-
Notifications
You must be signed in to change notification settings - Fork 8
/
Copy pathproject.go
264 lines (230 loc) · 6.04 KB
/
project.go
1
2
3
4
5
6
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
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
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
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
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
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
package ep
import (
"context"
"sync"
)
var _ = registerGob(project([]Runner{}), &dummyRunner{})
// Project returns a horizontal composite projection runner that dispatches
// its input to all of the internal runners, and joins the result into a single
// dataset to return. It is required that all runners produce Datasets of the
// same length.
func Project(runners ...Runner) Runner {
// flatten nested projects. note we should examine only first level, as any
// pre-created project was already flatten during its creation
var flat project
for _, r := range runners {
p, isProject := r.(project)
if isProject {
flat = append(flat, p...)
} else {
flat = append(flat, r)
}
}
if len(flat) == 0 {
panic("at least 1 runner is required for projecting")
} else if len(flat) == 1 {
return runners[0]
}
if cmpProj, ok := createComposeProjectRunner(runners); ok {
return cmpProj
}
return flat
}
// Placeholder returns a project with a placeholder of shift size length
func Placeholder(shift int, runners ...Runner) Runner {
dummies := make([]Runner, shift, shift+len(runners))
for i := 0; i < shift; i++ {
dummies[i] = dummyRunnerSingleton
}
runners = append(dummies, runners...)
return Project(runners...)
}
type project []Runner
func (rs project) Equals(other interface{}) bool {
otherP, ok := other.(project)
if !ok || len(rs) != len(otherP) {
return false
}
for i, r := range rs {
if !r.Equals(otherP[i]) {
return false
}
}
return true
}
// Returns a concatenation of all runners' return types
func (rs project) Returns() []Type {
var types []Type
for _, r := range rs {
types = append(types, r.Returns()...)
}
return types
}
func (rs project) Filter(keep []bool) {
currIdx := 0
for i, r := range rs {
returnLen := len(r.Returns())
// simplest (and most common for project) case - r return single value
if returnLen == 1 {
if !keep[currIdx] {
rs[i] = dummyRunnerSingleton
}
} else {
if r, isFilterable := r.(FilterRunner); isFilterable {
r.Filter(keep[currIdx : currIdx+returnLen])
}
}
currIdx += returnLen
}
}
// Run dispatches the same input to all inner runners, then collects and
// joins their results into a single dataset output
func (rs project) Run(ctx context.Context, inp, out chan Dataset) (err error) {
rs.useDummySingleton()
// set up the left and right channels
inps := make([]chan Dataset, len(rs))
outs := make([]chan Dataset, len(rs))
errs := make([]error, len(rs))
var wg sync.WaitGroup
defer func() {
wg.Wait()
// choose first error out from all errors, that isn't project internal error
for _, errI := range errs {
if err == nil && errI != nil {
err = errI
break
}
}
}()
ctx, cancel := context.WithCancel(ctx)
// run all runners in go-routines
for i := range rs {
inps[i] = make(chan Dataset)
outs[i] = make(chan Dataset)
wg.Add(1)
go func(idx int) {
defer wg.Done()
Run(ctx, rs[idx], inps[idx], outs[idx], cancel, &errs[idx])
}(i)
}
wg.Add(1)
go func() {
defer drain(inp)
defer wg.Done()
defer func() {
for i := range rs {
close(inps[i])
}
}()
for {
select {
case <-ctx.Done(): // listen to both new ctx and original ctx
cancel()
return
case data, ok := <-inp:
if !ok {
return
}
// dispatch (duplicate) input to all runners
for i := range rs {
inps[i] <- data
}
}
}
}()
// cancel all runners when we're done - just in case few still running
// NOTE: cancel must be first defer to be called, to allow internal runners to finish
defer func() {
cancel()
for i := range rs {
// drain all runners' output to allow them catch cancellation
go drain(outs[i])
}
}()
// collect & join the output from all runners, in order
var resultWidth int
for {
resultLen := -1
result := make([]Data, 0, resultWidth)
allOpen, allDummies := true, true
for i := range rs {
curr, open := <-outs[i]
if rs[i] == dummyRunnerSingleton {
// dummy runners shouldn't affect state check
curr, open = variadicDummiesBatch, allOpen
} else if allDummies {
allDummies = false
// init allOpen according to first out channel of non dummy runner
allOpen = open
} else if allOpen != open { // verify all still open or all closed
return nil
}
if open {
resultLen, err = verifySameLength(resultLen, curr.Len())
if err != nil {
return err
}
result = append(result, curr.(dataset)...)
}
}
if !allOpen {
return // all done
}
resultWidth = len(result)
out <- dataset(result)
if allDummies {
return
}
}
}
func (rs project) Scopes() StringsSet {
scopes := make(StringsSet)
for _, r := range rs {
if r, ok := r.(ScopesRunner); ok {
scopes.AddAll(r.Scopes())
}
}
return scopes
}
func (rs project) ApproxSize() int {
totalSize := 0
for _, r := range rs {
sizer, ok := r.(ApproxSizer)
if !ok {
return UnknownSize
}
size := sizer.ApproxSize()
if size == UnknownSize {
return UnknownSize
}
totalSize += size
}
return totalSize
}
// useDummySingleton replaces all dummies with pre-defined singleton to allow addresses comparison
// instead of casting for each batch.
// required for distribute runner that creates new dummy instances instead of using singleton
func (rs project) useDummySingleton() {
for i, r := range rs {
if _, isDummy := r.(*dummyRunner); isDummy {
rs[i] = dummyRunnerSingleton
}
}
}
// dummyRunnerSingleton is a runner that does nothing and just returns immediately
var dummyRunnerSingleton = &dummyRunner{}
// variadicDummiesBatch is used for replacing unused columns
var variadicDummiesBatch = NewDataset(dummy.Data(-1))
type dummyRunner struct{}
func (*dummyRunner) Equals(other interface{}) bool {
_, ok := other.(*dummyRunner)
return ok
}
func (*dummyRunner) Args() []Type { return []Type{Wildcard} }
func (*dummyRunner) Returns() []Type { return []Type{dummy} }
func (*dummyRunner) Run(_ context.Context, inp, out chan Dataset) error {
return nil
}
func (*dummyRunner) BatchFunction() BatchFunction {
return func(Dataset) (Dataset, error) { return variadicDummiesBatch, nil }
}