query.go 7.33 KB
Newer Older
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
1 2
package query

3
import (
Jakub Sztandera's avatar
Jakub Sztandera committed
4
	goprocess "github.com/jbenet/goprocess"
5 6
)

Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58
/*
Query represents storage for any key-value pair.

tl;dr:

  queries are supported across datastores.
  Cheap on top of relational dbs, and expensive otherwise.
  Pick the right tool for the job!

In addition to the key-value store get and set semantics, datastore
provides an interface to retrieve multiple records at a time through
the use of queries. The datastore Query model gleans a common set of
operations performed when querying. To avoid pasting here years of
database research, let’s summarize the operations datastore supports.

Query Operations:

  * namespace - scope the query, usually by object type
  * filters - select a subset of values by applying constraints
  * orders - sort the results by applying sort conditions
  * limit - impose a numeric limit on the number of results
  * offset - skip a number of results (for efficient pagination)

datastore combines these operations into a simple Query class that allows
applications to define their constraints in a simple, generic, way without
introducing datastore specific calls, languages, etc.

Of course, different datastores provide relational query support across a
wide spectrum, from full support in traditional databases to none at all in
most key-value stores. Datastore aims to provide a common, simple interface
for the sake of application evolution over time and keeping large code bases
free of tool-specific code. It would be ridiculous to claim to support high-
performance queries on architectures that obviously do not. Instead, datastore
provides the interface, ideally translating queries to their native form
(e.g. into SQL for MySQL).

However, on the wrong datastore, queries can potentially incur the high cost
of performing the aforemantioned query operations on the data set directly in
Go. It is the client’s responsibility to select the right tool for the job:
pick a data storage solution that fits the application’s needs now, and wrap
it with a datastore implementation. As the needs change, swap out datastore
implementations to support your new use cases. Some applications, particularly
in early development stages, can afford to incurr the cost of queries on non-
relational databases (e.g. using a FSDatastore and not worry about a database
at all). When it comes time to switch the tool for performance, updating the
application code can be as simple as swapping the datastore in one place, not
all over the application code base. This gain in engineering time, both at
initial development and during later iterations, can significantly offset the
cost of the layer of abstraction.

*/
type Query struct {
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
59 60 61 62 63 64
	Prefix   string   // namespaces the query to results whose keys have Prefix
	Filters  []Filter // filter results. apply sequentially
	Orders   []Order  // order results. apply sequentially
	Limit    int      // maximum number of results
	Offset   int      // skip given number of results
	KeysOnly bool     // return only keys.
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
65 66
}

67 68 69 70
// NotFetched is a special type that signals whether or not the value
// of an Entry has been fetched or not. This is needed because
// datastore implementations get to decide whether Query returns values
// or only keys. nil is not a good signal, as real values may be nil.
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
71
const NotFetched int = iota
72

Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
73 74 75 76 77 78
// Entry is a query result entry.
type Entry struct {
	Key   string // cant be ds.Key because circular imports ...!!!
	Value interface{}
}

79 80 81 82
// Result is a special entry that includes an error, so that the client
// may be warned about internal errors.
type Result struct {
	Entry
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
83

84
	Error error
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
85 86
}

87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119
// Results is a set of Query results. This is the interface for clients.
// Example:
//
//   qr, _ := myds.Query(q)
//   for r := range qr.Next() {
//     if r.Error != nil {
//       // handle.
//       break
//     }
//
//     fmt.Println(r.Entry.Key, r.Entry.Value)
//   }
//
// or, wait on all results at once:
//
//   qr, _ := myds.Query(q)
//   es, _ := qr.Rest()
//   for _, e := range es {
//     	fmt.Println(e.Key, e.Value)
//   }
//
type Results interface {
	Query() Query           // the query these Results correspond to
	Next() <-chan Result    // returns a channel to wait for the next result
	Rest() ([]Entry, error) // waits till processing finishes, returns all entries at once.
	Close() error           // client may call Close to signal early exit

	// Process returns a goprocess.Process associated with these results.
	// most users will not need this function (Close is all they want),
	// but it's here in case you want to connect the results to other
	// goprocess-friendly things.
	Process() goprocess.Process
}
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
120

121 122 123 124 125 126
// results implements Results
type results struct {
	query Query
	proc  goprocess.Process
	res   <-chan Result
}
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
127

128
func (r *results) Next() <-chan Result {
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
129 130 131
	return r.res
}

132 133
func (r *results) Rest() ([]Entry, error) {
	var es []Entry
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
134
	for e := range r.res {
135 136 137 138
		if e.Error != nil {
			return es, e.Error
		}
		es = append(es, e.Entry)
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
139
	}
140 141 142 143 144 145 146 147 148 149 150 151 152 153
	<-r.proc.Closed() // wait till the processing finishes.
	return es, nil
}

func (r *results) Process() goprocess.Process {
	return r.proc
}

func (r *results) Close() error {
	return r.proc.Close()
}

func (r *results) Query() Query {
	return r.query
Juan Batiz-Benet's avatar
Juan Batiz-Benet committed
154
}
155

156 157 158
// ResultBuilder is what implementors use to construct results
// Implementors of datastores and their clients must respect the
// Process of the Request:
159
//
160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249
//   * clients must call r.Process().Close() on an early exit, so
//     implementations can reclaim resources.
//   * if the Entries are read to completion (channel closed), Process
//     should be closed automatically.
//   * datastores must respect <-Process.Closing(), which intermediates
//     an early close signal from the client.
//
type ResultBuilder struct {
	Query   Query
	Process goprocess.Process
	Output  chan Result
}

// Results returns a Results to to this builder.
func (rb *ResultBuilder) Results() Results {
	return &results{
		query: rb.Query,
		proc:  rb.Process,
		res:   rb.Output,
	}
}

func NewResultBuilder(q Query) *ResultBuilder {
	b := &ResultBuilder{
		Query:  q,
		Output: make(chan Result),
	}
	b.Process = goprocess.WithTeardown(func() error {
		close(b.Output)
		return nil
	})
	return b
}

// ResultsWithChan returns a Results object from a channel
// of Result entries. Respects its own Close()
func ResultsWithChan(q Query, res <-chan Result) Results {
	b := NewResultBuilder(q)

	// go consume all the entries and add them to the results.
	b.Process.Go(func(worker goprocess.Process) {
		for {
			select {
			case <-worker.Closing(): // client told us to close early
				return
			case e, more := <-res:
				if !more {
					return
				}

				select {
				case b.Output <- e:
				case <-worker.Closing(): // client told us to close early
					return
				}
			}
		}
		return
	})

	go b.Process.CloseAfterChildren()
	return b.Results()
}

// ResultsWithEntries returns a Results object from a list of entries
func ResultsWithEntries(q Query, res []Entry) Results {
	b := NewResultBuilder(q)

	// go consume all the entries and add them to the results.
	b.Process.Go(func(worker goprocess.Process) {
		for _, e := range res {
			select {
			case b.Output <- Result{Entry: e}:
			case <-worker.Closing(): // client told us to close early
				return
			}
		}
		return
	})

	go b.Process.CloseAfterChildren()
	return b.Results()
}

func ResultsReplaceQuery(r Results, q Query) Results {
	return &results{
		query: q,
		proc:  r.Process(),
		res:   r.Next(),
	}
250
}