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
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
|
package catfile
import (
"context"
"fmt"
"strings"
"sync"
"time"
"github.com/prometheus/client_golang/prometheus"
"gitlab.com/gitlab-org/gitaly/v16/internal/git"
"gitlab.com/gitlab-org/gitaly/v16/internal/gitaly/config"
"gitlab.com/gitlab-org/gitaly/v16/internal/gitaly/storage"
"gitlab.com/gitlab-org/gitaly/v16/internal/grpc/metadata"
"gitlab.com/gitlab-org/gitaly/v16/internal/helper"
"gitlab.com/gitlab-org/gitaly/v16/internal/tracing"
"gitlab.com/gitlab-org/labkit/correlation"
)
const (
// defaultBatchfileTTL is the default ttl for batch files to live in the cache
defaultBatchfileTTL = 10 * time.Second
defaultEvictionInterval = 1 * time.Second
// The default maximum number of cache entries
defaultMaxLen = 100
// SessionIDField is the gRPC metadata field we use to store the gitaly session ID.
SessionIDField = "gitaly-session-id"
)
// Cache is a cache for git-cat-file(1) processes.
type Cache interface {
// ObjectReader either creates a new object reader or returns a cached one for the given
// repository.
ObjectReader(context.Context, git.RepositoryExecutor) (ObjectContentReader, func(), error)
// ObjectInfoReader either creates a new object info reader or returns a cached one for the
// given repository.
ObjectInfoReader(context.Context, git.RepositoryExecutor) (ObjectInfoReader, func(), error)
// Evict evicts all cached processes from the cache.
Evict()
}
type cacheable interface {
isClosed() bool
isDirty() bool
close()
}
// ProcessCache entries always get added to the back of the list. If the
// list gets too long, we evict entries from the front of the list. When
// an entry gets added it gets an expiry time based on a fixed TTL. A
// monitor goroutine periodically evicts expired entries.
type ProcessCache struct {
// ttl is the fixed ttl for cache entries
ttl time.Duration
// monitorTicker is the ticker used for the monitoring Goroutine.
monitorTicker helper.Ticker
monitorDone chan interface{}
objectReaders processes
objectContentReaders processes
objectInfoReaders processes
catfileCacheCounter *prometheus.CounterVec
currentCatfileProcesses prometheus.Gauge
totalCatfileProcesses prometheus.Counter
catfileLookupCounter *prometheus.CounterVec
catfileCacheMembers *prometheus.GaugeVec
}
// NewCache creates a new catfile process cache.
func NewCache(cfg config.Cfg) *ProcessCache {
return newCache(defaultBatchfileTTL, cfg.Git.CatfileCacheSize, helper.NewTimerTicker(defaultEvictionInterval))
}
func newCache(ttl time.Duration, maxLen int, monitorTicker helper.Ticker) *ProcessCache {
if maxLen <= 0 {
maxLen = defaultMaxLen
}
processCache := &ProcessCache{
ttl: ttl,
objectContentReaders: processes{
maxLen: maxLen,
},
objectInfoReaders: processes{
maxLen: maxLen,
},
objectReaders: processes{
maxLen: maxLen,
},
catfileCacheCounter: prometheus.NewCounterVec(
prometheus.CounterOpts{
Name: "gitaly_catfile_cache_total",
Help: "Counter of catfile cache hit/miss",
},
[]string{"type"},
),
currentCatfileProcesses: prometheus.NewGauge(
prometheus.GaugeOpts{
Name: "gitaly_catfile_processes",
Help: "Gauge of active catfile processes",
},
),
totalCatfileProcesses: prometheus.NewCounter(
prometheus.CounterOpts{
Name: "gitaly_catfile_processes_total",
Help: "Counter of catfile processes",
},
),
catfileLookupCounter: prometheus.NewCounterVec(
prometheus.CounterOpts{
Name: "gitaly_catfile_lookups_total",
Help: "Git catfile lookups by object type",
},
[]string{"type"},
),
catfileCacheMembers: prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Name: "gitaly_catfile_cache_members",
Help: "Gauge of catfile cache members by process type",
},
[]string{"type"},
),
monitorTicker: monitorTicker,
monitorDone: make(chan interface{}),
}
go processCache.monitor()
return processCache
}
// Describe describes all metrics exposed by ProcessCache.
