DAOS-15794 tools: Add --health-only flag to (dmg|daos) pool query

As a convenience, provide a "streamlined" version of the pool query that only performs the minimum amount of work to query the pool's health. Practically speaking, this means that it will query for disabled ranks and omit the space query, which is expensive. Features: pool Required-githooks: true Change-Id: I5abe26c2a9a449a9d7c9c0867ae1fff1de9685d5 Signed-off-by: Michael MacDonald <[email protected]>
daos-stack · May 4, 2024 · bd86994 · bd86994
1 parent 140c2b2
commit bd86994
Show file tree

Hide file tree

Showing 19 changed files with 2,668 additions and 1,473 deletions.
diff --git a/src/control/cmd/daos/pool.go b/src/control/cmd/daos/pool.go
@@ -194,6 +194,7 @@ type poolQueryCmd struct {
 	poolBaseCmd
 	ShowEnabledRanks  bool `short:"e" long:"show-enabled" description:"Show engine unique identifiers (ranks) which are enabled"`
 	ShowDisabledRanks bool `short:"b" long:"show-disabled" description:"Show engine unique identifiers (ranks) which are disabled"`
+	HealthOnly        bool `short:"t" long:"health-only" description:"Only perform pool health related queries"`
 }
 
 func convertPoolSpaceInfo(in *C.struct_daos_pool_space, mt C.uint) *daos.StorageUsageStats {
@@ -238,6 +239,7 @@ func convertPoolRebuildStatus(in *C.struct_daos_rebuild_status) *daos.PoolRebuil
 func convertPoolInfo(pinfo *C.daos_pool_info_t) (*daos.PoolInfo, error) {
 	poolInfo := new(daos.PoolInfo)
 
+	poolInfo.QueryMask = daos.PoolQueryMask(pinfo.pi_bits)
 	poolInfo.UUID = uuid.Must(uuidFromC(pinfo.pi_uuid))
 	poolInfo.TotalTargets = uint32(pinfo.pi_ntargets)
 	poolInfo.DisabledTargets = uint32(pinfo.pi_ndisabled)
@@ -251,20 +253,16 @@ func convertPoolInfo(pinfo *C.daos_pool_info_t) (*daos.PoolInfo, error) {
 	}
 
 	poolInfo.Rebuild = convertPoolRebuildStatus(&pinfo.pi_rebuild_st)
-	poolInfo.TierStats = []*daos.StorageUsageStats{
-		convertPoolSpaceInfo(&pinfo.pi_space, C.DAOS_MEDIA_SCM),
-		convertPoolSpaceInfo(&pinfo.pi_space, C.DAOS_MEDIA_NVME),
+	if poolInfo.QueryMask.HasOption("space") {
+		poolInfo.TierStats = []*daos.StorageUsageStats{
+			convertPoolSpaceInfo(&pinfo.pi_space, C.DAOS_MEDIA_SCM),
+			convertPoolSpaceInfo(&pinfo.pi_space, C.DAOS_MEDIA_NVME),
+		}
 	}
 
 	return poolInfo, nil
 }
 
-const (
-	dpiQuerySpace   = C.DPI_SPACE
-	dpiQueryRebuild = C.DPI_REBUILD_STATUS
-	dpiQueryAll     = C.uint64_t(^uint64(0)) // DPI_ALL is -1
-)
-
 func generateRankSet(ranklist *C.d_rank_list_t) string {
 	if ranklist.rl_nr == 0 {
 		return ""
@@ -283,9 +281,19 @@ func generateRankSet(ranklist *C.d_rank_list_t) string {
 }
 
 func (cmd *poolQueryCmd) Execute(_ []string) error {
+	queryMask := daos.DefaultPoolQueryMask
+	if cmd.HealthOnly {
+		if !cmd.ShowEnabledRanks {
+			cmd.ShowDisabledRanks = true // enable for health queries
+		}
+		queryMask.SetQuerySpace(false)
+	}
 	if cmd.ShowEnabledRanks && cmd.ShowDisabledRanks {
 		return errors.New("show-enabled and show-disabled can't be used at the same time.")
 	}
+	queryMask.SetQueryDisabledEngines(cmd.ShowDisabledRanks)
+	queryMask.SetQueryEnabledEngines(cmd.ShowEnabledRanks)
+
 	var rlPtr **C.d_rank_list_t = nil
 	var rl *C.d_rank_list_t = nil
 
