services/keep-web/cache.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package keepweb
   6
   7 import (
   8         "sync"
   9         "sync/atomic"
  10         "time"
  11
  12         "git.arvados.org/arvados.git/sdk/go/arvados"
  13         "git.arvados.org/arvados.git/sdk/go/arvadosclient"
  14         "git.arvados.org/arvados.git/sdk/go/keepclient"
  15         lru "github.com/hashicorp/golang-lru"
  16         "github.com/prometheus/client_golang/prometheus"
  17         "github.com/sirupsen/logrus"
  18 )
  19
  20 const metricsUpdateInterval = time.Second / 10
  21
  22 type cache struct {
  23         cluster     *arvados.Cluster
  24         logger      logrus.FieldLogger
  25         registry    *prometheus.Registry
  26         metrics     cacheMetrics
  27         pdhs        *lru.TwoQueueCache
  28         collections *lru.TwoQueueCache
  29         sessions    *lru.TwoQueueCache
  30         setupOnce   sync.Once
  31
  32         chPruneSessions    chan struct{}
  33         chPruneCollections chan struct{}
  34 }
  35
  36 type cacheMetrics struct {
  37         requests          prometheus.Counter
  38         collectionBytes   prometheus.Gauge
  39         collectionEntries prometheus.Gauge
  40         sessionEntries    prometheus.Gauge
  41         collectionHits    prometheus.Counter
  42         pdhHits           prometheus.Counter
  43         sessionHits       prometheus.Counter
  44         sessionMisses     prometheus.Counter
  45         apiCalls          prometheus.Counter
  46 }
  47
  48 func (m *cacheMetrics) setup(reg *prometheus.Registry) {
  49         m.requests = prometheus.NewCounter(prometheus.CounterOpts{
  50                 Namespace: "arvados",
  51                 Subsystem: "keepweb_collectioncache",
  52                 Name:      "requests",
  53                 Help:      "Number of targetID-to-manifest lookups handled.",
  54         })
  55         reg.MustRegister(m.requests)
  56         m.collectionHits = prometheus.NewCounter(prometheus.CounterOpts{
  57                 Namespace: "arvados",
  58                 Subsystem: "keepweb_collectioncache",
  59                 Name:      "hits",
  60                 Help:      "Number of pdh-to-manifest cache hits.",
  61         })
  62         reg.MustRegister(m.collectionHits)
  63         m.pdhHits = prometheus.NewCounter(prometheus.CounterOpts{
  64                 Namespace: "arvados",
  65                 Subsystem: "keepweb_collectioncache",
  66                 Name:      "pdh_hits",
  67                 Help:      "Number of uuid-to-pdh cache hits.",
  68         })
  69         reg.MustRegister(m.pdhHits)
  70         m.apiCalls = prometheus.NewCounter(prometheus.CounterOpts{
  71                 Namespace: "arvados",
  72                 Subsystem: "keepweb_collectioncache",
  73                 Name:      "api_calls",
  74                 Help:      "Number of outgoing API calls made by cache.",
  75         })
  76         reg.MustRegister(m.apiCalls)
  77         m.collectionBytes = prometheus.NewGauge(prometheus.GaugeOpts{
  78                 Namespace: "arvados",
  79                 Subsystem: "keepweb_sessions",
  80                 Name:      "cached_collection_bytes",
  81                 Help:      "Total size of all cached manifests and sessions.",
  82         })
  83         reg.MustRegister(m.collectionBytes)
  84         m.collectionEntries = prometheus.NewGauge(prometheus.GaugeOpts{
  85                 Namespace: "arvados",
  86                 Subsystem: "keepweb_collectioncache",
  87                 Name:      "cached_manifests",
  88                 Help:      "Number of manifests in cache.",
  89         })
  90         reg.MustRegister(m.collectionEntries)
  91         m.sessionEntries = prometheus.NewGauge(prometheus.GaugeOpts{
