open-consul/agent/consul/config_endpoint.go

591 lines
16 KiB
Go
Raw Permalink Normal View History

// Copyright (c) HashiCorp, Inc.
// SPDX-License-Identifier: MPL-2.0
package consul
import (
"fmt"
"reflect"
"time"
metrics "github.com/armon/go-metrics"
"github.com/armon/go-metrics/prometheus"
"github.com/hashicorp/go-bexpr"
"github.com/hashicorp/go-hclog"
memdb "github.com/hashicorp/go-memdb"
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
hashstructure_v2 "github.com/mitchellh/hashstructure/v2"
"github.com/hashicorp/consul/acl"
"github.com/hashicorp/consul/agent/configentry"
"github.com/hashicorp/consul/agent/consul/state"
"github.com/hashicorp/consul/agent/structs"
)
var ConfigSummaries = []prometheus.SummaryDefinition{
{
Name: []string{"config_entry", "apply"},
Help: "",
},
{
Name: []string{"config_entry", "get"},
Help: "",
},
{
Name: []string{"config_entry", "list"},
Help: "",
},
{
Name: []string{"config_entry", "listAll"},
Help: "",
},
{
Name: []string{"config_entry", "delete"},
Help: "",
},
{
Name: []string{"config_entry", "resolve_service_config"},
Help: "",
},
}
// The ConfigEntry endpoint is used to query centralized config information
type ConfigEntry struct {
srv *Server
logger hclog.Logger
}
// Apply does an upsert of the given config entry.
Centralized Config CLI (#5731) * Add HTTP endpoints for config entry management * Finish implementing decoding in the HTTP Config entry apply endpoint * Add CAS operation to the config entry apply endpoint Also use this for the bootstrapping and move the config entry decoding function into the structs package. * First pass at the API client for the config entries * Fixup some of the ConfigEntry APIs Return a singular response object instead of a list for the ConfigEntry.Get RPC. This gets plumbed through the HTTP API as well. Dont return QueryMeta in the JSON response for the config entry listing HTTP API. Instead just return a list of config entries. * Minor API client fixes * Attempt at some ConfigEntry api client tests These don’t currently work due to weak typing in JSON * Get some of the api client tests passing * Implement reflectwalk magic to correct JSON encoding a ProxyConfigEntry Also added a test for the HTTP endpoint that exposes the problem. However, since the test doesn’t actually do the JSON encode/decode its still failing. * Move MapWalk magic into a binary marshaller instead of JSON. * Add a MapWalk test * Get rid of unused func * Get rid of unused imports * Fixup some tests now that the decoding from msgpack coerces things into json compat types * Stub out most of the central config cli Fully implement the config read command. * Basic config delete command implementation * Implement config write command * Implement config list subcommand Not entirely sure about the output here. Its basically the read output indented with a line specifying the kind/name of each type which is also duplicated in the indented output. * Update command usage * Update some help usage formatting * Add the connect enable helper cli command * Update list command output * Rename the config entry API client methods. * Use renamed apis * Implement config write tests Stub the others with the noTabs tests. * Change list output format Now just simply output 1 line per named config * Add config read tests * Add invalid args write test. * Add config delete tests * Add config list tests * Add connect enable tests * Update some CLI commands to use CAS ops This also modifies the HTTP API for a write op to return a boolean indicating whether the value was written or not. * Fix up the HTTP API CAS tests as I realized they weren’t testing what they should. * Update config entry rpc tests to properly test CAS * Fix up a few more tests * Fix some tests that using ConfigEntries.Apply * Update config_write_test.go * Get rid of unused import
2019-04-30 23:27:16 +00:00
func (c *ConfigEntry) Apply(args *structs.ConfigEntryRequest, reply *bool) error {
if err := c.srv.validateEnterpriseRequest(args.Entry.GetEnterpriseMeta(), true); err != nil {
return err
}
err := gateWriteToSecondary(args.Datacenter, c.srv.config.Datacenter, c.srv.config.PrimaryDatacenter, args.Entry.GetKind())
if err != nil {
return err
}
// Ensure that all config entry writes go to the primary datacenter. These will then
// be replicated to all the other datacenters.
