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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361
|
package flightcontrol
import (
"context"
"io"
"math/rand"
"sort"
"sync"
"time"
"github.com/moby/buildkit/util/progress"
"github.com/pkg/errors"
)
// flightcontrol is like singleflight but with support for cancellation and
// nested progress reporting
var (
errRetry = errors.Errorf("retry")
errRetryTimeout = errors.Errorf("exceeded retry timeout")
)
type contextKeyT string
var contextKey = contextKeyT("buildkit/util/flightcontrol.progress")
// Group is a flightcontrol synchronization group
type Group[T any] struct {
mu sync.Mutex // protects m
m map[string]*call[T] // lazily initialized
}
// Do executes a context function syncronized by the key
func (g *Group[T]) Do(ctx context.Context, key string, fn func(ctx context.Context) (T, error)) (v T, err error) {
var backoff time.Duration
for {
v, err = g.do(ctx, key, fn)
if err == nil || !errors.Is(err, errRetry) {
return v, err
}
// backoff logic
if backoff >= 15*time.Second {
err = errors.Wrapf(errRetryTimeout, "flightcontrol")
return v, err
}
if backoff > 0 {
backoff = time.Duration(float64(backoff) * 1.2)
} else {
// randomize initial backoff to avoid all goroutines retrying at once
//nolint:gosec // using math/rand pseudo-randomness is acceptable here
backoff = time.Millisecond + time.Duration(rand.Intn(1e7))*time.Nanosecond
}
time.Sleep(backoff)
}
}
func (g *Group[T]) do(ctx context.Context, key string, fn func(ctx context.Context) (T, error)) (T, error) {
g.mu.Lock()
if g.m == nil {
g.m = make(map[string]*call[T])
}
if c, ok := g.m[key]; ok { // register 2nd waiter
g.mu.Unlock()
return c.wait(ctx)
}
c := newCall(fn)
g.m[key] = c
go func() {
// cleanup after a caller has returned
<-c.ready
g.mu.Lock()
delete(g.m, key)
g.mu.Unlock()
close(c.cleaned)
}()
g.mu.Unlock()
return c.wait(ctx)
}
type call[T any] struct {
mu sync.Mutex
result T
err error
ready chan struct{}
cleaned chan struct{}
ctx *sharedContext[T]
ctxs []context.Context
fn func(ctx context.Context) (T, error)
once sync.Once
closeProgressWriter func(error)
progressState *progressState
progressCtx context.Context
}
func newCall[T any](fn func(ctx context.Context) (T, error)) *call[T] {
c := &call[T]{
fn: fn,
ready: make(chan struct{}),
cleaned: make(chan struct{}),
progressState: newProgressState(),
}
ctx := newContext(c) // newSharedContext
pr, pctx, closeProgressWriter := progress.NewContext(context.Background())
c.progressCtx = pctx
c.ctx = ctx
c.closeProgressWriter = closeProgressWriter
go c.progressState.run(pr) // TODO: remove this, wrap writer instead
return c
}
func (c *call[T]) run() {
defer c.closeProgressWriter(errors.WithStack(context.Canceled))
ctx, cancel := context.WithCancelCause(c.ctx)
defer cancel(errors.WithStack(context.Canceled))
v, err := c.fn(ctx)
c.mu.Lock()
c.result = v
c.err = err
c.mu.Unlock()
close(c.ready)
}
func (c *call[T]) wait(ctx context.Context) (v T, err error) {
var empty T
c.mu.Lock()
// detect case where caller has just returned, let it clean up before
select {
case <-c.ready:
c.mu.Unlock()
if c.err != nil { // on error retry
<-c.cleaned
return empty, errRetry
}
pw, ok, _ := progress.NewFromContext(ctx)
if ok {
c.progressState.add(pw)
}
return c.result, nil
case <-c.ctx.done: // could return if no error
c.mu.Unlock()
<-c.cleaned
return empty, errRetry
default:
}
pw, ok, ctx := progress.NewFromContext(ctx)
