[backward compatibility] v3.22 and prior; bump CLI version

* keep v3.22 control structures - denote all of them - attach "v322" or "V322" suffix * REST APIs: node and cluster stats; node and cluster status * refactor `api` pkg - extract stats.go * remove redundant code; simplify Signed-off-by: Alex Aizman <[email protected]>
NVIDIA · Mar 24, 2024 · 85e15b9 · 85e15b9
1 parent bcd7dca
commit 85e15b9
Show file tree

Hide file tree

Showing 29 changed files with 324 additions and 212 deletions.
diff --git a/ais/htrun.go b/ais/htrun.go
@@ -23,6 +23,7 @@ import (
 	"time"
 
 	"github.com/NVIDIA/aistore/api/apc"
+	"github.com/NVIDIA/aistore/api/env"
 	"github.com/NVIDIA/aistore/cmn"
 	"github.com/NVIDIA/aistore/cmn/archive"
 	"github.com/NVIDIA/aistore/cmn/atomic"
@@ -1045,15 +1046,64 @@ func (h *htrun) httpdaeget(w http.ResponseWriter, r *http.Request, query url.Val
 		statsNode := h.statsT.GetStats()
 		statsNode.Snode = h.si
 		body = statsNode
+	case apc.WhatNodeStatsV322:
+		statsNode := h.statsT.GetStatsV322()
+		statsNode.Snode = h.si
+		body = statsNode
 	case apc.WhatMetricNames:
 		body = h.statsT.GetMetricNames()
+	case apc.WhatNodeStatsAndStatus:
+		ds := h.statsAndStatus()
+		daeStats := h.statsT.GetStats()
+		ds.Tracker = daeStats.Tracker
+		body = ds
+	case apc.WhatNodeStatsAndStatusV322:
+		ds := h.statsAndStatusV322()
+		daeStats := h.statsT.GetStatsV322()
+		ds.Tracker = daeStats.Tracker
+		body = ds
 	default:
 		h.writeErrf(w, r, "invalid GET /daemon request: unrecognized what=%s", what)
 		return
 	}
 	h.writeJSON(w, r, body, "httpdaeget-"+what)
 }
 
+func (h *htrun) statsAndStatus() (ds *stats.NodeStatus) {
+	smap := h.owner.smap.get()
+	ds = &stats.NodeStatus{
+		Node: stats.Node{
+			Snode: h.si,
+		},
+		SmapVersion:    smap.Version,
+		MemCPUInfo:     apc.GetMemCPU(),
+		DeploymentType: deploymentType(),
+		Version:        daemon.version,
+		BuildTime:      daemon.buildTime,
+		K8sPodName:     os.Getenv(env.AIS.K8sPod),
+		Status:         h._status(smap),
+	}
+	return ds
+}
+
+// [backward compatibility] v3.22 and prior
+func (h *htrun) statsAndStatusV322() (ds *stats.NodeStatusV322) {
+	smap := h.owner.smap.get()
+	ds = &stats.NodeStatusV322{
+		NodeV322: stats.NodeV322{
+			Snode: h.si,
+		},
+		SmapVersion:    smap.Version,
+		MemCPUInfo:     apc.GetMemCPU(),
+		DeploymentType: deploymentType(),
+		Version:        daemon.version,
+		BuildTime:      daemon.buildTime,
+		K8sPodName:     os.Getenv(env.AIS.K8sPod),
+		Status:         h._status(smap),
+	}
+	return ds
+}
+
 func (h *htrun) sendAllLogs(w http.ResponseWriter, r *http.Request, query url.Values) string {
 	sev := query.Get(apc.QparamLogSev)
 	tempdir, archname, err := h.targzLogs(sev)

diff --git a/ais/proxy.go b/ais/proxy.go
@@ -21,7 +21,6 @@ import (
 	"time"
 