func (c *ProcessCache) Describe(descs chan<- *prometheus.Desc) {
prometheus.DescribeByCollect(c, descs)
}
// Collect collects all metrics exposed by ProcessCache.
func (c *ProcessCache) Collect(metrics chan<- prometheus.Metric) {
c.catfileCacheCounter.Collect(metrics)
c.currentCatfileProcesses.Collect(metrics)
c.totalCatfileProcesses.Collect(metrics)
c.catfileLookupCounter.Collect(metrics)
c.catfileCacheMembers.Collect(metrics)
}
func (c *ProcessCache) monitor() {
c.monitorTicker.Reset()
for {
select {
case <-c.monitorTicker.C():
c.objectContentReaders.EnforceTTL(time.Now())
c.objectInfoReaders.EnforceTTL(time.Now())
c.objectReaders.EnforceTTL(time.Now())
c.monitorTicker.Reset()
case <-c.monitorDone:
close(c.monitorDone)
return
}
c.reportCacheMembers()
}
}
// Stop stops the monitoring Goroutine and evicts all cached processes. This must only be called
// once.
func (c *ProcessCache) Stop() {
c.monitorTicker.Stop()
c.monitorDone <- struct{}{}
<-c.monitorDone
c.Evict()
}
// ObjectReader creates a new ObjectReader process for the given repository.
func (c *ProcessCache) ObjectReader(ctx context.Context, repo git.RepositoryExecutor) (ObjectContentReader, func(), error) {
var cached cacheable
var err error
var cancel func()
version, err := repo.GitVersion(ctx)
if err != nil {
return nil, nil, fmt.Errorf("git version: %w", err)
}
if version.CatfileSupportsNulTerminatedOutput() {
cached, cancel, err = c.getOrCreateProcess(ctx, repo, &c.objectReaders, func(ctx context.Context) (cacheable, error) {
return newObjectReader(ctx, repo, c.catfileLookupCounter)
}, "catfile.ObjectReader")
} else {
cached, cancel, err = c.getOrCreateProcess(ctx, repo, &c.objectContentReaders, func(ctx context.Context) (cacheable, error) {
return newObjectContentReader(ctx, repo, c.catfileLookupCounter)
}, "catfile.ObjectContentReader")
}
if err != nil {
return nil, nil, err
}
objectReader, ok := cached.(ObjectContentReader)
if !ok {
return nil, nil, fmt.Errorf("expected object reader, got %T", cached)
}
return objectReader, cancel, nil
}
// ObjectInfoReader creates a new ObjectInfoReader process for the given repository.
func (c *ProcessCache) ObjectInfoReader(ctx context.Context, repo git.RepositoryExecutor) (ObjectInfoReader, func(), error) {
var cached cacheable
var err error
var cancel func()
version, err := repo.GitVersion(ctx)
if err != nil {
return nil, nil, fmt.Errorf("git version: %w", err)
}
if version.CatfileSupportsNulTerminatedOutput() {
cached, cancel, err = c.getOrCreateProcess(ctx, repo, &c.objectReaders, func(ctx context.Context) (cacheable, error) {
return newObjectReader(ctx, repo, c.catfileLookupCounter)
}, "catfile.ObjectReader")
} else {
cached, cancel, err = c.getOrCreateProcess(ctx, repo, &c.objectInfoReaders, func(ctx context.Context) (cacheable, error) {
return newObjectInfoReader(ctx, repo, c.catfileLookupCounter)
}, "catfile.ObjectInfoReader")
}
if err != nil {
return nil, nil, err
}
objectInfoReader, ok := cached.(ObjectInfoReader)
if !ok {
return nil, nil, fmt.Errorf("expected object info reader, got %T", cached)
}
return objectInfoReader, cancel, nil
}
func (c *ProcessCache) getOrCreateProcess(
ctx context.Context,
repo storage.Repository,
processes *processes,
create func(context.Context) (cacheable, error),
spanName string,
) (_ cacheable, _ func(), returnedErr error) {
defer c.reportCacheMembers()
span, ctx := tracing.StartSpanIfHasParent(ctx, spanName, nil)
defer span.Finish()
cacheKey, isCacheable := newCacheKey(metadata.GetValue(ctx, SessionIDField), repo)
if isCacheable {
// We only try to look up cached processes in case it is cacheable, which requires a
// session ID. This is mostly done such that git-cat-file(1) processes from one user
// cannot interfere with those from another user. The main intent is to disallow
// trivial denial of service attacks against other users in case it is possible to
// poison the cache with broken git-cat-file(1) processes.