if ok {
c.progressState.add(pw)
}
ctx, cancel := context.WithCancelCause(ctx)
defer cancel(errors.WithStack(context.Canceled))
c.ctxs = append(c.ctxs, ctx)
c.mu.Unlock()
go c.once.Do(c.run)
select {
case <-ctx.Done():
if c.ctx.checkDone() {
// if this cancelled the last context, then wait for function to shut down
// and don't accept any more callers
<-c.ready
return c.result, c.err
}
if ok {
c.progressState.close(pw)
}
return empty, context.Cause(ctx)
case <-c.ready:
return c.result, c.err // shared not implemented yet
}
}
func (c *call[T]) Deadline() (deadline time.Time, ok bool) {
c.mu.Lock()
defer c.mu.Unlock()
for _, ctx := range c.ctxs {
select {
case <-ctx.Done():
default:
dl, ok := ctx.Deadline()
if ok {
return dl, ok
}
}
}
return time.Time{}, false
}
func (c *call[T]) Done() <-chan struct{} {
return c.ctx.done
}
func (c *call[T]) Err() error {
select {
case <-c.ctx.Done():
return c.ctx.err
default:
return nil
}
}
func (c *call[T]) Value(key interface{}) interface{} {
if key == contextKey {
return c.progressState
}
c.mu.Lock()
defer c.mu.Unlock()
ctx := c.progressCtx
select {
case <-ctx.Done():
default:
if v := ctx.Value(key); v != nil {
return v
}
}
if len(c.ctxs) > 0 {
ctx = c.ctxs[0]
select {
case <-ctx.Done():
default:
if v := ctx.Value(key); v != nil {
return v
}
}
}
return nil
}
type sharedContext[T any] struct {
*call[T]
done chan struct{}
err error
}
func newContext[T any](c *call[T]) *sharedContext[T] {
return &sharedContext[T]{call: c, done: make(chan struct{})}
}
func (sc *sharedContext[T]) checkDone() bool {
sc.mu.Lock()
select {
case <-sc.done:
sc.mu.Unlock()
return true
default:
}
var err error
for _, ctx := range sc.ctxs {
select {
case <-ctx.Done():
// Cause can't be used here because this error is returned for Err() in custom context
// implementation and unfortunately stdlib does not allow defining Cause() for custom contexts
err = ctx.Err() //nolint: forbidigo
default:
sc.mu.Unlock()
return false
}
}
sc.err = err
close(sc.done)
sc.mu.Unlock()
return true
}
type rawProgressWriter interface {
WriteRawProgress(*progress.Progress) error
Close() error
}
type progressState struct {
mu sync.Mutex
items map[string]*progress.Progress
writers []rawProgressWriter
done bool
}
func newProgressState() *progressState {
return &progressState{
items: make(map[string]*progress.Progress),
}
}
func (ps *progressState) run(pr progress.Reader) {
for {
p, err := pr.Read(context.TODO())
if err != nil {
if err == io.EOF {
ps.mu.Lock()
ps.done = true
ps.mu.Unlock()
for _, w := range ps.writers {
w.Close()
}
}
return
}
ps.mu.Lock()
for _, p := range p {
for _, w := range ps.writers {
w.WriteRawProgress(p)
}
ps.items[p.ID] = p
}
ps.mu.Unlock()
}
}
func (ps *progressState) add(pw progress.Writer) {
rw, ok := pw.(rawProgressWriter)
if !ok {
return
}
ps.mu.Lock()
plist := make([]*progress.Progress, 0, len(ps.items))
for _, p := range ps.items {
plist = append(plist, p)
}
sort.Slice(plist, func(i, j int) bool {
return plist[i].Timestamp.Before(plist[j].Timestamp)
})
for _, p := range plist {
rw.WriteRawProgress(p)
}
if ps.done {
rw.Close()
} else {
ps.writers = append(ps.writers, rw)
}
ps.mu.Unlock()
}
func (ps *progressState) close(pw progress.Writer) {
rw, ok := pw.(rawProgressWriter)
if !ok {
return
}
ps.mu.Lock()
for i, w := range ps.writers {
if w == rw {
w.Close()
ps.writers = append(ps.writers[:i], ps.writers[i+1:]...)
break
}
}
ps.mu.Unlock()
}
|