// synchronizeState explicitly goes through each docker container stored in
// "state" and updates its KnownStatus appropriately, as well as queueing up
// events to push upstream.
func (engine *DockerTaskEngine) synchronizeState() {
engine.processTasks.Lock()
defer engine.processTasks.Unlock()
imageStates := engine.state.AllImageStates()
if len(imageStates) != 0 {
engine.imageManager.AddAllImageStates(imageStates)
}
tasks := engine.state.AllTasks()
for _, task := range tasks {
conts, ok := engine.state.ContainerMapByArn(task.Arn)
if !ok {
engine.startTask(task)
continue
}
for _, cont := range conts {
if cont.DockerId == "" {
log.Debug("Found container potentially created while we were down", "name", cont.DockerName)
// Figure out the dockerid
describedCont, err := engine.client.InspectContainer(cont.DockerName)
if err != nil {
log.Warn("Could not find matching container for expected", "name", cont.DockerName)
} else {
cont.DockerId = describedCont.ID
// update mappings that need dockerid
engine.state.AddContainer(cont, task)
engine.imageManager.AddContainerReferenceToImageState(cont.Container)
}
}
if cont.DockerId != "" {
currentState, metadata := engine.client.DescribeContainer(cont.DockerId)
if metadata.Error != nil {
currentState = api.ContainerStopped
if !cont.Container.KnownTerminal() {
cont.Container.ApplyingError = api.NewNamedError(&ContainerVanishedError{})
log.Warn("Could not describe previously known container; assuming dead", "err", metadata.Error, "id", cont.DockerId, "name", cont.DockerName)
engine.imageManager.RemoveContainerReferenceFromImageState(cont.Container)
}
} else {
engine.imageManager.AddContainerReferenceToImageState(cont.Container)
}
if currentState > cont.Container.GetKnownStatus() {
cont.Container.SetKnownStatus(currentState)
}
}
}
engine.startTask(task)
}
engine.saver.Save()
}
func (mtask *managedTask) handleContainerChange(containerChange dockerContainerChange) {
llog := log.New("task", mtask.Task)
// Handle container change updates a container's known status.
// In addition, if the change mentions interesting information (like
// exit codes or ports) this propegates them.
container := containerChange.container
found := false
for _, c := range mtask.Containers {
if container == c {
found = true
}
}
if !found {
llog.Crit("State error; task manager called with another task's container!", "container", container)
return
}
event := containerChange.event
llog.Debug("Handling container change", "change", containerChange)
// Cases: If this is a forward transition (else) update the container to be known to be at that status.
// If this is a backwards transition stopped->running, the first time set it
// to be known running so it will be stopped. Subsequently ignore these backward transitions
if event.Status <= container.KnownStatus && container.KnownStatus == api.ContainerStopped {
if event.Status == api.ContainerRunning {
// If the container becomes running after we've stopped it (possibly
// because we got an error running it and it ran anyways), the first time
// update it to 'known running' so that it will be driven back to stopped
mtask.unexpectedStart.Do(func() {
llog.Warn("Container that we thought was stopped came back; re-stopping it once")
go mtask.engine.transitionContainer(mtask.Task, container, api.ContainerStopped)
// This will not proceed afterwards because status <= knownstatus below
})
}
}
if event.Status <= container.KnownStatus {
seelog.Infof("Redundant container state change for task %s: %s to %s, but already %s", mtask.Task, container, event.Status, container.KnownStatus)
return
}
container.KnownStatus = event.Status
if event.Error != nil {
if container.ApplyingError == nil {
container.ApplyingError = api.NewNamedError(event.Error)
}
if event.Status == api.ContainerStopped {
// If we were trying to transition to stopped and had an error, we
// clearly can't just continue trying to transition it to stopped
// again and again... In this case, assume it's stopped (or close
// enough) and get on with it
// This actually happens a lot for the case of stopping something that was not running.
llog.Info("Error for 'docker stop' of container; assuming it's stopped anyways")
container.KnownStatus = api.ContainerStopped
container.DesiredStatus = api.ContainerStopped
} else if event.Status == api.ContainerPulled {
// Another special case; a failure to pull might not be fatal if e.g. the image already exists.
llog.Info("Error while pulling container; will try to run anyways", "err", event.Error)
} else {
llog.Warn("Error with docker; stopping container", "container", container, "err", event.Error)
container.DesiredStatus = api.ContainerStopped
// the above 'knownstatus' is not truthful because of the error
// No point in emitting it, just continue on to stopped
return
}
}
if event.ExitCode != nil && event.ExitCode != container.KnownExitCode {
container.KnownExitCode = event.ExitCode
}
if event.PortBindings != nil {
container.KnownPortBindings = event.PortBindings
}
if event.Volumes != nil {
mtask.UpdateMountPoints(container, event.Volumes)
}
mtask.engine.emitContainerEvent(mtask.Task, container, "")
if mtask.UpdateStatus() {
llog.Debug("Container change also resulted in task change")
// If knownStatus changed, let it be known
mtask.engine.emitTaskEvent(mtask.Task, "")
}
}
