DAOS-14408 common: enable NDCTL for DCPM

.github/workflows/landing-builds.yml

@@ -16,6 +16,7 @@ on:
       - ci/**
       - requirements-build.txt
       - requirements-utest.txt
+      - utils/build.config
 
 permissions: {}
 

debian/changelog

@@ -1,3 +1,19 @@
+daos (2.7.100-7) unstable; urgency=medium
+  [ Tomasz Gromadzki ]
+  * Add support of the PMDK package 2.1.0 with NDCTL enabled.
+    * Increase the default ULT stack size to 20KiB if the engine uses
+      the DCPM storage class.
+    * Prevent using the RAM storage class (simulated PMem) when
+      the shutdown state (SDS) is active.
+      * Automatically disable SDS for the RAM storage class on engine startup.
+      * Force explicitly setting the PMEMOBJ_CONF='sds.at_create=0'
+        environment variable to deactivate SDS for the DAOS tools
+        (ddb, daos_perf, vos_perf, etc.) when used WITHOUT DCPM.
+        Otherwise, a user is supposed to be stopped by an error
+        like: "Unsafe shutdown count is not supported for this source".
+
+ -- Tomasz Gromadzki <tomasz.gromadzki@intel.com>  Tue, 02 Oct 2024 12:00:00 +0200
+
 daos (2.7.100-6) unstable; urgency=medium
   [ Kris Jacque ]
   * Bump minimum golang-go version to 1.21

debian/control

@@ -18,7 +18,7 @@ Build-Depends: debhelper (>= 10),
                python3-distro,
                libabt-dev,
                libucx-dev,
-               libpmemobj-dev (>= 2.0.0),
+               libpmemobj-dev (>= 2.1.0),
                libfuse3-dev,
                libprotobuf-c-dev,
                libjson-c-dev,
@@ -118,7 +118,9 @@ Depends: python (>=3.8), python3, python-yaml, python3-yaml,
          daos-client (= ${binary:Version}),
          daos-admin (= ${binary:Version}),
          golang-go (>= 2:1.21),
-         libcapstone-dev
+         libcapstone-dev,
+         libndctl-dev,
+         libdaxctl-dev
 Description: The Distributed Asynchronous Object Storage (DAOS) is an open-source
  software-defined object store designed from the ground up for
  massively distributed Non Volatile Memory (NVM). DAOS takes advantage

site_scons/components/__init__.py

@@ -266,7 +266,6 @@ def define_components(reqs):
                 retriever=GitRepoRetriever(),
                 commands=[['make',
                            'all',
-                           'NDCTL_ENABLE=n',
                            'BUILD_EXAMPLES=n',
                            'BUILD_BENCHMARKS=n',
                            'DOC=n',

src/control/server/engine/config.go

@@ -9,6 +9,7 @@ package engine
 import (
 	"fmt"
 	"os"
+	"strconv"
 	"strings"
 
 	"github.com/pkg/errors"
@@ -28,6 +29,8 @@ const (
 	envLogMasks      = "D_LOG_MASK"
 	envLogDbgStreams = "DD_MASK"
 	envLogSubsystems = "DD_SUBSYS"
+
+	minABTThreadStackSizeDCPM = 20480
 )
 
