X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/843b4e5cf7727528fb5b45629030fd0d6a364ab7..3310da050bd763055a67b2395f122ab5c01cbdf9:/services/keep-web/cache.go diff --git a/services/keep-web/cache.go b/services/keep-web/cache.go index 59e8de3bc9..b9a1f3069f 100644 --- a/services/keep-web/cache.go +++ b/services/keep-web/cache.go @@ -6,14 +6,16 @@ package main import ( "sync" - "sync/atomic" "time" "git.curoverse.com/arvados.git/sdk/go/arvados" "git.curoverse.com/arvados.git/sdk/go/arvadosclient" "github.com/hashicorp/golang-lru" + "github.com/prometheus/client_golang/prometheus" ) +const metricsUpdateInterval = time.Second / 10 + type cache struct { TTL arvados.Duration UUIDTTL arvados.Duration @@ -22,21 +24,74 @@ type cache struct { MaxPermissionEntries int MaxUUIDEntries int - stats cacheStats + 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 { @@ -67,23 +122,26 @@ func (c *cache) setup() { 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() + } + }() } -var selectPDH = map[string]interface{}{ - "select": []string{"portable_data_hash"}, +func (c *cache) updateGauges() { + c.metrics.collectionBytes.Set(float64(c.collectionBytes())) + c.metrics.collectionEntries.Set(float64(c.collections.Len())) } -func (c *cache) Stats() cacheStats { - 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), - } +var selectPDH = map[string]interface{}{ + "select": []string{"portable_data_hash"}, } // Update saves a modified version (fs) to an existing collection @@ -99,7 +157,11 @@ func (c *cache) Update(client *arvados.Client, coll arvados.Collection, fs arvad } var updated arvados.Collection defer c.pdhs.Remove(coll.UUID) - err := client.RequestAndDecode(&updated, "PATCH", "arvados/v1/collections/"+coll.UUID, client.UpdateBody(coll), nil) + 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.TTL)), @@ -111,8 +173,7 @@ func (c *cache) Update(client *arvados.Client, coll arvados.Collection, fs arvad 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 @@ -123,7 +184,7 @@ 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() } } @@ -136,7 +197,7 @@ 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() } } @@ -152,7 +213,7 @@ 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, ¤t) if err != nil { @@ -180,7 +241,7 @@ func (c *cache) Get(arv *arvadosclient.ArvadosClient, targetID string, forceRelo } // 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 @@ -261,16 +322,15 @@ func (c *cache) collectionBytes() uint64 { } func (c *cache) lookupCollection(key string) *arvados.Collection { - if ent, cached := c.collections.Get(key); !cached { + e, cached := c.collections.Get(key) + if !cached { + return nil + } + 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(key) - return nil - } else { - atomic.AddUint64(&c.stats.CollectionHits, 1) - return ent.collection - } } + c.metrics.collectionHits.Inc() + return ent.collection }