[ARVADOS] created: 1.1.1-58-ge9ce829
Git user
git at public.curoverse.com
Thu Nov 30 15:45:53 EST 2017
at e9ce829350cf4beddc52f0a2c0312151b47b9026 (commit)
commit e9ce829350cf4beddc52f0a2c0312151b47b9026
Author: Peter Amstutz <pamstutz at veritasgenetics.com>
Date: Thu Nov 30 15:45:25 2017 -0500
12573: Translate Arvados priority to slurm niceness.
Arvados-DCO-1.1-Signed-off-by: Peter Amstutz <pamstutz at veritasgenetics.com>
diff --git a/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go b/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go
index 30cbb79..f379099 100644
--- a/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go
+++ b/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go
@@ -151,6 +151,16 @@ func checkSqueueForOrphans(dispatcher *dispatch.Dispatcher, sqCheck *SqueueCheck
}
}
+func niceness(priority int) int {
+ if priority > 1000 {
+ priority = 1000
+ }
+ if priority < 0 {
+ priority = 0
+ }
+ return (1000 - priority) * 1000
+}
+
// sbatchCmd
func sbatchFunc(container arvados.Container) *exec.Cmd {
mem := int64(math.Ceil(float64(container.RuntimeConstraints.RAM+container.RuntimeConstraints.KeepCacheRAM) / float64(1048576)))
@@ -169,6 +179,7 @@ func sbatchFunc(container arvados.Container) *exec.Cmd {
sbatchArgs = append(sbatchArgs, fmt.Sprintf("--mem=%d", mem))
sbatchArgs = append(sbatchArgs, fmt.Sprintf("--cpus-per-task=%d", container.RuntimeConstraints.VCPUs))
sbatchArgs = append(sbatchArgs, fmt.Sprintf("--tmp=%d", disk))
+ sbatchArgs = append(sbatchArgs, fmt.Sprintf("--nice=%d", niceness(container.Priority)))
if len(container.SchedulingParameters.Partitions) > 0 {
sbatchArgs = append(sbatchArgs, fmt.Sprintf("--partition=%s", strings.Join(container.SchedulingParameters.Partitions, ",")))
}
@@ -181,9 +192,15 @@ func scancelFunc(container arvados.Container) *exec.Cmd {
return exec.Command("scancel", "--name="+container.UUID)
}
+// scontrolCmd
+func scontrolFunc(container arvados.Container) *exec.Cmd {
+ return exec.Command("scontrol", "update", "JobName="+container.UUID, fmt.Sprintf("--nice=%d", niceness(container.Priority)))
+}
+
// Wrap these so that they can be overridden by tests
var sbatchCmd = sbatchFunc
var scancelCmd = scancelFunc
+var scontrolCmd = scontrolFunc
// Submit job to slurm using sbatch.
func submit(dispatcher *dispatch.Dispatcher, container arvados.Container, crunchRunCommand []string) error {
@@ -277,6 +294,9 @@ func run(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados
} else if updated.Priority == 0 {
log.Printf("Container %s has state %q, priority %d: cancel slurm job", ctr.UUID, updated.State, updated.Priority)
scancel(ctr)
+ } else if niceness(updated.Priority) != sqCheck.GetNiceness(ctr.UUID) {
+ // dynamically adjust priority
+ scontrolUpdate(updated)
}
}
}
@@ -297,6 +317,21 @@ func scancel(ctr arvados.Container) {
}
}
+func scontrolUpdate(ctr arvados.Container) {
+ sqCheck.L.Lock()
+ cmd := scontrolCmd(ctr)
+ msg, err := cmd.CombinedOutput()
+ sqCheck.L.Unlock()
+
+ if err != nil {
+ log.Printf("%q %q: %s %q", cmd.Path, cmd.Args, err, msg)
+ time.Sleep(time.Second)
+ } else if sqCheck.HasUUID(ctr.UUID) {
+ log.Printf("Container %s priority is now %v, niceness is now %v",
+ ctr.UUID, ctr.Priority, sqCheck.GetNiceness(ctr.UUID))
+ }
+}
+
func readConfig(dst interface{}, path string) error {
err := config.LoadFile(dst, path)
if err != nil && os.IsNotExist(err) && path == defaultConfigPath {
diff --git a/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go b/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go
index 5879f84..58af4a3 100644
--- a/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go
+++ b/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go
@@ -71,7 +71,7 @@ func (s *TestSuite) TestIntegrationNormal(c *C) {
if done {
return exec.Command("true")
} else {
- return exec.Command("echo", "zzzzz-dz642-queuedcontainer")
+ return exec.Command("echo", "zzzzz-dz642-queuedcontainer 999000")
}
},
nil,
@@ -96,7 +96,7 @@ func (s *TestSuite) TestIntegrationCancel(c *C) {
if cmd != nil && cmd.ProcessState != nil {
return exec.Command("true")
} else {
- return exec.Command("echo", "zzzzz-dz642-queuedcontainer")
+ return exec.Command("echo", "zzzzz-dz642-queuedcontainer 999000")
}
},
func(container arvados.Container) *exec.Cmd {
@@ -131,7 +131,8 @@ func (s *TestSuite) TestIntegrationMissingFromSqueue(c *C) {
fmt.Sprintf("--job-name=%s", "zzzzz-dz642-queuedcontainer"),
fmt.Sprintf("--mem=%d", 11445),
fmt.Sprintf("--cpus-per-task=%d", 4),
- fmt.Sprintf("--tmp=%d", 45777)},
+ fmt.Sprintf("--tmp=%d", 45777),
+ fmt.Sprintf("--nice=%d", 999000)},
func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
dispatcher.UpdateState(container.UUID, dispatch.Running)
time.Sleep(3 * time.Second)
@@ -204,6 +205,14 @@ func (s *TestSuite) integrationTest(c *C,
}(scancelCmd)
scancelCmd = newScancelCmd
+ // Override scontrol
+ defer func(orig func(arvados.Container) *exec.Cmd) {
+ scontrolCmd = orig
+ }(scontrolCmd)
+ scontrolCmd = func(container arvados.Container) *exec.Cmd {
+ return exec.Command("true")
+ }
+
// There should be one queued container
params := arvadosclient.Dict{
"filters": [][]string{{"state", "=", "Queued"}},
@@ -377,25 +386,32 @@ func (s *MockArvadosServerSuite) TestSbatchFuncWithConfigArgs(c *C) {
func testSbatchFuncWithArgs(c *C, args []string) {
theConfig.SbatchArguments = append(theConfig.SbatchArguments, args...)