if entry, ok := processes.Checkout(cacheKey); ok {
c.catfileCacheCounter.WithLabelValues("hit").Inc()
span.SetTag("hit", true)
return entry.value, func() {
c.returnToCache(processes, cacheKey, entry.value, entry.cancel)
}, nil
}
c.catfileCacheCounter.WithLabelValues("miss").Inc()
span.SetTag("hit", false)
// When cache misses, a new process is created. This process may be re-used later.
// In that case, the lifecycle of the process is stretched across multiple
// gorountines. We should not attribute the span of this shared process to the
// current trace.
ctx = tracing.DiscardSpanInContext(ctx)
// We have not found any cached process, so we need to create a new one. In this
// case, we need to detach the process from the current context such that it does
// not get killed when the parent context is cancelled.
//
// Note that we explicitly retain feature flags here, which means that cached
// processes may retain flags for some time which have been changed meanwhile. While
// not ideal, it feels better compared to just ignoring feature flags altogether.
// The latter would mean that we cannot use flags in the catfile code, but more
// importantly we also wouldn't be able to use feature-flagged Git version upgrades
// for catfile processes.
ctx = helper.SuppressCancellation(ctx)
// We have to decorrelate the process from the current context given that it
// may potentially be reused across different RPC calls.
ctx = correlation.ContextWithCorrelation(ctx, "")
}
// Create a new cancellable process context such that we can kill it on demand. If it's a
// cached process, then it will only be killed when the cache evicts the entry because we
// detached the background further up. If it's an uncached value, we either kill it manually
// or via the RPC context's cancellation function.
ctx, cancelProcessContext := context.WithCancel(ctx)
defer func() {
if returnedErr != nil {
cancelProcessContext()
}
}()
process, err := create(ctx)
if err != nil {
return nil, nil, err
}
defer func() {
// If we somehow fail after creating a new process, then we want to kill spawned
// processes right away.
if returnedErr != nil {
process.close()
}
}()
c.totalCatfileProcesses.Inc()
c.currentCatfileProcesses.Inc()
// Note that we must make sure that `cancel` and `closeProcess` are two different variables.
// Otherwise if we passed `cancel` to `returnToCache` and then set `cancel` itself to the
// function that calls it we would accidentally call ourselves and end up with a segfault.
closeProcess := func() {
cancelProcessContext()
process.close()
c.currentCatfileProcesses.Dec()
}
cancel := closeProcess
if isCacheable {
// If the process is cacheable, then we want to put the process into the cache when
// the current outer context is done.
cancel = func() {
c.returnToCache(processes, cacheKey, process, closeProcess)
}
}
return process, cancel, nil
}
func (c *ProcessCache) reportCacheMembers() {
c.catfileCacheMembers.WithLabelValues("object_content_reader").Set(float64(c.objectContentReaders.EntryCount()))
c.catfileCacheMembers.WithLabelValues("object_info_reader").Set(float64(c.objectInfoReaders.EntryCount()))
c.catfileCacheMembers.WithLabelValues("object_reader").Set(float64(c.objectReaders.EntryCount()))
}
// Evict evicts all cached processes from the cache.