args.Datacenter = c.srv.config.PrimaryDatacenter
if done, err := c.srv.ForwardRPC("ConfigEntry.Apply", args, reply); done {
return err
}
defer metrics.MeasureSince([]string{"config_entry", "apply"}, time.Now())
entMeta := args.Entry.GetEnterpriseMeta()
authz, err := c.srv.ResolveTokenAndDefaultMeta(args.Token, entMeta, nil)
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
if err := c.preflightCheck(args.Entry.GetKind()); 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
// Normalize and validate the incoming config entry as if it came from a user.
if err := args.Entry.Normalize(); err != nil {
return err
}
if err := args.Entry.Validate(); err != nil {
return err
}
// Log any applicable warnings about the contents of the config entry.
if warnEntry, ok := args.Entry.(structs.WarningConfigEntry); ok {
warnings := warnEntry.Warnings()
for _, warning := range warnings {
c.logger.Warn(warning)
}
}
if err := args.Entry.CanWrite(authz); err != nil {
return err
}
Centralized Config CLI (#5731) * Add HTTP endpoints for config entry management * Finish implementing decoding in the HTTP Config entry apply endpoint * Add CAS operation to the config entry apply endpoint Also use this for the bootstrapping and move the config entry decoding function into the structs package. * First pass at the API client for the config entries * Fixup some of the ConfigEntry APIs Return a singular response object instead of a list for the ConfigEntry.Get RPC. This gets plumbed through the HTTP API as well. Dont return QueryMeta in the JSON response for the config entry listing HTTP API. Instead just return a list of config entries. * Minor API client fixes * Attempt at some ConfigEntry api client tests These don’t currently work due to weak typing in JSON * Get some of the api client tests passing * Implement reflectwalk magic to correct JSON encoding a ProxyConfigEntry Also added a test for the HTTP endpoint that exposes the problem. However, since the test doesn’t actually do the JSON encode/decode its still failing. * Move MapWalk magic into a binary marshaller instead of JSON. * Add a MapWalk test * Get rid of unused func * Get rid of unused imports * Fixup some tests now that the decoding from msgpack coerces things into json compat types * Stub out most of the central config cli Fully implement the config read command. * Basic config delete command implementation * Implement config write command * Implement config list subcommand Not entirely sure about the output here. Its basically the read output indented with a line specifying the kind/name of each type which is also duplicated in the indented output. * Update command usage * Update some help usage formatting * Add the connect enable helper cli command * Update list command output * Rename the config entry API client methods. * Use renamed apis * Implement config write tests Stub the others with the noTabs tests. * Change list output format Now just simply output 1 line per named config * Add config read tests * Add invalid args write test. * Add config delete tests * Add config list tests * Add connect enable tests * Update some CLI commands to use CAS ops This also modifies the HTTP API for a write op to return a boolean indicating whether the value was written or not. * Fix up the HTTP API CAS tests as I realized they weren’t testing what they should. * Update config entry rpc tests to properly test CAS * Fix up a few more tests * Fix some tests that using ConfigEntries.Apply * Update config_write_test.go * Get rid of unused import
2019-04-30 23:27:16 +00:00
if args.Op != structs.ConfigEntryUpsert && args.Op != structs.ConfigEntryUpsertCAS {
args.Op = structs.ConfigEntryUpsert
}
if skip, err := c.shouldSkipOperation(args); err != nil {
return err
} else if skip {
*reply = true
return nil
}
resp, err := c.srv.raftApply(structs.ConfigEntryRequestType, args)
if err != nil {
return err
}
Centralized Config CLI (#5731) * Add HTTP endpoints for config entry management * Finish implementing decoding in the HTTP Config entry apply endpoint * Add CAS operation to the config entry apply endpoint Also use this for the bootstrapping and move the config entry decoding function into the structs package. * First pass at the API client for the config entries * Fixup some of the ConfigEntry APIs Return a singular response object instead of a list for the ConfigEntry.Get RPC. This gets plumbed through the HTTP API as well. Dont return QueryMeta in the JSON response for the config entry listing HTTP API. Instead just return a list of config entries. * Minor API client fixes * Attempt at some