- container := arvados.Container{UUID: "123", RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2}}
+ container := arvados.Container{
+ UUID: "123",
+ RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2},
+ Priority: 1}
sbatchCmd := sbatchFunc(container)
var expected []string
expected = append(expected, "sbatch")
expected = append(expected, theConfig.SbatchArguments...)
- expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=2", "--tmp=0")
+ expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=2", "--tmp=0", "--nice=999000")
c.Check(sbatchCmd.Args, DeepEquals, expected)
}
func (s *MockArvadosServerSuite) TestSbatchPartition(c *C) {
theConfig.SbatchArguments = nil
- container := arvados.Container{UUID: "123", RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 1}, SchedulingParameters: arvados.SchedulingParameters{Partitions: []string{"blurb", "b2"}}}
+ container := arvados.Container{
+ UUID: "123",
+ RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 1},
+ SchedulingParameters: arvados.SchedulingParameters{Partitions: []string{"blurb", "b2"}},
+ Priority: 1}
sbatchCmd := sbatchFunc(container)
var expected []string
expected = append(expected, "sbatch")
- expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=1", "--tmp=0", "--partition=blurb,b2")
+ expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=1", "--tmp=0", "--nice=999000", "--partition=blurb,b2")
c.Check(sbatchCmd.Args, DeepEquals, expected)
}
diff --git a/services/crunch-dispatch-slurm/squeue.go b/services/crunch-dispatch-slurm/squeue.go
index d22e025..3f73b3c 100644
--- a/services/crunch-dispatch-slurm/squeue.go
+++ b/services/crunch-dispatch-slurm/squeue.go
@@ -6,6 +6,7 @@ package main
import (
"bytes"
+ "fmt"
"log"
"os/exec"
"strings"
@@ -17,14 +18,14 @@ import (
// command 'squeue'.
type SqueueChecker struct {
Period time.Duration
- uuids map[string]bool
+ uuids map[string]int
startOnce sync.Once
done chan struct{}
sync.Cond
}
func squeueFunc() *exec.Cmd {
- return exec.Command("squeue", "--all", "--format=%j")
+ return exec.Command("squeue", "--all", "--format=%j %y")
}
var squeueCmd = squeueFunc
@@ -40,7 +41,23 @@ func (sqc *SqueueChecker) HasUUID(uuid string) bool {
// block until next squeue broadcast signaling an update.
sqc.Wait()
- return sqc.uuids[uuid]
+ _, exists := sqc.uuids[uuid]
+ return exists
+}
+
+// GetNiceness returns the niceness of a given uuid, or -1 if it doesn't exist.
+func (sqc *SqueueChecker) GetNiceness(uuid string) int {
+ sqc.startOnce.Do(sqc.start)
+
+ sqc.L.Lock()
+ defer sqc.L.Unlock()
+
+ n, exists := sqc.uuids[uuid]
+ if exists {
+ return n
+ } else {
+ return -1
+ }
}
// Stop stops the squeue monitoring goroutine. Do not call HasUUID
@@ -70,10 +87,15 @@ func (sqc *SqueueChecker) check() {
return
}
- uuids := strings.Split(stdout.String(), "\n")
- sqc.uuids = make(map[string]bool, len(uuids))
- for _, uuid := range uuids {
- sqc.uuids[uuid] = true
+ lines := strings.Split(stdout.String(), "\n")
+ sqc.uuids = make(map[string]int, len(lines))
+ for _, line := range lines {
+ var uuid string
+ var nice int
+ fmt.Sscan(line, &uuid, &nice)
+ if uuid != "" {
+ sqc.uuids[uuid] = nice
+ }
}
sqc.Broadcast()
}
commit 1b949ff836696034183c136a1a3540e284bf3e11
Author: Peter Amstutz <pamstutz at veritasgenetics.com>
Date: Thu Nov 30 14:39:41 2017 -0500
12573: Clamp range of priority to [0,1000] for container & container request.
