom binary encodings we don't have a preamble including
// the kind so we will not have a concrete type to decode into. In those cases we must
// first decode into a map[string]interface{} and then call this function to decode
// into a concrete type.
//
// There is an 'api' variation of this in
// command/config/write/config_write.go:newDecodeConfigEntry
func DecodeConfigEntry(raw map[string]interface{}) (ConfigEntry, error) {
var entry ConfigEntry
kindVal, ok := raw["Kind"]
if !ok {
kindVal, ok = raw["kind"]
}
if !ok {
return nil, fmt.Errorf("Payload does not contain a kind/Kind key at the top level")
}
if kindStr, ok := kindVal.(string); ok {
newEntry, err := MakeConfigEntry(kindStr, "")
if err != nil {
return nil, err
}
entry = newEntry
} else {
return nil, fmt.Errorf("Kind value in payload is not a string")
}
var md mapstructure.Metadata
decodeConf := &mapstructure.DecoderConfig{
DecodeHook: mapstructure.ComposeDecodeHookFunc(
decode.HookWeakDecodeFromSlice,
decode.HookTranslateKeys,
mapstructure.StringToTimeDurationHookFunc(),
connect: intentions are now managed as a new config entry kind "service-intentions" (#8834) - Upgrade the ConfigEntry.ListAll RPC to be kind-aware so that older copies of consul will not see new config entries it doesn't understand replicate down. - Add shim conversion code so that the old API/CLI method of interacting with intentions will continue to work so long as none of these are edited via config entry endpoints. Almost all of the read-only APIs will continue to function indefinitely. - Add new APIs that operate on individual intentions without IDs so that the UI doesn't need to implement CAS operations. - Add a new serf feature flag indicating support for intentions-as-config-entries. - The old line-item intentions way of interacting with the state store will transparently flip between the legacy memdb table and the config entry representations so that readers will never see a hiccup during migration where the results are incomplete. It uses a piece of system metadata to control the flip. - The primary datacenter will begin migrating intentions into config entries on startup once all servers in the datacenter are on a version of Consul with the intentions-as-config-entries feature flag. When it is complete the old state store representations will be cleared. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up. - The secondary datacenters continue to run the old intentions replicator until all servers in the secondary DC and primary DC support intentions-as-config-entries (via serf flag). Once this condition it met the old intentions replicator ceases. - The secondary datacenters replicate the new config entries as they are migrated in the primary. When they detect that the primary has zeroed it's old state store table it waits until all config entries up to that point are replicated and then zeroes its own copy of the old state store table. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up.
2020-10-06 18:24:05 +00:00
mapstructure.StringToTimeHookFunc(time.RFC3339),
),
Metadata: &md,
Result: &entry,
WeaklyTypedInput: true,
}
decoder, err := mapstructure.NewDecoder(decodeConf)
if err != nil {
return nil, err
}
if err := decoder.Decode(raw); err != nil {
return nil, err
}
if err := validateUnusedKeys(md.Unused); err != nil {
return nil, err
}
return entry, nil
}
type ConfigEntryOp string
const (
ConfigEntryUpsert ConfigEntryOp = "upsert"
ConfigEntryUpsertCAS ConfigEntryOp = "upsert-cas"
ConfigEntryDelete ConfigEntryOp = "delete"
)
// ConfigEntryRequest is used when creating/updating/deleting a ConfigEntry.
type ConfigEntryRequest struct {
Op ConfigEntryOp
Datacenter string
Entry ConfigEntry
WriteRequest
}
func (c *ConfigEntryRequest) RequestDatacenter() string {
return c.Datacenter
}
func (c *ConfigEntryRequest) MarshalBinary() (data []byte, err error) {
// bs will grow if needed but allocate enough to avoid reallocation in common
// case.
bs := make([]byte, 128)
enc := codec.NewEncoderBytes(&bs, MsgpackHandle)
// Encode kind first
err = enc.Encode(c.Entry.GetKind())
if err != nil {
return nil, err
}
// Then actual value using alias trick to avoid infinite recursion
type Alias ConfigEntryRequest
err = enc.Encode(struct {
*Alias
}{
Alias: (*Alias)(c),
})
if err != nil {
return nil, err
}
return bs, nil
}
func (c *ConfigEntryRequest) UnmarshalBinary(data []byte) error {
// First decode the kind prefix
var kind string
dec := codec.NewDecoderBytes(data, MsgpackHandle)
if err := dec.Decode(&kind); err != nil {
return err
}
// Then decode the real thing with appropriate kind of ConfigEntry
entry, err := MakeConfigEntry(kind, "")
if err != nil {
return err
}
c.Entry = entry
// Alias juggling to prevent infinite recursive calls back to this decode
// method.