  92                 Namespace: "arvados",
  93                 Subsystem: "keepweb_sessions",
  94                 Name:      "active",
  95                 Help:      "Number of active token sessions.",
  96         })
  97         reg.MustRegister(m.sessionEntries)
  98         m.sessionHits = prometheus.NewCounter(prometheus.CounterOpts{
  99                 Namespace: "arvados",
 100                 Subsystem: "keepweb_sessions",
 101                 Name:      "hits",
 102                 Help:      "Number of token session cache hits.",
 103         })
 104         reg.MustRegister(m.sessionHits)
 105         m.sessionMisses = prometheus.NewCounter(prometheus.CounterOpts{
 106                 Namespace: "arvados",
 107                 Subsystem: "keepweb_sessions",
 108                 Name:      "misses",
 109                 Help:      "Number of token session cache misses.",
 110         })
 111         reg.MustRegister(m.sessionMisses)
 112 }
 113
 114 type cachedPDH struct {
 115         expire  time.Time
 116         refresh time.Time
 117         pdh     string
 118 }
 119
 120 type cachedCollection struct {
 121         expire     time.Time
 122         collection *arvados.Collection
 123 }
 124
 125 type cachedPermission struct {
 126         expire time.Time
 127 }
 128
 129 type cachedSession struct {
 130         expire        time.Time
 131         fs            atomic.Value
 132         client        *arvados.Client
 133         arvadosclient *arvadosclient.ArvadosClient
 134         keepclient    *keepclient.KeepClient
 135         user          atomic.Value
 136 }
 137
 138 func (c *cache) setup() {
 139         var err error
 140         c.pdhs, err = lru.New2Q(c.cluster.Collections.WebDAVCache.MaxUUIDEntries)
 141         if err != nil {
 142                 panic(err)
 143         }
 144         c.collections, err = lru.New2Q(c.cluster.Collections.WebDAVCache.MaxCollectionEntries)
 145         if err != nil {
 146                 panic(err)
 147         }
 148         c.sessions, err = lru.New2Q(c.cluster.Collections.WebDAVCache.MaxSessions)
 149         if err != nil {
 150                 panic(err)
 151         }
 152
 153         reg := c.registry
 154         if reg == nil {
 155                 reg = prometheus.NewRegistry()
 156         }
 157         c.metrics.setup(reg)
 158         go func() {
 159                 for range time.Tick(metricsUpdateInterval) {
 160                         c.updateGauges()
 161                 }
 162         }()
 163         c.chPruneCollections = make(chan struct{}, 1)
 164         go func() {
 165                 for range c.chPruneCollections {
 166                         c.pruneCollections()
 167                 }
 168         }()
 169         c.chPruneSessions = make(chan struct{}, 1)
 170         go func() {
 171                 for range c.chPruneSessions {
 172                         c.pruneSessions()
 173                 }
 174         }()
 175 }
 176
 177 func (c *cache) updateGauges() {
 178         c.metrics.collectionBytes.Set(float64(c.collectionBytes()))
 179         c.metrics.collectionEntries.Set(float64(c.collections.Len()))
 180         c.metrics.sessionEntries.Set(float64(c.sessions.Len()))
 181 }
 182
 183 var selectPDH = map[string]interface{}{
 184         "select": []string{"portable_data_hash"},
 185 }
 186
 187 // Update saves a modified version (fs) to an existing collection
 188 // (coll) and, if successful, updates the relevant cache entries so
 189 // subsequent calls to Get() reflect the modifications.
 190 func (c *cache) Update(client *arvados.Client, coll arvados.Collection, fs arvados.CollectionFileSystem) error {
 191         c.setupOnce.Do(c.setup)
 192
 193         m, err := fs.MarshalManifest(".")
