| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package scrape |
| 4 | |
| 5 | import ( |
| 6 | "fmt" |
| 7 | "strconv" |
| 8 | "strings" |
| 9 | "sync" |
| 10 | "time" |
| 11 | |
| 12 | "github.com/netdata/netdata/go/plugins/logger" |
| 13 | rs "github.com/netdata/netdata/go/plugins/plugin/go.d/collector/vsphere/resources" |
| 14 | |
| 15 | "github.com/vmware/govmomi/performance" |
| 16 | "github.com/vmware/govmomi/vim25/types" |
| 17 | vsantypes "github.com/vmware/govmomi/vsan/types" |
| 18 | ) |
| 19 | |
| 20 | type Client interface { |
| 21 | Version() string |
| 22 | PerformanceMetrics([]types.PerfQuerySpec) ([]performance.EntityMetric, error) |
| 23 | VSANPerfMetrics(types.ManagedObjectReference, []vsantypes.VsanPerfQuerySpec) ([]vsantypes.VsanPerfEntityMetricCSV, error) |
| 24 | VSANSpaceUsage(types.ManagedObjectReference) (*vsantypes.VsanSpaceUsage, error) |
| 25 | VSANHealth(types.ManagedObjectReference) (string, error) |
| 26 | } |
| 27 | |
| 28 | func New(client Client) *Scraper { |
| 29 | v := &Scraper{ |
| 30 | Client: client, |
| 31 | vsanWarnings: make(map[string]bool), |
| 32 | } |
| 33 | v.calcMaxQuery() |
| 34 | return v |
| 35 | } |
| 36 | |
| 37 | type Scraper struct { |
| 38 | *logger.Logger |
| 39 | Client |
| 40 | maxQuery int |
| 41 | vsanWarnings map[string]bool |
| 42 | vsanWarningsLock sync.Mutex |
| 43 | } |
| 44 | |
| 45 | // Default settings for vCenter 6.5 and above is 256, prior versions of vCenter have this set to 64. |
| 46 | func (s *Scraper) calcMaxQuery() { |
| 47 | major, minor, err := parseVersion(s.Version()) |
| 48 | if err != nil || major < 6 || (major == 6 && minor < 5) { |
| 49 | s.maxQuery = 64 |
| 50 | return |
| 51 | } |
| 52 | s.maxQuery = 256 |
| 53 | } |
| 54 | |
| 55 | func (s *Scraper) ScrapeHosts(hosts rs.Hosts) []performance.EntityMetric { |
| 56 | t := time.Now() |
| 57 | pqs := newHostsPerfQuerySpecs(hosts) |
| 58 | ms := s.scrapeMetrics(pqs) |
| 59 | s.Debugf("scraping : scraped metrics for %d/%d hosts, process took %s", |
| 60 | len(ms), |
| 61 | len(hosts), |
| 62 | time.Since(t), |
| 63 | ) |
| 64 | return ms |
| 65 | } |
| 66 | |
| 67 | func (s *Scraper) ScrapeVMs(vms rs.VMs) []performance.EntityMetric { |
| 68 | t := time.Now() |
| 69 | pqs := newVMsPerfQuerySpecs(vms) |
| 70 | ms := s.scrapeMetrics(pqs) |
| 71 | s.Debugf("scraping : scraped metrics for %d/%d vms, process took %s", |
| 72 | len(ms), |
| 73 | len(vms), |
| 74 | time.Since(t), |
| 75 | ) |
| 76 | return ms |
| 77 | } |
| 78 | |
| 79 | func (s *Scraper) ScrapeDatastores(datastores rs.Datastores) []performance.EntityMetric { |
| 80 | t := time.Now() |
| 81 | pqs := newDatastoresPerfQuerySpecs(datastores) |
| 82 | ms := s.scrapeMetrics(pqs) |
| 83 | s.Debugf("scraping : scraped metrics for %d/%d datastores, process took %s", |
| 84 | len(ms), |
| 85 | len(datastores), |
| 86 | time.Since(t), |
| 87 | ) |
| 88 | return ms |
| 89 | } |
| 90 | |
| 91 | func (s *Scraper) ScrapeClusters(clusters rs.Clusters) []performance.EntityMetric { |