@@ -300,10 +308,7 @@ func (cmd *poolQueryCmd) Execute(_ []string) error {
 	defer cleanup()
 
 	cPoolInfo := C.daos_pool_info_t{
-		pi_bits: dpiQueryAll,
-	}
-	if cmd.ShowDisabledRanks {
-		cPoolInfo.pi_bits &= C.uint64_t(^(uint64(C.DPI_ENGINES_ENABLED)))
+		pi_bits: C.uint64_t(queryMask),
 	}
 
 	rc := C.daos_pool_query(cmd.cPoolHandle, rlPtr, &cPoolInfo, nil, nil)
@@ -336,6 +341,7 @@ func (cmd *poolQueryCmd) Execute(_ []string) error {
 		return err
 	}
 
+	cmd.Debugf("Pool query options: %s", poolInfo.QueryMask)
 	cmd.Info(bld.String())
 
 	return nil

diff --git a/src/control/cmd/daos/pretty/pool.go b/src/control/cmd/daos/pretty/pool.go
@@ -46,15 +46,25 @@ func PrintPoolInfo(pi *daos.PoolInfo, out io.Writer) error {
 		fmt.Fprintf(w, "Pool layout out of date (%d < %d) -- see `dmg pool upgrade` for details.\n",
 			pi.PoolLayoutVer, pi.UpgradeLayoutVer)
 	}
-	fmt.Fprintln(w, "Pool space info:")
-	if pi.EnabledRanks != nil {
+	fmt.Fprintln(w, "Pool health info:")
+	if pi.EnabledRanks != nil && pi.EnabledRanks.Count() > 0 {
 		fmt.Fprintf(w, "- Enabled ranks: %s\n", pi.EnabledRanks)
 	}
-	if pi.DisabledRanks != nil {
+	if pi.DisabledRanks != nil && pi.DisabledRanks.Count() > 0 {
 		fmt.Fprintf(w, "- Disabled ranks: %s\n", pi.DisabledRanks)
 	}
-	fmt.Fprintf(w, "- Target(VOS) count:%d\n", pi.ActiveTargets)
-	if pi.TierStats != nil {
+	if pi.Rebuild != nil {
+		if pi.Rebuild.Status == 0 {
+			fmt.Fprintf(w, "- Rebuild %s, %d objs, %d recs\n",
+				pi.Rebuild.State, pi.Rebuild.Objects, pi.Rebuild.Records)
+		} else {
+			fmt.Fprintf(w, "- Rebuild failed, status=%d\n", pi.Rebuild.Status)
+		}
+	}
+
+	if pi.QueryMask.HasOption("space") && pi.TierStats != nil {
+		fmt.Fprintln(w, "Pool space info:")
+		fmt.Fprintf(w, "- Target(VOS) count:%d\n", pi.ActiveTargets)
 		for tierIdx, tierStats := range pi.TierStats {
 			fmt.Fprintln(w, getTierNameText(tierIdx))
 			fmt.Fprintf(w, "  Total size: %s\n", humanize.Bytes(tierStats.Total))
@@ -63,15 +73,6 @@ func PrintPoolInfo(pi *daos.PoolInfo, out io.Writer) error {
 				humanize.Bytes(tierStats.Max), humanize.Bytes(tierStats.Mean))
 		}
 	}
-	if pi.Rebuild != nil {
-		if pi.Rebuild.Status == 0 {
-			fmt.Fprintf(w, "Rebuild %s, %d objs, %d recs\n",
-				pi.Rebuild.State, pi.Rebuild.Objects, pi.Rebuild.Records)
-		} else {
-			fmt.Fprintf(w, "Rebuild failed, status=%d\n", pi.Rebuild.Status)
-		}
-	}
-
 	return w.Err
 }
 

