mirror of
https://github.com/podman-container-tools/podman.git
synced 2026-08-14 20:59:40 +00:00
This was implemented by containers/netavark #1369; this commit completes the process by wiring it into Podman. We now respect the CLI order for configured networks - if a user passes `--net net1,net2` we guarantee that net1 will be configured before net2. For containers created before this patch, we don't retain enough information to configure networks in CLI order, so we use alphabetical order instead to still guarantee consistency. No breaking API changes have been made, but we do add a new field to supplement the existing map to (optionally) provide ordering information. The Podman CLI will always pass this. Existing applications that do not will, again, receive] deterministic ordering based on an alphabetical sort of network names. This requires the latest version of Netavark to work properly. Signed-off-by: Matthew Heon <matthew.heon@pm.me>
461 lines
13 KiB
Go
461 lines
13 KiB
Go
// Copyright 2014 go-dockerclient authors. All rights reserved.
|
|
// Use of this source code is governed by a BSD-style
|
|
// license that can be found in the LICENSE file.
|
|
|
|
package docker
|
|
|
|
import (
|
|
"encoding/json"
|
|
"errors"
|
|
"io"
|
|
"math"
|
|
"net"
|
|
"net/http"
|
|
"net/http/httputil"
|
|
"slices"
|
|
"strconv"
|
|
"sync"
|
|
"sync/atomic"
|
|
"time"
|
|
)
|
|
|
|
// EventsOptions to filter events
|
|
// See https://docs.docker.com/engine/api/v1.41/#operation/SystemEvents for more details.
|
|
type EventsOptions struct {
|
|
// Show events created since this timestamp then stream new events.
|
|
Since string
|
|
|
|
// Show events created until this timestamp then stop streaming.
|
|
Until string
|
|
|
|
// Filter for events. For example:
|
|
// map[string][]string{"type": {"container"}, "event": {"start", "die"}}
|
|
// will return events when container was started and stopped or killed
|
|
//
|
|
// Available filters:
|
|
// config=<string> config name or ID
|
|
// container=<string> container name or ID
|
|
// daemon=<string> daemon name or ID
|
|
// event=<string> event type
|
|
// image=<string> image name or ID
|
|
// label=<string> image or container label
|
|
// network=<string> network name or ID
|
|
// node=<string> node ID
|
|
// plugin= plugin name or ID
|
|
// scope= local or swarm
|
|
// secret=<string> secret name or ID
|
|
// service=<string> service name or ID
|
|
// type=<string> container, image, volume, network, daemon, plugin, node, service, secret or config
|
|
// volume=<string> volume name
|
|
Filters map[string][]string
|
|
}
|
|
|
|
// APIEvents represents events coming from the Docker API
|
|
// The fields in the Docker API changed in API version 1.22, and
|
|
// events for more than images and containers are now fired off.
|
|
// To maintain forward and backward compatibility, go-dockerclient
|
|
// replicates the event in both the new and old format as faithfully as possible.
|
|
//
|
|
// For events that only exist in 1.22 in later, `Status` is filled in as
|
|
// `"Type:Action"` instead of just `Action` to allow for older clients to
|
|
// differentiate and not break if they rely on the pre-1.22 Status types.
|
|
//
|
|
// The transformEvent method can be consulted for more information about how
|
|
// events are translated from new/old API formats
|
|
type APIEvents struct {
|
|
// New API Fields in 1.22
|
|
Action string `json:"action,omitempty"`
|
|
Type string `json:"type,omitempty"`
|
|
Actor APIActor `json:"actor"`
|
|
|
|
// Old API fields for < 1.22
|
|
Status string `json:"status,omitempty"`
|
|
ID string `json:"id,omitempty"`
|
|
From string `json:"from,omitempty"`
|
|
|
|
// Fields in both
|
|
Time int64 `json:"time,omitempty"`
|
|
TimeNano int64 `json:"timeNano,omitempty"`
|
|
}
|
|
|
|
// APIActor represents an actor that accomplishes something for an event
|
|
type APIActor struct {
|
|
ID string `json:"id,omitempty"`
|
|
Attributes map[string]string `json:"attributes,omitempty"`
|
|
}
|
|
|
|
type eventMonitoringState struct {
|
|
// `sync/atomic` expects the first word in an allocated struct to be 64-bit
|
|
// aligned on both ARM and x86-32. See https://goo.gl/zW7dgq for more details.
