18992: Fix broken-node-hook global flag.

Arvados-DCO-1.1-Signed-off-by: Tom Clegg <tom@curii.com>

diff --git a/lib/crunchrun/crunchrun.go b/lib/crunchrun/crunchrun.go
index 65f43e96440aa57508cb7e2a80af99e455420120..84b153554ae5e1a13b6c07024c75f9de95fec703 100644 (file)
--- a/lib/crunchrun/crunchrun.go
+++ b/lib/crunchrun/crunchrun.go
@@ -167,6 +167,7 @@ type ContainerRunner struct {
        enableMemoryLimit bool
        enableNetwork     string // one of "default" or "always"
        networkMode       string // "none", "host", or "" -- passed through to executor
        enableMemoryLimit bool
        enableNetwork     string // one of "default" or "always"
        networkMode       string // "none", "host", or "" -- passed through to executor

+       brokenNodeHook    string // script to run if node appears to be broken

        arvMountLog       *ThrottledLogger
 
        containerWatchdogInterval time.Duration
        arvMountLog       *ThrottledLogger
 
        containerWatchdogInterval time.Duration


@@ -210,10 +211,9 @@ var errorBlacklist = []string{
        "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
        "(?ms).*grpc: the connection is unavailable.*",
 }
        "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
        "(?ms).*grpc: the connection is unavailable.*",
 }

-var brokenNodeHook *string = flag.String("broken-node-hook", "", "Script to run if node is detected to be broken (for example, Docker daemon is not running)")

 
 func (runner *ContainerRunner) runBrokenNodeHook() {
 
 func (runner *ContainerRunner) runBrokenNodeHook() {

-       if *brokenNodeHook == "" {
+       if runner.brokenNodeHook == "" {

                path := filepath.Join(lockdir, brokenfile)
                runner.CrunchLog.Printf("Writing %s to mark node as broken", path)
                f, err := os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0700)
                path := filepath.Join(lockdir, brokenfile)
                runner.CrunchLog.Printf("Writing %s to mark node as broken", path)
                f, err := os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0700)


@@ -223,9 +223,9 @@ func (runner *ContainerRunner) runBrokenNodeHook() {
                }
                f.Close()
        } else {
                }
                f.Close()
        } else {

-               runner.CrunchLog.Printf("Running broken node hook %q", *brokenNodeHook)
+               runner.CrunchLog.Printf("Running broken node hook %q", runner.brokenNodeHook)

                // run killme script
                // run killme script

-               c := exec.Command(*brokenNodeHook)
+               c := exec.Command(runner.brokenNodeHook)

                c.Stdout = runner.CrunchLog
                c.Stderr = runner.CrunchLog
                err := c.Run()
                c.Stdout = runner.CrunchLog
                c.Stderr = runner.CrunchLog
                err := c.Run()


@@ -1730,6 +1730,7 @@ func (command) RunCommand(prog string, args []string, stdin io.Reader, stdout, s
        networkMode := flags.String("container-network-mode", "default", `Docker network mode for container (use any argument valid for docker --net)`)
        memprofile := flags.String("memprofile", "", "write memory profile to `file` after running container")
        runtimeEngine := flags.String("runtime-engine", "docker", "container runtime: docker or singularity")
        networkMode := flags.String("container-network-mode", "default", `Docker network mode for container (use any argument valid for docker --net)`)
        memprofile := flags.String("memprofile", "", "write memory profile to `file` after running container")
        runtimeEngine := flags.String("runtime-engine", "docker", "container runtime: docker or singularity")

+       brokenNodeHook := flags.String("broken-node-hook", "", "script to run if node is detected to be broken (for example, Docker daemon is not running)")

        flags.Duration("check-containerd", 0, "Ignored. Exists for compatibility with older versions.")
 
        ignoreDetachFlag := false
        flags.Duration("check-containerd", 0, "Ignored. Exists for compatibility with older versions.")
 
        ignoreDetachFlag := false


@@ -1883,6 +1884,8 @@ func (command) RunCommand(prog string, args []string, stdin io.Reader, stdout, s
        }
        defer cr.executor.Close()
 
        }
        defer cr.executor.Close()
 

+       cr.brokenNodeHook = *brokenNodeHook
+

        gwAuthSecret := os.Getenv("GatewayAuthSecret")
        os.Unsetenv("GatewayAuthSecret")
        if gwAuthSecret == "" {
        gwAuthSecret := os.Getenv("GatewayAuthSecret")
        os.Unsetenv("GatewayAuthSecret")
        if gwAuthSecret == "" {