11package runtime
22
33import (
4+ "bytes"
45 "context"
56 "errors"
67 "fmt"
@@ -15,58 +16,73 @@ import (
1516)
1617
1718const (
18- pidFile = "/var/run/nginx/nginx.pid"
19- pidFileTimeout = 10 * time .Second
19+ pidFile = "/var/run/nginx/nginx.pid"
20+ pidFileTimeout = 10000 * time .Millisecond
21+ childProcsTimeout = 1000 * time .Millisecond
22+ nginxReloadTimeout = 60000 * time .Millisecond
2023)
2124
2225type (
2326 readFileFunc func (string ) ([]byte , error )
2427 checkFileFunc func (string ) (fs.FileInfo , error )
2528)
2629
30+ var (
31+ noNewWorkersErrFmt = "reload unsuccessful: no new NGINX worker processes started for config version %d." +
32+ " Please check the NGINX container logs for possible configuration issues: %w"
33+ childProcPathFmt = "/proc/%[1]v/task/%[1]v/children"
34+ )
35+
2736//go:generate go run github.com/maxbrunsfeld/counterfeiter/v6 . Manager
2837
2938// Manager manages the runtime of NGINX.
3039type Manager interface {
3140 // Reload reloads NGINX configuration. It is a blocking operation.
32- Reload (ctx context.Context ) error
41+ Reload (ctx context.Context , configVersion int ) error
3342}
3443
3544// ManagerImpl implements Manager.
36- type ManagerImpl struct {}
45+ type ManagerImpl struct {
46+ verifyClient * verifyClient
47+ }
3748
3849// NewManagerImpl creates a new ManagerImpl.
3950func NewManagerImpl () * ManagerImpl {
40- return & ManagerImpl {}
51+ return & ManagerImpl {
52+ verifyClient : newVerifyClient (nginxReloadTimeout ),
53+ }
4154}
4255
43- func (m * ManagerImpl ) Reload (ctx context.Context ) error {
56+ func (m * ManagerImpl ) Reload (ctx context.Context , configVersion int ) error {
4457 // We find the main NGINX PID on every reload because it will change if the NGINX container is restarted.
4558 pid , err := findMainProcess (ctx , os .Stat , os .ReadFile , pidFileTimeout )
4659 if err != nil {
4760 return fmt .Errorf ("failed to find NGINX main process: %w" , err )
4861 }
4962
63+ childProcFile := fmt .Sprintf (childProcPathFmt , pid )
64+ previousChildProcesses , err := os .ReadFile (childProcFile )
65+ if err != nil {
66+ return err
67+ }
68+
5069 // send HUP signal to the NGINX main process reload configuration
5170 // See https://nginx.org/en/docs/control.html
5271 if err := syscall .Kill (pid , syscall .SIGHUP ); err != nil {
5372 return fmt .Errorf ("failed to send the HUP signal to NGINX main: %w" , err )
5473 }
5574
56- // FIXME(pleshakov)
57- // (1) ensure the reload actually happens.
58- // https://github.com/nginxinc/nginx-kubernetes-gateway/issues/664
59-
60- // for now, to prevent a subsequent reload starting before the in-flight reload finishes, we simply sleep.
61- // Fixing (1) will make the sleep unnecessary.
62-
63- select {
64- case <- ctx .Done ():
65- return nil
66- case <- time .After (1 * time .Second ):
75+ if err := ensureNewNginxWorkers (
76+ ctx ,
77+ childProcFile ,
78+ previousChildProcesses ,
79+ os .ReadFile ,
80+ childProcsTimeout ,
81+ ); err != nil {
82+ return fmt .Errorf (noNewWorkersErrFmt , configVersion , err )
6783 }
6884
69- return nil
85+ return m . verifyClient . waitForCorrectVersion ( ctx , configVersion )
7086}
7187
7288// EnsureNginxRunning ensures NGINX is running by locating the main process.
@@ -116,3 +132,30 @@ func findMainProcess(
116132
117133 return pid , nil
118134}
135+
136+ func ensureNewNginxWorkers (
137+ ctx context.Context ,
138+ childProcFile string ,
139+ previousContents []byte ,
140+ readFile readFileFunc ,
141+ timeout time.Duration ,
142+ ) error {
143+ ctx , cancel := context .WithTimeout (ctx , timeout )
144+ defer cancel ()
145+
146+ return wait .PollUntilContextCancel (
147+ ctx ,
148+ 25 * time .Millisecond ,
149+ true , /* poll immediately */
150+ func (ctx context.Context ) (bool , error ) {
151+ content , err := readFile (childProcFile )
152+ if err != nil {
153+ return false , err
154+ }
155+ if ! bytes .Equal (previousContents , content ) {
156+ return true , nil
157+ }
158+ return false , nil
159+ },
160+ )
161+ }
0 commit comments