|
|
lastSeen int64
|
|
sync.RWMutex
|
|
sync.WaitGroup
|
|
enabled bool
|
|
C chan *APIEvents
|
|
errC chan error
|
|
listeners []chan<- *APIEvents
|
|
closeConn func()
|
|
}
|
|
|
|
const (
|
|
maxMonitorConnRetries = 5
|
|
retryInitialWaitTime = 10.
|
|
)
|
|
|
|
var (
|
|
// ErrNoListeners is the error returned when no listeners are available
|
|
// to receive an event.
|
|
ErrNoListeners = errors.New("no listeners present to receive event")
|
|
|
|
// ErrListenerAlreadyExists is the error returned when the listerner already
|
|
// exists.
|
|
ErrListenerAlreadyExists = errors.New("listener already exists for docker events")
|
|
|
|
// ErrTLSNotSupported is the error returned when the client does not support
|
|
// TLS (this applies to the Windows named pipe client).
|
|
ErrTLSNotSupported = errors.New("tls not supported by this client")
|
|
|
|
// EOFEvent is sent when the event listener receives an EOF error.
|
|
EOFEvent = &APIEvents{
|
|
Type: "EOF",
|
|
Status: "EOF",
|
|
}
|
|
)
|
|
|
|
// AddEventListener adds a new listener to container events in the Docker API.
|
|
//
|
|
// The parameter is a channel through which events will be sent.
|
|
func (c *Client) AddEventListener(listener chan<- *APIEvents) error {
|
|
return c.AddEventListenerWithOptions(EventsOptions{}, listener)
|
|
}
|
|
|
|
// AddEventListener adds a new listener to container events in the Docker API.
|
|
// See https://docs.docker.com/engine/api/v1.41/#operation/SystemEvents for more details.
|
|
//
|
|
// The listener parameter is a channel through which events will be sent.
|
|
func (c *Client) AddEventListenerWithOptions(options EventsOptions, listener chan<- *APIEvents) error {
|
|
var err error
|
|
if !c.eventMonitor.isEnabled() {
|
|
err = c.eventMonitor.enableEventMonitoring(c, options)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return c.eventMonitor.addListener(listener)
|
|
}
|
|
|
|
// RemoveEventListener removes a listener from the monitor.
|
|
func (c *Client) RemoveEventListener(listener chan *APIEvents) error {
|
|
err := c.eventMonitor.removeListener(listener)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if c.eventMonitor.listernersCount() == 0 {
|
|
c.eventMonitor.disableEventMonitoring()
|
|
}
|
|
return nil
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) addListener(listener chan<- *APIEvents) error {
|
|
eventState.Lock()
|
|
defer eventState.Unlock()
|
|
if listenerExists(listener, &eventState.listeners) {
|
|
return ErrListenerAlreadyExists
|
|
}
|
|
eventState.Add(1)
|
|
eventState.listeners = append(eventState.listeners, listener)
|
|
return nil
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) removeListener(listener chan<- *APIEvents) error {
|
|
eventState.Lock()
|
|
defer eventState.Unlock()
|
|
if listenerExists(listener, &eventState.listeners) {
|
|
var newListeners []chan<- *APIEvents
|
|
for _, l := range eventState.listeners {
|
|
if l != listener {
|
|
newListeners = append(newListeners, l)
|
|
}
|
|
}
|
|
eventState.listeners = newListeners
|
|
eventState.Add(-1)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) closeListeners() {
|
|
for _, l := range eventState.listeners {
|
|
close(l)
|
|
eventState.Add(-1)
|
|
}
|
|
eventState.listeners = nil
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) listernersCount() int {
|
|
eventState.RLock()
|
|
defer eventState.RUnlock()
|
|
return len(eventState.listeners)
|
|
}
|
|
|
|
func listenerExists(a chan<- *APIEvents, list *[]chan<- *APIEvents) bool {
|
|
return slices.Contains(*list, a)
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) enableEventMonitoring(c *Client, opts EventsOptions) error {
|
|
eventState.Lock()
|
|
defer eventState.Unlock()
|
|
if !eventState.enabled {
|
|
eventState.enabled = true
|
|
atomic.StoreInt64(&eventState.lastSeen, 0)
|
|
eventState.C = make(chan *APIEvents, 100)
|
|
eventState.errC = make(chan error, 1)
|
|
go eventState.monitorEvents(c, opts)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) disableEventMonitoring() {
|
|
eventState.Lock()
|
|
defer eventState.Unlock()
|
|
|
|
eventState.closeListeners()
|
|
|
|
eventState.Wait()
|
|
|
|
if eventState.enabled {
|
|
eventState.enabled = false
|
|
close(eventState.C)
|
|
close(eventState.errC)
|
|
|
|
if eventState.closeConn != nil {
|
|
eventState.closeConn()
|
|
eventState.closeConn = nil
|
|
}
|
|
}
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) monitorEvents(c *Client, opts EventsOptions) {
|
|
const (
|
|
noListenersTimeout = 5 * time.Second
|
|
noListenersInterval = 10 * time.Millisecond
|
|
noListenersMaxTries = noListenersTimeout / noListenersInterval
|
|
)
|
|
|
|
var err error
|
|
for i := time.Duration(0); i < noListenersMaxTries && eventState.noListeners(); i++ {
|
|
time.Sleep(10 * time.Millisecond)
|
|
}
|
|
|
|
if eventState.noListeners() {
|
|
// terminate if no listener is available after 5 seconds.