ConfigEntry api client tests These don’t currently work due to weak typing in JSON * Get some of the api client tests passing * Implement reflectwalk magic to correct JSON encoding a ProxyConfigEntry Also added a test for the HTTP endpoint that exposes the problem. However, since the test doesn’t actually do the JSON encode/decode its still failing. * Move MapWalk magic into a binary marshaller instead of JSON. * Add a MapWalk test * Get rid of unused func * Get rid of unused imports * Fixup some tests now that the decoding from msgpack coerces things into json compat types * Stub out most of the central config cli Fully implement the config read command. * Basic config delete command implementation * Implement config write command * Implement config list subcommand Not entirely sure about the output here. Its basically the read output indented with a line specifying the kind/name of each type which is also duplicated in the indented output. * Update command usage * Update some help usage formatting * Add the connect enable helper cli command * Update list command output * Rename the config entry API client methods. * Use renamed apis * Implement config write tests Stub the others with the noTabs tests. * Change list output format Now just simply output 1 line per named config * Add config read tests * Add invalid args write test. * Add config delete tests * Add config list tests * Add connect enable tests * Update some CLI commands to use CAS ops This also modifies the HTTP API for a write op to return a boolean indicating whether the value was written or not. * Fix up the HTTP API CAS tests as I realized they weren’t testing what they should. * Update config entry rpc tests to properly test CAS * Fix up a few more tests * Fix some tests that using ConfigEntries.Apply * Update config_write_test.go * Get rid of unused import
2019-04-30 23:27:16 +00:00
if respBool, ok := resp.(bool); ok {
*reply = respBool
}
return nil
}
// shouldSkipOperation returns true if the result of the operation has
// already happened and is safe to skip.
//
// It is ok if this incorrectly detects something as changed when it
// in fact has not, the important thing is that it doesn't do
// the reverse and incorrectly detect a change as a no-op.
func (c *ConfigEntry) shouldSkipOperation(args *structs.ConfigEntryRequest) (bool, error) {
state := c.srv.fsm.State()
_, currentEntry, err := state.ConfigEntry(nil, args.Entry.GetKind(), args.Entry.GetName(), args.Entry.GetEnterpriseMeta())
if err != nil {
return false, fmt.Errorf("error reading current config entry value: %w", err)
}
switch args.Op {
case structs.ConfigEntryUpsert, structs.ConfigEntryUpsertCAS:
return c.shouldSkipUpsertOperation(currentEntry, args.Entry)
case structs.ConfigEntryDelete, structs.ConfigEntryDeleteCAS:
return (currentEntry == nil), nil
default:
return false, fmt.Errorf("invalid config entry operation type: %v", args.Op)
}
}
func (c *ConfigEntry) shouldSkipUpsertOperation(currentEntry, updatedEntry structs.ConfigEntry) (bool, error) {
if currentEntry == nil {
return false, nil
}
if currentEntry.GetKind() != updatedEntry.GetKind() ||
currentEntry.GetName() != updatedEntry.GetName() ||
!currentEntry.GetEnterpriseMeta().IsSame(updatedEntry.GetEnterpriseMeta()) {
return false, nil
}
// The only reason a fully Normalized and Validated config entry may
// legitimately differ from the persisted one is due to the embedded
// RaftIndex.
//
// So, to intercept more no-op upserts we temporarily set the new config
// entry's raft index field to that of the existing data for the purposes
// of comparison, and then restore it.
var (
currentRaftIndex = currentEntry.GetRaftIndex()
userProvidedRaftIndex = updatedEntry.GetRaftIndex()
currentRaftIndexCopy = *currentRaftIndex
userProvidedRaftIndexCopy = *userProvidedRaftIndex
)
*userProvidedRaftIndex = currentRaftIndexCopy // change
same := reflect.DeepEqual(currentEntry, updatedEntry) // compare
*userProvidedRaftIndex = userProvidedRaftIndexCopy // restore
return same, nil
}
// Get returns a single config entry by Kind/Name.
func (c *ConfigEntry) Get(args *structs.ConfigEntryQuery, reply *structs.ConfigEntryResponse) error {
if err := c.srv.validateEnterpriseRequest(&args.EnterpriseMeta, false); err != nil {
return err
}
if done, err := c.srv.ForwardRPC("ConfigEntry.Get", args, reply); done {
return err
}
defer metrics.MeasureSince([]string{"config_entry", "get"}, time.Now())
authz, err := c.srv.ResolveTokenAndDefaultMeta(args.Token, &args.EnterpriseMeta, nil)
if err != nil {
return err
}
// Create a dummy config entry to check the ACL permissions.
