open-consul/agent/consul/state/config_entry.go

1229 lines
36 KiB
Go
Raw Normal View History

package state
import (
"errors"
"fmt"
memdb "github.com/hashicorp/go-memdb"
"github.com/hashicorp/consul/agent/connect"
"github.com/hashicorp/consul/agent/consul/discoverychain"
"github.com/hashicorp/consul/agent/structs"
"github.com/hashicorp/consul/lib"
)
const (
configTableName = "config-entries"
)
type ConfigEntryLinkIndex struct {
}
type discoveryChainConfigEntry interface {
structs.ConfigEntry
// ListRelatedServices returns a list of other names of services referenced
// in this config entry.
ListRelatedServices() []structs.ServiceID
}
func (s *ConfigEntryLinkIndex) FromObject(obj interface{}) (bool, [][]byte, error) {
entry, ok := obj.(structs.ConfigEntry)
if !ok {
return false, nil, fmt.Errorf("object is not a ConfigEntry")
}
dcEntry, ok := entry.(discoveryChainConfigEntry)
if !ok {
return false, nil, nil
}
linkedServices := dcEntry.ListRelatedServices()
numLinks := len(linkedServices)
if numLinks == 0 {
return false, nil, nil
}
vals := make([][]byte, 0, numLinks)
for _, linkedService := range linkedServices {
vals = append(vals, []byte(linkedService.String()+"\x00"))
}
return true, vals, nil
}
func (s *ConfigEntryLinkIndex) FromArgs(args ...interface{}) ([]byte, error) {
if len(args) != 1 {
return nil, fmt.Errorf("must provide only a single argument")
}
arg, ok := args[0].(structs.ServiceID)
if !ok {
return nil, fmt.Errorf("argument must be a structs.ServiceID: %#v", args[0])
}
// Add the null character as a terminator
return []byte(arg.String() + "\x00"), nil
}
func (s *ConfigEntryLinkIndex) PrefixFromArgs(args ...interface{}) ([]byte, error) {
val, err := s.FromArgs(args...)
if err != nil {
return nil, err
}
// Strip the null terminator, the rest is a prefix
n := len(val)
if n > 0 {
return val[:n-1], nil
}
return val, nil
}
func init() {
registerSchema(configTableSchema)
}
// ConfigEntries is used to pull all the config entries for the snapshot.
func (s *Snapshot) ConfigEntries() ([]structs.ConfigEntry, error) {
entries, err := s.tx.Get(configTableName, "id")
if err != nil {
return nil, err
}
var ret []structs.ConfigEntry
for wrapped := entries.Next(); wrapped != nil; wrapped = entries.Next() {
ret = append(ret, wrapped.(structs.ConfigEntry))
}
return ret, nil
}
// ConfigEntry is used when restoring from a snapshot.
func (s *Restore) ConfigEntry(c structs.ConfigEntry) error {
return insertConfigEntryWithTxn(s.tx, c.GetRaftIndex().ModifyIndex, c)
}
// ConfigEntry is called to get a given config entry.
func (s *Store) ConfigEntry(ws memdb.WatchSet, kind, name string, entMeta *structs.EnterpriseMeta) (uint64, structs.ConfigEntry, error) {
tx := s.db.Txn(false)
defer tx.Abort()
return configEntryTxn(tx, ws, kind, name, entMeta)
}
func configEntryTxn(tx ReadTxn, ws memdb.WatchSet, kind, name string, entMeta *structs.EnterpriseMeta) (uint64, structs.ConfigEntry, error) {
// Get the index
idx := maxIndexTxn(tx, configTableName)
// Get the existing config entry.
watchCh, existing, err := firstWatchConfigEntryWithTxn(tx, kind, name, entMeta)
if err != nil {
return 0, nil, fmt.Errorf("failed config entry lookup: %s", err)
}
ws.Add(watchCh)
if existing == nil {
return idx, nil, nil
}
conf, ok := existing.(structs.ConfigEntry)
if !ok {
return 0, nil, fmt.Errorf("config entry %q (%s) is an invalid type: %T", name, kind, conf)
}
return idx, conf, nil
}
// ConfigEntries is called to get all config entry objects.
func (s *Store) ConfigEntries(ws memdb.WatchSet, entMeta *structs.EnterpriseMeta) (uint64, []structs.ConfigEntry, error) {
return s.ConfigEntriesByKind(ws, "", entMeta)
}
// ConfigEntriesByKind is called to get all config entry objects with the given kind.
// If kind is empty, all config entries will be returned.
func (s *Store) ConfigEntriesByKind(ws memdb.WatchSet, kind string, entMeta *structs.EnterpriseMeta) (uint64, []structs.ConfigEntry, error) {
tx := s.db.Txn(false)
defer tx.Abort()
return configEntriesByKindTxn(tx, ws, kind, entMeta)
}
func configEntriesByKindTxn(tx ReadTxn, ws memdb.WatchSet, kind string, entMeta *structs.EnterpriseMeta) (uint64, []structs.ConfigEntry, error) {
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
// Get the index and watch for updates
idx := maxIndexWatchTxn(tx, ws, configTableName)
// Lookup by kind, or all if kind is empty
var iter memdb.ResultIterator
var err error
if kind != "" {
iter, err = getConfigEntryKindsWithTxn(tx, kind, entMeta)
} else {
iter, err = getAllConfigEntriesWithTxn(tx, entMeta)
}
if err != nil {
return 0, nil, fmt.Errorf("failed config entry lookup: %s", err)
}
ws.Add(iter.WatchCh())
var results []structs.ConfigEntry
for v := iter.Next(); v != nil; v = iter.Next() {
results = append(results, v.(structs.ConfigEntry))
}
return idx, results, nil
}
// EnsureConfigEntry is called to do an upsert of a given config entry.