type Alias ConfigEntryRequest
as := struct {
*Alias
}{
Alias: (*Alias)(c),
}
if err := dec.Decode(&as); err != nil {
return err
}
return nil
}
func MakeConfigEntry(kind, name string) (ConfigEntry, error) {
switch kind {
case ServiceDefaults:
return &ServiceConfigEntry{Name: name}, nil
case ProxyDefaults:
return &ProxyConfigEntry{Name: name}, nil
case ServiceRouter:
return &ServiceRouterConfigEntry{Name: name}, nil
case ServiceSplitter:
return &ServiceSplitterConfigEntry{Name: name}, nil
case ServiceResolver:
return &ServiceResolverConfigEntry{Name: name}, nil
case IngressGateway:
return &IngressGatewayConfigEntry{Name: name}, nil
case TerminatingGateway:
return &TerminatingGatewayConfigEntry{Name: name}, nil
connect: intentions are now managed as a new config entry kind "service-intentions" (#8834) - Upgrade the ConfigEntry.ListAll RPC to be kind-aware so that older copies of consul will not see new config entries it doesn't understand replicate down. - Add shim conversion code so that the old API/CLI method of interacting with intentions will continue to work so long as none of these are edited via config entry endpoints. Almost all of the read-only APIs will continue to function indefinitely. - Add new APIs that operate on individual intentions without IDs so that the UI doesn't need to implement CAS operations. - Add a new serf feature flag indicating support for intentions-as-config-entries. - The old line-item intentions way of interacting with the state store will transparently flip between the legacy memdb table and the config entry representations so that readers will never see a hiccup during migration where the results are incomplete. It uses a piece of system metadata to control the flip. - The primary datacenter will begin migrating intentions into config entries on startup once all servers in the datacenter are on a version of Consul with the intentions-as-config-entries feature flag. When it is complete the old state store representations will be cleared. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up. - The secondary datacenters continue to run the old intentions replicator until all servers in the secondary DC and primary DC support intentions-as-config-entries (via serf flag). Once this condition it met the old intentions replicator ceases. - The secondary datacenters replicate the new config entries as they are migrated in the primary. When they detect that the primary has zeroed it's old state store table it waits until all config entries up to that point are replicated and then zeroes its own copy of the old state store table. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up.
2020-10-06 18:24:05 +00:00
case ServiceIntentions:
return &ServiceIntentionsConfigEntry{Name: name}, nil
case MeshConfig:
return &MeshConfigEntry{}, nil
default:
return nil, fmt.Errorf("invalid config entry kind: %s", kind)
}
}
// ConfigEntryQuery is used when requesting info about a config entry.
type ConfigEntryQuery struct {
Kind string
Name string
Datacenter string
EnterpriseMeta `hcl:",squash" mapstructure:",squash"`
QueryOptions
}
func (c *ConfigEntryQuery) RequestDatacenter() string {
return c.Datacenter
}
func (r *ConfigEntryQuery) CacheInfo() cache.RequestInfo {
info := cache.RequestInfo{
Token: r.Token,
Datacenter: r.Datacenter,
MinIndex: r.MinQueryIndex,
Timeout: r.MaxQueryTime,
MaxAge: r.MaxAge,
MustRevalidate: r.MustRevalidate,
}
v, err := hashstructure.Hash([]interface{}{
r.Kind,
r.Name,
r.Filter,
r.EnterpriseMeta,
}, nil)
if err == nil {
// If there is an error, we don't set the key. A blank key forces
// no cache for this request so the request is forwarded directly
// to the server.
