16306: Merge branch 'master'
[arvados.git] / services / keep-web / cache.go
index 174e370c13dd115859053120a7c40f1f68094f07..eeb78ad9058d6c35e8b544cbef1a5c6500c90bcf 100644 (file)
@@ -6,36 +6,86 @@ package main
 
 import (
        "sync"
-       "sync/atomic"
        "time"
 
-       "git.curoverse.com/arvados.git/sdk/go/arvados"
-       "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
+       "git.arvados.org/arvados.git/sdk/go/arvados"
+       "git.arvados.org/arvados.git/sdk/go/arvadosclient"
        "github.com/hashicorp/golang-lru"
+       "github.com/prometheus/client_golang/prometheus"
 )
 
-type cache struct {
-       TTL                  arvados.Duration
-       MaxCollectionEntries int
-       MaxCollectionBytes   int64
-       MaxPermissionEntries int
-       MaxUUIDEntries       int
+const metricsUpdateInterval = time.Second / 10
 
-       stats       cacheStats
+type cache struct {
+       config      *arvados.WebDAVCacheConfig
+       registry    *prometheus.Registry
+       metrics     cacheMetrics
        pdhs        *lru.TwoQueueCache
        collections *lru.TwoQueueCache
        permissions *lru.TwoQueueCache
        setupOnce   sync.Once
 }
 
-type cacheStats struct {
-       Requests          uint64 `json:"Cache.Requests"`
-       CollectionBytes   uint64 `json:"Cache.CollectionBytes"`
-       CollectionEntries int    `json:"Cache.CollectionEntries"`
-       CollectionHits    uint64 `json:"Cache.CollectionHits"`
-       PDHHits           uint64 `json:"Cache.UUIDHits"`
-       PermissionHits    uint64 `json:"Cache.PermissionHits"`
-       APICalls          uint64 `json:"Cache.APICalls"`
+type cacheMetrics struct {
+       requests          prometheus.Counter
+       collectionBytes   prometheus.Gauge
+       collectionEntries prometheus.Gauge
+       collectionHits    prometheus.Counter
+       pdhHits           prometheus.Counter
+       permissionHits    prometheus.Counter
+       apiCalls          prometheus.Counter
+}
+
+func (m *cacheMetrics) setup(reg *prometheus.Registry) {
+       m.requests = prometheus.NewCounter(prometheus.CounterOpts{
+               Namespace: "arvados",
+               Subsystem: "keepweb_collectioncache",
+               Name:      "requests",
+               Help:      "Number of targetID-to-manifest lookups handled.",
+       })
+       reg.MustRegister(m.requests)
+       m.collectionHits = prometheus.NewCounter(prometheus.CounterOpts{
+               Namespace: "arvados",
+               Subsystem: "keepweb_collectioncache",
+               Name:      "hits",
+               Help:      "Number of pdh-to-manifest cache hits.",
+       })
+       reg.MustRegister(m.collectionHits)
+       m.pdhHits = prometheus.NewCounter(prometheus.CounterOpts{
+               Namespace: "arvados",
+               Subsystem: "keepweb_collectioncache",
+               Name:      "pdh_hits",
+               Help:      "Number of uuid-to-pdh cache hits.",
+       })
+       reg.MustRegister(m.pdhHits)
+       m.permissionHits = prometheus.NewCounter(prometheus.CounterOpts{
+               Namespace: "arvados",
+               Subsystem: "keepweb_collectioncache",
+               Name:      "permission_hits",
+               Help:      "Number of targetID-to-permission cache hits.",
+       })
+       reg.MustRegister(m.permissionHits)
+       m.apiCalls = prometheus.NewCounter(prometheus.CounterOpts{
+               Namespace: "arvados",
+               Subsystem: "keepweb_collectioncache",
+               Name:      "api_calls",
+               Help:      "Number of outgoing API calls made by cache.",
+       })
+       reg.MustRegister(m.apiCalls)
+       m.collectionBytes = prometheus.NewGauge(prometheus.GaugeOpts{
+               Namespace: "arvados",
+               Subsystem: "keepweb_collectioncache",
+               Name:      "cached_manifest_bytes",
+               Help:      "Total size of all manifests in cache.",
+       })
+       reg.MustRegister(m.collectionBytes)
+       m.collectionEntries = prometheus.NewGauge(prometheus.GaugeOpts{
+               Namespace: "arvados",
+               Subsystem: "keepweb_collectioncache",
+               Name:      "cached_manifests",
+               Help:      "Number of manifests in cache.",
+       })
+       reg.MustRegister(m.collectionEntries)
 }
 