func (s *Store) EnsureConfigEntry(idx uint64, conf structs.ConfigEntry, entMeta *structs.EnterpriseMeta) error {
tx := s.db.WriteTxn(idx)
defer tx.Abort()
if err := ensureConfigEntryTxn(tx, idx, conf, entMeta); err != nil {
return err
}
2020-06-02 20:34:56 +00:00
return tx.Commit()
}
// ensureConfigEntryTxn upserts a config entry inside of a transaction.
func ensureConfigEntryTxn(tx WriteTxn, idx uint64, conf structs.ConfigEntry, entMeta *structs.EnterpriseMeta) error {
// Check for existing configuration.
existing, err := firstConfigEntryWithTxn(tx, conf.GetKind(), conf.GetName(), entMeta)
if err != nil {
return fmt.Errorf("failed configuration lookup: %s", err)
}
raftIndex := conf.GetRaftIndex()
if existing != nil {
existingIdx := existing.(structs.ConfigEntry).GetRaftIndex()
raftIndex.CreateIndex = existingIdx.CreateIndex
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
// Handle optional upsert logic.
if updatableConf, ok := conf.(structs.UpdatableConfigEntry); ok {
if err := updatableConf.UpdateOver(existing.(structs.ConfigEntry)); err != nil {
return err
}
}
} else {
raftIndex.CreateIndex = idx
}
raftIndex.ModifyIndex = idx
err = validateProposedConfigEntryInGraph(tx, conf.GetKind(), conf.GetName(), conf, entMeta)
if err != nil {
return err // Err is already sufficiently decorated.
}
if err := validateConfigEntryEnterprise(tx, conf); err != nil {
return err
}
return insertConfigEntryWithTxn(tx, idx, conf)
}
// EnsureConfigEntryCAS is called to do a check-and-set upsert of a given config entry.
func (s *Store) EnsureConfigEntryCAS(idx, cidx uint64, conf structs.ConfigEntry, entMeta *structs.EnterpriseMeta) (bool, error) {
tx := s.db.WriteTxn(idx)
defer tx.Abort()
// Check for existing configuration.
existing, err := firstConfigEntryWithTxn(tx, conf.GetKind(), conf.GetName(), entMeta)
if err != nil {
return false, fmt.Errorf("failed configuration lookup: %s", err)
}
// Check if the we should do the set. A ModifyIndex of 0 means that
// we are doing a set-if-not-exists.
var existingIdx structs.RaftIndex
if existing != nil {
existingIdx = *existing.(structs.ConfigEntry).GetRaftIndex()
}
if cidx == 0 && existing != nil {
return false, nil
}
if cidx != 0 && existing == nil {
return false, nil
}
if existing != nil && cidx != 0 && cidx != existingIdx.ModifyIndex {
return false, nil
}
if err := ensureConfigEntryTxn(tx, idx, conf, entMeta); err != nil {
return false, err
}
2020-06-02 20:34:56 +00:00
err = tx.Commit()
return err == nil, err
}
func (s *Store) DeleteConfigEntry(idx uint64, kind, name string, entMeta *structs.EnterpriseMeta) error {
tx := s.db.WriteTxn(idx)
defer tx.Abort()
if err := deleteConfigEntryTxn(tx, idx, kind, name, entMeta); err != nil {
return err
}
return tx.Commit()
}
func deleteConfigEntryTxn(tx WriteTxn, idx uint64, kind, name string, entMeta *structs.EnterpriseMeta) error {
// Try to retrieve the existing config entry.
existing, err := firstConfigEntryWithTxn(tx, kind, name, entMeta)
if err != nil {
return fmt.Errorf("failed config entry lookup: %s", err)
}
if existing == nil {
return nil
}
// If the config entry is for terminating or ingress gateways we delete entries from the memdb table
// that associates gateways <-> services.
sn := structs.NewServiceName(name, entMeta)
if kind == structs.TerminatingGateway || kind == structs.IngressGateway {
if _, err := tx.DeleteAll(gatewayServicesTableName, "gateway", sn); err != nil {
return fmt.Errorf("failed to truncate gateway services table: %v", err)
}
if err := indexUpdateMaxTxn(tx, idx, gatewayServicesTableName); err != nil {
return fmt.Errorf("failed updating gateway-services index: %v", err)
}
}
// Also clean up associations in the mesh topology table for ingress gateways
if kind == structs.IngressGateway {
if _, err := tx.DeleteAll(topologyTableName, "downstream", sn); err != nil {
return fmt.Errorf("failed to truncate %s table: %v", topologyTableName, err)
}
if err := indexUpdateMaxTxn(tx, idx, topologyTableName); err != nil {
return fmt.Errorf("failed updating %s index: %v", topologyTableName, err)
}
}
err = validateProposedConfigEntryInGraph(tx, kind, name, nil, entMeta)
if err != nil {
return err // Err is already sufficiently decorated.
