fix(dir): remove timeout and evaluate switch *before* removing entry

io/directory.go

@@ -3,13 +3,10 @@ package io
 import (
 	"context"
 	"fmt"
-	"os"
-	"time"
-
 	mdag "github.com/ipfs/go-merkledag"
-
 	format "github.com/ipfs/go-unixfs"
 	"github.com/ipfs/go-unixfs/hamt"
+	"os"
 
 	"github.com/ipfs/go-cid"
 	ipld "github.com/ipfs/go-ipld-format"
@@ -26,11 +23,6 @@ var log = logging.Logger("unixfs")
 // ProtoNode doesn't use the Data field so this estimate is pretty accurate).
 var HAMTShardingSize = 0
 
-// Time in seconds allowed to fetch the shards to compute the size before
-// returning an error.
-// FIXME: Adjust to sane value.
-var EvaluateHAMTTransitionTimeout = time.Duration(1)
-
 // DefaultShardWidth is the default value used for hamt sharding width.
 var DefaultShardWidth = 256
 
@@ -438,15 +430,13 @@ func (d *HAMTDirectory) removeFromSizeChange(name string, linkCid cid.Cid) {
 	d.sizeChange -= estimatedLinkSize(name, linkCid)
 }
 
-// Evaluate directory size and check if it's below HAMTShardingSize threshold
-// (to trigger a transition to a BasicDirectory). It returns two `bool`s:
-// * whether it's below (true) or equal/above (false)
-// * whether the passed timeout to compute the size has been exceeded
+// Evaluate directory size and a future sizeChange and check if it will be below
+// HAMTShardingSize threshold (to trigger a transition to a BasicDirectory).
 // Instead of enumerating the entire tree we eagerly call EnumLinksAsync
-// until we either reach a value above the threshold (in that case no need)
-// to keep counting or the timeout runs out in which case the `below` return
-// value is not to be trusted as we didn't have time to count enough shards.
-func (d *HAMTDirectory) sizeBelowThreshold(timeout time.Duration) (below bool, timeoutExceeded bool) {
+// until we either reach a value above the threshold (in that case no need
+// to keep counting) or an error occurs (like the context being canceled
+// if we take too much time fetching the necessary shards).
+func (d *HAMTDirectory) sizeBelowThreshold(ctx context.Context, sizeChange int) (below bool, err error) {
 	if HAMTShardingSize == 0 {
 		panic("asked to compute HAMT size with HAMTShardingSize option off (0)")
 	}
@@ -455,57 +445,54 @@ func (d *HAMTDirectory) sizeBelowThreshold(timeout time.Duration) (below bool, t
 	// end early if we already know we're above the threshold or run out of time.
 	partialSize := 0
 
-	ctx, cancel := context.WithTimeout(context.Background(), time.Second*timeout)
+	// We stop the enumeration once we have enough information and exit this function.
+	ctx, cancel := context.WithCancel(ctx)
 	defer cancel()
+
 	for linkResult := range d.EnumLinksAsync(ctx) {
 		if linkResult.Err != nil {
-			continue
-			// The timeout exceeded errors will be coming through here but I'm
-			// not sure if we can just compare against a generic DeadlineExceeded
-			// error here to return early and avoid iterating the entire loop.
-			// (We might confuse a specific DeadlineExceeded of an internal function
-			//  with our context here.)
-			// Since *our* DeadlineExceeded will quickly propagate to any other
-			// pending fetches it seems that iterating the entire loop won't add
-			// much more cost anyway.
-			// FIXME: Check the above reasoning.
+			return false, linkResult.Err
 		}
-		if linkResult.Link == nil {
-			panic("empty link result (both values nil)")
-			// FIXME: Is this *ever* possible?
-		}
-		partialSize += estimatedLinkSize(linkResult.Link.Name, linkResult.Link.Cid)
 
-		if partialSize >= HAMTShardingSize {
+		partialSize += estimatedLinkSize(linkResult.Link.Name, linkResult.Link.Cid)
+		if partialSize+sizeChange >= HAMTShardingSize {
 			// We have already fetched enough shards to assert we are
 			//  above the threshold, so no need to keep fetching.
-			return false, false
+			return false, nil
 		}
 	}
-	// At this point either we enumerated all shards or run out of time.
-	// Figure out which.
 
