| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package hdfs |
| 4 | |
| 5 | import "github.com/netdata/netdata/go/plugins/plugin/framework/collectorapi" |
| 6 | |
| 7 | type ( |
| 8 | Charts = collectorapi.Charts |
| 9 | Dims = collectorapi.Dims |
| 10 | Vars = collectorapi.Vars |
| 11 | ) |
| 12 | |
| 13 | var jvmCharts = Charts{ |
| 14 | { |
| 15 | ID: "jvm_heap_memory", |
| 16 | Title: "Heap Memory", |
| 17 | Units: "MiB", |
| 18 | Fam: "jvm", |
| 19 | Ctx: "hdfs.heap_memory", |
| 20 | Type: collectorapi.Area, |
| 21 | Dims: Dims{ |
| 22 | {ID: "jvm_mem_heap_committed", Name: "committed", Div: 1000}, |
| 23 | {ID: "jvm_mem_heap_used", Name: "used", Div: 1000}, |
| 24 | }, |
| 25 | Vars: Vars{ |
| 26 | {ID: "jvm_mem_heap_max"}, |
| 27 | }, |
| 28 | }, |
| 29 | { |
| 30 | ID: "jvm_gc_count_total", |
| 31 | Title: "GC Events", |
| 32 | Units: "events/s", |
| 33 | Fam: "jvm", |
| 34 | Ctx: "hdfs.gc_count_total", |
| 35 | Dims: Dims{ |
| 36 | {ID: "jvm_gc_count", Name: "gc", Algo: collectorapi.Incremental}, |
| 37 | }, |
| 38 | }, |
| 39 | { |
| 40 | ID: "jvm_gc_time_total", |
| 41 | Title: "GC Time", |
| 42 | Units: "ms", |
| 43 | Fam: "jvm", |
| 44 | Ctx: "hdfs.gc_time_total", |
| 45 | Dims: Dims{ |
| 46 | {ID: "jvm_gc_time_millis", Name: "time", Algo: collectorapi.Incremental}, |
| 47 | }, |
| 48 | }, |
| 49 | { |
| 50 | ID: "jvm_gc_threshold", |
| 51 | Title: "Number of Times That the GC Threshold is Exceeded", |
| 52 | Units: "events/s", |
| 53 | Fam: "jvm", |
| 54 | Ctx: "hdfs.gc_threshold", |
| 55 | Dims: Dims{ |
| 56 | {ID: "jvm_gc_num_info_threshold_exceeded", Name: "info", Algo: collectorapi.Incremental}, |
| 57 | {ID: "jvm_gc_num_warn_threshold_exceeded", Name: "warn", Algo: collectorapi.Incremental}, |
| 58 | }, |
| 59 | }, |
| 60 | { |
| 61 | ID: "jvm_threads", |
| 62 | Title: "Number of Threads", |
| 63 | Units: "num", |
| 64 | Fam: "jvm", |
| 65 | Ctx: "hdfs.threads", |
| 66 | Type: collectorapi.Stacked, |
| 67 | Dims: Dims{ |
| 68 | {ID: "jvm_threads_new", Name: "new"}, |
| 69 | {ID: "jvm_threads_runnable", Name: "runnable"}, |
| 70 | {ID: "jvm_threads_blocked", Name: "blocked"}, |
| 71 | {ID: "jvm_threads_waiting", Name: "waiting"}, |
| 72 | {ID: "jvm_threads_timed_waiting", Name: "timed_waiting"}, |
| 73 | {ID: "jvm_threads_terminated", Name: "terminated"}, |
| 74 | }, |
| 75 | }, |
| 76 | { |
| 77 | ID: "jvm_logs_total", |
| 78 | Title: "Number of Logs", |
| 79 | Units: "logs/s", |
| 80 | Fam: "jvm", |
| 81 | Ctx: "hdfs.logs_total", |
| 82 | Type: collectorapi.Stacked, |
| 83 | Dims: Dims{ |
| 84 | {ID: "jvm_log_info", Name: "info", Algo: collectorapi.Incremental}, |