diff --git a/src/control/cmd/daos/pretty/pool_test.go b/src/control/cmd/daos/pretty/pool_test.go
@@ -31,12 +31,12 @@ func TestPretty_PrintPoolInfo(t *testing.T) {
 			pi: &daos.PoolInfo{},
 			expPrintStr: fmt.Sprintf(`
 Pool %s, ntarget=0, disabled=0, leader=0, version=0, state=Creating
-Pool space info:
-- Target(VOS) count:0
+Pool health info:
 `, uuid.Nil.String()),
 		},
 		"normal response": {
 			pi: &daos.PoolInfo{
+				QueryMask:        daos.DefaultPoolQueryMask,
 				State:            daos.PoolServiceStateDegraded,
 				UUID:             poolUUID,
 				TotalTargets:     2,
@@ -65,6 +65,8 @@ Pool space info:
 			expPrintStr: fmt.Sprintf(`
 Pool %s, ntarget=2, disabled=1, leader=42, version=100, state=Degraded
 Pool layout out of date (1 < 2) -- see `+backtickStr+` for details.
+Pool health info:
+- Rebuild busy, 42 objs, 21 recs
 Pool space info:
 - Target(VOS) count:1
 - Storage tier 0 (SCM):
@@ -73,11 +75,11 @@ Pool space info:
 - Storage tier 1 (NVMe):
   Total size: 2 B
   Free: 1 B, min:0 B, max:0 B, mean:0 B
-Rebuild busy, 42 objs, 21 recs
 `, poolUUID.String()),
 		},
 		"normal response; enabled ranks": {
 			pi: &daos.PoolInfo{
+				QueryMask:        daos.DefaultPoolQueryMask,
 				State:            daos.PoolServiceStateDegraded,
 				UUID:             poolUUID,
 				TotalTargets:     2,
@@ -107,20 +109,22 @@ Rebuild busy, 42 objs, 21 recs
 			expPrintStr: fmt.Sprintf(`
 Pool %s, ntarget=2, disabled=1, leader=42, version=100, state=Degraded
 Pool layout out of date (1 < 2) -- see `+backtickStr+` for details.
-Pool space info:
+Pool health info:
 - Enabled ranks: 0-2
+- Rebuild busy, 42 objs, 21 recs
+Pool space info:
 - Target(VOS) count:1
 - Storage tier 0 (SCM):
   Total size: 2 B
   Free: 1 B, min:0 B, max:0 B, mean:0 B
 - Storage tier 1 (NVMe):
   Total size: 2 B
   Free: 1 B, min:0 B, max:0 B, mean:0 B
-Rebuild busy, 42 objs, 21 recs
 `, poolUUID.String()),
 		},
 		"normal response; disabled ranks": {
 			pi: &daos.PoolInfo{
+				QueryMask:        daos.DefaultPoolQueryMask,
 				State:            daos.PoolServiceStateDegraded,
 				UUID:             poolUUID,
 				TotalTargets:     2,
@@ -150,20 +154,22 @@ Rebuild busy, 42 objs, 21 recs
 			expPrintStr: fmt.Sprintf(`
 Pool %s, ntarget=2, disabled=1, leader=42, version=100, state=Degraded
 Pool layout out of date (1 < 2) -- see `+backtickStr+` for details.
-Pool space info:
+Pool health info:
 - Disabled ranks: 0-1,3
+- Rebuild busy, 42 objs, 21 recs
+Pool space info:
 - Target(VOS) count:1
 - Storage tier 0 (SCM):
   Total size: 2 B
   Free: 1 B, min:0 B, max:0 B, mean:0 B
 - Storage tier 1 (NVMe):
   Total size: 2 B
   Free: 1 B, min:0 B, max:0 B, mean:0 B
-Rebuild busy, 42 objs, 21 recs
 `, poolUUID.String()),
 		},
 		"unknown/invalid rebuild state response": {
 			pi: &daos.PoolInfo{
+				QueryMask:        daos.DefaultPoolQueryMask,
 				State:            daos.PoolServiceStateDegraded,
 				UUID:             poolUUID,
 				TotalTargets:     2,
@@ -193,20 +199,22 @@ Rebuild busy, 42 objs, 21 recs
 			expPrintStr: fmt.Sprintf(`
 Pool %s, ntarget=2, disabled=1, leader=42, version=100, state=Degraded
 Pool layout out of date (1 < 2) -- see `+backtickStr+` for details.
-Pool space info:
+Pool health info:
 - Disabled ranks: 0-1,3
+- Rebuild unknown, 42 objs, 21 recs
+Pool space info:
 - Target(VOS) count:1
 - Storage tier 0 (SCM):
   Total size: 2 B
   Free: 1 B, min:0 B, max:0 B, mean:0 B
 - Storage tier 1 (NVMe):
   Total size: 2 B
   Free: 1 B, min:0 B, max:0 B, mean:0 B
-Rebuild unknown, 42 objs, 21 recs
 `, poolUUID.String()),
 		},
 		"rebuild failed": {
 			pi: &daos.PoolInfo{
+				QueryMask:        daos.DefaultPoolQueryMask,
 				State:            daos.PoolServiceStateDegraded,
 				UUID:             poolUUID,
 				TotalTargets:     2,
@@ -236,6 +244,8 @@ Rebuild unknown, 42 objs, 21 recs
 			expPrintStr: fmt.Sprintf(`
 Pool %s, ntarget=2, disabled=1, leader=42, version=100, state=Degraded
 Pool layout out of date (1 < 2) -- see `+backtickStr+` for details.
+Pool health info:
+- Rebuild failed, status=2
 Pool space info:
 - Target(VOS) count:1
 - Storage tier 0 (SCM):
@@ -244,7 +254,6 @@ Pool space info:
 - Storage tier 1 (NVMe):
   Total size: 2 B
   Free: 1 B, min:0 B, max:0 B, mean:0 B
-Rebuild failed, status=2
 `, poolUUID.String()),
 		},
 	} {

