Skip to content

Commit

Permalink
image manager: cleanup 'dead' and 'created' containers
Browse files Browse the repository at this point in the history 
also cleanup of 'dangling' images that have no tags or names associated
with them (ie, they show as <none> in 'docker images')

closes aws#1684

unit tests
  • Loading branch information
@sparrc
sparrc committed May 2, 2019
1 parent df83ed3 commit be2d686
Show file tree
Hide file tree
Showing 3 changed files with 255 additions and 47 deletions.
4 changes: 2 additions & 2 deletions agent/config/config.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -306,7 +306,7 @@ func (cfg *Config) validateAndOverrideBounds() error {
// If a value has been set for taskCleanupWaitDuration and the value is less than the minimum allowed cleanup duration,
// print a warning and override it
if cfg.TaskCleanupWaitDuration < minimumTaskCleanupWaitDuration {
seelog.Warnf("Invalid value for image cleanup duration, will be overridden with the default value: %s. Parsed value: %v, minimum value: %v.", DefaultTaskCleanupWaitDuration.String(), cfg.TaskCleanupWaitDuration, minimumTaskCleanupWaitDuration)
seelog.Warnf("Invalid value for ECS_ENGINE_TASK_CLEANUP_WAIT_DURATION, will be overridden with the default value: %s. Parsed value: %v, minimum value: %v.", DefaultTaskCleanupWaitDuration.String(), cfg.TaskCleanupWaitDuration, minimumTaskCleanupWaitDuration)
cfg.TaskCleanupWaitDuration = DefaultTaskCleanupWaitDuration
}

Expand All @@ -316,7 +316,7 @@ func (cfg *Config) validateAndOverrideBounds() error {
}

if cfg.ImageCleanupInterval < minimumImageCleanupInterval {
seelog.Warnf("Invalid value for image cleanup duration, will be overridden with the default value: %s. Parsed value: %v, minimum value: %v.", DefaultImageCleanupTimeInterval.String(), cfg.ImageCleanupInterval, minimumImageCleanupInterval)
seelog.Warnf("Invalid value for ECS_IMAGE_CLEANUP_INTERVAL, will be overridden with the default value: %s. Parsed value: %v, minimum value: %v.", DefaultImageCleanupTimeInterval.String(), cfg.ImageCleanupInterval, minimumImageCleanupInterval)
cfg.ImageCleanupInterval = DefaultImageCleanupTimeInterval
}

Expand Down
81 changes: 46 additions & 35 deletions agent/engine/docker_image_manager.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -312,6 +312,7 @@ func (imageManager *dockerImageManager) removeUnusedImages(ctx context.Context)

var numECSImagesDeleted int
imageManager.imageStatesConsideredForDeletion = imageManager.imagesConsiderForDeletion(imageManager.getAllImageStates())

for i := 0; i < imageManager.numImagesToDelete; i++ {
err := imageManager.removeLeastRecentlyUsedImage(ctx)
numECSImagesDeleted = i
Expand Down Expand Up @@ -342,9 +343,13 @@ func (imageManager *dockerImageManager) removeNonECSContainers(ctx context.Conte
continue
}

finishedTime, _ := time.Parse(time.Now().String(), response.State.FinishedAt)
seelog.Infof("Inspecting Non-ECS Container ID [%s] for removal, Finished [%s] Status [%s]", id, response.State.FinishedAt, response.State.Status)
finishedTime, err := time.Parse(time.RFC3339Nano, response.State.FinishedAt)
if err != nil {
seelog.Errorf("Error parsing time string for container. id: %s, time: %s err: %s", id, response.State.FinishedAt, err)
}

if response.State.Status == "exited" && time.Now().Sub(finishedTime) > imageManager.nonECSContainerCleanupWaitDuration {
if (response.State.Status == "exited" || response.State.Status == "dead" || response.State.Status == "created") && time.Now().Sub(finishedTime) > imageManager.nonECSContainerCleanupWaitDuration {
nonECSContainerRemoveAvailableIDs = append(nonECSContainerRemoveAvailableIDs, id)
}
}
Expand All @@ -353,13 +358,13 @@ func (imageManager *dockerImageManager) removeNonECSContainers(ctx context.Conte
if numNonECSContainerDeleted == imageManager.numNonECSContainersToDelete {
break
}
seelog.Infof("Removing non-ECS container id: %s", id)
seelog.Debugf("Removing non-ECS container id: %s", id)
err := imageManager.client.RemoveContainer(ctx, id, dockerclient.RemoveContainerTimeout)
if err == nil {
seelog.Infof("Image removed: %s", id)
numNonECSContainerDeleted++
} else {
seelog.Errorf("Error removing Image %s - %v", id, err)
seelog.Errorf("Error Removing Container %s - %s", id, err)
continue
}
}
Expand All @@ -379,70 +384,75 @@ func (imageManager *dockerImageManager) getNonECSContainerIDs(ctx context.Contex
return nonECSContainersIDs, nil
}

type ImageWithSize struct {
type ImageWithSizeID struct {
ImageName string
ImageID string
Size int64
}