 	"github.com/NVIDIA/aistore/api/apc"
-	"github.com/NVIDIA/aistore/api/env"
 	"github.com/NVIDIA/aistore/cmn"
 	"github.com/NVIDIA/aistore/cmn/archive"
 	"github.com/NVIDIA/aistore/cmn/atomic"
@@ -2450,7 +2449,8 @@ func (p *proxy) httpdaeget(w http.ResponseWriter, r *http.Request) {
 		}
 		fallthrough // fallthrough
 	case apc.WhatNodeConfig, apc.WhatSmapVote, apc.WhatSnode, apc.WhatLog,
-		apc.WhatNodeStats, apc.WhatMetricNames:
+		apc.WhatNodeStats, apc.WhatNodeStatsV322, apc.WhatMetricNames,
+		apc.WhatNodeStatsAndStatus, apc.WhatNodeStatsAndStatusV322:
 		p.htrun.httpdaeget(w, r, query, nil /*htext*/)
 	case apc.WhatSysInfo:
 		p.writeJSON(w, r, apc.GetMemCPU(), what)
@@ -2478,24 +2478,6 @@ func (p *proxy) httpdaeget(w http.ResponseWriter, r *http.Request) {
 			return
 		}
 		p.writeJSON(w, r, smap, what)
-	case apc.WhatNodeStatsAndStatus:
-		smap := p.owner.smap.get()
-		msg := &stats.NodeStatus{
-			Node: stats.Node{
-				Snode: p.htrun.si,
-			},
-			SmapVersion:    smap.Version,
-			MemCPUInfo:     apc.GetMemCPU(),
-			DeploymentType: deploymentType(),
-			Version:        daemon.version,
-			BuildTime:      daemon.buildTime,
-			K8sPodName:     os.Getenv(env.AIS.K8sPod),
-			Status:         p._status(smap),
-		}
-		daeStats := p.statsT.GetStats()
-		msg.Tracker = daeStats.Tracker
-
-		p.writeJSON(w, r, msg, what)
 	default:
 		p.htrun.httpdaeget(w, r, query, nil /*htext*/)
 	}

diff --git a/ais/prxclu.go b/ais/prxclu.go
@@ -74,7 +74,7 @@ func (p *proxy) httpcluget(w http.ResponseWriter, r *http.Request) {
 		p.xquery(w, r, what, query)
 	case apc.WhatAllRunningXacts:
 		p.xgetRunning(w, r, what, query)
-	case apc.WhatNodeStats:
+	case apc.WhatNodeStats, apc.WhatNodeStatsV322:
 		p.qcluStats(w, r, what, query)
 	case apc.WhatSysInfo:
 		p.qcluSysinfo(w, r, what, query)

diff --git a/ais/tgtcp.go b/ais/tgtcp.go
@@ -9,15 +9,13 @@ import (
 	"fmt"
 	"net/http"
 	"net/url"
-	"os"
 	"path/filepath"
 	"strconv"
 	"sync"
 	"time"
 