func (c *ProcessCache) Evict() {
c.objectContentReaders.Evict()
c.objectInfoReaders.Evict()
c.objectReaders.Evict()
}
func (c *ProcessCache) returnToCache(p *processes, cacheKey key, value cacheable, cancel func()) {
defer func() {
c.reportCacheMembers()
}()
if value == nil || value.isClosed() {
cancel()
return
}
if value.isDirty() {
cancel()
c.catfileCacheCounter.WithLabelValues("dirty").Inc()
value.close()
return
}
if replaced := p.Add(cacheKey, value, time.Now().Add(c.ttl), cancel); replaced {
c.catfileCacheCounter.WithLabelValues("duplicate").Inc()
}
}
type key struct {
sessionID string
repoStorage string
repoRelPath string
repoObjDir string
repoAltDir string
}
func newCacheKey(sessionID string, repo storage.Repository) (key, bool) {
if sessionID == "" {
return key{}, false
}
return key{
sessionID: sessionID,
repoStorage: repo.GetStorageName(),
repoRelPath: repo.GetRelativePath(),
repoObjDir: repo.GetGitObjectDirectory(),
repoAltDir: strings.Join(repo.GetGitAlternateObjectDirectories(), ","),
}, true
}
type entry struct {
key
value cacheable
expiry time.Time
cancel func()
}
type processes struct {
maxLen int
entriesMutex sync.Mutex
entries []*entry
}
// Add adds a key, value pair to p. If there are too many keys in p
// already add will evict old keys until the length is OK again.
func (p *processes) Add(k key, value cacheable, expiry time.Time, cancel func()) bool {
p.entriesMutex.Lock()
defer p.entriesMutex.Unlock()
replacedExisting := false
if i, ok := p.lookup(k); ok {
p.delete(i, true)
replacedExisting = true
}
ent := &entry{
key: k,
value: value,
expiry: expiry,
cancel: cancel,
}
p.entries = append(p.entries, ent)
for len(p.entries) > p.maxLen {
p.evictHead()
}
return replacedExisting
}
// Checkout removes a value from p. After use the caller can re-add the value with p.Add.
func (p *processes) Checkout(k key) (*entry, bool) {
p.entriesMutex.Lock()
defer p.entriesMutex.Unlock()
i, ok := p.lookup(k)
if !ok {
return nil, false
}
entry := p.entries[i]
p.delete(i, false)
return entry, true
}
// EnforceTTL evicts all entries older than now, assuming the entry
// expiry times are increasing.
func (p *processes) EnforceTTL(now time.Time) {
p.entriesMutex.Lock()
defer p.entriesMutex.Unlock()
for len(p.entries) > 0 && now.After(p.head().expiry) {
p.evictHead()
}
}
// Evict evicts all cached processes from the cache.
func (p *processes) Evict() {
p.entriesMutex.Lock()
defer p.entriesMutex.Unlock()
for len(p.entries) > 0 {
p.evictHead()
}
}
// EntryCount returns the number of cached entries. This function will locks the ProcessCache to
// avoid races.
func (p *processes) EntryCount() int {
p.entriesMutex.Lock()
defer p.entriesMutex.Unlock()
return len(p.entries)
}
func (p *processes) head() *entry { return p.entries[0] }
func (p *processes) evictHead() { p.delete(0, true) }
func (p *processes) lookup(k key) (int, bool) {
for i, ent := range p.entries {
if ent.key == k {
return i, true
}
}
return -1, false
}
func (p *processes) delete(i int, wantClose bool) {
ent := p.entries[i]
if wantClose {
// We first cancel the context such that the process gets a SIGKILL signal. Calling
// `close()` first may lead to a deadlock given that it waits for the process to
// exit, which may not happen if it hangs writing data to stdout.
ent.cancel()
ent.value.close()
}
p.entries = append(p.entries[:i], p.entries[i+1:]...)
}
|