|
|
// Prevents goroutine leak when RemoveEventListener is called
|
|
// right after AddEventListener.
|
|
eventState.disableEventMonitoring()
|
|
return
|
|
}
|
|
|
|
if err = eventState.connectWithRetry(c, opts); err != nil {
|
|
// terminate if connect failed
|
|
eventState.disableEventMonitoring()
|
|
return
|
|
}
|
|
for eventState.isEnabled() {
|
|
timeout := time.After(100 * time.Millisecond)
|
|
select {
|
|
case ev, ok := <-eventState.C:
|
|
if !ok {
|
|
return
|
|
}
|
|
if ev == EOFEvent {
|
|
go eventState.disableEventMonitoring()
|
|
return
|
|
}
|
|
go func(ev *APIEvents) {
|
|
eventState.updateLastSeen(ev)
|
|
eventState.sendEvent(ev)
|
|
}(ev)
|
|
case err = <-eventState.errC:
|
|
if errors.Is(err, ErrNoListeners) {
|
|
eventState.disableEventMonitoring()
|
|
return
|
|
} else if err != nil {
|
|
defer func() { go eventState.monitorEvents(c, opts) }()
|
|
return
|
|
}
|
|
case <-timeout:
|
|
continue
|
|
}
|
|
}
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) connectWithRetry(c *Client, opts EventsOptions) error {
|
|
var retries int
|
|
eventState.RLock()
|
|
eventChan := eventState.C
|
|
errChan := eventState.errC
|
|
eventState.RUnlock()
|
|
closeConn, err := c.eventHijack(opts, atomic.LoadInt64(&eventState.lastSeen), eventChan, errChan)
|
|
for ; err != nil && retries < maxMonitorConnRetries; retries++ {
|
|
waitTime := int64(retryInitialWaitTime * math.Pow(2, float64(retries)))
|
|
time.Sleep(time.Duration(waitTime) * time.Millisecond)
|
|
eventState.RLock()
|
|
eventChan = eventState.C
|
|
errChan = eventState.errC
|
|
eventState.RUnlock()
|
|
closeConn, err = c.eventHijack(opts, atomic.LoadInt64(&eventState.lastSeen), eventChan, errChan)
|
|
}
|
|
eventState.Lock()
|
|
defer eventState.Unlock()
|
|
eventState.closeConn = closeConn
|
|
return err
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) noListeners() bool {
|
|
eventState.RLock()
|
|
defer eventState.RUnlock()
|
|
return len(eventState.listeners) == 0
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) isEnabled() bool {
|
|
eventState.RLock()
|
|
defer eventState.RUnlock()
|
|
return eventState.enabled
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) sendEvent(event *APIEvents) {
|
|
eventState.RLock()
|
|
defer eventState.RUnlock()
|
|
eventState.Add(1)
|
|
defer eventState.Done()
|
|
if eventState.enabled {
|
|
if len(eventState.listeners) == 0 {
|
|
eventState.errC <- ErrNoListeners
|
|
return
|
|
}
|
|
|
|
for _, listener := range eventState.listeners {
|
|
select {
|
|
case listener <- event:
|
|
default:
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
func (eventState *eventMonitoringState) updateLastSeen(e *APIEvents) {
|
|
eventState.Lock()
|
|
defer eventState.Unlock()
|
|
if atomic.LoadInt64(&eventState.lastSeen) < e.Time {
|
|
atomic.StoreInt64(&eventState.lastSeen, e.Time)
|
|
}
|
|
}
|
|
|
|
func (c *Client) eventHijack(opts EventsOptions, startTime int64, eventChan chan *APIEvents, errChan chan error) (closeConn func(), err error) {
|
|
// on reconnect override initial Since with last event seen time
|
|
if startTime != 0 {
|
|
opts.Since = strconv.FormatInt(startTime, 10)
|
|
}
|
|