info.Key = strconv.FormatUint(v, 10)
}
return info
}
connect: intentions are now managed as a new config entry kind "service-intentions" (#8834) - Upgrade the ConfigEntry.ListAll RPC to be kind-aware so that older copies of consul will not see new config entries it doesn't understand replicate down. - Add shim conversion code so that the old API/CLI method of interacting with intentions will continue to work so long as none of these are edited via config entry endpoints. Almost all of the read-only APIs will continue to function indefinitely. - Add new APIs that operate on individual intentions without IDs so that the UI doesn't need to implement CAS operations. - Add a new serf feature flag indicating support for intentions-as-config-entries. - The old line-item intentions way of interacting with the state store will transparently flip between the legacy memdb table and the config entry representations so that readers will never see a hiccup during migration where the results are incomplete. It uses a piece of system metadata to control the flip. - The primary datacenter will begin migrating intentions into config entries on startup once all servers in the datacenter are on a version of Consul with the intentions-as-config-entries feature flag. When it is complete the old state store representations will be cleared. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up. - The secondary datacenters continue to run the old intentions replicator until all servers in the secondary DC and primary DC support intentions-as-config-entries (via serf flag). Once this condition it met the old intentions replicator ceases. - The secondary datacenters replicate the new config entries as they are migrated in the primary. When they detect that the primary has zeroed it's old state store table it waits until all config entries up to that point are replicated and then zeroes its own copy of the old state store table. We also record a piece of system metadata indicating this has occurred. We use this metadata to skip ALL of this code the next time the leader starts up.
2020-10-06 18:24:05 +00:00
// ConfigEntryListAllRequest is used when requesting to list all config entries
// of a set of kinds.
type ConfigEntryListAllRequest struct {
// Kinds should always be set. For backwards compatibility with versions
// prior to 1.9.0, if this is omitted or left empty it is assumed to mean
// the subset of config entry kinds that were present in 1.8.0:
//
// proxy-defaults, service-defaults, service-resolver, service-splitter,
// service-router, terminating-gateway, and ingress-gateway.
Kinds []string
Datacenter string
EnterpriseMeta `hcl:",squash" mapstructure:",squash"`
QueryOptions
}
func (r *ConfigEntryListAllRequest) RequestDatacenter() string {
return r.Datacenter
}
// ServiceConfigRequest is used when requesting the resolved configuration
// for a service.
type ServiceConfigRequest struct {
Name string
Datacenter string
// MeshGateway contains the mesh gateway configuration from the requesting proxy's registration
MeshGateway MeshGatewayConfig
// Mode indicates how the requesting proxy's listeners are dialed
Mode ProxyMode
UpstreamIDs []ServiceID
// DEPRECATED
// Upstreams is a list of upstream service names to use for resolving the service config
// UpstreamIDs should be used instead which can encode more than just the name to
// uniquely identify a service.
Upstreams []string
EnterpriseMeta `hcl:",squash" mapstructure:",squash"`
QueryOptions
}
func (s *ServiceConfigRequest) RequestDatacenter() string {
return s.Datacenter
}
func (r *ServiceConfigRequest) CacheInfo() cache.RequestInfo {
info := cache.RequestInfo{
Token: r.Token,
Datacenter: r.Datacenter,
MinIndex: r.MinQueryIndex,
Timeout: r.MaxQueryTime,
MaxAge: r.MaxAge,
MustRevalidate: r.MustRevalidate,
}
// To calculate the cache key we only hash the service name and upstream set.
// We don't want ordering of the upstreams to affect the outcome so use an
// anonymous struct field with hash:set behavior. Note the order of fields in
// the slice would affect cache keys if we ever persist between agent restarts
// and change it.
v, err := hashstructure.Hash(struct {
Name string
EnterpriseMeta EnterpriseMeta
Upstreams []string `hash:"set"`
}{
Name: r.Name,
EnterpriseMeta: r.EnterpriseMeta,
Upstreams: r.Upstreams,
}, nil)
if err == nil {
// If there is an error, we don't set the key. A blank key forces
// no cache for this request so the request is forwarded directly
// to the server.
info.Key = strconv.FormatUint(v, 10)
}
return info
}
type UpstreamConfig struct {
// Name is only accepted within a service-defaults config entry.
Name string `json:",omitempty"`
// EnterpriseMeta is only accepted within a service-defaults config entry.
EnterpriseMeta `hcl:",squash" mapstructure:",squash"`
// EnvoyListenerJSON is a complete override ("escape hatch") for the upstream's
// listener.
//
// Note: This escape hatch is NOT compatible with the discovery chain and
// will be ignored if a discovery chain is active.
EnvoyListenerJSON string `json:",omitempty" alias:"envoy_listener_json"`
// EnvoyClusterJSON is a complete override ("escape hatch") for the upstream's
// cluster. The Connect client TLS certificate and context will be injected
// overriding any TLS settings present.
//
// Note: This escape hatch is NOT compatible with the discovery chain and
// will be ignored if a discovery chain is active.