lookupEntry, err := structs.MakeConfigEntry(args.Kind, args.Name)
if err != nil {
return err
}
lookupEntry.GetEnterpriseMeta().Merge(&args.EnterpriseMeta)
if err := lookupEntry.CanRead(authz); err != nil {
return err
}
return c.srv.blockingQuery(
&args.QueryOptions,
&reply.QueryMeta,
func(ws memdb.WatchSet, state *state.Store) error {
index, entry, err := state.ConfigEntry(ws, args.Kind, args.Name, &args.EnterpriseMeta)
if err != nil {
return err
}
reply.Index, reply.Entry = index, entry
if entry == nil {
return errNotFound
}
return nil
})
}
// List returns all the config entries of the given kind. If Kind is blank,
// all existing config entries will be returned.
func (c *ConfigEntry) List(args *structs.ConfigEntryQuery, reply *structs.IndexedConfigEntries) error {
if err := c.srv.validateEnterpriseRequest(&args.EnterpriseMeta, false); err != nil {
return err
}
if done, err := c.srv.ForwardRPC("ConfigEntry.List", args, reply); done {
return err
}
defer metrics.MeasureSince([]string{"config_entry", "list"}, time.Now())
authz, err := c.srv.ResolveTokenAndDefaultMeta(args.Token, &args.EnterpriseMeta, nil)
if err != nil {
return err
}
if args.Kind != "" {
if _, err := structs.MakeConfigEntry(args.Kind, ""); err != nil {
return fmt.Errorf("invalid config entry kind: %s", args.Kind)
}
}
// Filtering.
// This is only supported for certain config entries.
var filter *bexpr.Filter
if args.Filter != "" {
switch args.Kind {
case structs.ServiceDefaults:
f, err := bexpr.CreateFilter(args.Filter, nil, []*structs.ServiceConfigEntry{})
if err != nil {
return err
}
filter = f
default:
return fmt.Errorf("filtering not supported for config entry kind=%v", args.Kind)
}
}
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
var (
priorHash uint64
ranOnce bool
)
return c.srv.blockingQuery(
&args.QueryOptions,
&reply.QueryMeta,
func(ws memdb.WatchSet, state *state.Store) error {
index, entries, err := state.ConfigEntriesByKind(ws, args.Kind, &args.EnterpriseMeta)
if err != nil {
return err
}
// Filter the entries returned by ACL permissions.
filteredEntries := make([]structs.ConfigEntry, 0, len(entries))
for _, entry := range entries {
if err := entry.CanRead(authz); err != nil {
// TODO we may wish to extract more details from this error to aid user comprehension
reply.QueryMeta.ResultsFilteredByACLs = true
continue
}
filteredEntries = append(filteredEntries, entry)
}
reply.Kind = args.Kind
reply.Index = index
reply.Entries = filteredEntries
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
// Generate a hash of the content driving this response. Use it to
// determine if the response is identical to a prior wakeup.
newHash, err := hashstructure_v2.Hash(filteredEntries, hashstructure_v2.FormatV2, nil)
if err != nil {
return fmt.Errorf("error hashing reply for spurious wakeup suppression: %w", err)
}
if filter != nil {
raw, err := filter.Execute(reply.Entries)
if err != nil {
return err
}
reply.Entries = raw.([]structs.ConfigEntry)
}
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
if ranOnce && priorHash == newHash {
priorHash = newHash
return errNotChanged
} else {
priorHash = newHash
ranOnce = true
}
if len(reply.Entries) == 0 {
return errNotFound
}
return 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
var configEntryKindsFromConsul_1_8_0 = []string{
structs.ServiceDefaults,
structs.ProxyDefaults,
structs.ServiceRouter,
structs.ServiceSplitter,
structs.ServiceResolver,
structs.IngressGateway,
structs.TerminatingGateway,
}
// ListAll returns all the known configuration entries
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
func (c *ConfigEntry) ListAll(args *structs.ConfigEntryListAllRequest, reply *structs.IndexedGenericConfigEntries) error {
if err := c.srv.validateEnterpriseRequest(&args.EnterpriseMeta, false); err != nil {
return err
}
if done, err := c.srv.ForwardRPC("ConfigEntry.ListAll", args, reply); done {
return err
}
defer metrics.MeasureSince([]string{"config_entry", "listAll"}, time.Now())
authz, err := c.srv.ResolveTokenAndDefaultMeta(args.Token, &args.EnterpriseMeta, nil)
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
if len(args.Kinds) == 0 {
args.Kinds = configEntryKindsFromConsul_1_8_0
}
kindMap := make(map[string]struct{})
for _, kind := range args.Kinds {
kindMap[kind] = struct{}{}
}
return c.srv.blockingQuery(
&args.QueryOptions,
&reply.QueryMeta,
func(ws memdb.WatchSet, state *state.Store) error {
index, entries, err := state.ConfigEntries(ws, &args.EnterpriseMeta)
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
// Filter the entries returned by ACL permissions or by the provided kinds.