 // FabricConfig encapsulates networking fabric configuration.
@@ -342,7 +345,80 @@ func (c *Config) Validate() error {
 	if err := ValidateLogSubsystems(subsystems); err != nil {
 		return errors.Wrap(err, "validate engine log subsystems")
 	}
+	return nil
+}
+
+// Ensure at least 20KiB ABT stack size for an engine with DCPM storage class.
+func (c *Config) UpdatePMDKEnvarsStackSizeDCPM() error {
+	stackSizeStr, err := c.GetEnvVar("ABT_THREAD_STACKSIZE")
+	if err != nil {
+		c.EnvVars = append(c.EnvVars, fmt.Sprintf("ABT_THREAD_STACKSIZE=%d",
+			minABTThreadStackSizeDCPM))
+		return nil
+	}
+	// Ensure at least 20KiB ABT stack size for an engine with DCPM storage class.
+	stackSizeValue, err := strconv.Atoi(stackSizeStr)
+	if err != nil {
+		return errors.Errorf("env_var ABT_THREAD_STACKSIZE has invalid value: %s",
+			stackSizeStr)
+	}
+	if stackSizeValue < minABTThreadStackSizeDCPM {
+		return errors.Errorf("env_var ABT_THREAD_STACKSIZE should be >= %d "+
+			"for DCPM storage class, found %d", minABTThreadStackSizeDCPM,
+			stackSizeValue)
+	}
+	return nil
+}
+
+// Ensure proper configuration of shutdown (SDS) state
+func (c *Config) UpdatePMDKEnvarsPMemobjConf(isDCPM bool) error {
+	pmemobjConfStr, pmemobjConfErr := c.GetEnvVar("PMEMOBJ_CONF")
+	//also work for empty string
+	hasSdsAtCreate := strings.Contains(pmemobjConfStr, "sds.at_create")
+	if isDCPM {
+		if !hasSdsAtCreate {
+			return nil
+		}
+		// Confirm default handling of shutdown state (SDS) for DCPM storage class.
+		return errors.New("env_var PMEMOBJ_CONF should NOT contain 'sds.at_create=?' " +
+			"for DCPM storage class, found '" + pmemobjConfStr + "'")
+	}
+
+	// Disable shutdown state (SDS) (part of RAS) for RAM-based simulated SCM.
+	if pmemobjConfErr != nil {
+		c.EnvVars = append(c.EnvVars, "PMEMOBJ_CONF=sds.at_create=0")
+		return nil
+	}
+	if !hasSdsAtCreate {
+		envVars, _ := common.DeleteKeyValue(c.EnvVars, "PMEMOBJ_CONF")
+		c.EnvVars = append(envVars, "PMEMOBJ_CONF="+pmemobjConfStr+
+			";sds.at_create=0")
+		return nil
+	}
+	if strings.Contains(pmemobjConfStr, "sds.at_create=1") {
+		return errors.New("env_var PMEMOBJ_CONF should contain 'sds.at_create=0' " +
+			"for non-DCPM storage class, found '" + pmemobjConfStr + "'")
+	}
+	return nil
+}
+
+// Ensure proper environment variables for PMDK w/ NDCTL enabled based on
+// the actual configuration of the storage class.
+func (c *Config) UpdatePMDKEnvars() error {
+
+	if len(c.Storage.Tiers) == 0 {
+		return errors.New("Invalid config - no tier 0 defined")
+	}
+
+	isDCPM := c.Storage.Tiers[0].Class == storage.ClassDcpm
 
+	if err := c.UpdatePMDKEnvarsPMemobjConf(isDCPM); err != nil {
+		return err
+	}
+
+	if isDCPM {
+		return c.UpdatePMDKEnvarsStackSizeDCPM()
+	}
 	return nil
 }
 
@@ -690,3 +766,13 @@ func (c *Config) WithStorageIndex(i uint32) *Config {
 	c.Storage.EngineIdx = uint(i)
 	return c
 }
+
+// WithEnvVarAbtThreadStackSize sets environment variable ABT_THREAD_STACKSIZE.
+func (c *Config) WithEnvVarAbtThreadStackSize(stack_size uint16) *Config {
+	return c.WithEnvVars(fmt.Sprintf("ABT_THREAD_STACKSIZE=%d", stack_size))
+}
+
+// WithEnvVarPMemObjSdsAtCreate sets PMEMOBJ_CONF env. var. to sds.at_create=0/1 value
+func (c *Config) WithEnvVarPMemObjSdsAtCreate(value uint8) *Config {
+	return c.WithEnvVars(fmt.Sprintf("PMEMOBJ_CONF=sds.at_create=%d", value))
+}