EnvoyClusterJSON string `json:",omitempty" alias:"envoy_cluster_json"`
// Protocol describes the upstream's service protocol. Valid values are "tcp",
// "http" and "grpc". Anything else is treated as tcp. The enables protocol
// aware features like per-request metrics and connection pooling, tracing,
// routing etc.
Protocol string `json:",omitempty"`
// ConnectTimeoutMs is the number of milliseconds to timeout making a new
// connection to this upstream. Defaults to 5000 (5 seconds) if not set.
ConnectTimeoutMs int `json:",omitempty" alias:"connect_timeout_ms"`
// Limits are the set of limits that are applied to the proxy for a specific upstream of a
// service instance.
Limits *UpstreamLimits `json:",omitempty"`
// PassiveHealthCheck configuration determines how upstream proxy instances will
// be monitored for removal from the load balancing pool.
PassiveHealthCheck *PassiveHealthCheck `json:",omitempty" alias:"passive_health_check"`
// MeshGatewayConfig controls how Mesh Gateways are configured and used
MeshGateway MeshGatewayConfig `json:",omitempty" alias:"mesh_gateway" `
}
func (cfg UpstreamConfig) Clone() UpstreamConfig {
cfg2 := cfg
cfg2.Limits = cfg.Limits.Clone()
cfg2.PassiveHealthCheck = cfg.PassiveHealthCheck.Clone()
return cfg2
}
func (cfg *UpstreamConfig) ServiceID() ServiceID {
if cfg.Name == "" {
return ServiceID{}
}
return NewServiceID(cfg.Name, &cfg.EnterpriseMeta)
}
func (cfg *UpstreamConfig) ServiceName() ServiceName {
if cfg.Name == "" {
return ServiceName{}
}
return NewServiceName(cfg.Name, &cfg.EnterpriseMeta)
}
func (cfg UpstreamConfig) MergeInto(dst map[string]interface{}) {
// Avoid storing empty values in the map, since these can act as overrides
if cfg.EnvoyListenerJSON != "" {
dst["envoy_listener_json"] = cfg.EnvoyListenerJSON
}
if cfg.EnvoyClusterJSON != "" {
dst["envoy_cluster_json"] = cfg.EnvoyClusterJSON
}
if cfg.Protocol != "" {
dst["protocol"] = cfg.Protocol
}
if cfg.ConnectTimeoutMs != 0 {
dst["connect_timeout_ms"] = cfg.ConnectTimeoutMs
}
if !cfg.MeshGateway.IsZero() {
dst["mesh_gateway"] = cfg.MeshGateway
}
if cfg.Limits != nil {
dst["limits"] = cfg.Limits
}
if cfg.PassiveHealthCheck != nil {
dst["passive_health_check"] = cfg.PassiveHealthCheck
}
}
func (cfg *UpstreamConfig) NormalizeWithoutName() error {
return cfg.normalize(false, nil)
}
func (cfg *UpstreamConfig) NormalizeWithName(entMeta *EnterpriseMeta) error {
return cfg.normalize(true, entMeta)
}
func (cfg *UpstreamConfig) normalize(named bool, entMeta *EnterpriseMeta) error {
if named {
// If the upstream namespace is omitted it inherits that of the enclosing
// config entry.
cfg.EnterpriseMeta.MergeNoWildcard(entMeta)
cfg.EnterpriseMeta.Normalize()
}
cfg.Protocol = strings.ToLower(cfg.Protocol)
if cfg.ConnectTimeoutMs < 0 {
cfg.ConnectTimeoutMs = 0
}
return nil
}
func (cfg UpstreamConfig) ValidateWithoutName() error {
return cfg.validate(false)
}
func (cfg UpstreamConfig) ValidateWithName() error {
return cfg.validate(true)
}
func (cfg UpstreamConfig) validate(named bool) error {
if named {
if cfg.Name == "" {
return fmt.Errorf("Name is required")
}
} else {
if cfg.Name != "" {
return fmt.Errorf("Name must be empty")
}
if cfg.EnterpriseMeta.NamespaceOrEmpty() != "" {
return fmt.Errorf("Namespace must be empty")
}
}
var validationErr error
if cfg.PassiveHealthCheck != nil {
err := cfg.PassiveHealthCheck.Validate()
if err != nil {
validationErr = multierror.Append(validationErr, err)
}
}
if cfg.Limits != nil {
err := cfg.Limits.Validate()
if err != nil {
validationErr = multierror.Append(validationErr, err)
}
}
return validationErr
}
func ParseUpstreamConfigNoDefaults(m map[string]interface{}) (UpstreamConfig, error) {
var cfg UpstreamConfig
config := &mapstructure.DecoderConfig{
DecodeHook: mapstructure.ComposeDecodeHookFunc(
decode.HookWeakDecodeFromSlice,
decode.HookTranslateKeys,
mapstructure.StringToTimeDurationHookFunc(),
),
Result: &cfg,
WeaklyTypedInput: true,
}
decoder, err := mapstructure.NewDecoder(config)
if err != nil {
return cfg, err
}
if err := decoder.Decode(m); err != nil {
return cfg, err
}
err = cfg.NormalizeWithoutName()
return cfg, err
}
// ParseUpstreamConfig returns the UpstreamConfig parsed from an opaque map.