}
// Delete the config entry from the DB and update the index.
if err := tx.Delete(configTableName, existing); err != 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
return fmt.Errorf("failed removing config entry: %s", err)
}
if err := tx.Insert("index", &IndexEntry{configTableName, idx}); err != nil {
return fmt.Errorf("failed updating index: %s", err)
}
return nil
}
func insertConfigEntryWithTxn(tx WriteTxn, idx uint64, conf structs.ConfigEntry) error {
if conf == nil {
return fmt.Errorf("cannot insert nil config entry")
}
// If the config entry is for a terminating or ingress gateway we update the memdb table
// that associates gateways <-> services.
if conf.GetKind() == structs.TerminatingGateway || conf.GetKind() == structs.IngressGateway {
err := updateGatewayServices(tx, idx, conf, conf.GetEnterpriseMeta())
if err != nil {
return fmt.Errorf("failed to associate services to gateway: %v", err)
}
}
// Insert the config entry and update the index
if err := tx.Insert(configTableName, conf); err != nil {
return fmt.Errorf("failed inserting config entry: %s", err)
}
if err := indexUpdateMaxTxn(tx, idx, configTableName); err != nil {
return fmt.Errorf("failed updating index: %v", err)
}
return nil
}
// validateProposedConfigEntryInGraph can be used to verify graph integrity for
// a proposed graph create/update/delete.
//
// This must be called before any mutations occur on the config entries table!
//
// May return *ConfigEntryGraphValidationError if there is a concern to surface
// to the caller that they can correct.
func validateProposedConfigEntryInGraph(
tx ReadTxn,
kind, name string,
proposedEntry structs.ConfigEntry,
entMeta *structs.EnterpriseMeta,
) error {
validateAllChains := false
switch kind {
case structs.ProxyDefaults:
if name != structs.ProxyConfigGlobal {
return nil
}
validateAllChains = true
case structs.ServiceDefaults:
case structs.ServiceRouter:
case structs.ServiceSplitter:
case structs.ServiceResolver:
case structs.IngressGateway:
err := checkGatewayClash(tx, name, structs.IngressGateway, structs.TerminatingGateway, entMeta)
if err != nil {
return err
}
case structs.TerminatingGateway:
err := checkGatewayClash(tx, name, structs.TerminatingGateway, structs.IngressGateway, entMeta)
if err != nil {
return err
}
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 structs.ServiceIntentions:
default:
return fmt.Errorf("unhandled kind %q during validation of %q", kind, name)
}
return validateProposedConfigEntryInServiceGraph(tx, kind, name, proposedEntry, validateAllChains, entMeta)
}
func checkGatewayClash(
tx ReadTxn,
name, selfKind, otherKind string,
entMeta *structs.EnterpriseMeta,
) error {
_, entry, err := configEntryTxn(tx, nil, otherKind, name, entMeta)
if err != nil {
return err
}
if entry != nil {
return fmt.Errorf("cannot create a %q config entry with name %q, "+
"a %q config entry with that name already exists", selfKind, name, otherKind)
}
return nil
}
var serviceGraphKinds = []string{
structs.ServiceRouter,
structs.ServiceSplitter,
structs.ServiceResolver,
}
2020-09-29 01:42:03 +00:00
// discoveryChainTargets will return a list of services listed as a target for the input's discovery chain
func (s *Store) discoveryChainTargetsTxn(tx ReadTxn, ws memdb.WatchSet, dc, service string, entMeta *structs.EnterpriseMeta) (uint64, []structs.ServiceName, error) {
source := structs.NewServiceName(service, entMeta)
req := discoverychain.CompileRequest{
ServiceName: source.Name,
EvaluateInNamespace: source.NamespaceOrDefault(),
EvaluateInDatacenter: dc,
UseInDatacenter: dc,
}
idx, chain, err := s.serviceDiscoveryChainTxn(tx, ws, source.Name, entMeta, req)
if err != nil {
return 0, nil, fmt.Errorf("failed to fetch discovery chain for %q: %v", source.String(), err)
}
var resp []structs.ServiceName
for _, t := range chain.Targets {
em := structs.EnterpriseMetaInitializer(t.Namespace)
target := structs.NewServiceName(t.Service, &em)
// TODO (freddy): Allow upstream DC and encode in response
if t.Datacenter == dc {
resp = append(resp, target)
}
}
return idx, resp, nil
}
// discoveryChainSourcesTxn will return a list of services whose discovery chains have the given service as a target
func (s *Store) discoveryChainSourcesTxn(tx ReadTxn, ws memdb.WatchSet, dc string, destination structs.ServiceName) (uint64, []structs.ServiceName, error) {
seenLink := map[structs.ServiceName]bool{destination: true}
queue := []structs.ServiceName{destination}
for len(queue) > 0 {
// The "link" index returns config entries that reference a service
iter, err := tx.Get(configTableName, "link", queue[0].ToServiceID())
if err != nil {
return 0, nil, err
}
ws.Add(iter.WatchCh())
for raw := iter.Next(); raw != nil; raw = iter.Next() {
entry := raw.(structs.ConfigEntry)
sn := structs.NewServiceName(entry.GetName(), entry.GetEnterpriseMeta())
if !seenLink[sn] {
seenLink[sn] = true
queue = append(queue, sn)
}
}
queue = queue[1:]
}
var (
maxIdx uint64 = 1
resp []structs.ServiceName
)
// Only return the services that target the destination anywhere in their discovery chains.
seenSource := make(map[structs.ServiceName]bool)
2020-09-30 14:23:19 +00:00
for sn := range seenLink {
req := discoverychain.CompileRequest{
ServiceName: sn.Name,
EvaluateInNamespace: sn.NamespaceOrDefault(),
EvaluateInDatacenter: dc,
UseInDatacenter: dc,
}
idx, chain, err := s.serviceDiscoveryChainTxn(tx, ws, sn.Name, &sn.EnterpriseMeta, req)
if err != nil {
return 0, nil, fmt.Errorf("failed to fetch discovery chain for %q: %v", sn.String(), err)
}
for _, t := range chain.Targets {
em := structs.EnterpriseMetaInitializer(t.Namespace)
candidate := structs.NewServiceName(t.Service, &em)
if !candidate.Matches(destination) {
continue
}
if idx > maxIdx {
maxIdx = idx
}
if !seenSource[sn] {
seenSource[sn] = true
resp = append(resp, sn)
}
}
}
return maxIdx, resp, nil
}
func validateProposedConfigEntryInServiceGraph(
tx ReadTxn,
kind, name string,
proposedEntry structs.ConfigEntry,
validateAllChains bool,
entMeta *structs.EnterpriseMeta,
) error {
// Collect all of the chains that could be affected by this change
// including our own.