src/control/server/engine/config_test.go

@@ -1104,3 +1104,210 @@ func TestFabricConfig_Update(t *testing.T) {
 		})
 	}
 }
+
+func TestConfig_UpdatePMDKEnvarsStackSizeDCPM(t *testing.T) {
+	validConfig := func() *Config {
+		return MockConfig().WithStorage(
+			storage.NewTierConfig().
+				WithStorageClass("dcpm"))
+	}
+
+	for name, tc := range map[string]struct {
+		cfg                   *Config
+		expErr                error
+		expABTthreadStackSize int
+	}{
+		"empty config should not fail": {
+			cfg:                   MockConfig(),
+			expABTthreadStackSize: minABTThreadStackSizeDCPM,
+		},
+		"valid config for DCPM should not fail": {
+			cfg:                   validConfig().WithEnvVarAbtThreadStackSize(minABTThreadStackSizeDCPM),
+			expABTthreadStackSize: minABTThreadStackSizeDCPM,
+		},
+		"config for DCPM without thread size should not fail": {
+			cfg:                   validConfig(),
+			expABTthreadStackSize: minABTThreadStackSizeDCPM,
+		},
+		"config for DCPM with stack size big enough should not fail": {
+			cfg: validConfig().
+				WithEnvVarAbtThreadStackSize(minABTThreadStackSizeDCPM + 1),
+			expABTthreadStackSize: minABTThreadStackSizeDCPM + 1,
+		},
+		"config for DCPM with stack size too small should fail": {
+			cfg: validConfig().
+				WithEnvVarAbtThreadStackSize(minABTThreadStackSizeDCPM - 1),
+			expErr: errors.New(fmt.Sprintf("env_var ABT_THREAD_STACKSIZE "+
+				"should be >= %d for DCPM storage class, found %d",
+				minABTThreadStackSizeDCPM, minABTThreadStackSizeDCPM-1)),
+		},
+		"config for DCPM with invalid ABT_THREAD_STACKSIZE value should fail": {
+			cfg:    validConfig().WithEnvVars("ABT_THREAD_STACKSIZE=foo_bar"),
+			expErr: errors.New("env_var ABT_THREAD_STACKSIZE has invalid value: foo_bar"),
+		},
+	} {
+		t.Run(name, func(t *testing.T) {
+			err := tc.cfg.UpdatePMDKEnvarsStackSizeDCPM()
+			test.CmpErr(t, tc.expErr, err)
+			if err == nil {
+				stackSizeStr, err := tc.cfg.GetEnvVar("ABT_THREAD_STACKSIZE")
+				test.AssertTrue(t, err == nil, "Missing env var ABT_THREAD_STACKSIZE")
+				stackSizeVal, err := strconv.Atoi(stackSizeStr)
+				test.AssertTrue(t, err == nil, "Invalid env var ABT_THREAD_STACKSIZE")
+				test.AssertEqual(t, tc.expABTthreadStackSize, stackSizeVal,
+					"Invalid ABT_THREAD_STACKSIZE value")
+			}
+		})
+	}
+}
+
+func TestConfig_UpdatePMDKEnvarsPMemobjConfDCPM(t *testing.T) {
+	validConfig := func() *Config {
+		return MockConfig().WithStorage(
+			storage.NewTierConfig().WithStorageClass("dcpm"))
+	}
+
+	for name, tc := range map[string]struct {
+		cfg    *Config
+		expErr error
+	}{
+		"empty config should not fail": {
+			cfg: MockConfig(),
+		},
+		"valid config for DCPM should not fail": {
+			cfg: validConfig(),
+		},
+		"config for DCPM with forced sds.at_create (1) should fail": {
+			cfg: validConfig().WithEnvVarPMemObjSdsAtCreate(1),
+			expErr: errors.New("env_var PMEMOBJ_CONF should NOT contain " +
+				"'sds.at_create=?' for DCPM storage class, found 'sds.at_create=1'"),
+		},
+		"config for DCPM with forced sds.at_create (0) should fail": {
+			cfg: validConfig().WithEnvVarPMemObjSdsAtCreate(0),
+			expErr: errors.New("env_var PMEMOBJ_CONF should NOT contain " +
+				"'sds.at_create=?' for DCPM storage class, found 'sds.at_create=0'"),
+		},
+	} {
+		t.Run(name, func(t *testing.T) {
+			test.CmpErr(t, tc.expErr, tc.cfg.UpdatePMDKEnvarsPMemobjConf(true))
+		})
+	}
+}
+
+func TestConfig_UpdatePMDKEnvarsPMemobjConfNRam(t *testing.T) {
+	validConfig := func() *Config {
+		return MockConfig().WithStorage(
+			storage.NewTierConfig().
+				WithStorageClass("dcpm"))
+	}
+
+	for name, tc := range map[string]struct {
+		cfg             *Config
+		expErr          error
+		expPMEMOBJ_CONF string
+	}{
+		"empty config should not fail": {
+			cfg:             validConfig(),
+			expPMEMOBJ_CONF: "sds.at_create=0",
+		},
+		"config for ram without PMEMOBJ_CONF should not fail": {
+			cfg:             MockConfig(),
+			expPMEMOBJ_CONF: "sds.at_create=0",
+		},
+		"valid config for should not fail": {