 	"github.com/NVIDIA/aistore/ais/backend"
 	"github.com/NVIDIA/aistore/api/apc"
-	"github.com/NVIDIA/aistore/api/env"
 	"github.com/NVIDIA/aistore/cmn"
 	"github.com/NVIDIA/aistore/cmn/cos"
 	"github.com/NVIDIA/aistore/cmn/debug"
@@ -32,7 +30,6 @@ import (
 	"github.com/NVIDIA/aistore/nl"
 	"github.com/NVIDIA/aistore/reb"
 	"github.com/NVIDIA/aistore/res"
-	"github.com/NVIDIA/aistore/stats"
 	"github.com/NVIDIA/aistore/xact"
 	"github.com/NVIDIA/aistore/xact/xreg"
 	jsoniter "github.com/json-iterator/go"
@@ -287,40 +284,40 @@ func (t *target) httpdaeget(w http.ResponseWriter, r *http.Request) {
 	)
 	switch getWhat {
 	case apc.WhatNodeConfig, apc.WhatSmap, apc.WhatBMD, apc.WhatSmapVote,
-		apc.WhatSnode, apc.WhatLog, apc.WhatNodeStats, apc.WhatMetricNames:
+		apc.WhatSnode, apc.WhatLog, apc.WhatMetricNames:
 		t.htrun.httpdaeget(w, r, query, t /*htext*/)
 	case apc.WhatSysInfo:
 		tsysinfo := apc.TSysInfo{MemCPUInfo: apc.GetMemCPU(), CapacityInfo: fs.CapStatusGetWhat()}
 		t.writeJSON(w, r, tsysinfo, httpdaeWhat)
 	case apc.WhatMountpaths:
 		t.writeJSON(w, r, fs.MountpathsToLists(), httpdaeWhat)
+
+	case apc.WhatNodeStats:
+		ds := t.statsAndStatus()
+		daeStats := t.statsT.GetStats()
+		ds.Tracker = daeStats.Tracker
+		t.writeJSON(w, r, ds, httpdaeWhat)
+	case apc.WhatNodeStatsV322: // [backward compatibility] v3.22 and prior
+		ds := t.statsAndStatusV322()
+		daeStats := t.statsT.GetStatsV322()
+		ds.Tracker = daeStats.Tracker
+		t.writeJSON(w, r, ds, httpdaeWhat)
 	case apc.WhatNodeStatsAndStatus:
-		var rebSnap *core.Snap
-		if entry := xreg.GetLatest(xreg.Flt{Kind: apc.ActRebalance}); entry != nil {
-			if xctn := entry.Get(); xctn != nil {
-				rebSnap = xctn.Snap()
-			}
-		}
-		smap := t.owner.smap.get()
-		msg := &stats.NodeStatus{
-			Node: stats.Node{
-				Snode: t.htrun.si,
-			},
-			SmapVersion:    smap.Version,
-			MemCPUInfo:     apc.GetMemCPU(),
-			RebSnap:        rebSnap,
-			DeploymentType: deploymentType(),
-			Version:        daemon.version,
-			BuildTime:      daemon.buildTime,
-			K8sPodName:     os.Getenv(env.AIS.K8sPod),
-			Status:         t._status(smap),
-		}
-		// stats and capacity
+		ds := t.statsAndStatus()
+		ds.RebSnap = _rebSnap()
 		daeStats := t.statsT.GetStats()
-		msg.Tracker = daeStats.Tracker
-		msg.TargetCDF = daeStats.TargetCDF
+		ds.Tracker = daeStats.Tracker
+		ds.TargetCDF = daeStats.TargetCDF
+		nlog.Errorln(ds.TargetCDF.TotalUsed, ds.TargetCDF.TotalAvail) // DEBUG
+		t.writeJSON(w, r, ds, httpdaeWhat)
+	case apc.WhatNodeStatsAndStatusV322: // [ditto]
+		ds := t.statsAndStatusV322()
+		ds.RebSnap = _rebSnap()
+		daeStats := t.statsT.GetStatsV322()
+		ds.Tracker = daeStats.Tracker
+		ds.TargetCDF = daeStats.TargetCDF
+		t.writeJSON(w, r, ds, httpdaeWhat)
 
-		t.writeJSON(w, r, msg, httpdaeWhat)
 	case apc.WhatDiskStats:
 		diskStats := make(ios.AllDiskStats)
 		fs.FillDiskStats(diskStats)
@@ -349,6 +346,15 @@ func (t *target) httpdaeget(w http.ResponseWriter, r *http.Request) {
 	}
 }
 
+func _rebSnap() (rebSnap *core.Snap) {
+	if entry := xreg.GetLatest(xreg.Flt{Kind: apc.ActRebalance}); entry != nil {
+		if xctn := entry.Get(); xctn != nil {
+			rebSnap = xctn.Snap()
+		}
+	}
+	return rebSnap
+}
+
 // admin-join target | enable/disable mountpath
 func (t *target) httpdaepost(w http.ResponseWriter, r *http.Request) {
 	apiItems, err := t.parseURL(w, r, apc.URLPathDae.L, 0, true)

diff --git a/api/apc/query.go b/api/apc/query.go
@@ -184,11 +184,15 @@ const (
 	// config
 	WhatNodeConfig    = "config" // query specific node for (cluster config + overrides, local config)
 	WhatClusterConfig = "cluster_config"
-	// stats
-	WhatNodeStats          = "stats"
-	WhatNodeStatsAndStatus = "status"
-	WhatMetricNames        = "metrics"
-	WhatDiskStats          = "disk"
+
+	// stats and status
+	WhatNodeStatsV322          = "stats"  // [ backward compatibility ]
+	WhatNodeStatsAndStatusV322 = "status" // [ ditto ]
+	WhatNodeStats              = "node_stats"
+	WhatNodeStatsAndStatus     = "node_status"
+
+	WhatMetricNames = "metrics"
+	WhatDiskStats   = "disk"
 	// assorted
 	WhatMountpaths = "mountpaths"
 	WhatRemoteAIS  = "remote"

diff --git a/api/blob.go b/api/blob.go
@@ -1,4 +1,4 @@
-// Package api provides Go based AIStore API/SDK over HTTP(S)
+// Package api provides native Go-based API/SDK over HTTP(S).
 /*
  * Copyright (c) 2024, NVIDIA CORPORATION. All rights reserved.
  */