Arvados-DCO-1.1-Signed-off-by: Peter Amstutz <pamstutz at veritasgenetics.com>
diff --git a/services/api/app/models/container.rb b/services/api/app/models/container.rb
index 83765fb..176ac45 100644
--- a/services/api/app/models/container.rb
+++ b/services/api/app/models/container.rb
@@ -24,6 +24,7 @@ class Container < ArvadosModel
before_validation :fill_field_defaults, :if => :new_record?
before_validation :set_timestamps
validates :command, :container_image, :output_path, :cwd, :priority, :presence => true
+ validates :priority, numericality: { only_integer: true, greater_than_or_equal_to: 0, less_than_or_equal_to: 1000 }
validate :validate_state_change
validate :validate_change
validate :validate_lock
diff --git a/services/api/app/models/container_request.rb b/services/api/app/models/container_request.rb
index 94e4e1f..178c156 100644
--- a/services/api/app/models/container_request.rb
+++ b/services/api/app/models/container_request.rb
@@ -23,6 +23,7 @@ class ContainerRequest < ArvadosModel
before_validation :set_container
validates :command, :container_image, :output_path, :cwd, :presence => true
validates :output_ttl, numericality: { only_integer: true, greater_than_or_equal_to: 0 }
+ validates :priority, numericality: { only_integer: true, greater_than_or_equal_to: 0, less_than_or_equal_to: 1000 }
validate :validate_state_change
validate :check_update_whitelist
after_save :update_priority
@@ -158,6 +159,7 @@ class ContainerRequest < ArvadosModel
self.container_count_max ||= Rails.configuration.container_count_max
self.scheduling_parameters ||= {}
self.output_ttl ||= 0
+ self.priority ||= 500
end
def set_container
diff --git a/services/api/test/unit/container_request_test.rb b/services/api/test/unit/container_request_test.rb
index e751d61..1744584 100644
--- a/services/api/test/unit/container_request_test.rb
+++ b/services/api/test/unit/container_request_test.rb
@@ -41,7 +41,7 @@ class ContainerRequestTest < ActiveSupport::TestCase
cr = create_minimal_req!
assert_nil cr.container_uuid
- assert_nil cr.priority
+ assert_equal 500, cr.priority
check_bogus_states cr
@@ -109,6 +109,7 @@ class ContainerRequestTest < ActiveSupport::TestCase
test "Container request priority must be non-nil" do
set_user_from_auth :active
cr = create_minimal_req!(priority: nil)
+ cr.priority = nil
cr.state = "Committed"
assert_raises(ActiveRecord::RecordInvalid) do
cr.save!
@@ -774,4 +775,35 @@ class ContainerRequestTest < ActiveSupport::TestCase
assert_nothing_raised {cr.destroy}
end
end
+
+ test "Container request valid priority" do
+ set_user_from_auth :active
+ cr = create_minimal_req!
+
+ assert_raises(ActiveRecord::RecordInvalid) do
+ cr.priority = -1
+ cr.save!
+ end
+
+ cr.priority = 0
+ cr.save!
+
+ cr.priority = 1
+ cr.save!
+
+ cr.priority = 500
+ cr.save!
+
+ cr.priority = 999
+ cr.save!
+
+ cr.priority = 1000
+ cr.save!
+
+ assert_raises(ActiveRecord::RecordInvalid) do
+ cr.priority = 1001
+ cr.save!
+ end
+ end
+
end
diff --git a/services/api/test/unit/container_test.rb b/services/api/test/unit/container_test.rb
index 09373fd..ab53240 100644
--- a/services/api/test/unit/container_test.rb
+++ b/services/api/test/unit/container_test.rb
@@ -95,6 +95,42 @@ class ContainerTest < ActiveSupport::TestCase
end
end
+ test "Container request valid priority" do
+ act_as_system_user do
+ cr, _ = minimal_new(environment: {},
+ mounts: {"BAR" => "FOO"},
+ output_path: "/tmp",
+ priority: 1,
+ runtime_constraints: {"vcpus" => 1, "ram" => 1})
+
+ assert_raises(ActiveRecord::RecordInvalid) do
+ cr.priority = -1
+ cr.save!
+ end
+
+ cr.priority = 0
+ cr.save!
+
+ cr.priority = 1
+ cr.save!
+
+ cr.priority = 500
+ cr.save!
+
+ cr.priority = 999
+ cr.save!
+
+ cr.priority = 1000
+ cr.save!
+
+ assert_raises(ActiveRecord::RecordInvalid) do
+ cr.priority = 1001
+ cr.save!
+ end
+ end
+ end
+
+
test "Container serialized hash attributes sorted before save" do
env = {"C" => 3, "B" => 2, "A" => 1}
m = {"F" => {"kind" => 3}, "E" => {"kind" => 2}, "D" => {"kind" => 1}}
-----------------------------------------------------------------------
hooks/post-receive
--
More information about the arvados-commits
mailing list