filteredEntries := make([]structs.ConfigEntry, 0, len(entries))
for _, entry := range entries {
if err := entry.CanRead(authz); err != nil {
// TODO we may wish to extract more details from this error to aid user comprehension
reply.QueryMeta.ResultsFilteredByACLs = true
continue
}
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
// Doing this filter outside of memdb isn't terribly
// performant. This kind filter is currently only used across
// version upgrades, so in the common case we are going to
// always return all of the data anyway, so it should be fine.
// If that changes at some point, then we should move this down
// into memdb.
if _, ok := kindMap[entry.GetKind()]; !ok {
continue
}
filteredEntries = append(filteredEntries, entry)
}
reply.Entries = filteredEntries
reply.Index = index
return nil
})
}
// Delete deletes a config entry.
func (c *ConfigEntry) Delete(args *structs.ConfigEntryRequest, reply *structs.ConfigEntryDeleteResponse) error {
if err := c.srv.validateEnterpriseRequest(args.Entry.GetEnterpriseMeta(), true); err != nil {
return err
}
// Ensure that all config entry writes go to the primary datacenter. These will then
// be replicated to all the other datacenters.
args.Datacenter = c.srv.config.PrimaryDatacenter
if done, err := c.srv.ForwardRPC("ConfigEntry.Delete", args, reply); done {
return err
}
defer metrics.MeasureSince([]string{"config_entry", "delete"}, time.Now())
authz, err := c.srv.ResolveTokenAndDefaultMeta(args.Token, args.Entry.GetEnterpriseMeta(), nil)
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
if err := c.preflightCheck(args.Entry.GetKind()); err != nil {
return err
}
// Normalize the incoming entry.
if err := args.Entry.Normalize(); err != nil {
return err
}
if err := args.Entry.CanWrite(authz); err != nil {
return err
}
// Only delete and delete-cas ops are supported. If the caller erroneously
// sent something else, we assume they meant delete.
switch args.Op {
case structs.ConfigEntryDelete, structs.ConfigEntryDeleteCAS:
default:
args.Op = structs.ConfigEntryDelete
}
if skip, err := c.shouldSkipOperation(args); err != nil {
return err
} else if skip {
reply.Deleted = true
return nil
}
rsp, err := c.srv.raftApply(structs.ConfigEntryRequestType, args)
if err != nil {
return err
}
if args.Op == structs.ConfigEntryDeleteCAS {
// In CAS deletions the FSM will return a boolean value indicating whether the
// operation was successful.
deleted, _ := rsp.(bool)
reply.Deleted = deleted
} else {
// For non-CAS deletions any non-error result indicates a successful deletion.
reply.Deleted = true
}
return nil
}
// ResolveServiceConfig
func (c *ConfigEntry) ResolveServiceConfig(args *structs.ServiceConfigRequest, reply *structs.ServiceConfigResponse) error {
if err := c.srv.validateEnterpriseRequest(&args.EnterpriseMeta, false); err != nil {
return err
}
if done, err := c.srv.ForwardRPC("ConfigEntry.ResolveServiceConfig", args, reply); done {
return err
}
defer metrics.MeasureSince([]string{"config_entry", "resolve_service_config"}, time.Now())
var authzContext acl.AuthorizerContext
authz, err := c.srv.ResolveTokenAndDefaultMeta(args.Token, &args.EnterpriseMeta, &authzContext)
if err != nil {
return err
}
if err := authz.ToAllowAuthorizer().ServiceReadAllowed(args.Name, &authzContext); err != nil {
return err
}
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
var (
priorHash uint64
ranOnce bool
)
return c.srv.blockingQuery(
&args.QueryOptions,
&reply.QueryMeta,
func(ws memdb.WatchSet, state *state.Store) error {
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
// Fetch all relevant config entries.