| 85 | {ID: "jvm_log_error", Name: "error", Algo: collectorapi.Incremental}, |
| 86 | {ID: "jvm_log_warn", Name: "warn", Algo: collectorapi.Incremental}, |
| 87 | {ID: "jvm_log_fatal", Name: "fatal", Algo: collectorapi.Incremental}, |
| 88 | }, |
| 89 | }, |
| 90 | } |
| 91 | |
| 92 | var rpcActivityCharts = Charts{ |
| 93 | { |
| 94 | ID: "rpc_bandwidth", |
| 95 | Title: "RPC Bandwidth", |
| 96 | Units: "kilobits/s", |
| 97 | Fam: "rpc", |
| 98 | Ctx: "hdfs.rpc_bandwidth", |
| 99 | Type: collectorapi.Area, |
| 100 | Dims: Dims{ |
| 101 | {ID: "rpc_received_bytes", Name: "received", Div: 1000, Algo: collectorapi.Incremental}, |
| 102 | {ID: "rpc_sent_bytes", Name: "sent", Div: -1000, Algo: collectorapi.Incremental}, |
| 103 | }, |
| 104 | }, |
| 105 | { |
| 106 | ID: "rpc_calls", |
| 107 | Title: "RPC Calls", |
| 108 | Units: "calls/s", |
| 109 | Fam: "rpc", |
| 110 | Ctx: "hdfs.rpc_calls", |
| 111 | Dims: Dims{ |
| 112 | {ID: "rpc_queue_time_num_ops", Name: "calls", Algo: collectorapi.Incremental}, |
| 113 | }, |
| 114 | }, |
| 115 | { |
| 116 | ID: "rpc_open_connections", |
| 117 | Title: "RPC Open Connections", |
| 118 | Units: "connections", |
| 119 | Fam: "rpc", |
| 120 | Ctx: "hdfs.open_connections", |
| 121 | Dims: Dims{ |
| 122 | {ID: "rpc_num_open_connections", Name: "open"}, |
| 123 | }, |
| 124 | }, |
| 125 | { |
| 126 | ID: "rpc_call_queue_length", |
| 127 | Title: "RPC Call Queue Length", |
| 128 | Units: "num", |
| 129 | Fam: "rpc", |
| 130 | Ctx: "hdfs.call_queue_length", |
| 131 | Dims: Dims{ |
| 132 | {ID: "rpc_call_queue_length", Name: "length"}, |
| 133 | }, |
| 134 | }, |
| 135 | { |
| 136 | ID: "rpc_avg_queue_time", |
| 137 | Title: "RPC Avg Queue Time", |
| 138 | Units: "ms", |
| 139 | Fam: "rpc", |
| 140 | Ctx: "hdfs.avg_queue_time", |
| 141 | Dims: Dims{ |
| 142 | {ID: "rpc_queue_time_avg_time", Name: "time", Div: 1000}, |
| 143 | }, |
| 144 | }, |
| 145 | { |
| 146 | ID: "rpc_avg_processing_time", |
| 147 | Title: "RPC Avg Processing Time", |
| 148 | Units: "ms", |
| 149 | Fam: "rpc", |
| 150 | Ctx: "hdfs.avg_processing_time", |
| 151 | Dims: Dims{ |
| 152 | {ID: "rpc_processing_time_avg_time", Name: "time", Div: 1000}, |
| 153 | }, |
| 154 | }, |
| 155 | } |
| 156 | |
| 157 | var fsNameSystemCharts = Charts{ |
| 158 | { |
| 159 | ID: "fs_name_system_capacity", |
| 160 | Title: "Capacity Across All Datanodes", |
| 161 | Units: "KiB", |
| 162 | Fam: "fs name system", |
| 163 | Ctx: "hdfs.capacity", |
| 164 | Type: collectorapi.Stacked, |
| 165 | Dims: Dims{ |
| 166 | {ID: "fsns_capacity_remaining", Name: "remaining", Div: 1024}, |
| 167 | {ID: "fsns_capacity_used", Name: "used", Div: 1024}, |
| 168 | }, |