 type cachedPDH struct {
@@ -54,41 +104,70 @@ type cachedPermission struct {
 
 func (c *cache) setup() {
        var err error
-       c.pdhs, err = lru.New2Q(c.MaxUUIDEntries)
+       c.pdhs, err = lru.New2Q(c.config.MaxUUIDEntries)
        if err != nil {
                panic(err)
        }
-       c.collections, err = lru.New2Q(c.MaxCollectionEntries)
+       c.collections, err = lru.New2Q(c.config.MaxCollectionEntries)
        if err != nil {
                panic(err)
        }
-       c.permissions, err = lru.New2Q(c.MaxPermissionEntries)
+       c.permissions, err = lru.New2Q(c.config.MaxPermissionEntries)
        if err != nil {
                panic(err)
        }
+
+       reg := c.registry
+       if reg == nil {
+               reg = prometheus.NewRegistry()
+       }
+       c.metrics.setup(reg)
+       go func() {
+               for range time.Tick(metricsUpdateInterval) {
+                       c.updateGauges()
+               }
+       }()
+}
+
+func (c *cache) updateGauges() {
+       c.metrics.collectionBytes.Set(float64(c.collectionBytes()))
+       c.metrics.collectionEntries.Set(float64(c.collections.Len()))
 }
 
 var selectPDH = map[string]interface{}{
        "select": []string{"portable_data_hash"},
 }
 
-func (c *cache) Stats() cacheStats {
+// Update saves a modified version (fs) to an existing collection
+// (coll) and, if successful, updates the relevant cache entries so
+// subsequent calls to Get() reflect the modifications.
+func (c *cache) Update(client *arvados.Client, coll arvados.Collection, fs arvados.CollectionFileSystem) error {
        c.setupOnce.Do(c.setup)
-       return cacheStats{
-               Requests:          atomic.LoadUint64(&c.stats.Requests),
-               CollectionBytes:   c.collectionBytes(),
-               CollectionEntries: c.collections.Len(),
-               CollectionHits:    atomic.LoadUint64(&c.stats.CollectionHits),
-               PDHHits:           atomic.LoadUint64(&c.stats.PDHHits),
-               PermissionHits:    atomic.LoadUint64(&c.stats.PermissionHits),
-               APICalls:          atomic.LoadUint64(&c.stats.APICalls),
+
+       m, err := fs.MarshalManifest(".")
+       if err != nil || m == coll.ManifestText {
+               return err
        }
+       coll.ManifestText = m
+       var updated arvados.Collection
+       defer c.pdhs.Remove(coll.UUID)
+       err = client.RequestAndDecode(&updated, "PATCH", "arvados/v1/collections/"+coll.UUID, nil, map[string]interface{}{
+               "collection": map[string]string{
+                       "manifest_text": coll.ManifestText,
+               },
+       })
+       if err == nil {
+               c.collections.Add(client.AuthToken+"\000"+coll.PortableDataHash, &cachedCollection{
+                       expire:     time.Now().Add(time.Duration(c.config.TTL)),
+                       collection: &updated,
+               })
+       }
+       return err
 }
 
 func (c *cache) Get(arv *arvadosclient.ArvadosClient, targetID string, forceReload bool) (*arvados.Collection, error) {
        c.setupOnce.Do(c.setup)
-
-       atomic.AddUint64(&c.stats.Requests, 1)
+       c.metrics.requests.Inc()
 
        permOK := false
        permKey := arv.ApiToken + "\000" + targetID
@@ -99,7 +178,7 @@ func (c *cache) Get(arv *arvadosclient.ArvadosClient, targetID string, forceRelo
                        c.permissions.Remove(permKey)
                } else {
                        permOK = true
-                       atomic.AddUint64(&c.stats.PermissionHits, 1)
+                       c.metrics.permissionHits.Inc()
                }
        }
 