index, entries, err := state.ReadResolvedServiceConfigEntries(
ws,
args.Name,
&args.EnterpriseMeta,
args.GetLocalUpstreamIDs(),
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
args.Mode,
)
if err != nil {
return err
}
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
// Generate a hash of the config entry content driving this
// response. Use it to determine if the response is identical to a
// prior wakeup.
newHash, err := hashstructure_v2.Hash(entries, hashstructure_v2.FormatV2, nil)
if err != nil {
return fmt.Errorf("error hashing reply for spurious wakeup suppression: %w", err)
}
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
if ranOnce && priorHash == newHash {
priorHash = newHash
reply.Index = index
// NOTE: the prior response is still alive inside of *reply, which
// is desirable
return errNotChanged
} else {
priorHash = newHash
ranOnce = true
}
thisReply, err := configentry.ComputeResolvedServiceConfig(
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
args,
entries,
c.logger,
)
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
if err != nil {
return err
}
thisReply.Index = index
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
*reply = *thisReply
if entries.IsEmpty() {
// No config entries factored into this reply; it's a default.
return errNotFound
}
server: suppress spurious blocking query returns where multiple config entries are involved (#12362) Starting from and extending the mechanism introduced in #12110 we can specially handle the 3 main special Consul RPC endpoints that react to many config entries in a single blocking query in Connect: - `DiscoveryChain.Get` - `ConfigEntry.ResolveServiceConfig` - `Intentions.Match` All of these will internally watch for many config entries, and at least one of those will likely be not found in any given query. Because these are blends of multiple reads the exact solution from #12110 isn't perfectly aligned, but we can tweak the approach slightly and regain the utility of that mechanism. ### No Config Entries Found In this case, despite looking for many config entries none may be found at all. Unlike #12110 in this scenario we do not return an empty reply to the caller, but instead synthesize a struct from default values to return. This can be handled nearly identically to #12110 with the first 1-2 replies being non-empty payloads followed by the standard spurious wakeup suppression mechanism from #12110. ### No Change Since Last Wakeup Once a blocking query loop on the server has completed and slept at least once, there is a further optimization we can make here to detect if any of the config entries that were present at specific versions for the prior execution of the loop are identical for the loop we just woke up for. In that scenario we can return a slightly different internal sentinel error and basically externally handle it similar to #12110. This would mean that even if 20 discovery chain read RPC handling goroutines wakeup due to the creation of an unrelated config entry, the only ones that will terminate and reply with a blob of data are those that genuinely have new data to report. ### Extra Endpoints Since this pattern is pretty reusable, other key config-entry-adjacent endpoints used by `agent/proxycfg` also were updated: - `ConfigEntry.List` - `Internal.IntentionUpstreams` (tproxy)
2022-02-25 21:46:34 +00:00
return nil
})
}
func gateWriteToSecondary(targetDC, localDC, primaryDC, kind string) error {
// ExportedServices entries are gated from interactions from secondary DCs
// because non-default partitions cannot be created in secondaries
// and services cannot be exported to another datacenter.
if kind != structs.ExportedServices {
return nil
}
if localDC == "" {
// This should not happen because the datacenter is defaulted in DefaultConfig.
return fmt.Errorf("unknown local datacenter")
}
if primaryDC == "" {
primaryDC = localDC
}
switch {
case targetDC == "" && localDC != primaryDC:
return fmt.Errorf("exported-services writes in secondary datacenters must target the primary datacenter explicitly.")
case targetDC != "" && targetDC != primaryDC:
return fmt.Errorf("exported-services writes must not target secondary datacenters.")
}
return 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
// preflightCheck is meant to have kind-specific system validation outside of
// content validation. The initial use case is restricting the ability to do
// writes of service-intentions until the system is finished migration.
func (c *ConfigEntry) preflightCheck(kind string) error {
switch kind {
case structs.ServiceIntentions:
// Exit early if Connect hasn't been enabled.
if !c.srv.config.ConnectEnabled {
return ErrConnectNotEnabled
}
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
usingConfigEntries, err := c.srv.fsm.State().AreIntentionsInConfigEntries()
if err != nil {
return fmt.Errorf("system metadata lookup failed: %v", err)
}
if !usingConfigEntries {
return ErrIntentionsNotUpgradedYet
}
}
return nil
}