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

312 lines
9.8 KiB
Go
Raw Normal View History

package state
import (
"context"
"errors"
"fmt"
"time"
memdb "github.com/hashicorp/go-memdb"
"github.com/hashicorp/consul/agent/consul/stream"
"github.com/hashicorp/consul/agent/structs"
)
var (
// ErrMissingNode is the error returned when trying an operation
// which requires a node registration but none exists.
ErrMissingNode = errors.New("Missing node registration")
// ErrMissingService is the error we return if trying an
// operation which requires a service but none exists.
ErrMissingService = errors.New("Missing service registration")
// ErrMissingSessionID is returned when a session registration
// is attempted with an empty session ID.
ErrMissingSessionID = errors.New("Missing session ID")
2015-09-07 04:13:45 +00:00
2019-02-13 17:57:29 +00:00
// ErrMissingACLTokenSecret is returned when a token set is called on a
// token with an empty SecretID.
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
ErrMissingACLTokenSecret = errors.New("Missing ACL Token SecretID")
2019-02-13 17:57:29 +00:00
// ErrMissingACLTokenAccessor is returned when a token set is called on a
// token with an empty AccessorID.
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
ErrMissingACLTokenAccessor = errors.New("Missing ACL Token AccessorID")
// ErrTokenHasNoPrivileges is returned when a token set is called on a
// token with no policies, roles, or service identities and the caller
// requires at least one to be set.
ErrTokenHasNoPrivileges = errors.New("Token has no privileges")
2019-02-13 17:57:29 +00:00
// ErrMissingACLPolicyID is returned when a policy set is called on a
// policy with an empty ID.
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
ErrMissingACLPolicyID = errors.New("Missing ACL Policy ID")
2019-02-13 17:57:29 +00:00
// ErrMissingACLPolicyName is returned when a policy set is called on a
// policy with an empty Name.
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
ErrMissingACLPolicyName = errors.New("Missing ACL Policy Name")
// ErrMissingACLRoleID is returned when a role set is called on
// a role with an empty ID.
ErrMissingACLRoleID = errors.New("Missing ACL Role ID")
// ErrMissingACLRoleName is returned when a role set is called on
// a role with an empty Name.
ErrMissingACLRoleName = errors.New("Missing ACL Role Name")
// ErrMissingACLBindingRuleID is returned when a binding rule set
// is called on a binding rule with an empty ID.
ErrMissingACLBindingRuleID = errors.New("Missing ACL Binding Rule ID")
// ErrMissingACLBindingRuleAuthMethod is returned when a binding rule set
// is called on a binding rule with an empty AuthMethod.
ErrMissingACLBindingRuleAuthMethod = errors.New("Missing ACL Binding Rule Auth Method")
// ErrMissingACLAuthMethodName is returned when an auth method set is
// called on an auth method with an empty Name.
ErrMissingACLAuthMethodName = errors.New("Missing ACL Auth Method Name")
// ErrMissingACLAuthMethodType is returned when an auth method set is
// called on an auth method with an empty Type.
ErrMissingACLAuthMethodType = errors.New("Missing ACL Auth Method Type")
// ErrMissingQueryID is returned when a Query set is called on
// a Query with an empty ID.
ErrMissingQueryID = errors.New("Missing Query ID")
// ErrMissingCARootID is returned when an CARoot set is called
// with an CARoot with an empty ID.
ErrMissingCARootID = errors.New("Missing CA Root ID")
// ErrMissingIntentionID is returned when an Intention set is called
// with an Intention with an empty ID.
ErrMissingIntentionID = errors.New("Missing Intention ID")
)
var (
// watchLimit is used as a soft limit to cap how many watches we allow
// for a given blocking query. If this is exceeded, then we will use a
// higher-level watch that's less fine-grained. Choosing the perfect
// value is impossible given how different deployments and workload
// are. This value was recommended by customers with many servers. We
// expect streaming to arrive soon and that should help a lot with
// blocking queries. Please see
// https://github.com/hashicorp/consul/pull/7200 and linked issues/prs
// for more context
watchLimit = 8192
)
2017-04-21 00:46:29 +00:00
// Store is where we store all of Consul's state, including
// records of node registrations, services, checks, key/value
// pairs and more. The DB is entirely in-memory and is constructed
// from the Raft log through the FSM.