diff --git a/src/control/cmd/dmg/pool.go b/src/control/cmd/dmg/pool.go
@@ -1,5 +1,5 @@
 //
-// (C) Copyright 2019-2023 Intel Corporation.
+// (C) Copyright 2019-2024 Intel Corporation.
 //
 // SPDX-License-Identifier: BSD-2-Clause-Patent
 //
@@ -633,20 +633,28 @@ type PoolQueryCmd struct {
 	poolCmd
 	ShowEnabledRanks  bool `short:"e" long:"show-enabled" description:"Show engine unique identifiers (ranks) which are enabled"`
 	ShowDisabledRanks bool `short:"b" long:"show-disabled" description:"Show engine unique identifiers (ranks) which are disabled"`
+	HealthOnly        bool `short:"t" long:"health-only" description:"Only perform pool health related queries"`
 }
 
 // Execute is run when PoolQueryCmd subcommand is activated
 func (cmd *PoolQueryCmd) Execute(args []string) error {
 	req := &control.PoolQueryReq{
-		ID: cmd.PoolID().String(),
+		ID:        cmd.PoolID().String(),
+		QueryMask: daos.DefaultPoolQueryMask,
 	}
 
+	if cmd.HealthOnly {
+		if !cmd.ShowEnabledRanks {
+			cmd.ShowDisabledRanks = true // enable for health queries
+		}
+		req.QueryMask.SetQuerySpace(false)
+	}
 	// TODO (DAOS-10250) The two options should not be incompatible (i.e. engine limitation)
 	if cmd.ShowEnabledRanks && cmd.ShowDisabledRanks {
 		return errIncompatFlags("show-enabled-ranks", "show-disabled-ranks")
 	}
-	req.IncludeEnabledRanks = cmd.ShowEnabledRanks
-	req.IncludeDisabledRanks = cmd.ShowDisabledRanks
+	req.QueryMask.SetQueryEnabledEngines(cmd.ShowEnabledRanks)
+	req.QueryMask.SetQueryDisabledEngines(cmd.ShowDisabledRanks)
 
 	resp, err := control.PoolQuery(cmd.MustLogCtx(), cmd.ctlInvoker, req)
 
@@ -662,6 +670,8 @@ func (cmd *PoolQueryCmd) Execute(args []string) error {
 	if err := pretty.PrintPoolQueryResponse(resp, &bld); err != nil {
 		return err
 	}
+
+	cmd.Debugf("Pool query options: %s", resp.PoolInfo.QueryMask)
 	cmd.Info(bld.String())
 	return nil
 }

diff --git a/src/control/cmd/dmg/pool_test.go b/src/control/cmd/dmg/pool_test.go
@@ -187,6 +187,12 @@ func TestPoolCommands(t *testing.T) {
 		return prop
 	}
 