diff --git a/api/bsumm.go b/api/bsumm.go
@@ -1,6 +1,6 @@
-// Package api provides Go based AIStore API/SDK over HTTP(S)
+// Package api provides native Go-based API/SDK over HTTP(S).
 /*
- * Copyright (c) 2018-2023, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2018-2024, NVIDIA CORPORATION. All rights reserved.
  */
 package api
 

diff --git a/api/bucket.go b/api/bucket.go
@@ -1,6 +1,6 @@
-// Package api provides Go based AIStore API/SDK over HTTP(S)
+// Package api provides native Go-based API/SDK over HTTP(S).
 /*
- * Copyright (c) 2018-2023, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2018-2024, NVIDIA CORPORATION. All rights reserved.
  */
 package api
 

diff --git a/api/client.go b/api/client.go
@@ -1,6 +1,6 @@
-// Package api provides Go based AIStore API/SDK over HTTP(S)
+// Package api provides native Go-based API/SDK over HTTP(S).
 /*
- * Copyright (c) 2018-2023, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2018-2024, NVIDIA CORPORATION. All rights reserved.
  */
 package api
 

diff --git a/api/cluster.go b/api/cluster.go
@@ -1,6 +1,6 @@
-// Package api provides Go based AIStore API/SDK over HTTP(S)
+// Package api provides native Go-based API/SDK over HTTP(S).
 /*
- * Copyright (c) 2018-2023, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2018-2024, NVIDIA CORPORATION. All rights reserved.
  */
 package api
 
@@ -13,8 +13,6 @@ import (
 	"github.com/NVIDIA/aistore/cmn"
 	"github.com/NVIDIA/aistore/cmn/cos"
 	"github.com/NVIDIA/aistore/core/meta"
-	"github.com/NVIDIA/aistore/stats"
-	jsoniter "github.com/json-iterator/go"
 )
 
 // to be used by external watchdogs (Kubernetes, etc.)
@@ -158,46 +156,6 @@ func GetClusterSysInfo(bp BaseParams) (info apc.ClusterSysInfo, err error) {
 	return
 }
 
-// How to compute throughputs:
-//
-// - AIS supports several enumerated metric "kinds", including `KindThroughput`
-// (for complete enumeration, see stats/api.go)
-// - By convention, metrics that have `KindThroughput` kind are named with ".bps"
-// ("bytes per second") suffix.
-// - ".bps" metrics reported by api.GetClusterStats and api.GetDaemonStats are,
-// in fact, cumulative byte numbers.
-// - It is the client's responsibility to compute the actual throughputs
-// as only the client knows _when_ exactly the same ".bps" metric was queried
-// the previous time.
-//
-// - See also: `api.GetDaemonStats`, stats/api.go
-func GetClusterStats(bp BaseParams) (res stats.Cluster, err error) {
-	bp.Method = http.MethodGet
-	reqParams := AllocRp()
-	{
-		reqParams.BaseParams = bp
-		reqParams.Path = apc.URLPathClu.S
-		reqParams.Query = url.Values{apc.QparamWhat: []string{apc.WhatNodeStats}}
-	}
-
-	var rawStats stats.ClusterRaw
-	_, err = reqParams.DoReqAny(&rawStats)
-	FreeRp(reqParams)
-	if err != nil {
-		return
-	}
-
-	res.Proxy = rawStats.Proxy
-	res.Target = make(map[string]*stats.Node)
-	for tid := range rawStats.Target {
-		var ts stats.Node
-		if err := jsoniter.Unmarshal(rawStats.Target[tid], &ts); err == nil {
-			res.Target[tid] = &ts
-		}
-	}
-	return
-}
-
 func GetRemoteAIS(bp BaseParams) (remais meta.RemAisVec, err error) {
 	bp.Method = http.MethodGet
 	reqParams := AllocRp()
@@ -290,11 +248,6 @@ func SetClusterConfigUsingMsg(bp BaseParams, configToUpdate *cmn.ConfigToSet, tr
 	return err
 }
 
-// zero out: all metrics _or_ only error counters
-func ResetClusterStats(bp BaseParams, errorsOnly bool) (err error) {
-	return _putCluster(bp, apc.ActMsg{Action: apc.ActResetStats, Value: errorsOnly})
-}
-
 // all nodes: reset configuration to cluster defaults
 func ResetClusterConfig(bp BaseParams) error {
 	return _putCluster(bp, apc.ActMsg{Action: apc.ActResetConfig})