@@ -112,13 +191,13 @@ func (c *cache) Get(arv *arvadosclient.ArvadosClient, targetID string, forceRelo
                        c.pdhs.Remove(targetID)
                } else {
                        pdh = ent.pdh
-                       atomic.AddUint64(&c.stats.PDHHits, 1)
+                       c.metrics.pdhHits.Inc()
                }
        }
 
        var collection *arvados.Collection
        if pdh != "" {
-               collection = c.lookupCollection(pdh)
+               collection = c.lookupCollection(arv.ApiToken + "\000" + pdh)
        }
 
        if collection != nil && permOK {
@@ -128,53 +207,51 @@ func (c *cache) Get(arv *arvadosclient.ArvadosClient, targetID string, forceRelo
                // likely, the cached PDH is still correct; if so,
                // _and_ the current token has permission, we can
                // use our cached manifest.
-               atomic.AddUint64(&c.stats.APICalls, 1)
+               c.metrics.apiCalls.Inc()
                var current arvados.Collection
                err := arv.Get("collections", targetID, selectPDH, &current)
                if err != nil {
                        return nil, err
                }
                if current.PortableDataHash == pdh {
-                       exp := time.Now().Add(time.Duration(c.TTL))
                        c.permissions.Add(permKey, &cachedPermission{
-                               expire: exp,
+                               expire: time.Now().Add(time.Duration(c.config.TTL)),
                        })
                        if pdh != targetID {
                                c.pdhs.Add(targetID, &cachedPDH{
-                                       expire: exp,
+                                       expire: time.Now().Add(time.Duration(c.config.UUIDTTL)),
                                        pdh:    pdh,
                                })
                        }
                        return collection, err
-               } else {
-                       // PDH changed, but now we know we have
-                       // permission -- and maybe we already have the
-                       // new PDH in the cache.
-                       if coll := c.lookupCollection(current.PortableDataHash); coll != nil {
-                               return coll, nil
-                       }
+               }
+               // PDH changed, but now we know we have
+               // permission -- and maybe we already have the
+               // new PDH in the cache.
+               if coll := c.lookupCollection(arv.ApiToken + "\000" + current.PortableDataHash); coll != nil {
+                       return coll, nil
                }
        }
 
        // Collection manifest is not cached.
-       atomic.AddUint64(&c.stats.APICalls, 1)
+       c.metrics.apiCalls.Inc()
        err := arv.Get("collections", targetID, nil, &collection)
        if err != nil {
                return nil, err
        }
-       exp := time.Now().Add(time.Duration(c.TTL))
+       exp := time.Now().Add(time.Duration(c.config.TTL))
        c.permissions.Add(permKey, &cachedPermission{
                expire: exp,
        })
        c.pdhs.Add(targetID, &cachedPDH{
-               expire: exp,
+               expire: time.Now().Add(time.Duration(c.config.UUIDTTL)),
                pdh:    collection.PortableDataHash,
        })
-       c.collections.Add(collection.PortableDataHash, &cachedCollection{
+       c.collections.Add(arv.ApiToken+"\000"+collection.PortableDataHash, &cachedCollection{
                expire:     exp,
                collection: collection,
        })
-       if int64(len(collection.ManifestText)) > c.MaxCollectionBytes/int64(c.MaxCollectionEntries) {
+       if int64(len(collection.ManifestText)) > c.config.MaxCollectionBytes/int64(c.config.MaxCollectionEntries) {
                go c.pruneCollections()
        }
        return collection, nil
@@ -211,7 +288,7 @@ func (c *cache) pruneCollections() {
                }
        }
        for i, k := range keys {
-               if size <= c.MaxCollectionBytes {
+               if size <= c.config.MaxCollectionBytes {
                        break
                }
                if expired[i] {
@@ -237,19 +314,16 @@ func (c *cache) collectionBytes() uint64 {
        return size
 }
 
-func (c *cache) lookupCollection(pdh string) *arvados.Collection {
-       if pdh == "" {
+func (c *cache) lookupCollection(key string) *arvados.Collection {
+       e, cached := c.collections.Get(key)
+       if !cached {
                return nil
-       } else if ent, cached := c.collections.Get(pdh); !cached {
+       }
+       ent := e.(*cachedCollection)
+       if ent.expire.Before(time.Now()) {
+               c.collections.Remove(key)
                return nil
-       } else {
-               ent := ent.(*cachedCollection)
-               if ent.expire.Before(time.Now()) {
-                       c.collections.Remove(pdh)
-                       return nil
-               } else {
-                       atomic.AddUint64(&c.stats.CollectionHits, 1)
-                       return ent.collection
-               }
        }
+       c.metrics.collectionHits.Inc()
+       return ent.collection
 }