Tests and bug fixes

util/formats/oldfmt/decode_samples.go

@@ -60,7 +60,7 @@ Reader:
 			m[key] = value
 		}
 
-		if _, found := m["version"]; !found {
+		if _, found := m["v"]; !found {
 			continue
 		}
 		t, found := m["time"]
@@ -132,8 +132,8 @@ Reader:
 			if err != nil {
 				continue Reader
 			}
-			envelope.Samples = append(envelope.Samples, sample)
 		}
+		envelope.Samples = append(envelope.Samples, sample)
 	}
 	if envelope != nil {
 		consume(envelope)

util/formats/oldfmt_test.go

@@ -0,0 +1,111 @@
+// Test the decoders.
+
+package formats
+
+import (
+	"os"
+	"strings"
+	"testing"
+
+	"formats/oldfmt"
+)
+
+func TestSysinfo(t *testing.T) {
+	f, err := os.Open("testdata/sysinfo.json")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer f.Close()
+	// There are two records in the file.
+	var iter int
+	err = oldfmt.ConsumeSysinfo(f, func(info *oldfmt.SysinfoEnvelope) {
+		switch iter {
+		case 0:
+			assert(t, info.Version == "0.13.100", "#0 version")
+			assert(t, info.Timestamp == "2025-03-01T00:00:01+01:00", "#0 time")
+			assert(t, info.Hostname == "ml1.hpc.uio.no", "#0 host")
+			assert(t, strings.HasPrefix(info.Description, "2x14 (hyperthreaded) Intel(R) Xeon(R)"), "#0 desc")
+			assert(t, info.CpuCores == 56, "#0 cores")
+			assert(t, info.MemGB == 125, "#0 memory")
+			assert(t, info.GpuCards == 3, "#0 gpu-cards")
+			assert(t, info.GpuMemGB == 33, "#0 gpu-mem")
+			assert(t, len(info.GpuInfo) == 3, "#0 gpu-info len")
+			g := info.GpuInfo[1]
+			assert(t, g.BusAddress == "00000000:3B:00.0", "#0 addr")
+			assert(t, g.Index == 1, "#0 index")
+			assert(t, g.UUID == "GPU-be013a01-364d-ca23-f871-206fe3f259ba", "#0 UUID")
+			assert(t, g.Manufacturer == "NVIDIA", "#0 manufacturer")
+			assert(t, g.Model == "NVIDIA GeForce RTX 2080 Ti", "#0 model")
+			assert(t, g.Architecture == "Turing", "#0 arch")
+			assert(t, g.Driver == "550.127.08", "#0 driver")
+			assert(t, g.Firmware == "12.4", "#0 firmware")
+			assert(t, g.MemKB == 11534336, "#0 card mem")
+			assert(t, g.PowerLimit == 250, "#0 power limit")
+			assert(t, g.MaxPowerLimit == 280, "#0 max power limit")
+			assert(t, g.MinPowerLimit == 100, "#0 min power limit")
+			assert(t, g.MaxCEClock == 2100, "#0 max ce clock")
+			assert(t, g.MaxMemClock == 7000, "#0 max memory clock")
+		case 1:
+			assert(t, info.Version == "0.13.200", "#1 version")
+			assert(t, info.Timestamp == "2025-02-28T00:00:01+01:00", "#1 time")
+			// The rest tested adequately above
+		}
+		iter++
+	})
+	assert(t, iter == 2, "Iteration count")
+	if err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestSamples(t *testing.T) {
+	f, err := os.Open("testdata/samples.csv")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer f.Close()
+	// There are three *groups* of records in the file.
+	// FIXME: Currently none of them have cpuload or gpuload information.
+	var iter int
+	err = oldfmt.ConsumeCSVSamples(f, func(info *oldfmt.SampleEnvelope) {
+		switch iter {
+		case 0:
+			assert(t, info.Timestamp == "2025-02-27T01:00:01+01:00", "#0 time")
+			assert(t, info.Hostname == "c1-6.fox", "#0 host")
+			assert(t, len(info.Samples) == 4, "#0 samples")
+		case 1:
+			assert(t, info.Timestamp == "2025-02-27T01:00:01+01:00", "#1 time")
+			assert(t, info.Hostname == "gpu-11.fox", "#1 host")
+			assert(t, len(info.Samples) == 8, "#1 samples")
+		case 2:
+			assert(t, info.Timestamp == "2025-02-27T01:05:01+01:00", "#2 time")
+			assert(t, info.Hostname == "gpu-11.fox", "#2 host")
+			assert(t, len(info.Samples) == 8, "#2 samples")
+			s := info.Samples[3]
+			assert(t, s.User == "ec-dhananjt", "#2 user")
+			assert(t, s.Cmd == "python", "#2 cmd")
+			assert(t, s.JobId == 1345347, "#2 job")
+			assert(t, s.Pid == 2164020, "#2 pid")
+			assert(t, s.ParentPid == 2163996, "#2 ppid")
+			assert(t, s.CpuPct == 898.6, "#2 cpu%")
+			assert(t, s.CpuKib == 77012712, "#2 cpukib")
+			assert(t, s.RssAnonKib == 70950544, "#2 rss")
+			assert(t, s.Gpus == "3", "#2 gpus")
+			assert(t, s.GpuPct == 47, "#2 gpu%")
+			assert(t, s.GpuMemPct == 7, "#2 gpumem%")
+			assert(t, s.GpuKib == 23418880, "#2 gpukib")
+			assert(t, s.CpuTimeSec == 5369665, "#2 cputime")
+		}
+		iter++
+	})
+	assert(t, iter == 3, "Iteration count")
+	if err != nil {
+		t.Fatal(err)
+	}
+}
+
+func assert(t *testing.T, c bool, msg string) {
+	if !c {
+		t.Fatal(msg)
+	}
+}