+	setQueryMask := func(xfrm func(qm *daos.PoolQueryMask)) daos.PoolQueryMask {
+		qm := daos.DefaultPoolQueryMask
+		xfrm(&qm)
+		return qm
+	}
+
 	runCmdTests(t, []cmdTest{
 		{
 			"Pool create with extra argument",
@@ -956,7 +962,8 @@ func TestPoolCommands(t *testing.T) {
 			"pool query 12345678-1234-1234-1234-1234567890ab",
 			strings.Join([]string{
 				printRequest(t, &control.PoolQueryReq{
-					ID: "12345678-1234-1234-1234-1234567890ab",
+					ID:        "12345678-1234-1234-1234-1234567890ab",
+					QueryMask: daos.DefaultPoolQueryMask,
 				}),
 			}, " "),
 			nil,
@@ -966,8 +973,8 @@ func TestPoolCommands(t *testing.T) {
 			"pool query --show-enabled 12345678-1234-1234-1234-1234567890ab",
 			strings.Join([]string{
 				printRequest(t, &control.PoolQueryReq{
-					ID:                  "12345678-1234-1234-1234-1234567890ab",
-					IncludeEnabledRanks: true,
+					ID:        "12345678-1234-1234-1234-1234567890ab",
+					QueryMask: setQueryMask(func(qm *daos.PoolQueryMask) { qm.SetQueryEnabledEngines(true) }),
 				}),
 			}, " "),
 			nil,
@@ -977,8 +984,8 @@ func TestPoolCommands(t *testing.T) {
 			"pool query -e 12345678-1234-1234-1234-1234567890ab",
 			strings.Join([]string{
 				printRequest(t, &control.PoolQueryReq{
-					ID:                  "12345678-1234-1234-1234-1234567890ab",
-					IncludeEnabledRanks: true,
+					ID:        "12345678-1234-1234-1234-1234567890ab",
+					QueryMask: setQueryMask(func(qm *daos.PoolQueryMask) { qm.SetQueryEnabledEngines(true) }),
 				}),
 			}, " "),
 			nil,
@@ -988,8 +995,8 @@ func TestPoolCommands(t *testing.T) {
 			"pool query --show-disabled 12345678-1234-1234-1234-1234567890ab",
 			strings.Join([]string{
 				printRequest(t, &control.PoolQueryReq{
-					ID:                   "12345678-1234-1234-1234-1234567890ab",
-					IncludeDisabledRanks: true,
+					ID:        "12345678-1234-1234-1234-1234567890ab",
+					QueryMask: setQueryMask(func(qm *daos.PoolQueryMask) { qm.SetQueryDisabledEngines(true) }),
 				}),
 			}, " "),
 			nil,
@@ -999,8 +1006,19 @@ func TestPoolCommands(t *testing.T) {
 			"pool query -b 12345678-1234-1234-1234-1234567890ab",
 			strings.Join([]string{
 				printRequest(t, &control.PoolQueryReq{
-					ID:                   "12345678-1234-1234-1234-1234567890ab",
-					IncludeDisabledRanks: true,
+					ID:        "12345678-1234-1234-1234-1234567890ab",
+					QueryMask: setQueryMask(func(qm *daos.PoolQueryMask) { qm.SetQueryDisabledEngines(true) }),
+				}),
+			}, " "),
+			nil,
+		},
+		{
+			"Query pool for health only",
+			"pool query --health-only 12345678-1234-1234-1234-1234567890ab",
+			strings.Join([]string{
+				printRequest(t, &control.PoolQueryReq{
+					ID:        "12345678-1234-1234-1234-1234567890ab",
+					QueryMask: setQueryMask(func(qm *daos.PoolQueryMask) { qm.SetQueryDisabledEngines(true); qm.SetQuerySpace(false) }),
 				}),
 			}, " "),
 			nil,
@@ -1010,7 +1028,8 @@ func TestPoolCommands(t *testing.T) {
 			"pool query test_label",
 			strings.Join([]string{
 				printRequest(t, &control.PoolQueryReq{
-					ID: "test_label",
+					ID:        "test_label",
+					QueryMask: daos.DefaultPoolQueryMask,
 				}),
 			}, " "),
 			nil,