cache

package
v0.9.8 Latest Latest
Warning

This package is not in the latest version of its module.

Go to latest
Published: Dec 7, 2020 License: Apache-2.0 Imports: 19 Imported by: 0

Documentation

Overview

Package cache defines a configuration cache for the server.

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

func GetResourceName

func GetResourceName(res types.Resource) string

GetResourceName returns the resource name for a valid xDS response type.

func GetResourceReferences

func GetResourceReferences(resources map[string]types.Resource) map[string]bool

GetResourceReferences returns the names for dependent resources (EDS cluster names for CDS, RDS routes names for LDS).

func GetResponseType

func GetResponseType(typeURL string) types.ResponseType

GetResponseType returns the enumeration for a valid xDS type URL

func IndexResourcesByName

func IndexResourcesByName(items []types.Resource) map[string]types.Resource

IndexResourcesByName creates a map from the resource name to the resource.

func MarshalResource

func MarshalResource(resource types.Resource) (types.MarshaledResource, error)

MarshalResource converts the Resource to MarshaledResource

Types

type Cache

type Cache interface {
	ConfigWatcher
	ConfigFetcher
}

Cache is a generic config cache with a watcher.

type ConfigFetcher added in v0.9.7

type ConfigFetcher interface {
	// Fetch implements the polling method of the config cache using a non-empty request.
	Fetch(context.Context, *Request) (Response, error)
}

ConfigFetcher fetches configuration resources from cache

type ConfigWatcher

type ConfigWatcher interface {
	// CreateWatch returns a new open watch from a non-empty request.
	// An individual consumer normally issues a single open watch by each type URL.
	//
	// Value channel produces requested resources, once they are available.  If
	// the channel is closed prior to cancellation of the watch, an unrecoverable
	// error has occurred in the producer, and the consumer should close the
	// corresponding stream.
	//
	// Cancel is an optional function to release resources in the producer. If
	// provided, the consumer may call this function multiple times.
	CreateWatch(*Request) (value chan Response, cancel func())
}

ConfigWatcher requests watches for configuration resources by a node, last applied version identifier, and resource names hint. The watch should send the responses when they are ready. The watch can be cancelled by the consumer, in effect terminating the watch for the request. ConfigWatcher implementation must be thread-safe.

type IDHash

type IDHash struct{}

IDHash uses ID field as the node hash.

func (IDHash) ID

func (IDHash) ID(node *core.Node) string

ID uses the node ID field

type LinearCache added in v0.9.7

type LinearCache struct {
	// contains filtered or unexported fields
}

LinearCache supports collectons of opaque resources. This cache has a single collection indexed by resource names and manages resource versions internally. It implements the cache interface for a single type URL and should be combined with other caches via type URL muxing. It can be used to supply EDS entries, for example, uniformly across a fleet of proxies.

func NewLinearCache added in v0.9.7

func NewLinearCache(typeURL string, opts ...LinearCacheOption) *LinearCache

NewLinearCache creates a new cache. See the comments on the struct definition.

func (*LinearCache) CreateWatch added in v0.9.7

func (cache *LinearCache) CreateWatch(request *Request) (chan Response, func())

func (*LinearCache) DeleteResource added in v0.9.7

func (cache *LinearCache) DeleteResource(name string) error

DeleteResource removes a resource in the collection.

func (*LinearCache) Fetch added in v0.9.7

func (cache *LinearCache) Fetch(ctx context.Context, request *Request) (Response, error)

func (*LinearCache) NumWatches added in v0.9.7

func (cache *LinearCache) NumWatches(name string) int

Number of active watches for a resource name.

func (*LinearCache) UpdateResource added in v0.9.7

func (cache *LinearCache) UpdateResource(name string, res types.Resource) error

UpdateResource updates a resource in the collection.

type LinearCacheOption added in v0.9.7

type LinearCacheOption func(*LinearCache)

Options for modifying the behavior of the linear cache.