 194         if err != nil || m == coll.ManifestText {
 195                 return err
 196         }
 197         coll.ManifestText = m
 198         var updated arvados.Collection
 199         err = client.RequestAndDecode(&updated, "PATCH", "arvados/v1/collections/"+coll.UUID, nil, map[string]interface{}{
 200                 "collection": map[string]string{
 201                         "manifest_text": coll.ManifestText,
 202                 },
 203         })
 204         if err != nil {
 205                 c.pdhs.Remove(coll.UUID)
 206                 return err
 207         }
 208         c.collections.Add(client.AuthToken+"\000"+updated.PortableDataHash, &cachedCollection{
 209                 expire:     time.Now().Add(time.Duration(c.cluster.Collections.WebDAVCache.TTL)),
 210                 collection: &updated,
 211         })
 212         c.pdhs.Add(coll.UUID, &cachedPDH{
 213                 expire:  time.Now().Add(time.Duration(c.cluster.Collections.WebDAVCache.TTL)),
 214                 refresh: time.Now().Add(time.Duration(c.cluster.Collections.WebDAVCache.UUIDTTL)),
 215                 pdh:     updated.PortableDataHash,
 216         })
 217         return nil
 218 }
 219
 220 // ResetSession unloads any potentially stale state. Should be called
 221 // after write operations, so subsequent reads don't return stale
 222 // data.
 223 func (c *cache) ResetSession(token string) {
 224         c.setupOnce.Do(c.setup)
 225         c.sessions.Remove(token)
 226 }
 227
 228 // Get a long-lived CustomFileSystem suitable for doing a read operation
 229 // with the given token.
 230 func (c *cache) GetSession(token string) (arvados.CustomFileSystem, *cachedSession, error) {
 231         c.setupOnce.Do(c.setup)
 232         now := time.Now()
 233         ent, _ := c.sessions.Get(token)
 234         sess, _ := ent.(*cachedSession)
 235         expired := false
 236         if sess == nil {
 237                 c.metrics.sessionMisses.Inc()
 238                 sess = &cachedSession{
 239                         expire: now.Add(c.cluster.Collections.WebDAVCache.TTL.Duration()),
 240                 }
 241                 var err error
 242                 sess.client, err = arvados.NewClientFromConfig(c.cluster)
 243                 if err != nil {
 244                         return nil, nil, err
 245                 }
 246                 sess.client.AuthToken = token
 247                 sess.arvadosclient, err = arvadosclient.New(sess.client)
 248                 if err != nil {
 249                         return nil, nil, err
 250                 }
 251                 sess.keepclient = keepclient.New(sess.arvadosclient)
 252                 c.sessions.Add(token, sess)
 253         } else if sess.expire.Before(now) {
 254                 c.metrics.sessionMisses.Inc()
 255                 expired = true
 256         } else {
 257                 c.metrics.sessionHits.Inc()
 258         }
 259         select {
 260         case c.chPruneSessions <- struct{}{}:
 261         default:
 262         }
 263         fs, _ := sess.fs.Load().(arvados.CustomFileSystem)
 264         if fs != nil && !expired {
 265                 return fs, sess, nil
 266         }
 267         fs = sess.client.SiteFileSystem(sess.keepclient)
 268         fs.ForwardSlashNameSubstitution(c.cluster.Collections.ForwardSlashNameSubstitution)
 269         sess.fs.Store(fs)
 270         return fs, sess, nil
 271 }
 272
 273 // Remove all expired session cache entries, then remove more entries
 274 // until approximate remaining size <= maxsize/2
 275 func (c *cache) pruneSessions() {
 276         now := time.Now()
 277         var size int64
 278         keys := c.sessions.Keys()
 279         for _, token := range keys {
 280                 ent, ok := c.sessions.Peek(token)
 281                 if !ok {
 282                         continue
 283                 }
 284                 s := ent.(*cachedSession)
 285                 if s.expire.Before(now) {
 286                         c.sessions.Remove(token)
 287                         continue
 288                 }
 289                 if fs, ok := s.fs.Load().(arvados.CustomFileSystem); ok {
 290                         size += fs.MemorySize()
 291                 }
 292         }
 293         // Remove tokens until reaching size limit, starting with the
 294         // least frequently used entries (which Keys() returns last).