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
var (
checkChains = make(map[structs.ServiceID]struct{})
checkIngress []*structs.IngressGatewayConfigEntry
checkIntentions []*structs.ServiceIntentionsConfigEntry
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
enforceIngressProtocolsMatch bool
)
if validateAllChains {
// Must be proxy-defaults/global.
// Check anything that has a discovery chain entry. In the future we could
// somehow omit the ones that have a default protocol configured.
for _, kind := range serviceGraphKinds {
_, entries, err := configEntriesByKindTxn(tx, nil, kind, structs.WildcardEnterpriseMeta())
if err != nil {
return err
}
for _, entry := range entries {
checkChains[structs.NewServiceID(entry.GetName(), entry.GetEnterpriseMeta())] = struct{}{}
}
}
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
_, ingressEntries, err := configEntriesByKindTxn(tx, nil, structs.IngressGateway, structs.WildcardEnterpriseMeta())
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
if err != nil {
return err
}
for _, entry := range ingressEntries {
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
ingress, ok := entry.(*structs.IngressGatewayConfigEntry)
if !ok {
return fmt.Errorf("type %T is not an ingress gateway config entry", entry)
}
checkIngress = append(checkIngress, ingress)
}
_, ixnEntries, err := configEntriesByKindTxn(tx, nil, structs.ServiceIntentions, structs.WildcardEnterpriseMeta())
if err != nil {
return err
}
for _, entry := range ixnEntries {
ixn, ok := entry.(*structs.ServiceIntentionsConfigEntry)
if !ok {
return fmt.Errorf("type %T is not a service intentions config entry", entry)
}
checkIntentions = append(checkIntentions, ixn)
}
} else if kind == structs.ServiceIntentions {
// Check that the protocols match.
// This is the case for deleting a config entry
if proposedEntry == nil {
return nil
}
ixn, ok := proposedEntry.(*structs.ServiceIntentionsConfigEntry)
if !ok {
return fmt.Errorf("type %T is not a service intentions config entry", proposedEntry)
}
checkIntentions = append(checkIntentions, ixn)
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
} else if kind == structs.IngressGateway {
// Checking an ingress pointing to multiple chains.
// This is the case for deleting a config entry
if proposedEntry == nil {
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
return nil
}
ingress, ok := proposedEntry.(*structs.IngressGatewayConfigEntry)
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
if !ok {
return fmt.Errorf("type %T is not an ingress gateway config entry", proposedEntry)
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
}
checkIngress = append(checkIngress, ingress)
// When editing an ingress-gateway directly we are stricter about
// validating the protocol equivalence.
enforceIngressProtocolsMatch = true
} else {
// Must be a single chain.
// Check to see if we should ensure L7 intentions have an L7 protocol.
_, ixn, err := getServiceIntentionsConfigEntryTxn(
tx, nil, name, nil, entMeta,
)
if err != nil {
return err
} else if ixn != nil {
checkIntentions = append(checkIntentions, ixn)
}
_, ixnEntries, err := configEntriesByKindTxn(tx, nil, structs.ServiceIntentions, structs.WildcardEnterpriseMeta())
if err != nil {
return err
}
for _, entry := range ixnEntries {
ixn, ok := entry.(*structs.ServiceIntentionsConfigEntry)
if !ok {
return fmt.Errorf("type %T is not a service intentions config entry", entry)
}
checkIntentions = append(checkIntentions, ixn)
}
sid := structs.NewServiceID(name, entMeta)
checkChains[sid] = struct{}{}
iter, err := tx.Get(configTableName, "link", sid)
2020-06-23 17:18:22 +00:00
if err != nil {
return err
}
for raw := iter.Next(); raw != nil; raw = iter.Next() {
entry := raw.(structs.ConfigEntry)
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
switch entry.GetKind() {
case structs.ServiceRouter, structs.ServiceSplitter, structs.ServiceResolver:
svcID := structs.NewServiceID(entry.GetName(), entry.GetEnterpriseMeta())
checkChains[svcID] = struct{}{}
case structs.IngressGateway:
ingress, ok := entry.(*structs.IngressGatewayConfigEntry)
if !ok {
return fmt.Errorf("type %T is not an ingress gateway config entry", entry)
}
checkIngress = append(checkIngress, ingress)
}
}
}
// Ensure if any ingress or intention is affected that we fetch all of the
// chains needed to fully validate them.