| 169 | Vars: Vars{ |
| 170 | {ID: "fsns_capacity_total"}, |
| 171 | }, |
| 172 | }, |
| 173 | { |
| 174 | ID: "fs_name_system_used_capacity", |
| 175 | Title: "Used Capacity Across All Datanodes", |
| 176 | Units: "KiB", |
| 177 | Fam: "fs name system", |
| 178 | Ctx: "hdfs.used_capacity", |
| 179 | Type: collectorapi.Stacked, |
| 180 | Dims: Dims{ |
| 181 | {ID: "fsns_capacity_used_dfs", Name: "dfs", Div: 1024}, |
| 182 | {ID: "fsns_capacity_used_non_dfs", Name: "non_dfs", Div: 1024}, |
| 183 | }, |
| 184 | }, |
| 185 | { |
| 186 | ID: "fs_name_system_load", |
| 187 | Title: "Number of Concurrent File Accesses (read/write) Across All DataNodes", |
| 188 | Units: "load", |
| 189 | Fam: "fs name system", |
| 190 | Ctx: "hdfs.load", |
| 191 | Dims: Dims{ |
| 192 | {ID: "fsns_total_load", Name: "load"}, |
| 193 | }, |
| 194 | }, |
| 195 | { |
| 196 | ID: "fs_name_system_volume_failures_total", |
| 197 | Title: "Number of Volume Failures Across All Datanodes", |
| 198 | Units: "events/s", |
| 199 | Fam: "fs name system", |
| 200 | Ctx: "hdfs.volume_failures_total", |
| 201 | Dims: Dims{ |
| 202 | {ID: "fsns_volume_failures_total", Name: "failures", Algo: collectorapi.Incremental}, |
| 203 | }, |
| 204 | }, |
| 205 | { |
| 206 | ID: "fs_files_total", |
| 207 | Title: "Number of Tracked Files", |
| 208 | Units: "num", |
| 209 | Fam: "fs name system", |
| 210 | Ctx: "hdfs.files_total", |
| 211 | Dims: Dims{ |
| 212 | {ID: "fsns_files_total", Name: "files"}, |
| 213 | }, |
| 214 | }, |
| 215 | { |
| 216 | ID: "fs_blocks_total", |
| 217 | Title: "Number of Allocated Blocks in the System", |
| 218 | Units: "num", |
| 219 | Fam: "fs name system", |
| 220 | Ctx: "hdfs.blocks_total", |
| 221 | Dims: Dims{ |
| 222 | {ID: "fsns_blocks_total", Name: "blocks"}, |
| 223 | }, |
| 224 | }, |
| 225 | { |
| 226 | ID: "fs_problem_blocks", |
| 227 | Title: "Number of Problem Blocks (can point to an unhealthy cluster)", |
| 228 | Units: "num", |
| 229 | Fam: "fs name system", |
| 230 | Ctx: "hdfs.blocks", |
| 231 | Dims: Dims{ |
| 232 | {ID: "fsns_corrupt_blocks", Name: "corrupt"}, |
| 233 | {ID: "fsns_missing_blocks", Name: "missing"}, |
| 234 | {ID: "fsns_under_replicated_blocks", Name: "under_replicated"}, |
| 235 | }, |
| 236 | }, |
| 237 | { |
| 238 | ID: "fs_name_system_data_nodes", |
| 239 | Title: "Number of Data Nodes By Status", |
| 240 | Units: "num", |
| 241 | Fam: "fs name system", |
| 242 | Ctx: "hdfs.data_nodes", |
| 243 | Type: collectorapi.Stacked, |
| 244 | Dims: Dims{ |
| 245 | {ID: "fsns_num_live_data_nodes", Name: "live"}, |
| 246 | {ID: "fsns_num_dead_data_nodes", Name: "dead"}, |
| 247 | {ID: "fsns_stale_data_nodes", Name: "stale"}, |
