Merge branch '21678-installer-diagnostics-internal'. Closes #21678
[arvados.git] / services / crunch-dispatch-slurm / squeue_test.go
index c9329fdf95bf87028346fb727b8521dc8edfa1cd..d41e1982b4a91986259e09ea5fc985d7de6e45ce 100644 (file)
@@ -2,11 +2,12 @@
 //
 // SPDX-License-Identifier: AGPL-3.0
 
-package main
+package dispatchslurm
 
 import (
        "time"
 
+       "github.com/sirupsen/logrus"
        . "gopkg.in/check.v1"
 )
 
@@ -24,6 +25,7 @@ func (s *SqueueSuite) TestReleasePending(c *C) {
                queue: uuids[0] + " 10000 4294000000 PENDING Resources\n" + uuids[1] + " 10000 4294000111 PENDING Resources\n" + uuids[2] + " 10000 0 PENDING BadConstraints\n",
        }
        sqc := &SqueueChecker{
+               Logger: logrus.StandardLogger(),
                Slurm:  slurm,
                Period: time.Hour,
        }
@@ -88,6 +90,7 @@ func (s *SqueueSuite) TestReniceAll(c *C) {
                        queue: test.squeue,
                }
                sqc := &SqueueChecker{
+                       Logger:         logrus.StandardLogger(),
                        Slurm:          slurm,
                        PrioritySpread: test.spread,
                        Period:         time.Hour,
@@ -103,6 +106,51 @@ func (s *SqueueSuite) TestReniceAll(c *C) {
        }
 }
 
+// If a limited nice range prevents desired priority adjustments, give
+// up and clamp nice to 10K.
+func (s *SqueueSuite) TestReniceInvalidNiceValue(c *C) {
+       uuids := []string{"zzzzz-dz642-fake0fake0fake0", "zzzzz-dz642-fake1fake1fake1", "zzzzz-dz642-fake2fake2fake2"}
+       slurm := &slurmFake{
+               queue:         uuids[0] + " 0 4294000222 PENDING Resources\n" + uuids[1] + " 0 4294555222 PENDING Resources\n",
+               rejectNice10K: true,
+       }
+       sqc := &SqueueChecker{
+               Logger:         logrus.StandardLogger(),
+               Slurm:          slurm,
+               PrioritySpread: 1,
+               Period:         time.Hour,
+       }
+       sqc.startOnce.Do(sqc.start)
+       sqc.check()
+       sqc.SetPriority(uuids[0], 2)
+       sqc.SetPriority(uuids[1], 1)
+
+       // First attempt should renice to 555001, which will fail
+       sqc.reniceAll()
+       c.Check(slurm.didRenice, DeepEquals, [][]string{{uuids[1], "555001"}})
+
+       // Next attempt should renice to 10K, which will succeed
+       sqc.reniceAll()
+       c.Check(slurm.didRenice, DeepEquals, [][]string{{uuids[1], "555001"}, {uuids[1], "10000"}})
+       // ...so we'll change the squeue response to reflect the
+       // updated priority+nice, and make sure sqc sees that...
+       slurm.queue = uuids[0] + " 0 4294000222 PENDING Resources\n" + uuids[1] + " 10000 4294545222 PENDING Resources\n"
+       sqc.check()
+
+       // Next attempt should leave nice alone because it's already
+       // at the 10K limit
+       sqc.reniceAll()
+       c.Check(slurm.didRenice, DeepEquals, [][]string{{uuids[1], "555001"}, {uuids[1], "10000"}})
+
+       // Back to normal if desired nice value falls below 10K
+       slurm.queue = uuids[0] + " 0 4294000222 PENDING Resources\n" + uuids[1] + " 10000 4294000111 PENDING Resources\n"
+       sqc.check()
+       sqc.reniceAll()
+       c.Check(slurm.didRenice, DeepEquals, [][]string{{uuids[1], "555001"}, {uuids[1], "10000"}, {uuids[1], "9890"}})
+
+       sqc.Stop()
+}
+
 // If the given UUID isn't in the slurm queue yet, SetPriority()
 // should wait for it to appear on the very next poll, then give up.
 func (s *SqueueSuite) TestSetPriorityBeforeQueued(c *C) {
@@ -111,6 +159,7 @@ func (s *SqueueSuite) TestSetPriorityBeforeQueued(c *C) {
 
        slurm := &slurmFake{}
        sqc := &SqueueChecker{
+               Logger: logrus.StandardLogger(),
                Slurm:  slurm,
                Period: time.Hour,
        }