 295         for i := len(keys) - 1; i >= 0; i-- {
 296                 token := keys[i]
 297                 if size <= c.cluster.Collections.WebDAVCache.MaxCollectionBytes/2 {
 298                         break
 299                 }
 300                 ent, ok := c.sessions.Peek(token)
 301                 if !ok {
 302                         continue
 303                 }
 304                 s := ent.(*cachedSession)
 305                 fs, _ := s.fs.Load().(arvados.CustomFileSystem)
 306                 if fs == nil {
 307                         continue
 308                 }
 309                 c.sessions.Remove(token)
 310                 size -= fs.MemorySize()
 311         }
 312 }
 313
 314 func (c *cache) Get(arv *arvadosclient.ArvadosClient, targetID string, forceReload bool) (*arvados.Collection, error) {
 315         c.setupOnce.Do(c.setup)
 316         c.metrics.requests.Inc()
 317
 318         var pdhRefresh bool
 319         var pdh string
 320         if arvadosclient.PDHMatch(targetID) {
 321                 pdh = targetID
 322         } else if ent, cached := c.pdhs.Get(targetID); cached {
 323                 ent := ent.(*cachedPDH)
 324                 if ent.expire.Before(time.Now()) {
 325                         c.pdhs.Remove(targetID)
 326                 } else {
 327                         pdh = ent.pdh
 328                         pdhRefresh = forceReload || time.Now().After(ent.refresh)
 329                         c.metrics.pdhHits.Inc()
 330                 }
 331         }
 332
 333         if pdh == "" {
 334                 // UUID->PDH mapping is not cached, might as well get
 335                 // the whole collection record and be done (below).
 336                 c.logger.Debugf("cache(%s): have no pdh", targetID)
 337         } else if cached := c.lookupCollection(arv.ApiToken + "\000" + pdh); cached == nil {
 338                 // PDH->manifest is not cached, might as well get the
 339                 // whole collection record (below).
 340                 c.logger.Debugf("cache(%s): have pdh %s but manifest is not cached", targetID, pdh)
 341         } else if !pdhRefresh {
 342                 // We looked up UUID->PDH very recently, and we still
 343                 // have the manifest for that PDH.
 344                 c.logger.Debugf("cache(%s): have pdh %s and refresh not needed", targetID, pdh)
 345                 return cached, nil
 346         } else {
 347                 // Get current PDH for this UUID (and confirm we still
 348                 // have read permission).  Most likely, the cached PDH
 349                 // is still correct, in which case we can use our
 350                 // cached manifest.
 351                 c.metrics.apiCalls.Inc()
 352                 var current arvados.Collection
 353                 err := arv.Get("collections", targetID, selectPDH, &current)
 354                 if err != nil {
 355                         return nil, err
 356                 }
 357                 if current.PortableDataHash == pdh {
 358                         // PDH has not changed, cached manifest is
 359                         // correct.
 360                         c.logger.Debugf("cache(%s): verified cached pdh %s is still correct", targetID, pdh)
 361                         return cached, nil
 362                 }
 363                 if cached := c.lookupCollection(arv.ApiToken + "\000" + current.PortableDataHash); cached != nil {
 364                         // PDH changed, and we already have the
 365                         // manifest for that new PDH.
 366                         c.logger.Debugf("cache(%s): cached pdh %s was stale, new pdh is %s and manifest is already in cache", targetID, pdh, current.PortableDataHash)
 367                         return cached, nil
 368                 }
 369         }
 370
 371         // Either UUID->PDH is not cached, or PDH->manifest is not
 372         // cached.
 373         var retrieved arvados.Collection
 374         c.metrics.apiCalls.Inc()
 375         err := arv.Get("collections", targetID, nil, &retrieved)
 376         if err != nil {
 377                 return nil, err
 378         }
 379         c.logger.Debugf("cache(%s): retrieved manifest, caching with pdh %s", targetID, retrieved.PortableDataHash)
 380         exp := time.Now().Add(time.Duration(c.cluster.Collections.WebDAVCache.TTL))
 381         if targetID != retrieved.PortableDataHash {
 382                 c.pdhs.Add(targetID, &cachedPDH{
 383                         expire:  exp,
 384                         refresh: time.Now().Add(time.Duration(c.cluster.Collections.WebDAVCache.UUIDTTL)),
 385                         pdh:     retrieved.PortableDataHash,
 386                 })
 387         }
 388         c.collections.Add(arv.ApiToken+"\000"+retrieved.PortableDataHash, &cachedCollection{
 389                 expire:     exp,
 390                 collection: &retrieved,
 391         })
 392         if int64(len(retrieved.ManifestText)) > c.cluster.Collections.WebDAVCache.MaxCollectionBytes/int64(c.cluster.Collections.WebDAVCache.MaxCollectionEntries) {
 393                 select {
 394                 case c.chPruneCollections <- struct{}{}:
 395                 default:
 396                 }
 397         }
 398         return &retrieved, nil
 399 }
 400
 401 // pruneCollections checks the total bytes occupied by manifest_text
 402 // in the collection cache and removes old entries as needed to bring
 403 // the total size down to CollectionBytes. It also deletes all expired
 404 // entries.
