X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/8ed521f7fd1e48e1e415125745ed8c6627a62c91..ff3bb22d4b2bff5666907a6eeb6cd68cd3cbe22b:/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go

diff --git a/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go b/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go
index 617b076da2..d84d461c30 100644
--- a/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go
+++ b/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go
@@ -11,6 +11,7 @@ import (
 	"math"
 	"os"
 	"os/exec"
+	"regexp"
 	"strings"
 	"time"
 
@@ -122,9 +123,30 @@ func doMain() error {
 		log.Printf("Error notifying init daemon: %v", err)
 	}
 
+	go checkSqueueForOrphans(dispatcher, sqCheck)
+
 	return dispatcher.Run(context.Background())
 }
 
+var containerUuidPattern = regexp.MustCompile(`^[a-z0-9]{5}-dz642-[a-z0-9]{15}$`)
+
+// Check the next squeue report, and invoke TrackContainer for all the
+// containers in the report. This gives us a chance to cancel slurm
+// jobs started by a previous dispatch process that never released
+// their slurm allocations even though their container states are
+// Cancelled or Complete. See https://dev.arvados.org/issues/10979
+func checkSqueueForOrphans(dispatcher *dispatch.Dispatcher, sqCheck *SqueueChecker) {
+	for _, uuid := range sqCheck.All() {
+		if !containerUuidPattern.MatchString(uuid) {
+			continue
+		}
+		err := dispatcher.TrackContainer(uuid)
+		if err != nil {
+			log.Printf("checkSqueueForOrphans: TrackContainer(%s): %s", uuid, err)
+		}
+	}
+}
+
 // sbatchCmd
 func sbatchFunc(container arvados.Container) *exec.Cmd {
 	memPerCPU := math.Ceil(float64(container.RuntimeConstraints.RAM) / (float64(container.RuntimeConstraints.VCPUs) * 1048576))