// If an error occurs during parsing it is returned along with the default
// config this allows caller to choose whether and how to report the error.
func ParseUpstreamConfig(m map[string]interface{}) (UpstreamConfig, error) {
cfg, err := ParseUpstreamConfigNoDefaults(m)
// Set default (even if error is returned)
if cfg.Protocol == "" {
cfg.Protocol = "tcp"
}
if cfg.ConnectTimeoutMs == 0 {
cfg.ConnectTimeoutMs = 5000
}
return cfg, err
}
type PassiveHealthCheck struct {
// Interval between health check analysis sweeps. Each sweep may remove
// hosts or return hosts to the pool.
Interval time.Duration `json:",omitempty"`
// MaxFailures is the count of consecutive failures that results in a host
// being removed from the pool.
MaxFailures uint32 `json:",omitempty" alias:"max_failures"`
}
func (chk *PassiveHealthCheck) Clone() *PassiveHealthCheck {
if chk == nil {
return nil
}
chk2 := *chk
return &chk2
}
func (chk *PassiveHealthCheck) IsZero() bool {
zeroVal := PassiveHealthCheck{}
return *chk == zeroVal
}
func (chk PassiveHealthCheck) Validate() error {
if chk.Interval < 0*time.Second {
return fmt.Errorf("passive health check interval cannot be negative")
}
return nil
}
// UpstreamLimits describes the limits that are associated with a specific
// upstream of a service instance.
type UpstreamLimits struct {
// MaxConnections is the maximum number of connections the local proxy can
// make to the upstream service.
MaxConnections *int `json:",omitempty" alias:"max_connections"`
// MaxPendingRequests is the maximum number of requests that will be queued
// waiting for an available connection. This is mostly applicable to HTTP/1.1
// clusters since all HTTP/2 requests are streamed over a single
// connection.
MaxPendingRequests *int `json:",omitempty" alias:"max_pending_requests"`
// MaxConcurrentRequests is the maximum number of in-flight requests that will be allowed
// to the upstream cluster at a point in time. This is mostly applicable to HTTP/2
// clusters since all HTTP/1.1 requests are limited by MaxConnections.
MaxConcurrentRequests *int `json:",omitempty" alias:"max_concurrent_requests"`
}
func (ul *UpstreamLimits) Clone() *UpstreamLimits {
if ul == nil {
return nil
}
return &UpstreamLimits{
MaxConnections: intPointerCopy(ul.MaxConnections),
MaxPendingRequests: intPointerCopy(ul.MaxPendingRequests),
MaxConcurrentRequests: intPointerCopy(ul.MaxConcurrentRequests),
}
}
func intPointerCopy(v *int) *int {
if v == nil {
return nil
}
v2 := *v
return &v2
}
func (ul *UpstreamLimits) IsZero() bool {
zeroVal := UpstreamLimits{}
return *ul == zeroVal
}
func (ul UpstreamLimits) Validate() error {
if ul.MaxConnections != nil && *ul.MaxConnections < 0 {
return fmt.Errorf("max connections cannot be negative")
}
if ul.MaxPendingRequests != nil && *ul.MaxPendingRequests < 0 {
return fmt.Errorf("max pending requests cannot be negative")
}
if ul.MaxConcurrentRequests != nil && *ul.MaxConcurrentRequests < 0 {
return fmt.Errorf("max concurrent requests cannot be negative")
}
return nil
}
type OpaqueUpstreamConfig struct {
Upstream ServiceID
Config map[string]interface{}
}
type OpaqueUpstreamConfigs []OpaqueUpstreamConfig
func (configs OpaqueUpstreamConfigs) GetUpstreamConfig(sid ServiceID) (config map[string]interface{}, found bool) {
for _, usconf := range configs {
if usconf.Upstream.Matches(sid) {
return usconf.Config, true
}
}
return nil, false
}
type ServiceConfigResponse struct {
ProxyConfig map[string]interface{}
UpstreamConfigs map[string]map[string]interface{}
UpstreamIDConfigs OpaqueUpstreamConfigs
MeshGateway MeshGatewayConfig `json:",omitempty"`
Expose ExposeConfig `json:",omitempty"`
TransparentProxy TransparentProxyConfig `json:",omitempty"`
Mode ProxyMode `json:",omitempty"`
QueryMeta
}
func (r *ServiceConfigResponse) Reset() {
r.ProxyConfig = nil
r.UpstreamConfigs = nil
r.MeshGateway = MeshGatewayConfig{}
}
// MarshalBinary writes ServiceConfigResponse as msgpack encoded. It's only here
// because we need custom decoding of the raw interface{} values.