 405 //
 406 // pruneCollections does not aim to be perfectly correct when there is
 407 // concurrent cache activity.
 408 func (c *cache) pruneCollections() {
 409         var size int64
 410         now := time.Now()
 411         keys := c.collections.Keys()
 412         entsize := make([]int, len(keys))
 413         expired := make([]bool, len(keys))
 414         for i, k := range keys {
 415                 v, ok := c.collections.Peek(k)
 416                 if !ok {
 417                         continue
 418                 }
 419                 ent := v.(*cachedCollection)
 420                 n := len(ent.collection.ManifestText)
 421                 size += int64(n)
 422                 entsize[i] = n
 423                 expired[i] = ent.expire.Before(now)
 424         }
 425         for i, k := range keys {
 426                 if expired[i] {
 427                         c.collections.Remove(k)
 428                         size -= int64(entsize[i])
 429                 }
 430         }
 431         for i, k := range keys {
 432                 if size <= c.cluster.Collections.WebDAVCache.MaxCollectionBytes/2 {
 433                         break
 434                 }
 435                 if expired[i] {
 436                         // already removed this entry in the previous loop
 437                         continue
 438                 }
 439                 c.collections.Remove(k)
 440                 size -= int64(entsize[i])
 441         }
 442 }
 443
 444 // collectionBytes returns the approximate combined memory size of the
 445 // collection cache and session filesystem cache.
 446 func (c *cache) collectionBytes() uint64 {
 447         var size uint64
 448         for _, k := range c.collections.Keys() {
 449                 v, ok := c.collections.Peek(k)
 450                 if !ok {
 451                         continue
 452                 }
 453                 size += uint64(len(v.(*cachedCollection).collection.ManifestText))
 454         }
 455         for _, token := range c.sessions.Keys() {
 456                 ent, ok := c.sessions.Peek(token)
 457                 if !ok {
 458                         continue
 459                 }
 460                 if fs, ok := ent.(*cachedSession).fs.Load().(arvados.CustomFileSystem); ok {
 461                         size += uint64(fs.MemorySize())
 462                 }
 463         }
 464         return size
 465 }
 466
 467 func (c *cache) lookupCollection(key string) *arvados.Collection {
 468         e, cached := c.collections.Get(key)
 469         if !cached {
 470                 return nil
 471         }
 472         ent := e.(*cachedCollection)
 473         if ent.expire.Before(time.Now()) {
 474                 c.collections.Remove(key)
 475                 return nil
 476         }
 477         c.metrics.collectionHits.Inc()
 478         return ent.collection
 479 }
 480
 481 func (c *cache) GetTokenUser(token string) (*arvados.User, error) {
 482         // Get and cache user record associated with this
 483         // token.  We need to know their UUID for logging, and
 484         // whether they are an admin or not for certain
 485         // permission checks.
 486
 487         // Get/create session entry
 488         _, sess, err := c.GetSession(token)
 489         if err != nil {
 490                 return nil, err
 491         }
 492
 493         // See if the user is already set, and if so, return it
 494         user, _ := sess.user.Load().(*arvados.User)
 495         if user != nil {
 496                 return user, nil
 497         }
 498
 499         // Fetch the user record
 500         c.metrics.apiCalls.Inc()
 501         var current arvados.User
 502
 503         err = sess.client.RequestAndDecode(&current, "GET", "/arvados/v1/users/current", nil, nil)
 504         if err != nil {
 505                 return nil, err
 506         }
 507
 508         // Stash the user record for next time
 509         sess.user.Store(&current)
 510         return &current, nil
 511 }