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
for _, ingress := range checkIngress {
for _, svcID := range ingress.ListRelatedServices() {
checkChains[svcID] = struct{}{}
}
}
for _, ixn := range checkIntentions {
sn := ixn.DestinationServiceName()
checkChains[sn.ToServiceID()] = struct{}{}
}
overrides := map[structs.ConfigEntryKindName]structs.ConfigEntry{
structs.NewConfigEntryKindName(kind, name, entMeta): proposedEntry,
}
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
var (
svcProtocols = make(map[structs.ServiceID]string)
svcTopNodeType = make(map[structs.ServiceID]string)
)
for chain := range checkChains {
protocol, topNode, err := testCompileDiscoveryChain(tx, chain.ID, overrides, &chain.EnterpriseMeta)
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
if err != nil {
return err
}
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
svcProtocols[chain] = protocol
svcTopNodeType[chain] = topNode.Type
}
// Now validate all of our ingress gateways.
for _, e := range checkIngress {
for _, listener := range e.Listeners {
expectedProto := listener.Protocol
for _, service := range listener.Services {
if service.Name == structs.WildcardSpecifier {
continue
}
svcID := structs.NewServiceID(service.Name, &service.EnterpriseMeta)
svcProto := svcProtocols[svcID]
if svcProto != expectedProto {
// The only time an ingress gateway and its upstreams can
// have differing protocols is when:
//
// 1. ingress is tcp and the target is not-tcp
// AND
// 2. the disco chain has a resolver as the top node
topNodeType := svcTopNodeType[svcID]
if enforceIngressProtocolsMatch ||
(expectedProto != "tcp") ||
(expectedProto == "tcp" && topNodeType != structs.DiscoveryGraphNodeTypeResolver) {
return fmt.Errorf(
"service %q has protocol %q, which does not match defined listener protocol %q",
svcID.String(),
svcProto,
expectedProto,
)
}
}
}
}
}
// Now validate that intentions with L7 permissions reference HTTP services
for _, e := range checkIntentions {
// We only have to double check things that try to use permissions
if e.HasWildcardDestination() || !e.HasAnyPermissions() {
continue
}
sn := e.DestinationServiceName()
svcID := sn.ToServiceID()
svcProto := svcProtocols[svcID]
if !structs.IsProtocolHTTPLike(svcProto) {
return fmt.Errorf(
"service %q has protocol %q, which is incompatible with L7 intentions permissions",
svcID.String(),
svcProto,
)
}
}
return nil
}
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
// testCompileDiscoveryChain speculatively compiles a discovery chain with
// pending modifications to see if it would be valid. Also returns the computed
// protocol and topmost discovery chain node.
func testCompileDiscoveryChain(
tx ReadTxn,
chainName string,
overrides map[structs.ConfigEntryKindName]structs.ConfigEntry,
entMeta *structs.EnterpriseMeta,
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
) (string, *structs.DiscoveryGraphNode, error) {
_, speculativeEntries, err := readDiscoveryChainConfigEntriesTxn(tx, nil, chainName, overrides, entMeta)
if err != nil {
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
return "", nil, err
}
// Note we use an arbitrary namespace and datacenter as those would not
// currently affect the graph compilation in ways that matter here.
connect: fix failover through a mesh gateway to a remote datacenter (#6259) Failover is pushed entirely down to the data plane by creating envoy clusters and putting each successive destination in a different load assignment priority band. For example this shows that normally requests go to 1.2.3.4:8080 but when that fails they go to 6.7.8.9:8080: - name: foo load_assignment: cluster_name: foo policy: overprovisioning_factor: 100000 endpoints: - priority: 0 lb_endpoints: - endpoint: address: socket_address: address: 1.2.3.4 port_value: 8080 - priority: 1 lb_endpoints: - endpoint: address: socket_address: address: 6.7.8.9 port_value: 8080 Mesh gateways route requests based solely on the SNI header tacked onto the TLS layer. Envoy currently only lets you configure the outbound SNI header at the cluster layer. If you try to failover through a mesh gateway you ideally would configure the SNI value per endpoint, but that's not possible in envoy today. This PR introduces a simpler way around the problem for now: 1. We identify any target of failover that will use mesh gateway mode local or remote and then further isolate any resolver node in the compiled discovery chain that has a failover destination set to one of those targets. 2. For each of these resolvers we will perform a small measurement of comparative healths of the endpoints that come back from the health API for the set of primary target and serial failover targets. We walk the list of targets in order and if any endpoint is healthy we return that target, otherwise we move on to the next target. 3. The CDS and EDS endpoints both perform the measurements in (2) for the affected resolver nodes. 4. For CDS this measurement selects which TLS SNI field to use for the cluster (note the cluster is always going to be named for the primary target) 5. For EDS this measurement selects which set of endpoints will populate the cluster. Priority tiered failover is ignored. One of the big downsides to this approach to failover is that the failover detection and correction is going to be controlled by consul rather than deferring that entirely to the data plane as with the prior version. This also means that we are bound to only failover using official health signals and cannot make use of data plane signals like outlier detection to affect failover. In this specific scenario the lack of data plane signals is ok because the effectiveness is already muted by the fact that the ultimate destination endpoints will have their data plane signals scrambled when they pass through the mesh gateway wrapper anyway so we're not losing much. Another related fix is that we now use the endpoint health from the underlying service, not the health of the gateway (regardless of failover mode).