func (r *ServiceConfigResponse) MarshalBinary() (data []byte, err error) {
// bs will grow if needed but allocate enough to avoid reallocation in common
// case.
bs := make([]byte, 128)
enc := codec.NewEncoderBytes(&bs, MsgpackHandle)
type Alias ServiceConfigResponse
if err := enc.Encode((*Alias)(r)); err != nil {
return nil, err
}
return bs, nil
}
// UnmarshalBinary decodes msgpack encoded ServiceConfigResponse. It used
// default msgpack encoding but fixes up the uint8 strings and other problems we
// have with encoding map[string]interface{}.
func (r *ServiceConfigResponse) UnmarshalBinary(data []byte) error {
dec := codec.NewDecoderBytes(data, MsgpackHandle)
type Alias ServiceConfigResponse
var a Alias
if err := dec.Decode(&a); err != nil {
return err
}
*r = ServiceConfigResponse(a)
var err error
// Fix strings and maps in the returned maps
r.ProxyConfig, err = lib.MapWalk(r.ProxyConfig)
if err != nil {
return err
}
for k := range r.UpstreamConfigs {
r.UpstreamConfigs[k], err = lib.MapWalk(r.UpstreamConfigs[k])
if err != nil {
return err
}
}
for k := range r.UpstreamIDConfigs {
r.UpstreamIDConfigs[k].Config, err = lib.MapWalk(r.UpstreamIDConfigs[k].Config)
if err != nil {
return err
}
}
return nil
}
// ConfigEntryResponse returns a single ConfigEntry
type ConfigEntryResponse struct {
Entry ConfigEntry
QueryMeta
}
func (c *ConfigEntryResponse) MarshalBinary() (data []byte, err error) {
// bs will grow if needed but allocate enough to avoid reallocation in common
// case.
bs := make([]byte, 128)
enc := codec.NewEncoderBytes(&bs, MsgpackHandle)
if c.Entry != nil {
if err := enc.Encode(c.Entry.GetKind()); err != nil {
return nil, err
}
if err := enc.Encode(c.Entry); err != nil {
return nil, err
}
} else {
if err := enc.Encode(""); err != nil {
return nil, err
}
}
if err := enc.Encode(c.QueryMeta); err != nil {
return nil, err
}
return bs, nil
}
func (c *ConfigEntryResponse) UnmarshalBinary(data []byte) error {
dec := codec.NewDecoderBytes(data, MsgpackHandle)
var kind string
if err := dec.Decode(&kind); err != nil {
return err
}
if kind != "" {
entry, err := MakeConfigEntry(kind, "")
if err != nil {
return err
}
if err := dec.Decode(entry); err != nil {
return err
}
c.Entry = entry
} else {
c.Entry = nil
}
if err := dec.Decode(&c.QueryMeta); err != nil {
return err
}
return nil
}
func validateConfigEntryMeta(meta map[string]string) error {
var err error
if len(meta) > metaMaxKeyPairs {
err = multierror.Append(err, fmt.Errorf(
"Meta exceeds maximum element count %d", metaMaxKeyPairs))
}
for k, v := range meta {
if len(k) > metaKeyMaxLength {
err = multierror.Append(err, fmt.Errorf(
"Meta key %q exceeds maximum length %d", k, metaKeyMaxLength))
}
if len(v) > metaValueMaxLength {
err = multierror.Append(err, fmt.Errorf(
"Meta value for key %q exceeds maximum length %d", k, metaValueMaxLength))
}
}
return err
}