| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package powervault |
| 4 | |
| 5 | // collectHardwareHealth classifies hardware components by health state. |
| 6 | // Uses cached discovery data — no API calls. |
| 7 | // |
| 8 | // health-numeric mapping: 0=OK, 1=Degraded, 2=Fault, 3=Unknown. |
| 9 | // Special cases: |
| 10 | // - Drives with usage-numeric=3 (Global Hot Spare) report health=Unknown but are functionally OK. |
| 11 | // - FRUs use fru-status-numeric: 0=OK, 1=Degraded, 2=Fault, 3=Unknown, 5=N/A (treated as OK). |
| 12 | // - Ports with health-numeric=4 (Not Available) are excluded from counts. |
| 13 | func (c *Collector) collectHardwareHealth() { |
| 14 | var ctrlOK, ctrlDeg, ctrlFault, ctrlUnk float64 |
| 15 | var drvOK, drvDeg, drvFault, drvUnk float64 |
| 16 | var fanOK, fanDeg, fanFault, fanUnk float64 |
| 17 | var psuOK, psuDeg, psuFault, psuUnk float64 |
| 18 | var fruOK, fruDeg, fruFault, fruUnk float64 |
| 19 | var portOK, portDeg, portFault, portUnk float64 |
| 20 | |
| 21 | for _, ctrl := range c.discovered.controllers { |
| 22 | ok, deg, fault, unk := classifyHealth(ctrl.HealthNumeric) |
| 23 | ctrlOK += ok |
| 24 | ctrlDeg += deg |
| 25 | ctrlFault += fault |
| 26 | ctrlUnk += unk |
| 27 | } |
| 28 | |
| 29 | for _, d := range c.discovered.drives { |
| 30 | // Global Hot Spare drives report health=Unknown but are functionally OK. |
| 31 | if d.UsageNumeric == 3 && d.HealthNumeric == 3 { |
| 32 | drvOK++ |
| 33 | continue |
| 34 | } |
| 35 | ok, deg, fault, unk := classifyHealth(d.HealthNumeric) |
| 36 | drvOK += ok |
| 37 | drvDeg += deg |
| 38 | drvFault += fault |
| 39 | drvUnk += unk |
| 40 | } |
| 41 | |
| 42 | for _, f := range c.discovered.fans { |
| 43 | ok, deg, fault, unk := classifyHealth(f.HealthNumeric) |
| 44 | fanOK += ok |
| 45 | fanDeg += deg |
| 46 | fanFault += fault |
| 47 | fanUnk += unk |
| 48 | } |
| 49 | |
| 50 | for _, p := range c.discovered.psus { |
| 51 | ok, deg, fault, unk := classifyHealth(p.HealthNumeric) |
| 52 | psuOK += ok |
| 53 | psuDeg += deg |
| 54 | psuFault += fault |
| 55 | psuUnk += unk |
| 56 | } |
| 57 | |
| 58 | for _, f := range c.discovered.frus { |
| 59 | ok, deg, fault, unk := classifyFRUHealth(f.FRUStatusNum) |
| 60 | fruOK += ok |
| 61 | fruDeg += deg |
| 62 | fruFault += fault |
| 63 | fruUnk += unk |
| 64 | } |
| 65 | |
| 66 | for _, p := range c.discovered.ports { |
| 67 | // 4 = Not Available — exclude from health counts. |
| 68 | if p.HealthNumeric == 4 { |
| 69 | continue |
| 70 | } |
| 71 | ok, deg, fault, unk := classifyHealth(p.HealthNumeric) |
| 72 | portOK += ok |
| 73 | portDeg += deg |
| 74 | portFault += fault |
| 75 | portUnk += unk |
| 76 | } |
| 77 | |
| 78 | hw := c.mx.hardware |
| 79 | hw.controllerOK.Observe(ctrlOK) |
| 80 | hw.controllerDegraded.Observe(ctrlDeg) |
| 81 | hw.controllerFault.Observe(ctrlFault) |
| 82 | hw.controllerUnknown.Observe(ctrlUnk) |
| 83 | hw.driveOK.Observe(drvOK) |
| 84 | hw.driveDegraded.Observe(drvDeg) |
| 85 | hw.driveFault.Observe(drvFault) |
| 86 | hw.driveUnknown.Observe(drvUnk) |
| 87 | hw.fanOK.Observe(fanOK) |
| 88 | hw.fanDegraded.Observe(fanDeg) |
| 89 | hw.fanFault.Observe(fanFault) |
| 90 | hw.fanUnknown.Observe(fanUnk) |
| 91 | hw.psuOK.Observe(psuOK) |
| 92 | hw.psuDegraded.Observe(psuDeg) |
| 93 | hw.psuFault.Observe(psuFault) |
| 94 | hw.psuUnknown.Observe(psuUnk) |
| 95 | hw.fruOK.Observe(fruOK) |
| 96 | hw.fruDegraded.Observe(fruDeg) |
| 97 | hw.fruFault.Observe(fruFault) |
| 98 | hw.fruUnknown.Observe(fruUnk) |
| 99 | hw.portOK.Observe(portOK) |
| 100 | hw.portDegraded.Observe(portDeg) |
| 101 | hw.portFault.Observe(portFault) |
| 102 | hw.portUnknown.Observe(portUnk) |
| 103 | } |
| 104 | |
| 105 | // classifyHealth maps health-numeric (0=OK, 1=Degraded, 2=Fault, 3=Unknown). |
| 106 | func classifyHealth(h int) (ok, degraded, fault, unknown float64) { |
| 107 | switch h { |
| 108 | case 0: |
| 109 | return 1, 0, 0, 0 |
| 110 | case 1: |
| 111 | return 0, 1, 0, 0 |
| 112 | case 2: |
| 113 | return 0, 0, 1, 0 |
| 114 | default: |
| 115 | return 0, 0, 0, 1 |
| 116 | } |
| 117 | } |
| 118 | |
| 119 | // classifyFRUHealth maps fru-status-numeric (0=OK, 1=Degraded, 2=Fault, 3=Unknown, 5=N/A→OK). |
| 120 | func classifyFRUHealth(h int) (ok, degraded, fault, unknown float64) { |
| 121 | switch h { |
| 122 | case 0, 5: |
| 123 | return 1, 0, 0, 0 |
| 124 | case 1: |
| 125 | return 0, 1, 0, 0 |
| 126 | case 2: |
| 127 | return 0, 0, 1, 0 |
| 128 | default: |
| 129 | return 0, 0, 0, 1 |
| 130 | } |
| 131 | } |