GODT-1356 GODT-1302: Cache on disk concurency and API retries

- GODT-1302: Change maximum resty retries from 0 to 30
- GODT-1302: Make sure we are closing GetAttachmen io.ReadCloser on error
- GODT-1356: Do not use attachmentPool - it was useless anyway
- GODT-1356: Increase cache watcher limit to 10min
- GODT-1356: Start cache watcher right after start (do not wait first 10 min)
- GODT-1356: Limit number of buildJobs (memory allocation) in BuildAndCacheMessage
- Other: Pass context from job options (message builder) to fetcher (both message and attachments)
- Other: BuildJob contains same function as returned buildDone (proper map locking)

internal/app/bridge/bridge.go

@@ -32,6 +32,7 @@ import (
 	"github.com/ProtonMail/proton-bridge/internal/frontend/types"
 	"github.com/ProtonMail/proton-bridge/internal/imap"
 	"github.com/ProtonMail/proton-bridge/internal/smtp"
+	"github.com/ProtonMail/proton-bridge/internal/store"
 	"github.com/ProtonMail/proton-bridge/internal/store/cache"
 	"github.com/ProtonMail/proton-bridge/internal/updater"
 	"github.com/ProtonMail/proton-bridge/pkg/message"
@@ -74,7 +75,7 @@ func run(b *base.Base, c *cli.Context) error { // nolint[funlen]
 		return err
 	}
 
-	cache, err := loadCache(b)
+	cache, err := loadMessageCache(b)
 	if err != nil {
 		return err
 	}
@@ -247,16 +248,20 @@ func checkAndHandleUpdate(u types.Updater, f frontend.Frontend, autoUpdate bool)
 	f.NotifySilentUpdateInstalled()
 }
 
-// NOTE(GODT-1158): How big should in-memory cache be?
-// NOTE(GODT-1158): How to handle cache location migration if user changes custom path?
-func loadCache(b *base.Base) (cache.Cache, error) {
+func loadMessageCache(b *base.Base) (cache.Cache, error) {
 	if !b.Settings.GetBool(settings.CacheEnabledKey) {
-		return cache.NewInMemoryCache(100 * (1 << 20)), nil
+		// Memory cache was estimated by empirical usage in past and it
+		// was set to 100MB.
+		// NOTE: This value must not be less than maximal size of one
+		// email (~30MB).
+		return cache.NewInMemoryCache(100 << 20), nil
 	}
 
 	var compressor cache.Compressor
 
-	// NOTE(GODT-1158): If user changes compression setting we have to nuke the cache.
+	// NOTE(GODT-1158): Changing compression is not an option currently
+	// available for user but, if user changes compression setting we have
+	// to nuke the cache.
 	if b.Settings.GetBool(settings.CacheCompressionKey) {
 		compressor = &cache.GZipCompressor{}
 	} else {
@@ -269,10 +274,15 @@ func loadCache(b *base.Base) (cache.Cache, error) {
 		path = customPath
 	} else {
 		path = b.Cache.GetDefaultMessageCacheDir()
-		// Store path so it will allways persist if default location will be changed in new version.
+		// Store path so it will allways persist if default location
+		// will be changed in new version.
 		b.Settings.Set(settings.CacheLocationKey, path)
 	}
 
+	// To prevent memory peaks we set maximal write concurency for store
+	// build jobs.
+	store.SetBuildAndCacheJobLimit(b.Settings.GetInt(settings.CacheConcurrencyWrite))
+
 	return cache.NewOnDiskCache(path, compressor, cache.Options{
 		MinFreeAbs:      uint64(b.Settings.GetInt(settings.CacheMinFreeAbsKey)),
 		MinFreeRat:      b.Settings.GetFloat64(settings.CacheMinFreeRatKey),

internal/store/cache.go

@@ -89,6 +89,24 @@ func (store *Store) clearCachePassphrase() error {
 	})
 }
 
+// buildAndCacheJobs is used to limit the number of parallel background build
+// jobs by using a buffered channel. When channel is blocking the go routines
+// is running but the download didn't started yet and hence no space needs to
+// be allocated. Once other instances are finished the job can continue. The
+// bottleneck is `store.cache.Set` which can be take some time to write all
+// downloaded bytes. Therefore, it is not effective to start fetching and
+// building the message for more than maximum of possible parallel cache
+// writers.
+//
+// Default buildAndCacheJobs vaule is 16, it can be changed by SetBuildAndCacheJobLimit.
+var (
+	buildAndCacheJobs = make(chan struct{}, 16) //nolint[gochecknoglobals]
+)
+
+func SetBuildAndCacheJobLimit(maxJobs int) {
+	buildAndCacheJobs = make(chan struct{}, maxJobs)
+}
+
 func (store *Store) getCachedMessage(messageID string) ([]byte, error) {
 	if store.cache.Has(store.user.ID(), messageID) {
 		return store.cache.Get(store.user.ID(), messageID)
@@ -118,6 +136,9 @@ func (store *Store) IsCached(messageID string) bool {
 // BuildAndCacheMessage builds the given message (with background priority) and puts it in the cache.
 // It builds with background priority.
 func (store *Store) BuildAndCacheMessage(messageID string) error {
+	buildAndCacheJobs <- struct{}{}
+	defer func() { <-buildAndCacheJobs }()
+
 	job, done := store.newBuildJob(messageID, message.BackgroundPriority)
 	defer done()
 

internal/store/cache_watcher.go

@@ -23,26 +23,27 @@ func (store *Store) StartWatcher() {
 	store.done = make(chan struct{})
 
 	go func() {
-		ticker := time.NewTicker(3 * time.Minute)
+		ticker := time.NewTicker(10 * time.Minute)
 		defer ticker.Stop()
 
 		for {
+			// NOTE(GODT-1158): Race condition here? What if DB was already closed?
+			messageIDs, err := store.getAllMessageIDs()
+			if err != nil {
+				return
+			}
+
+			for _, messageID := range messageIDs {
+				if !store.IsCached(messageID) {
+					store.cacher.newJob(messageID)
+				}
+			}
+
 			select {
-			case <-ticker.C:
-				// NOTE(GODT-1158): Race condition here? What if DB was already closed?
-				messageIDs, err := store.getAllMessageIDs()
-				if err != nil {
-					return
-				}
-
-				for _, messageID := range messageIDs {
-					if !store.IsCached(messageID) {
-						store.cacher.newJob(messageID)
-					}
-				}
-
 			case <-store.done:
 				return
+			case <-ticker.C:
+				continue
 			}
 		}
 	}()