func WithInitialResources added in v0.9.7

func WithInitialResources(resources map[string]types.Resource) LinearCacheOption

WithInitialResources initializes the initial set of resources.

func WithVersionPrefix added in v0.9.7

func WithVersionPrefix(prefix string) LinearCacheOption

WithVersionPrefix sets a version prefix of the form "prefixN" in the version info. Version prefix can be used to distinguish replicated instances of the cache, in case a client re-connects to another instance.

type MuxCache added in v0.9.7

type MuxCache struct {
	// Classification functions.
	Classify func(Request) string
	// Muxed caches.
	Caches map[string]Cache
}

MuxCache multiplexes across several caches using a classification function. If there is no matching cache for a classification result, the cache responds with an empty closed channel, which effectively terminates the stream on the server. It might be preferred to respond with a "nil" channel instead which will leave the stream open in case the stream is aggregated by making sure there is always a matching cache.

func (*MuxCache) CreateWatch added in v0.9.7

func (mux *MuxCache) CreateWatch(request *Request) (chan Response, func())

func (*MuxCache) Fetch added in v0.9.7

func (mux *MuxCache) Fetch(ctx context.Context, request *Request) (Response, error)

type NodeHash

type NodeHash interface {
	// ID function defines a unique string identifier for the remote Envoy node.
	ID(node *core.Node) string
}

NodeHash computes string identifiers for Envoy nodes.

type PassthroughResponse added in v0.9.6

type PassthroughResponse struct {
	// Request is the original request.
	Request *discovery.DiscoveryRequest

	// The discovery response that needs to be sent as is, without any marshalling transformations.
	DiscoveryResponse *discovery.DiscoveryResponse
}

PassthroughResponse is a pre constructed xDS response that need not go through marshalling transformations.

func (*PassthroughResponse) GetDiscoveryResponse added in v0.9.6

func (r *PassthroughResponse) GetDiscoveryResponse() (*discovery.DiscoveryResponse, error)

GetDiscoveryResponse returns the final passthrough Discovery Response.

func (*PassthroughResponse) GetRequest added in v0.9.6

GetRequest returns the original Discovery Request

func (*PassthroughResponse) GetVersion added in v0.9.6

func (r *PassthroughResponse) GetVersion() (string, error)

GetVersion returns the response version.

type RawResponse added in v0.9.6

type RawResponse struct {
	// Request is the original request.
	Request *discovery.DiscoveryRequest

	// Version of the resources as tracked by the cache for the given type.
	// Proxy responds with this version as an acknowledgement.
	Version string

	// Resources to be included in the response.
	Resources []types.Resource
	// contains filtered or unexported fields
}

RawResponse is a pre-serialized xDS response containing the raw resources to be included in the final Discovery Response.

func (*RawResponse) GetDiscoveryResponse added in v0.9.6

func (r *RawResponse) GetDiscoveryResponse() (*discovery.DiscoveryResponse, error)

GetDiscoveryResponse performs the marshalling the first time its called and uses the cached response subsequently. This is necessary because the marshalled response does not change across the calls. This caching behavior is important in high throughput scenarios because grpc marshalling has a cost and it drives the cpu utilization under load.

func (*RawResponse) GetRequest added in v0.9.6

func (r *RawResponse) GetRequest() *discovery.DiscoveryRequest

GetRequest returns the original Discovery Request.

func (*RawResponse) GetVersion added in v0.9.6

func (r *RawResponse) GetVersion() (string, error)

GetVersion returns the response version.

type Request

type Request = discovery.DiscoveryRequest

Request is an alias for the discovery request type.

type Resources

type Resources struct {
	// Version information.
	Version string

	// Items in the group indexed by name.
	Items map[string]types.Resource
}

Resources is a versioned group of resources.

func NewResources

func NewResources(version string, items []types.Resource) Resources

NewResources creates a new resource group.

type Response

type Response interface {
	// Get the Constructed DiscoveryResponse
	GetDiscoveryResponse() (*discovery.DiscoveryResponse, error)

	// Get the original Request for the Response.
	GetRequest() *discovery.DiscoveryRequest

