archiver: restructure a little bit to facilitate testing

This introduces two sync.Cond pointers to the archiver package. If they're
non-nil when we go to process a request, we'll wait until signalled (at all)
to proceed. The tests will then create the sync.Cond so that it can signal
at-will and sanity-check the state of the queue at different phases.

The author believes that nil-checking these two sync.Cond pointers on every
archive processing will introduce minimal overhead with no impact on
maintainability.

services/archiver/archiver.go

@@ -41,6 +41,12 @@ type ArchiveRequest struct {
 var archiveInProgress []*ArchiveRequest
 var archiveMutex sync.Mutex
 
+// These facilitate testing, by allowing the unit tests to control (to some extent)
+// the goroutine used for processing the queue.
+var archiveQueueMutex *sync.Mutex
+var archiveQueueStartCond *sync.Cond
+var archiveQueueReleaseCond *sync.Cond
+
 // GetArchivePath returns the path from which we can serve this archive.
 func (aReq *ArchiveRequest) GetArchivePath() string {
 	return aReq.archivePath
@@ -216,11 +222,26 @@ func ArchiveRepository(request *ArchiveRequest) {
 		archiveInProgress = append(archiveInProgress, request)
 		archiveMutex.Unlock()
 
+		// Wait to start, if we have the Cond for it.  This is currently only
+		// useful for testing, so that the start and release of queued entries
+		// can be controlled to examine the queue.
+		if archiveQueueStartCond != nil {
+			archiveQueueMutex.Lock()
+			archiveQueueStartCond.Wait()
+			archiveQueueMutex.Unlock()
+		}
+
 		// Drop the mutex while we process the request.  This may take a long
 		// time, and it's not necessary now that we've added the reequest to
 		// archiveInProgress.
 		doArchive(request)
 
+		if archiveQueueReleaseCond != nil {
+			archiveQueueMutex.Lock()
+			archiveQueueReleaseCond.Wait()
+			archiveQueueMutex.Unlock()
+		}
+
 		// Purge this request from the list.  To do so, we'll just take the
 		// index at which we ended up at and swap the final element into that
 		// position, then chop off the now-redundant final element.  The slice

services/archiver/archiver_test.go

@@ -6,6 +6,7 @@ package archiver
 
 import (
 	"path/filepath"
+	"sync"
 	"testing"
 	"time"
 
@@ -16,8 +17,46 @@ import (
 	"github.com/unknwon/com"
 )
 
+var queueMutex sync.Mutex
+
 func TestMain(m *testing.M) {
 	models.MainTest(m, filepath.Join("..", ".."))
+
+	archiveQueueMutex = &queueMutex
+	archiveQueueStartCond = sync.NewCond(&queueMutex)
+	archiveQueueReleaseCond = sync.NewCond(&queueMutex)
+}
+
+func allComplete(inFlight []*ArchiveRequest) bool {
+	for _, req := range inFlight {
+		if !req.IsComplete() {
+			return false
+		}
+	}
+
+	return true
+}
+
+func releaseOneEntry(t *testing.T, inFlight []*ArchiveRequest) {
+	var nowQueued, numQueued int
+
+	numQueued = len(archiveInProgress)
+
+	// Release one, then wait up to 3 seconds for it to complete.
+	archiveQueueReleaseCond.Signal()
+	timeout := time.Now().Add(3 * time.Second)
+	for {
+		nowQueued = len(archiveInProgress)
+		if nowQueued != numQueued || time.Now().After(timeout) {
+			break
+		}
+	}
+
+	// Make sure we didn't just timeout.
+	assert.NotEqual(t, nowQueued, numQueued)
+
+	// Also make sure that we released only one.
+	assert.Equal(t, nowQueued, numQueued + 1)
 }
 
 func TestArchive_Basic(t *testing.T) {
@@ -59,14 +98,31 @@ func TestArchive_Basic(t *testing.T) {
 	secondReq := DeriveRequestFrom(ctx, secondCommit+".zip")
 	assert.NotNil(t, secondReq)
 
-	ArchiveRepository(zipReq)
+	inFlight := make([]*ArchiveRequest, 3)
-	ArchiveRepository(tgzReq)
+	inFlight[0] = zipReq
-	ArchiveRepository(secondReq)
+	inFlight[1] = tgzReq
+	inFlight[2] = secondReq
 
-	// Wait for those requests to complete, time out after 8 seconds.
+	ArchiveRepository(zipReq)
+	assert.Equal(t, len(archiveInProgress), 1)
+	ArchiveRepository(tgzReq)
+	assert.Equal(t, len(archiveInProgress), 2)
+	ArchiveRepository(secondReq)
+	assert.Equal(t, len(archiveInProgress), 3)
+
+	// Make sure sending an unprocessed request through doesn't affect the queue
+	// count.
+	ArchiveRepository(zipReq)
+	assert.Equal(t, len(archiveInProgress), 3)
+
+	// Release them all, they'll then stall at the archiveQueueReleaseCond while
+	// we examine the queue state.
+	archiveQueueStartCond.Broadcast()
+
+	// 8 second timeout for them all to complete.
 	timeout := time.Now().Add(8 * time.Second)
 	for {
-		if zipReq.IsComplete() && tgzReq.IsComplete() && secondReq.IsComplete() {
+		if allComplete(inFlight) {
 			break
 		} else if time.Now().After(timeout) {
 			break
@@ -80,8 +136,9 @@ func TestArchive_Basic(t *testing.T) {
 	assert.True(t, com.IsExist(tgzReq.GetArchivePath()))
 	assert.True(t, com.IsExist(secondReq.GetArchivePath()))
 
-	// The queue should also be drained, if all requests have completed.
+	// Queues should not have drained yet, because we haven't released them.
-	assert.Equal(t, len(archiveInProgress), 0)
+	// Do so now.
+	assert.Equal(t, len(archiveInProgress), 3)
 
 	zipReq2 := DeriveRequestFrom(ctx, firstCommit+".zip")
 	// After completion, zipReq should have dropped out of the queue.  Make sure
@@ -90,10 +147,25 @@ func TestArchive_Basic(t *testing.T) {
 	assert.Equal(t, zipReq, zipReq2)
 	assert.False(t, zipReq == zipReq2)
 
-	// Make sure we can submit this follow-up request with no side-effects, to
+	// We still have the other three stalled at completion, waiting to remove
-	// the extent that we can.
+	// from archiveInProgress.  Try to submit this new one before its
+	// predecessor has cleared out of the queue.
 	ArchiveRepository(zipReq2)
-	assert.Equal(t, zipReq, zipReq2)
+
+	// Make sure we didn't enqueue anything from this new one, and that the
+	// queue hasn't changed.
+	assert.Equal(t, len(archiveInProgress), 3)
+
+	for _, req := range archiveInProgress {
+		assert.False(t, req == zipReq2)
+	}
+
+	// Make sure the queue drains properly
+	releaseOneEntry(t, inFlight)
+	assert.Equal(t, len(archiveInProgress), 2)
+	releaseOneEntry(t, inFlight)
+	assert.Equal(t, len(archiveInProgress), 1)
+	releaseOneEntry(t, inFlight)
 	assert.Equal(t, len(archiveInProgress), 0)
 
 	// Same commit, different compression formats should have different names.