util/formats/testdata/samples.csv

@@ -0,0 +1,20 @@
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=c1-6.fox,user=ec-tsauren,cmd=python3,job=1351930,ppid=2200718,cpu%=51.3,cpukib=194835888,rssanonkib=74255936,cputime_sec=9534,rolledup=255
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=c1-6.fox,user=ec-tsauren,cmd=python3,job=1351930,pid=2200663,ppid=2200654,cpu%=53.5,cpukib=1039564,rssanonkib=532600,cputime_sec=11213
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=c1-6.fox,user=ec-lgcharpe,cmd=pt_main_thread,job=1352526,pid=2202575,ppid=2200641,cpu%=87.4,cpukib=166299984,rssanonkib=163705808,cputime_sec=57058
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=c1-6.fox,user=ec-marcink,cmd=mpihello,pid=2165818,ppid=1,cpu%=1.5,cpukib=6020,rssanonkib=1024,cputime_sec=40613
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=gpu-11.fox,user=ec-dhananjt,cmd=python,job=1345348,ppid=2164018,cpu%=118.9,cpukib=757555664,rssanonkib=691719512,cputime_sec=1010,rolledup=9
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=gpu-11.fox,user=ec-sindrre,cmd=python,job=1350861,pid=2877884,ppid=2877803,cpu%=0.4,cpukib=1010364,rssanonkib=549792,cputime_sec=502
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=gpu-11.fox,user=ec-dhananjt,cmd=python,job=1345347,pid=2164020,ppid=2163996,cpu%=898.6,cpukib=74012704,rssanonkib=68236740,gpus=3,gpu%=44,gpumem%=5,gpukib=23418880,cputime_sec=5367095
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=gpu-11.fox,user=ec-dhananjt,cmd=python,job=1345347,ppid=2164020,cpu%=98.2,cpukib=843997864,rssanonkib=780110372,cputime_sec=8290,rolledup=9
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=gpu-11.fox,user=ec-sindrre,cmd=ollama,job=1350861,pid=2877829,ppid=2877803,cpu%=4.6,cpukib=607000,rssanonkib=537008,cputime_sec=117271
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=gpu-11.fox,user=ec-sindrre,cmd=ollama_llama_se,job=1350861,ppid=2877829,cpu%=105.2,cpukib=1084072,rssanonkib=601468,"gpus=4,5,6,0",gpu%=70,gpumem%=45,gpukib=68816896,cputime_sec=4316,rolledup=1
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=gpu-11.fox,user=ec-dhananjt,cmd=python,job=1345348,pid=2164018,ppid=2163998,cpu%=917.6,cpukib=75652164,rssanonkib=69142664,gpus=2,gpu%=47,gpumem%=6,gpukib=23418880,cputime_sec=5486448