	// Get the version in the Response.
	GetVersion() (string, error)
}

Response is a wrapper around Envoy's DiscoveryResponse.

type ResponseWatch

type ResponseWatch struct {
	// Request is the original request for the watch.
	Request *Request

	// Response is the channel to push responses to.
	Response chan Response
}

ResponseWatch is a watch record keeping both the request and an open channel for the response.

type Snapshot

type Snapshot struct {
	Resources [types.UnknownType]Resources
}

Snapshot is an internally consistent snapshot of xDS resources. Consistency is important for the convergence as different resource types from the snapshot may be delivered to the proxy in arbitrary order.

func NewSnapshot

func NewSnapshot(version string,
	endpoints []types.Resource,
	clusters []types.Resource,
	routes []types.Resource,
	listeners []types.Resource,
	runtimes []types.Resource,
	secrets []types.Resource) Snapshot

NewSnapshot creates a snapshot from response types and a version.

func (*Snapshot) Consistent

func (s *Snapshot) Consistent() error

Consistent check verifies that the dependent resources are exactly listed in the snapshot: - all EDS resources are listed by name in CDS resources - all RDS resources are listed by name in LDS resources

Note that clusters and listeners are requested without name references, so Envoy will accept the snapshot list of clusters as-is even if it does not match all references found in xDS.

func (*Snapshot) GetResources

func (s *Snapshot) GetResources(typeURL string) map[string]types.Resource

GetResources selects snapshot resources by type.

func (*Snapshot) GetVersion

func (s *Snapshot) GetVersion(typeURL string) string

GetVersion returns the version for a resource type.

type SnapshotCache

type SnapshotCache interface {
	Cache

	// SetSnapshot sets a response snapshot for a node. For ADS, the snapshots
	// should have distinct versions and be internally consistent (e.g. all
	// referenced resources must be included in the snapshot).
	//
	// This method will cause the server to respond to all open watches, for which
	// the version differs from the snapshot version.
	SetSnapshot(node string, snapshot Snapshot) error

	// GetSnapshots gets the snapshot for a node.
	GetSnapshot(node string) (Snapshot, error)

	// ClearSnapshot removes all status and snapshot information associated with a node.
	ClearSnapshot(node string)

	// GetStatusInfo retrieves status information for a node ID.
	GetStatusInfo(string) StatusInfo

	// GetStatusKeys retrieves node IDs for all statuses.
	GetStatusKeys() []string
}

SnapshotCache is a snapshot-based cache that maintains a single versioned snapshot of responses per node. SnapshotCache consistently replies with the latest snapshot. For the protocol to work correctly in ADS mode, EDS/RDS requests are responded only when all resources in the snapshot xDS response are named as part of the request. It is expected that the CDS response names all EDS clusters, and the LDS response names all RDS routes in a snapshot, to ensure that Envoy makes the request for all EDS clusters or RDS routes eventually.

SnapshotCache can operate as a REST or regular xDS backend. The snapshot can be partial, e.g. only include RDS or EDS resources.

func NewSnapshotCache

func NewSnapshotCache(ads bool, hash NodeHash, logger log.Logger) SnapshotCache

NewSnapshotCache initializes a simple cache.

ADS flag forces a delay in responding to streaming requests until all resources are explicitly named in the request. This avoids the problem of a partial request over a single stream for a subset of resources which would require generating a fresh version for acknowledgement. ADS flag requires snapshot consistency. For non-ADS case (and fetch), multiple partial requests are sent across multiple streams and re-using the snapshot version is OK.

Logger is optional.

type StatusInfo

type StatusInfo interface {
	// GetNode returns the node metadata.
	GetNode() *core.Node

	// GetNumWatches returns the number of open watches.
	GetNumWatches() int

	// GetLastWatchRequestTime returns the timestamp of the last discovery watch request.
	GetLastWatchRequestTime() time.Time
}

StatusInfo tracks the server state for the remote Envoy node. Not all fields are used by all cache implementations.

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL