// Copyright (c) HashiCorp, Inc. // SPDX-License-Identifier: MPL-2.0 package inmem import ( "bytes" "fmt" "strings" "github.com/hashicorp/go-memdb" "github.com/hashicorp/consul/internal/storage" "github.com/hashicorp/consul/proto-public/pbresource" ) const ( tableNameMetadata = "metadata" tableNameResources = "resources" indexNameID = "id" indexNameOwner = "owner" metaKeyEventIndex = "index" ) func newDB() (*memdb.MemDB, error) { return memdb.NewMemDB(&memdb.DBSchema{ Tables: map[string]*memdb.TableSchema{ tableNameMetadata: { Name: tableNameMetadata, Indexes: map[string]*memdb.IndexSchema{ indexNameID: { Name: indexNameID, AllowMissing: false, Unique: true, Indexer: &memdb.StringFieldIndex{Field: "Key"}, }, }, }, tableNameResources: { Name: tableNameResources, Indexes: map[string]*memdb.IndexSchema{ indexNameID: { Name: indexNameID, AllowMissing: false, Unique: true, Indexer: idIndexer{}, }, indexNameOwner: { Name: indexNameOwner, AllowMissing: true, Unique: false, Indexer: ownerIndexer{}, }, }, }, }, }) } // indexSeparator delimits the segments of our radix tree keys. const indexSeparator = "\x00" // idIndexer implements the memdb.Indexer, memdb.SingleIndexer and // memdb.PrefixIndexer interfaces. It is used for indexing resources // by their IDs. type idIndexer struct{} // FromArgs constructs a radix tree key from an ID for lookup. func (i idIndexer) FromArgs(args ...any) ([]byte, error) { if l := len(args); l != 1 { return nil, fmt.Errorf("expected 1 arg, got: %d", l) } id, ok := args[0].(*pbresource.ID) if !ok { return nil, fmt.Errorf("expected *pbresource.ID, got: %T", args[0]) } return indexFromID(id, false), nil } // FromObject constructs a radix tree key from a Resource at write-time, or an // ID at delete-time. func (i idIndexer) FromObject(raw any) (bool, []byte, error) { switch t := raw.(type) { case *pbresource.ID: return true, indexFromID(t, false), nil case *pbresource.Resource: return true, indexFromID(t.Id, false), nil } return false, nil, fmt.Errorf("expected *pbresource.Resource or *pbresource.ID, got: %T", raw) } // PrefixFromArgs constructs a radix tree key prefix from a query for listing. func (i idIndexer) PrefixFromArgs(args ...any) ([]byte, error) { if l := len(args); l != 1 { return nil, fmt.Errorf("expected 1 arg, got: %d", l) } q, ok := args[0].(query) if !ok { return nil, fmt.Errorf("expected query, got: %T", args[0]) } return q.indexPrefix(), nil } // ownerIndexer implements the memdb.Indexer and memdb.SingleIndexer interfaces. // It is used for indexing resources by their owners. type ownerIndexer struct{} // FromArgs constructs a radix tree key from an ID for lookup. func (i ownerIndexer) FromArgs(args ...any) ([]byte, error) { if l := len(args); l != 1 { return nil, fmt.Errorf("expected 1 arg, got: %d", l) } id, ok := args[0].(*pbresource.ID) if !ok { return nil, fmt.Errorf("expected *pbresource.ID, got: %T", args[0]) } return indexFromID(id, true), nil } // FromObject constructs a radix key tree from a Resource at write-time. func (i ownerIndexer) FromObject(raw any) (bool, []byte, error) { res, ok := raw.(*pbresource.Resource) if !ok { return false, nil, fmt.Errorf("expected *pbresource.Resource, got: %T", raw) } if res.Owner == nil { return false, nil, nil } return true, indexFromID(res.Owner, true), nil } func indexFromType(t storage.UnversionedType) []byte { var b indexBuilder b.String(t.Group) b.String(t.Kind) return b.Bytes() } func indexFromTenancy(t *pbresource.Tenancy) []byte { var b indexBuilder b.String(t.Partition) b.String(t.PeerName) b.String(t.Namespace) return b.Bytes() } func indexFromID(id *pbresource.ID, includeUid bool) []byte { var b indexBuilder b.Raw(indexFromType(storage.UnversionedTypeFrom(id.Type))) b.Raw(indexFromTenancy(id.Tenancy)) b.String(id.Name) if includeUid { b.String(id.Uid) } return b.Bytes() } type indexBuilder bytes.Buffer func (i *indexBuilder) Raw(v []byte) { (*bytes.Buffer)(i).Write(v) } func (i *indexBuilder) String(s string) { (*bytes.Buffer)(i).WriteString(s) (*bytes.Buffer)(i).WriteString(indexSeparator) } func (i *indexBuilder) Bytes() []byte { return (*bytes.Buffer)(i).Bytes() } type query struct { resourceType storage.UnversionedType tenancy *pbresource.Tenancy namePrefix string } // indexPrefix is called by idIndexer.PrefixFromArgs to construct a radix tree // key prefix for list queries. // // Our radix tree keys are structured like so: // // // // Where each segment is followed by a NULL terminator. // // In order to handle wildcard queries, we return a prefix up to the wildcarded // field. For example: // // Query: type={mesh,v1,service}, partition=default, peer=*, namespace=default // Prefix: mesh[NULL]v1[NULL]service[NULL]default[NULL] // // Which means that we must manually apply filters after the wildcard (i.e. // namespace in the above example) in the matches method. func (q query) indexPrefix() []byte { var b indexBuilder b.Raw(indexFromType(q.resourceType)) if v := q.tenancy.Partition; v == storage.Wildcard { return b.Bytes() } else { b.String(v) } if v := q.tenancy.PeerName; v == storage.Wildcard { return b.Bytes() } else { b.String(v) } if v := q.tenancy.Namespace; v == storage.Wildcard { return b.Bytes() } else { b.String(v) } if q.namePrefix != "" { b.Raw([]byte(q.namePrefix)) } return b.Bytes() } // matches applies filters that couldn't be applied by just doing a radix tree // prefix scan, because an earlier segment of the key prefix was wildcarded. // // See docs on query.indexPrefix for an example. func (q query) matches(res *pbresource.Resource) bool { if q.tenancy.Partition != storage.Wildcard && res.Id.Tenancy.Partition != q.tenancy.Partition { return false } if q.tenancy.PeerName != storage.Wildcard && res.Id.Tenancy.PeerName != q.tenancy.PeerName { return false } if q.tenancy.Namespace != storage.Wildcard && res.Id.Tenancy.Namespace != q.tenancy.Namespace { return false } if len(q.namePrefix) != 0 && !strings.HasPrefix(res.Id.Name, q.namePrefix) { return false } return true }