| 92 | t := time.Now() |
| 93 | pqs := newClustersPerfQuerySpecs(clusters) |
| 94 | ms := s.scrapeMetrics(pqs) |
| 95 | s.Debugf("scraping : scraped metrics for %d/%d clusters, process took %s", |
| 96 | len(ms), |
| 97 | len(clusters), |
| 98 | time.Since(t), |
| 99 | ) |
| 100 | return ms |
| 101 | } |
| 102 | |
| 103 | func (s *Scraper) scrapeMetrics(pqs []types.PerfQuerySpec) []performance.EntityMetric { |
| 104 | tc := newThrottledCaller(5) |
| 105 | var ms []performance.EntityMetric |
| 106 | lock := &sync.Mutex{} |
| 107 | |
| 108 | chunks := chunkify(pqs, s.maxQuery) |
| 109 | for _, chunk := range chunks { |
| 110 | pqs := chunk |
| 111 | job := func() { |
| 112 | s.scrape(&ms, lock, pqs) |
| 113 | } |
| 114 | tc.call(job) |
| 115 | } |
| 116 | tc.wait() |
| 117 | |
| 118 | return ms |
| 119 | } |
| 120 | |
| 121 | func (s *Scraper) scrape(metrics *[]performance.EntityMetric, lock *sync.Mutex, pqs []types.PerfQuerySpec) { |
| 122 | m, err := s.PerformanceMetrics(pqs) |
| 123 | if err != nil { |
| 124 | s.Limit(logKeyPerfQueryError+perfQuerySpecEntityType(pqs), 1, recurringLogEvery). |
| 125 | Errorf("scrape vSphere performance metrics: query_specs=%d entities=[%s]: %v", len(pqs), describePerfQuerySpecs(pqs), err) |
| 126 | return |
| 127 | } |
| 128 | |
| 129 | lock.Lock() |
| 130 | *metrics = append(*metrics, m...) |
| 131 | lock.Unlock() |
| 132 | } |
| 133 | |
| 134 | func chunkify(pqs []types.PerfQuerySpec, chunkSize int) (chunks [][]types.PerfQuerySpec) { |
| 135 | for i := 0; i < len(pqs); i += chunkSize { |
| 136 | end := min(i+chunkSize, len(pqs)) |
| 137 | chunks = append(chunks, pqs[i:end]) |
| 138 | } |
| 139 | return chunks |
| 140 | } |
| 141 | |
| 142 | func describePerfQuerySpecs(pqs []types.PerfQuerySpec) string { |
| 143 | if len(pqs) == 0 { |
| 144 | return "none" |
| 145 | } |
| 146 | |
| 147 | const limit = 5 |
| 148 | refs := make([]string, 0, min(len(pqs), limit)) |
| 149 | for _, pq := range pqs[:min(len(pqs), limit)] { |
| 150 | refs = append(refs, fmt.Sprintf("%s/%s", pq.Entity.Type, pq.Entity.Value)) |
| 151 | } |
| 152 | if len(pqs) > limit { |
| 153 | refs = append(refs, fmt.Sprintf("+%d more", len(pqs)-limit)) |
| 154 | } |
| 155 | |
| 156 | return strings.Join(refs, ",") |
| 157 | } |
| 158 | |
| 159 | func perfQuerySpecEntityType(pqs []types.PerfQuerySpec) string { |
| 160 | if len(pqs) == 0 { |
| 161 | return "none" |
| 162 | } |
| 163 | return pqs[0].Entity.Type |
| 164 | } |
| 165 | |
| 166 | const ( |
| 167 | pqsMaxSample = 1 |
| 168 | pqsIntervalID = 20 |
| 169 | pqsFormat = "normal" |
| 170 | |
| 171 | recurringLogEvery = time.Hour |
| 172 | logKeyPerfQueryError = "vsphere:perf-query-error:" |
| 173 | ) |
| 174 | |
| 175 | func newHostsPerfQuerySpecs(hosts rs.Hosts) []types.PerfQuerySpec { |
| 176 | pqs := make([]types.PerfQuerySpec, 0, len(hosts)) |
| 177 | for _, host := range hosts { |
| 178 | if !host.IsPoweredOn() || len(host.MetricList) == 0 { |