2017-04-21 00:46:29 +00:00
type Store struct {
schema *memdb.DBSchema
2020-06-03 16:59:10 +00:00
db *changeTrackerDB
// abandonCh is used to signal watchers that this state store has been
// abandoned (usually during a restore). This is only ever closed.
abandonCh chan struct{}
// TODO: refactor abondonCh to use a context so that both can use the same
// cancel mechanism.
stopEventPublisher func()
// kvsGraveyard manages tombstones for the key value store.
kvsGraveyard *Graveyard
// lockDelay holds expiration times for locks associated with keys.
lockDelay *Delay
2015-09-20 08:36:39 +00:00
}
2017-04-21 00:46:29 +00:00
// Snapshot is used to provide a point-in-time snapshot. It
2015-09-20 08:36:39 +00:00
// works by starting a read transaction against the whole state store.
2017-04-21 00:46:29 +00:00
type Snapshot struct {
store *Store
tx AbortTxn
2015-09-20 08:36:39 +00:00
lastIndex uint64
}
2017-04-21 00:46:29 +00:00
// Restore is used to efficiently manage restoring a large amount of
// data to a state store.
2017-04-21 00:46:29 +00:00
type Restore struct {
store *Store
2020-06-03 17:21:00 +00:00
tx *txn
}
// sessionCheck is used to create a many-to-one table such that
// each check registered by a session can be mapped back to the
// session table. This is only used internally in the state
// store and thus it is not exported.
type sessionCheck struct {
Node string
Session string
CheckID structs.CheckID
structs.EnterpriseMeta
}
// NewStateStore creates a new in-memory state storage layer.
func NewStateStore(gc *TombstoneGC) *Store {
2015-09-20 08:36:39 +00:00
// Create the in-memory DB.
schema := newDBSchema()
2015-09-20 08:36:39 +00:00
db, err := memdb.NewMemDB(schema)
if err != nil {
// the only way for NewMemDB to error is if the schema is invalid. The
// scheme is static and tested to be correct, so any failure here would
// be a programming error, which should panic.
panic(fmt.Sprintf("failed to create state store: %v", err))
}
2017-04-21 00:46:29 +00:00
s := &Store{
schema: schema,
abandonCh: make(chan struct{}),
kvsGraveyard: NewGraveyard(gc),
lockDelay: NewDelay(),
stopEventPublisher: func() {},
db: &changeTrackerDB{
db: db,
publisher: stream.NoOpEventPublisher{},
processChanges: processDBChanges,
},
}
return s
}
func NewStateStoreWithEventPublisher(gc *TombstoneGC) *Store {
store := NewStateStore(gc)
ctx, cancel := context.WithCancel(context.TODO())
store.stopEventPublisher = cancel
pub := stream.NewEventPublisher(newSnapshotHandlers((*readDB)(store.db.db)), 10*time.Second)
store.db.publisher = pub
go pub.Run(ctx)
return store
}
// EventPublisher returns the stream.EventPublisher used by the Store to
// publish events.
func (s *Store) EventPublisher() EventPublisher {
return s.db.publisher
}
2015-09-20 08:36:39 +00:00
// Snapshot is used to create a point-in-time snapshot of the entire db.
2017-04-21 00:46:29 +00:00
func (s *Store) Snapshot() *Snapshot {
2015-09-20 08:36:39 +00:00
tx := s.db.Txn(false)
var tables []string
for table := range s.schema.Tables {
2015-09-20 08:36:39 +00:00
tables = append(tables, table)
}
idx := maxIndexTxn(tx, tables...)
2017-04-21 00:46:29 +00:00
return &Snapshot{s, tx, idx}
2015-09-20 08:36:39 +00:00
}
// LastIndex returns that last index that affects the snapshotted data.
2017-04-21 00:46:29 +00:00
func (s *Snapshot) LastIndex() uint64 {
2015-09-20 08:36:39 +00:00
return s.lastIndex
}
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
func (s *Snapshot) Indexes() (memdb.ResultIterator, error) {
2021-02-05 22:53:08 +00:00
return s.tx.Get(tableIndex, indexID)
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
}
// IndexRestore is used to restore an index
func (s *Restore) IndexRestore(idx *IndexEntry) error {
2021-02-05 22:53:08 +00:00
if err := s.tx.Insert(tableIndex, idx); err != nil {
New ACLs (#4791) This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week. Description At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers. On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though. Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though. All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management. Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are: A server running the new system must still support other clients using the legacy system. A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system. The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode. So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
2018-10-19 16:04:07 +00:00
return fmt.Errorf("index insert failed: %v", err)
}
return nil
}
2015-09-20 08:36:39 +00:00
// Close performs cleanup of a state snapshot.
2017-04-21 00:46:29 +00:00
func (s *Snapshot) Close() {
2015-09-20 08:36:39 +00:00
s.tx.Abort()
}
// Restore is used to efficiently manage restoring a large amount of data into
// the state store. It works by doing all the restores inside of a single
// transaction.
2017-04-21 00:46:29 +00:00
func (s *Store) Restore() *Restore {
tx := s.db.WriteTxnRestore()
2017-04-21 00:46:29 +00:00
return &Restore{s, tx}
}
// Abort abandons the changes made by a restore. This or Commit should always be
// called.
2017-04-21 00:46:29 +00:00
func (s *Restore) Abort() {
s.tx.Abort()
}
// Commit commits the changes made by a restore. This or Abort should always be
// called.
2020-06-02 20:34:56 +00:00
func (s *Restore) Commit() error {
return s.tx.Commit()
}
// AbandonCh returns a channel you can wait on to know if the state store was
// abandoned.
2017-04-21 00:46:29 +00:00
func (s *Store) AbandonCh() <-chan struct{} {
return s.abandonCh
}
// Abandon is used to signal that the given state store has been abandoned.
// Calling this more than one time will panic.
2017-04-21 00:46:29 +00:00
func (s *Store) Abandon() {
s.stopEventPublisher()
close(s.abandonCh)
}
// maxIndex is a helper used to retrieve the highest known index
// amongst a set of tables in the db.
2017-04-21 00:46:29 +00:00
func (s *Store) maxIndex(tables ...string) uint64 {
tx := s.db.Txn(false)
defer tx.Abort()
2015-09-20 08:36:39 +00:00
return maxIndexTxn(tx, tables...)
}
2015-09-20 08:36:39 +00:00
// maxIndexTxn is a helper used to retrieve the highest known index
// amongst a set of tables in the db.
func maxIndexTxn(tx ReadTxn, tables ...string) uint64 {
return maxIndexWatchTxn(tx, nil, tables...)
}
func maxIndexWatchTxn(tx ReadTxn, ws memdb.WatchSet, tables ...string) uint64 {
var lindex uint64
for _, table := range tables {
2021-02-05 22:53:08 +00:00
ch, ti, err := tx.FirstWatch(tableIndex, "id", table)
if err != nil {
panic(fmt.Sprintf("unknown index: %s err: %s", table, err))
}
if idx, ok := ti.(*IndexEntry); ok && idx.Value > lindex {
lindex = idx.Value
}
ws.Add(ch)
}
return lindex
}
2015-09-20 08:36:39 +00:00
// indexUpdateMaxTxn is used when restoring entries and sets the table's index to
// the given idx only if it's greater than the current index.
func indexUpdateMaxTxn(tx WriteTxn, idx uint64, table string) error {
2021-02-05 22:53:08 +00:00
ti, err := tx.First(tableIndex, indexID, table)
2015-09-20 08:36:39 +00:00
if err != nil {
return fmt.Errorf("failed to retrieve existing index: %s", err)
}
// Always take the first update, otherwise do the > check.
if ti == nil {
2021-02-05 22:53:08 +00:00
if err := tx.Insert(tableIndex, &IndexEntry{table, idx}); err != nil {
return fmt.Errorf("failed updating index %s", err)
}
2021-02-05 22:53:08 +00:00
return nil
}
if cur, ok := ti.(*IndexEntry); ok && idx > cur.Value {
if err := tx.Insert(tableIndex, &IndexEntry{table, idx}); err != nil {
2015-09-20 08:36:39 +00:00
return fmt.Errorf("failed updating index %s", err)
}
}
return nil
}