2019-08-05 18:30:35 +00:00
//
// TODO(rb): we should thread a better value than "dc1" and the throwaway trust domain down here as that is going to sometimes show up in user facing errors
req := discoverychain.CompileRequest{
ServiceName: chainName,
EvaluateInNamespace: entMeta.NamespaceOrDefault(),
EvaluateInDatacenter: "dc1",
EvaluateInTrustDomain: "b6fc9da3-03d4-4b5a-9134-c045e9b20152.consul",
UseInDatacenter: "dc1",
Entries: speculativeEntries,
}
connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams (#8470) Fixes #8466 Since Consul 1.8.0 there was a bug in how ingress gateway protocol compatibility was enforced. At the point in time that an ingress-gateway config entry was modified the discovery chain for each upstream was checked to ensure the ingress gateway protocol matched. Unfortunately future modifications of other config entries were not validated against existing ingress-gateway definitions, such as: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. create service-defaults for 'api' setting protocol=http (worked, but not ok) 3. create service-splitter or service-router for 'api' (worked, but caused an agent panic) If you were to do these in a different order, it would fail without a crash: 1. create service-defaults for 'api' setting protocol=http (ok) 2. create service-splitter or service-router for 'api' (ok) 3. create tcp ingress-gateway pointing to 'api' (fail with message about protocol mismatch) This PR introduces the missing validation. The two new behaviors are: 1. create tcp ingress-gateway pointing to 'api' (ok) 2. (NEW) create service-defaults for 'api' setting protocol=http ("ok" for back compat) 3. (NEW) create service-splitter or service-router for 'api' (fail with message about protocol mismatch) In consideration for any existing users that may be inadvertently be falling into item (2) above, that is now officiall a valid configuration to be in. For anyone falling into item (3) above while you cannot use the API to manufacture that scenario anymore, anyone that has old (now bad) data will still be able to have the agent use them just enough to generate a new agent/proxycfg error message rather than a panic. Unfortunately we just don't have enough information to properly fix the config entries.
2020-08-12 16:19:20 +00:00
chain, err := discoverychain.Compile(req)
if err != nil {
return "", nil, err
}
return chain.Protocol, chain.Nodes[chain.StartNode], nil
}
func (s *Store) ServiceDiscoveryChain(
ws memdb.WatchSet,
serviceName string,
entMeta *structs.EnterpriseMeta,
req discoverychain.CompileRequest,
) (uint64, *structs.CompiledDiscoveryChain, error) {
tx := s.db.ReadTxn()
defer tx.Abort()
return s.serviceDiscoveryChainTxn(tx, ws, serviceName, entMeta, req)
}
func (s *Store) serviceDiscoveryChainTxn(
tx ReadTxn,
ws memdb.WatchSet,
serviceName string,
entMeta *structs.EnterpriseMeta,
req discoverychain.CompileRequest,
) (uint64, *structs.CompiledDiscoveryChain, error) {
index, entries, err := readDiscoveryChainConfigEntriesTxn(tx, ws, serviceName, nil, entMeta)
if err != nil {
return 0, nil, err
}
req.Entries = entries
_, config, err := s.CAConfig(ws)
if err != nil {
return 0, nil, err
} else if config == nil {
return 0, nil, errors.New("no cluster ca config setup")
}
// Build TrustDomain based on the ClusterID stored.
signingID := connect.SpiffeIDSigningForCluster(config)
if signingID == nil {
// If CA is bootstrapped at all then this should never happen but be
// defensive.
return 0, nil, errors.New("no cluster trust domain setup")
}
req.EvaluateInTrustDomain = signingID.Host()
// Then we compile it into something useful.
chain, err := discoverychain.Compile(req)
if err != nil {
return 0, nil, fmt.Errorf("failed to compile discovery chain: %v", err)
}
return index, chain, nil
}
// ReadDiscoveryChainConfigEntries will query for the full discovery chain for
// the provided service name. All relevant config entries will be recursively
// fetched and included in the result.
//
// Once returned, the caller still needs to assemble these into a useful graph
// structure.
func (s *Store) ReadDiscoveryChainConfigEntries(
ws memdb.WatchSet,
serviceName string,
entMeta *structs.EnterpriseMeta,
) (uint64, *structs.DiscoveryChainConfigEntries, error) {
return s.readDiscoveryChainConfigEntries(ws, serviceName, nil, entMeta)
}
// readDiscoveryChainConfigEntries will query for the full discovery chain for
// the provided service name. All relevant config entries will be recursively
// fetched and included in the result.
//
// If 'overrides' is provided then it will use entries in that map instead of
// the database to simulate the entries that go into a modified discovery chain
// without actually modifying it yet. Nil values are tombstones to simulate
// deleting an entry.
//
// Overrides is not mutated.
func (s *Store) readDiscoveryChainConfigEntries(
ws memdb.WatchSet,
serviceName string,
overrides map[structs.ConfigEntryKindName]structs.ConfigEntry,
entMeta *structs.EnterpriseMeta,
) (uint64, *structs.DiscoveryChainConfigEntries, error) {
tx := s.db.Txn(false)
defer tx.Abort()
return readDiscoveryChainConfigEntriesTxn(tx, ws, serviceName, overrides, entMeta)
}
func readDiscoveryChainConfigEntriesTxn(
tx ReadTxn,
ws memdb.WatchSet,
serviceName string,
overrides map[structs.ConfigEntryKindName]structs.ConfigEntry,
entMeta *structs.EnterpriseMeta,
) (uint64, *structs.DiscoveryChainConfigEntries, error) {
res := structs.NewDiscoveryChainConfigEntries()
// Note that below we always look up splitters and resolvers in pairs, even
// in some circumstances where both are not strictly necessary.
//
// For now we'll just eat the cost of fetching pairs of splitter/resolver
// config entries even though we may not always need both. In the common
// case we will need the pair so there's not a big drive to optimize this
// here at this time.