| 248 | }, |
| 249 | }, |
| 250 | } |
| 251 | |
| 252 | var fsDatasetStateCharts = Charts{ |
| 253 | { |
| 254 | ID: "fs_dataset_state_capacity", |
| 255 | Title: "Capacity", |
| 256 | Units: "KiB", |
| 257 | Fam: "fs dataset", |
| 258 | Ctx: "hdfs.datanode_capacity", |
| 259 | Type: collectorapi.Stacked, |
| 260 | Dims: Dims{ |
| 261 | {ID: "fsds_capacity_remaining", Name: "remaining", Div: 1024}, |
| 262 | {ID: "fsds_capacity_used", Name: "used", Div: 1024}, |
| 263 | }, |
| 264 | Vars: Vars{ |
| 265 | {ID: "fsds_capacity_total"}, |
| 266 | }, |
| 267 | }, |
| 268 | { |
| 269 | ID: "fs_dataset_state_used_capacity", |
| 270 | Title: "Used Capacity", |
| 271 | Units: "KiB", |
| 272 | Fam: "fs dataset", |
| 273 | Ctx: "hdfs.datanode_used_capacity", |
| 274 | Type: collectorapi.Stacked, |
| 275 | Dims: Dims{ |
| 276 | {ID: "fsds_capacity_used_dfs", Name: "dfs", Div: 1024}, |
| 277 | {ID: "fsds_capacity_used_non_dfs", Name: "non_dfs", Div: 1024}, |
| 278 | }, |
| 279 | }, |
| 280 | { |
| 281 | ID: "fs_dataset_state_num_failed_volumes", |
| 282 | Title: "Number of Failed Volumes", |
| 283 | Units: "num", |
| 284 | Fam: "fs dataset", |
| 285 | Ctx: "hdfs.datanode_failed_volumes", |
| 286 | Dims: Dims{ |
| 287 | {ID: "fsds_num_failed_volumes", Name: "failed volumes"}, |
| 288 | }, |
| 289 | }, |
| 290 | } |
| 291 | |
| 292 | var fsDataNodeActivityCharts = Charts{ |
| 293 | { |
| 294 | ID: "dna_bandwidth", |
| 295 | Title: "Bandwidth", |
| 296 | Units: "KiB/s", |
| 297 | Fam: "activity", |
| 298 | Ctx: "hdfs.datanode_bandwidth", |
| 299 | Type: collectorapi.Area, |
| 300 | Dims: Dims{ |
| 301 | {ID: "dna_bytes_read", Name: "reads", Div: 1024, Algo: collectorapi.Incremental}, |
| 302 | {ID: "dna_bytes_written", Name: "writes", Div: -1024, Algo: collectorapi.Incremental}, |
| 303 | }, |
| 304 | }, |
| 305 | } |
| 306 | |
| 307 | func dataNodeCharts() *Charts { |
| 308 | charts := Charts{} |
| 309 | panicIfError(charts.Add(*jvmCharts.Copy()...)) |
| 310 | panicIfError(charts.Add(*rpcActivityCharts.Copy()...)) |
| 311 | panicIfError(charts.Add(*fsDatasetStateCharts.Copy()...)) |
| 312 | panicIfError(charts.Add(*fsDataNodeActivityCharts.Copy()...)) |
| 313 | return &charts |
| 314 | } |
| 315 | |
| 316 | func nameNodeCharts() *Charts { |
| 317 | charts := Charts{} |
| 318 | panicIfError(charts.Add(*jvmCharts.Copy()...)) |
| 319 | panicIfError(charts.Add(*rpcActivityCharts.Copy()...)) |
| 320 | panicIfError(charts.Add(*fsNameSystemCharts.Copy()...)) |
| 321 | return &charts |
| 322 | } |
| 323 | |
| 324 | func panicIfError(err error) { |
| 325 | if err != nil { |
| 326 | panic(err) |
| 327 | } |
| 328 | } |