-	if ctx.Err() == context.Canceled {
-		panic("the context was canceled but we're still evaluating a possible switch")
+	// We enumerated *all* links in all shards and didn't reach the threshold.
+	return true, nil
+}
+
+// FIXME: Will be extended later to the `AddEntry` case.
+func (d *HAMTDirectory) needsToSwitchToBasicDir(ctx context.Context, nameToRemove string) (switchToBasic bool, err error) {
+	if HAMTShardingSize == 0 { // Option disabled.
+		return false, nil
 	}
-	if partialSize >= HAMTShardingSize {
-		panic("we reach the threshold but we're still evaluating a possible switch")
+
+	entryToRemove, err := d.shard.Find(ctx, nameToRemove)
+	if err == os.ErrNotExist {
+		// Nothing to remove, no point in evaluating a switch.
+		return false, nil
+	} else if err != nil {
+		return false, err
 	}
+	sizeToRemove := estimatedLinkSize(nameToRemove, entryToRemove.Cid)
 
-	if ctx.Err() == context.DeadlineExceeded {
-		return false, true
+	if d.sizeChange-sizeToRemove >= 0 {
+		// We won't have reduced the HAMT net size.
+		return false, nil
 	}
 
-	// If we reach this then:
-	// * We are below the threshold (we didn't return inside the EnumLinksAsync
-	//   loop).
-	// * The context wasn't cancelled so we iterated *all* shards
-	//   and are sure that we have the full size.
-	// FIXME: Can we actually verify the last claim here to be sure?
-	//  (Iterating all the shards in the HAMT as a plumbing function maybe.
-	//   If they're in memory it shouldn't be that expensive, we won't be
-	//   switching that often, probably.)
-	return true, false
+	// We have reduced the directory size, check if went below the
+	// HAMTShardingSize threshold to trigger a switch.
+	belowThreshold, err := d.sizeBelowThreshold(ctx, -sizeToRemove)
+	if err != nil {
+		return false, err
+	}
+	return belowThreshold, nil
 }
 
 // UpgradeableDirectory wraps a Directory interface and provides extra logic
@@ -573,42 +560,21 @@ func (d *UpgradeableDirectory) getDagService() ipld.DAGService {
 //  sure we make good on the value). Finding the right margin can be tricky
 //  and very dependent on the use case so it might not be worth it.
 func (d *UpgradeableDirectory) RemoveChild(ctx context.Context, name string) error {
-	if err := d.Directory.RemoveChild(ctx, name); err != nil {
-		return err
-	}
-
 	hamtDir, ok := d.Directory.(*HAMTDirectory)
-	if !ok { // BasicDirectory
-		return nil
-	}
-
-	if HAMTShardingSize == 0 || // Option disabled.
-		hamtDir.sizeChange >= 0 { // We haven't reduced the HAMT net size.
-		return nil
-	}
-
-	// We have reduced the directory size, check if it didn't go under
-	// the HAMTShardingSize threshold.
-	belowThreshold, timeoutExceeded := hamtDir.sizeBelowThreshold(EvaluateHAMTTransitionTimeout)
-
-	if timeoutExceeded {
-		// We run out of time before confirming if we're indeed below the
-		// threshold. When in doubt error to not return inconsistent structures.
-		// FIXME: We could allow this to return without error and enforce this
-		//  timeout on a GetNode() call when we need to actually commit to a
-		//  structure/CID. (The downside is that GetNode() doesn't have a
-		//  context argument and we would have to break the API.)
-		return fmt.Errorf("not enought time to fetch shards")
-		// FIXME: Abstract in new error for testing.
-	}
-
-	if belowThreshold { // Switch.
-		basicDir, err := hamtDir.switchToBasic(ctx)
+	if ok {
+		switchToBasic, err := hamtDir.needsToSwitchToBasicDir(ctx, name)
 		if err != nil {
 			return err
 		}
-		d.Directory = basicDir
+
+		if switchToBasic {
+			basicDir, err := hamtDir.switchToBasic(ctx)
+			if err != nil {
+				return err
+			}
+			d.Directory = basicDir
+		}
 	}
 
-	return nil
+	return d.Directory.RemoveChild(ctx, name)
 }