// Both Splitters and Resolvers maps will contain placeholder nils until
// the end of this function to indicate "no such entry".
var (
todoSplitters = make(map[structs.ServiceID]struct{})
todoResolvers = make(map[structs.ServiceID]struct{})
todoDefaults = make(map[structs.ServiceID]struct{})
)
sid := structs.NewServiceID(serviceName, entMeta)
// Grab the proxy defaults if they exist.
idx, proxy, err := getProxyConfigEntryTxn(tx, ws, structs.ProxyConfigGlobal, overrides, structs.DefaultEnterpriseMeta())
if err != nil {
return 0, nil, err
} else if proxy != nil {
res.GlobalProxy = proxy
}
// At every step we'll need service defaults.
todoDefaults[sid] = struct{}{}
// first fetch the router, of which we only collect 1 per chain eval
_, router, err := getRouterConfigEntryTxn(tx, ws, serviceName, overrides, entMeta)
if err != nil {
return 0, nil, err
} else if router != nil {
res.Routers[sid] = router
}
if router != nil {
for _, svc := range router.ListRelatedServices() {
todoSplitters[svc] = struct{}{}
}
} else {
// Next hop in the chain is the splitter.
todoSplitters[sid] = struct{}{}
}
for {
splitID, ok := anyKey(todoSplitters)
if !ok {
break
}
delete(todoSplitters, splitID)
if _, ok := res.Splitters[splitID]; ok {
continue // already fetched
}
// Yes, even for splitters.
todoDefaults[splitID] = struct{}{}
_, splitter, err := getSplitterConfigEntryTxn(tx, ws, splitID.ID, overrides, &splitID.EnterpriseMeta)
if err != nil {
return 0, nil, err
}
if splitter == nil {
res.Splitters[splitID] = nil
// Next hop in the chain is the resolver.
todoResolvers[splitID] = struct{}{}
continue
}
res.Splitters[splitID] = splitter
todoResolvers[splitID] = struct{}{}
for _, svc := range splitter.ListRelatedServices() {
// If there is no splitter, this will end up adding a resolver
// after another iteration.
todoSplitters[svc] = struct{}{}
}
}
for {
resolverID, ok := anyKey(todoResolvers)
if !ok {
break
}
delete(todoResolvers, resolverID)
if _, ok := res.Resolvers[resolverID]; ok {
continue // already fetched
}
// And resolvers, too.
todoDefaults[resolverID] = struct{}{}
_, resolver, err := getResolverConfigEntryTxn(tx, ws, resolverID.ID, overrides, &resolverID.EnterpriseMeta)
if err != nil {
return 0, nil, err
}
if resolver == nil {
res.Resolvers[resolverID] = nil
continue
}
res.Resolvers[resolverID] = resolver
for _, svc := range resolver.ListRelatedServices() {
todoResolvers[svc] = struct{}{}
}
}
for {
svcID, ok := anyKey(todoDefaults)
if !ok {
break
}
delete(todoDefaults, svcID)
if _, ok := res.Services[svcID]; ok {
continue // already fetched
}
_, entry, err := getServiceConfigEntryTxn(tx, ws, svcID.ID, overrides, &svcID.EnterpriseMeta)
if err != nil {
return 0, nil, err
}
if entry == nil {
res.Services[svcID] = nil
continue
}
res.Services[svcID] = entry
}
// Strip nils now that they are no longer necessary.
for sid, entry := range res.Routers {
if entry == nil {
delete(res.Routers, sid)
}
}
for sid, entry := range res.Splitters {
if entry == nil {
delete(res.Splitters, sid)
}
}
for sid, entry := range res.Resolvers {
if entry == nil {
delete(res.Resolvers, sid)
}
}
for sid, entry := range res.Services {
if entry == nil {
delete(res.Services, sid)
}
}
return idx, res, nil
}
// anyKey returns any key from the provided map if any exist. Useful for using
// a map as a simple work queue of sorts.
func anyKey(m map[structs.ServiceID]struct{}) (structs.ServiceID, bool) {
if len(m) == 0 {
return structs.ServiceID{}, false
}
for k := range m {
return k, true
}
return structs.ServiceID{}, false
}
// getProxyConfigEntryTxn is a convenience method for fetching a
// proxy-defaults kind of config entry.
//
// If an override is returned the index returned will be 0.
func getProxyConfigEntryTxn(
tx ReadTxn,
ws memdb.WatchSet,
name string,
overrides map[structs.ConfigEntryKindName]structs.ConfigEntry,
entMeta *structs.EnterpriseMeta,
) (uint64, *structs.ProxyConfigEntry, error) {
idx, entry, err := configEntryWithOverridesTxn(tx, ws, structs.ProxyDefaults, name, overrides, entMeta)
if err != nil {
return 0, nil, err
} else if entry == nil {
return idx, nil, nil
}
proxy, ok := entry.(*structs.ProxyConfigEntry)
if !ok {
return 0, nil, fmt.Errorf("invalid service config type %T", entry)
}
return idx, proxy, nil
}
// getServiceConfigEntryTxn is a convenience method for fetching a
// service-defaults kind of config entry.
//
// If an override is returned the index returned will be 0.
func getServiceConfigEntryTxn(
tx ReadTxn,
ws memdb.WatchSet,
serviceName string,
overrides map[structs.ConfigEntryKindName]structs.ConfigEntry,
entMeta *structs.EnterpriseMeta,
) (uint64, *structs.ServiceConfigEntry, error) {
idx, entry, err := configEntryWithOverridesTxn(tx, ws, structs.ServiceDefaults, serviceName, overrides, entMeta)
if err != nil {
return 0, nil, err
} else if entry == nil {
return idx, nil, nil
}
service, ok := entry.(*structs.ServiceConfigEntry)
if !ok {
return 0, nil, fmt.Errorf("invalid service config type %T", entry)
}
return idx, service, nil
}
// getRouterConfigEntryTxn is a convenience method for fetching a
// service-router kind of config entry.