+			cfg:             validConfig().WithEnvVarPMemObjSdsAtCreate(0),
+			expPMEMOBJ_CONF: "sds.at_create=0",
+		},
+		"config for ram w/ PMEMOBJ_CONF w/o sds.at_create should should be updated": {
+			cfg:             validConfig().WithEnvVars("PMEMOBJ_CONF=foo_bar"),
+			expPMEMOBJ_CONF: "foo_bar;sds.at_create=0",
+		},
+		"config for ram with sds.at_create set to 1 should fail": {
+			cfg: validConfig().WithEnvVarPMemObjSdsAtCreate(1),
+			expErr: errors.New("env_var PMEMOBJ_CONF should contain " +
+				"'sds.at_create=0' for non-DCPM storage class" +
+				", found 'sds.at_create=1'"),
+		},
+		"config for ram w/ PMEMOBJ_CONF w/ sds.at_create=1 should fail": {
+			cfg: validConfig().
+				WithEnvVars("PMEMOBJ_CONF=sds.at_create=1;foo-bar"),
+			expErr: errors.New("env_var PMEMOBJ_CONF should contain " +
+				"'sds.at_create=0' for non-DCPM storage class" +
+				", found 'sds.at_create=1;foo-bar'"),
+		},
+	} {
+		t.Run(name, func(t *testing.T) {
+			test.CmpErr(t, tc.expErr, tc.cfg.UpdatePMDKEnvarsPMemobjConf(false))
+			if len(tc.expPMEMOBJ_CONF) > 0 {
+				sds_at_create, err := tc.cfg.GetEnvVar("PMEMOBJ_CONF")
+				test.AssertTrue(t, err == nil, "Missing env var PMEMOBJ_CONF")
+				test.AssertEqual(t, tc.expPMEMOBJ_CONF, sds_at_create,
+					"Invalid PMEMOBJ_CONF")
+			}
+
+		})
+	}
+}
+
+func TestConfig_UpdatePMDKEnvars(t *testing.T) {
+	validConfig := func(storageclas string) *Config {
+		return MockConfig().WithStorage(
+			storage.NewTierConfig().
+				WithStorageClass(storageclas))
+	}
+	for name, tc := range map[string]struct {
+		cfg                   *Config
+		expErr                error
+		expPMEMOBJ_CONF       string
+		expABTthreadStackSize int
+	}{
+		"empty config should fail": {
+			cfg:                   MockConfig(),
+			expErr:                errors.New("Invalid config - no tier 0 defined"),
+			expABTthreadStackSize: -1,
+		},
+		"valid config for RAM should not fail": {
+			cfg: validConfig("ram").
+				WithEnvVarAbtThreadStackSize(minABTThreadStackSizeDCPM - 1),
+			expPMEMOBJ_CONF:       "sds.at_create=0",
+			expABTthreadStackSize: minABTThreadStackSizeDCPM - 1,
+		},
+		"invalid config for RAM should fail": {
+			cfg: validConfig("ram").WithEnvVarPMemObjSdsAtCreate(1),
+			expErr: errors.New("env_var PMEMOBJ_CONF should contain " +
+				"'sds.at_create=0' for non-DCPM storage class, " +
+				"found 'sds.at_create=1'"),
+			expABTthreadStackSize: -1,
+		},
+		"valid config for DCPM should not fail": {
+			cfg:                   validConfig("dcpm"),
+			expABTthreadStackSize: minABTThreadStackSizeDCPM,
+		},
+		"invalid config for DCPM should not fail": {
+			cfg: validConfig("dcpm").
+				WithEnvVarAbtThreadStackSize(minABTThreadStackSizeDCPM - 1),
+			expErr: errors.New("env_var ABT_THREAD_STACKSIZE should be >= 20480 " +
+				"for DCPM storage class, found 20479"),
+			expABTthreadStackSize: minABTThreadStackSizeDCPM - 1,
+		},
+	} {
+		t.Run(name, func(t *testing.T) {
+			errTc := tc.cfg.UpdatePMDKEnvars()
+			test.CmpErr(t, tc.expErr, errTc)
+			if len(tc.expPMEMOBJ_CONF) > 0 {
+				sds_at_create, err := tc.cfg.GetEnvVar("PMEMOBJ_CONF")
+				test.AssertTrue(t, err == nil, "Missing env var PMEMOBJ_CONF")
+				test.AssertEqual(t, tc.expPMEMOBJ_CONF, sds_at_create,
+					"Invalid PMEMOBJ_CONF")
+			}
+			if tc.expABTthreadStackSize >= 0 {
+				stackSizeStr, err := tc.cfg.GetEnvVar("ABT_THREAD_STACKSIZE")
+				test.AssertTrue(t, err == nil, "Missing env var ABT_THREAD_STACKSIZE")
+				stackSizeVal, err := strconv.Atoi(stackSizeStr)
+				test.AssertTrue(t, err == nil, "Invalid env var ABT_THREAD_STACKSIZE")
+				test.AssertEqual(t, tc.expABTthreadStackSize, stackSizeVal,
+					"Invalid ABT_THREAD_STACKSIZE value")
+			}
+		})
+	}
+}

src/control/server/server.go

@@ -104,6 +104,12 @@ func processConfig(log logging.Logger, cfg *config.Server, fis *hardware.FabricI
 		return err
 	}
 
+	for _, ec := range cfg.Engines {
+		if err := ec.UpdatePMDKEnvars(); err != nil {
+			return err
+		}
+	}
+
 	return nil
 }