14203: Mkdir failing with os.ErrExist is not a fatal error

Don't fail CaptureOutput and cancel the container if a Mkdir returns
os.ErrExist because a directory already exists.  Fixes bcbio failing
postprocess_variants step.

Arvados-DCO-1.1-Signed-off-by: Peter Amstutz <pamstutz@veritasgenetics.com>

diff --git a/services/crunch-run/copier.go b/services/crunch-run/copier.go
index 4c45f6acb9bed4162dc9eaad28f7f6af82715b81..4a6253c459de53808203bd97499da561407b07a5 100644 (file)
--- a/services/crunch-run/copier.go
+++ b/services/crunch-run/copier.go
@@ -70,22 +70,22 @@ type copier struct {
 func (cp *copier) Copy() (string, error) {
        err := cp.walkMount("", cp.ctrOutputDir, limitFollowSymlinks, true)
        if err != nil {
-               return "", err
+               return "", fmt.Errorf("in walkMount: %v", err)
        }
        fs, err := (&arvados.Collection{ManifestText: cp.manifest}).FileSystem(cp.client, cp.keepClient)
        if err != nil {
-               return "", err
+               return "", fmt.Errorf("creating Collection.FileSystem: %v", err)
        }
        for _, d := range cp.dirs {
                err = fs.Mkdir(d, 0777)
-               if err != nil {
-                       return "", err
+               if err != nil && err != os.ErrExist {
+                       return "", fmt.Errorf("Could not Mkdir %v: %v", d, err)
                }
        }
        for _, f := range cp.files {
                err = cp.copyFile(fs, f)
                if err != nil {
-                       return "", err
+                       return "", fmt.Errorf("Could not copyFile %v: %v", f, err)
                }
        }
        return fs.MarshalManifest(".")

diff --git a/services/crunch-run/crunchrun.go b/services/crunch-run/crunchrun.go
index 0a980b9ce9359cde4529928e78eba06ac74644f3..df7b68833ac7e71022fb95d4f21337708848b1bd 100644 (file)
--- a/services/crunch-run/crunchrun.go
+++ b/services/crunch-run/crunchrun.go
@@ -1428,11 +1428,11 @@ func (runner *ContainerRunner) Run() (err error) {
                // hasn't already been assigned when Run() returns,
                // this cleanup func will cause Run() to return the
                // first non-nil error that is passed to checkErr().
-               checkErr := func(e error) {
+               checkErr := func(prefix string, e error) {
                        if e == nil {
                                return
                        }
-                       runner.CrunchLog.Print(e)
+                       runner.CrunchLog.Printf("%s error: %v", prefix, e)
                        if err == nil {
                                err = e
                        }
@@ -1443,7 +1443,7 @@ func (runner *ContainerRunner) Run() (err error) {
                }
 
                // Log the error encountered in Run(), if any
-               checkErr(err)
+               checkErr("Run", err)
 
                if runner.finalState == "Queued" {
                        runner.UpdateContainerFinal()
@@ -1456,10 +1456,10 @@ func (runner *ContainerRunner) Run() (err error) {
                        // capture partial output and write logs
                }
 
-               checkErr(runner.CaptureOutput())
-               checkErr(runner.stopHoststat())
-               checkErr(runner.CommitLogs())
-               checkErr(runner.UpdateContainerFinal())
+               checkErr("CaptureOutput", runner.CaptureOutput())
+               checkErr("stopHoststat", runner.stopHoststat())
+               checkErr("CommitLogs", runner.CommitLogs())
+               checkErr("UpdateContainerFinal", runner.UpdateContainerFinal())
        }()
 
        err = runner.fetchContainerRecord()