//
// If an override is returned the index returned will be 0.
func getRouterConfigEntryTxn(
tx ReadTxn,
ws memdb.WatchSet,
serviceName string,
overrides map[structs.ConfigEntryKindName]structs.ConfigEntry,
entMeta *structs.EnterpriseMeta,
) (uint64, *structs.ServiceRouterConfigEntry, error) {
idx, entry, err := configEntryWithOverridesTxn(tx, ws, structs.ServiceRouter, serviceName, overrides, entMeta)
if err != nil {
return 0, nil, err
} else if entry == nil {
return idx, nil, nil
}
router, ok := entry.(*structs.ServiceRouterConfigEntry)
if !ok {
return 0, nil, fmt.Errorf("invalid service config type %T", entry)
}
return idx, router, nil
}
// getSplitterConfigEntryTxn is a convenience method for fetching a
// service-splitter kind of config entry.
//
// If an override is returned the index returned will be 0.
func getSplitterConfigEntryTxn(
tx ReadTxn,
ws memdb.WatchSet,
serviceName string,
overrides map[structs.ConfigEntryKindName]structs.ConfigEntry,
entMeta *structs.EnterpriseMeta,
) (uint64, *structs.ServiceSplitterConfigEntry, error) {
idx, entry, err := configEntryWithOverridesTxn(tx, ws, structs.ServiceSplitter, serviceName, overrides, entMeta)
if err != nil {
return 0, nil, err
} else if entry == nil {
return idx, nil, nil
}
splitter, ok := entry.(*structs.ServiceSplitterConfigEntry)
if !ok {
return 0, nil, fmt.Errorf("invalid service config type %T", entry)
}
return idx, splitter, nil
}
// getResolverConfigEntryTxn is a convenience method for fetching a
// service-resolver kind of config entry.
//
// If an override is returned the index returned will be 0.
func getResolverConfigEntryTxn(
tx ReadTxn,
ws memdb.WatchSet,
serviceName string,
overrides map[structs.ConfigEntryKindName]structs.ConfigEntry,
entMeta *structs.EnterpriseMeta,
) (uint64, *structs.ServiceResolverConfigEntry, error) {
idx, entry, err := configEntryWithOverridesTxn(tx, ws, structs.ServiceResolver, serviceName, overrides, entMeta)
if err != nil {
return 0, nil, err
} else if entry == nil {
return idx, nil, nil
}
resolver, ok := entry.(*structs.ServiceResolverConfigEntry)
if !ok {
return 0, nil, fmt.Errorf("invalid service config type %T", entry)
}
return idx, resolver, 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
// getServiceIntentionsConfigEntryTxn is a convenience method for fetching a
// service-intentions kind of config entry.
//
// If an override is returned the index returned will be 0.
func getServiceIntentionsConfigEntryTxn(
tx ReadTxn,
ws memdb.WatchSet,
name string,
overrides map[structs.ConfigEntryKindName]structs.ConfigEntry,
entMeta *structs.EnterpriseMeta,
) (uint64, *structs.ServiceIntentionsConfigEntry, error) {
idx, entry, err := configEntryWithOverridesTxn(tx, ws, structs.ServiceIntentions, name, overrides, entMeta)
if err != nil {
return 0, nil, err
} else if entry == nil {
return idx, nil, nil
}
ixn, ok := entry.(*structs.ServiceIntentionsConfigEntry)
if !ok {
return 0, nil, fmt.Errorf("invalid service config type %T", entry)
}
return idx, ixn, nil
}
func configEntryWithOverridesTxn(
tx ReadTxn,
ws memdb.WatchSet,
kind string,
name string,
overrides map[structs.ConfigEntryKindName]structs.ConfigEntry,
entMeta *structs.EnterpriseMeta,
) (uint64, structs.ConfigEntry, error) {
if len(overrides) > 0 {
kn := structs.NewConfigEntryKindName(kind, name, entMeta)
entry, ok := overrides[kn]
if ok {
return 0, entry, nil // a nil entry implies it should act like it is erased
}
}
return configEntryTxn(tx, ws, kind, name, entMeta)
}
// protocolForService returns the service graph protocol associated to the
// provided service, checking all relevant config entries.
func protocolForService(
tx ReadTxn,
ws memdb.WatchSet,
svc structs.ServiceName,
) (uint64, string, error) {
// Get the global proxy defaults (for default protocol)
maxIdx, proxyConfig, err := configEntryTxn(tx, ws, structs.ProxyDefaults, structs.ProxyConfigGlobal, structs.DefaultEnterpriseMeta())
if err != nil {
return 0, "", err
}
idx, serviceDefaults, err := configEntryTxn(tx, ws, structs.ServiceDefaults, svc.Name, &svc.EnterpriseMeta)
if err != nil {
return 0, "", err
}
maxIdx = lib.MaxUint64(maxIdx, idx)
entries := structs.NewDiscoveryChainConfigEntries()
if proxyConfig != nil {
entries.AddEntries(proxyConfig)
}
if serviceDefaults != nil {
entries.AddEntries(serviceDefaults)
}
req := discoverychain.CompileRequest{
ServiceName: svc.Name,
EvaluateInNamespace: svc.NamespaceOrDefault(),
EvaluateInDatacenter: "dc1",
// Use a dummy trust domain since that won't affect the protocol here.
EvaluateInTrustDomain: "b6fc9da3-03d4-4b5a-9134-c045e9b20152.consul",
UseInDatacenter: "dc1",
Entries: entries,
}
chain, err := discoverychain.Compile(req)
if err != nil {
return 0, "", err
}
return maxIdx, chain.Protocol, nil
}