uri := "/events?" + queryString(opts)
|
|
protocol := c.endpointURL.Scheme
|
|
address := c.endpointURL.Path
|
|
if protocol != "unix" && protocol != "npipe" {
|
|
protocol = "tcp"
|
|
address = c.endpointURL.Host
|
|
}
|
|
var dial net.Conn
|
|
if c.TLSConfig == nil {
|
|
dial, err = c.Dialer.Dial(protocol, address)
|
|
} else {
|
|
netDialer, ok := c.Dialer.(*net.Dialer)
|
|
if !ok {
|
|
return nil, ErrTLSNotSupported
|
|
}
|
|
dial, err = tlsDialWithDialer(netDialer, protocol, address, c.TLSConfig)
|
|
}
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
//lint:ignore SA1019 the alternative doesn't quite work, so keep using the deprecated thing.
|
|
conn := httputil.NewClientConn(dial, nil)
|
|
req, err := http.NewRequest(http.MethodGet, uri, nil)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
res, err := conn.Do(req)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
keepRunning := int32(1)
|
|
//lint:ignore SA1019 the alternative doesn't quite work, so keep using the deprecated thing.
|
|
go func(res *http.Response, conn *httputil.ClientConn) {
|
|
defer conn.Close()
|
|
defer res.Body.Close()
|
|
decoder := json.NewDecoder(res.Body)
|
|
for atomic.LoadInt32(&keepRunning) == 1 {
|
|
var event APIEvents
|
|
if err := decoder.Decode(&event); err != nil {
|
|
if errors.Is(err, io.EOF) || errors.Is(err, io.ErrUnexpectedEOF) {
|
|
c.eventMonitor.RLock()
|
|
if c.eventMonitor.enabled && c.eventMonitor.C == eventChan {
|
|
// Signal that we're exiting.
|
|
eventChan <- EOFEvent
|
|
}
|
|
c.eventMonitor.RUnlock()
|
|
break
|
|
}
|
|
errChan <- err
|
|
}
|
|
if event.Time == 0 {
|
|
continue
|
|
}
|
|
transformEvent(&event)
|
|
c.eventMonitor.RLock()
|
|
if c.eventMonitor.enabled && c.eventMonitor.C == eventChan {
|
|
eventChan <- &event
|
|
}
|
|
c.eventMonitor.RUnlock()
|
|
}
|
|
}(res, conn)
|
|
return func() {
|
|
atomic.StoreInt32(&keepRunning, 0)
|
|
}, nil
|
|
}
|
|
|
|
// transformEvent takes an event and determines what version it is from
|
|
// then populates both versions of the event
|
|
func transformEvent(event *APIEvents) {
|
|
// if event version is <= 1.21 there will be no Action and no Type
|
|
if event.Action == "" && event.Type == "" {
|
|
event.Action = event.Status
|
|
event.Actor.ID = event.ID
|
|
event.Actor.Attributes = map[string]string{}
|
|
switch event.Status {
|
|
case "delete", "import", "pull", "push", "tag", "untag":
|
|
event.Type = "image"
|
|
default:
|
|
event.Type = "container"
|
|
if event.From != "" {
|
|
event.Actor.Attributes["image"] = event.From
|
|
}
|
|
}
|
|
} else {
|
|
if event.Status == "" {
|
|
if event.Type == "image" || event.Type == "container" {
|
|
event.Status = event.Action
|
|
} else {
|
|
// Because just the Status has been overloaded with different Types
|
|
// if an event is not for an image or a container, we prepend the type
|
|
// to avoid problems for people relying on actions being only for
|
|
// images and containers
|
|
event.Status = event.Type + ":" + event.Action
|
|
}
|
|
}
|
|
if event.ID == "" {
|
|
event.ID = event.Actor.ID
|
|
}
|
|
if event.From == "" {
|
|
event.From = event.Actor.Attributes["image"]
|
|
}
|
|
}
|
|
}
|