func (imageManager *dockerImageManager) removeNonECSImages(ctx context.Context, nonECSImagesNumToDelete int) {
if nonECSImagesNumToDelete == 0 {
return
}
var nonECSImageNames = imageManager.getNonECSImageNames(ctx)
var nonECSImageNamesRemoveEligible []string
for _, nonECSImage := range nonECSImageNames {
if !isInExclusionList(nonECSImage, imageManager.imageCleanupExclusionList) {
nonECSImageNamesRemoveEligible = append(nonECSImageNamesRemoveEligible, nonECSImage)
nonECSImages := imageManager.getNonECSImages(ctx)
var nonECSImagesRmEligible []ImageWithSizeID
for _, nonECSImage := range nonECSImages {
if !isInExclusionList(nonECSImage.ImageName, imageManager.imageCleanupExclusionList) {
nonECSImagesRmEligible = append(nonECSImagesRmEligible, nonECSImage)
}
}

var imageWithSizeList []ImageWithSize
for _, imageName := range nonECSImageNamesRemoveEligible {
resp, iiErr := imageManager.client.InspectImage(imageName)
if iiErr != nil {
seelog.Errorf("Error inspecting non-ECS image name: %s - %v", imageName, iiErr)
// Get the all image sizes
for _, image := range nonECSImagesRmEligible {
resp, err := imageManager.client.InspectImage(image.ImageID)
if err != nil {
seelog.Errorf("Error inspecting non-ECS image: %s (%s), %s", image.ImageName, image.ImageID, err)
continue
}
imageWithSizeList = append(imageWithSizeList, ImageWithSize{imageName, resp.Size})
image.Size = resp.Size
}
// we want to sort images with size ascending
sort.Slice(imageWithSizeList, func(i, j int) bool {
return imageWithSizeList[i].Size < imageWithSizeList[j].Size
sort.Slice(nonECSImagesRmEligible, func(i, j int) bool {
return nonECSImagesRmEligible[i].Size < nonECSImagesRmEligible[j].Size
})

// we will remove the remaining nonECSImages in each performPeriodicImageCleanup call()
var numImagesAlreadyDeleted = 0
for _, kv := range imageWithSizeList {
for _, kv := range nonECSImagesRmEligible {
if numImagesAlreadyDeleted == nonECSImagesNumToDelete {
break
}
seelog.Infof("Removing non-ECS Image: %s", kv.ImageName)
err := imageManager.client.RemoveImage(ctx, kv.ImageName, dockerclient.RemoveImageTimeout)
seelog.Infof("Removing non-ECS Image: %s (%s)", kv.ImageName, kv.ImageID)
err := imageManager.client.RemoveImage(ctx, kv.ImageID, dockerclient.RemoveImageTimeout)
if err != nil {
seelog.Errorf("Error removing Image %s - %v", kv.ImageName, err)
seelog.Errorf("Error removing Image %s (%s) - %v", kv.ImageName, kv.ImageID, err)
continue
} else {
seelog.Infof("Image removed: %s", kv.ImageName)
seelog.Infof("Image removed: %s (%s)", kv.ImageName, kv.ImageID)
numImagesAlreadyDeleted++
}
}
}

func (imageManager *dockerImageManager) getNonECSImageNames(ctx context.Context) []string {
response := imageManager.client.ListImages(ctx, dockerclient.ListImagesTimeout)
var allImagesNames []string
for _, name := range response.RepoTags {
allImagesNames = append(allImagesNames, name)
// getNonECSImages returns type ImageWithSizeID but does NOT populate the Size field.
func (imageManager *dockerImageManager) getNonECSImages(ctx context.Context) []ImageWithSizeID {
r := imageManager.client.ListImages(ctx, dockerclient.ListImagesTimeout)
var allImages []ImageWithSizeID
for i := 0; i < len(r.RepoTags); i++ {
allImages = append(allImages, ImageWithSizeID{ImageName: r.RepoTags[i], ImageID: r.ImageIDs[i]})
}
var ecsImageNames []string
var ecsImageIDs []string
for _, imageState := range imageManager.getAllImageStates() {
for _, imageName := range imageState.Image.Names {
ecsImageNames = append(ecsImageNames, imageName)
}
ecsImageIDs = append(ecsImageIDs, imageState.Image.ImageID)
}

var nonECSImageNames = exclude(allImagesNames, ecsImageNames)
return nonECSImageNames
var nonECSImages []ImageWithSizeID
for _, image := range allImages {
if !isInExclusionList(image.ImageID, ecsImageIDs) {
nonECSImages = append(nonECSImages, image)
}
}
return nonECSImages
}

func isInExclusionList(imageName string, imageExclusionList []string) bool {
Expand Down Expand Up @@ -477,8 +487,9 @@ func (imageManager *dockerImageManager) imagesConsiderForDeletion(allImageStates
for _, imageState := range allImageStates {
if imageManager.isExcludedFromCleanup(imageState) {
//imageState that we want to keep
seelog.Infof("Image excluded from deletion: [%s]", imageState.String())
seelog.Debugf("Image excluded from deletion: [%s]", imageState.String())
} else {
seelog.Debugf("Image going to be considered for deletion: [%s]", imageState.String())
imagesConsiderForDeletionMap[imageState.Image.ImageID] = imageState
}
}
Expand Down
Loading

0 comments on commit be2d686

Please sign in to comment.