| 179 | continue |
| 180 | } |
| 181 | pq := types.PerfQuerySpec{ |
| 182 | Entity: host.Ref, |
| 183 | MaxSample: pqsMaxSample, |
| 184 | MetricId: host.MetricList, |
| 185 | IntervalId: pqsIntervalID, |
| 186 | Format: pqsFormat, |
| 187 | } |
| 188 | pqs = append(pqs, pq) |
| 189 | } |
| 190 | return pqs |
| 191 | } |
| 192 | |
| 193 | func newVMsPerfQuerySpecs(vms rs.VMs) []types.PerfQuerySpec { |
| 194 | pqs := make([]types.PerfQuerySpec, 0, len(vms)) |
| 195 | for _, vm := range vms { |
| 196 | if !vm.IsPoweredOn() || len(vm.MetricList) == 0 { |
| 197 | continue |
| 198 | } |
| 199 | pq := types.PerfQuerySpec{ |
| 200 | Entity: vm.Ref, |
| 201 | MaxSample: pqsMaxSample, |
| 202 | MetricId: vm.MetricList, |
| 203 | IntervalId: pqsIntervalID, |
| 204 | Format: pqsFormat, |
| 205 | } |
| 206 | pqs = append(pqs, pq) |
| 207 | } |
| 208 | return pqs |
| 209 | } |
| 210 | |
| 211 | // Datastores, clusters, and resource pools do not support real-time (20s) collection. |
| 212 | // Minimum supported interval is 300s (5 min historical). |
| 213 | const pqsHistoricalIntervalID = 300 |
| 214 | |
| 215 | func newDatastoresPerfQuerySpecs(datastores rs.Datastores) []types.PerfQuerySpec { |
| 216 | pqs := make([]types.PerfQuerySpec, 0, len(datastores)) |
| 217 | for _, ds := range datastores { |
| 218 | if !ds.Accessible || len(ds.MetricList) == 0 { |
| 219 | continue |
| 220 | } |
| 221 | pq := types.PerfQuerySpec{ |
| 222 | Entity: ds.Ref, |
| 223 | MaxSample: pqsMaxSample, |
| 224 | MetricId: ds.MetricList, |
| 225 | IntervalId: pqsHistoricalIntervalID, |
| 226 | Format: pqsFormat, |
| 227 | } |
| 228 | pqs = append(pqs, pq) |
| 229 | } |
| 230 | return pqs |
| 231 | } |
| 232 | |
| 233 | func newClustersPerfQuerySpecs(clusters rs.Clusters) []types.PerfQuerySpec { |
| 234 | pqs := make([]types.PerfQuerySpec, 0, len(clusters)) |
| 235 | for _, c := range clusters { |
| 236 | if len(c.MetricList) == 0 { |
| 237 | continue |
| 238 | } |
| 239 | pq := types.PerfQuerySpec{ |
| 240 | Entity: c.Ref, |
| 241 | MaxSample: pqsMaxSample, |
| 242 | MetricId: c.MetricList, |
| 243 | IntervalId: pqsHistoricalIntervalID, |
| 244 | Format: pqsFormat, |
| 245 | } |
| 246 | pqs = append(pqs, pq) |
| 247 | } |
| 248 | return pqs |
| 249 | } |
| 250 | |
| 251 | func parseVersion(version string) (major, minor int, err error) { |
| 252 | parts := strings.Split(version, ".") |
| 253 | if len(parts) < 2 { |
| 254 | return 0, 0, fmt.Errorf("parse vSphere API version %q: expected <major>.<minor>", version) |
| 255 | } |
| 256 | if major, err = strconv.Atoi(parts[0]); err != nil { |
| 257 | return 0, 0, fmt.Errorf("parse vSphere API version major component %q from %q: %w", parts[0], version, err) |
| 258 | } |
| 259 | if minor, err = strconv.Atoi(parts[1]); err != nil { |
| 260 | return 0, 0, fmt.Errorf("parse vSphere API version minor component %q from %q: %w", parts[1], version, err) |
| 261 | } |
| 262 | return major, minor, nil |
| 263 | } |