+v=0.13.200,time=2025-02-27T01:00:01+01:00,host=gpu-11.fox,user=ec-sindrre,cmd=nvidia-smi,job=1350861,pid=2877827,ppid=2877803,cpu%=0.3,cpukib=14588,rssanonkib=14096,cputime_sec=337
+v=0.13.200,time=2025-02-27T01:05:01+01:00,host=gpu-11.fox,user=ec-dhananjt,cmd=python,job=1345348,ppid=2164018,cpu%=114.7,cpukib=761488280,rssanonkib=695407112,cputime_sec=1319,rolledup=9
+v=0.13.200,time=2025-02-27T01:05:01+01:00,host=gpu-11.fox,user=ec-dhananjt,cmd=python,job=1345347,ppid=2164020,cpu%=98.1,cpukib=848105108,rssanonkib=784025960,cputime_sec=8573,rolledup=9
+v=0.13.200,time=2025-02-27T01:05:01+01:00,host=gpu-11.fox,user=ec-sindrre,cmd=ollama,job=1350861,pid=2877829,ppid=2877803,cpu%=4.6,cpukib=606416,rssanonkib=538532,cputime_sec=117595
+v=0.13.200,time=2025-02-27T01:05:01+01:00,host=gpu-11.fox,user=ec-dhananjt,cmd=python,job=1345347,pid=2164020,ppid=2163996,cpu%=898.6,cpukib=77012712,rssanonkib=70950544,gpus=3,gpu%=47,gpumem%=7,gpukib=23418880,cputime_sec=5369665
+v=0.13.200,time=2025-02-27T01:05:01+01:00,host=gpu-11.fox,user=ec-sindrre,cmd=ollama_llama_se,job=1350861,ppid=2877829,cpu%=107.1,cpukib=805776,rssanonkib=464144,"gpus=4,0,6,5",gpu%=102,gpukib=64337920,cputime_sec=4305,rolledup=1
+v=0.13.200,time=2025-02-27T01:05:01+01:00,host=gpu-11.fox,user=ec-sindrre,cmd=python,job=1350861,pid=2877884,ppid=2877803,cpu%=0.4,cpukib=1010364,rssanonkib=549792,cputime_sec=503
+v=0.13.200,time=2025-02-27T01:05:01+01:00,host=gpu-11.fox,user=ec-dhananjt,cmd=python,job=1345348,pid=2164018,ppid=2163998,cpu%=917.6,cpukib=76652164,rssanonkib=70141132,gpus=2,gpu%=51,gpumem%=8,gpukib=23418880,cputime_sec=5488962
+v=0.13.200,time=2025-02-27T01:05:01+01:00,host=gpu-11.fox,user=ec-sindrre,cmd=nvidia-smi,job=1350861,pid=2877827,ppid=2877803,cpu%=0.3,cpukib=14588,rssanonkib=14096,cputime_sec=338

util/formats/testdata/sysinfo.json

@@ -0,0 +1,27 @@
+{
+  "version": "0.13.100",
+  "timestamp": "2025-03-01T00:00:01+01:00",
+  "hostname": "ml1.hpc.uio.no",
+  "description": "2x14 (hyperthreaded) Intel(R) Xeon(R) Gold 5120 CPU @ 2.20GHz, 125 GiB, 3x NVIDIA GeForce RTX 2080 Ti @ 11GiB",
+  "cpu_cores": 56,
+  "mem_gb": 125,
+  "gpu_cards": 3,
+  "gpumem_gb": 33,
+  "gpu_info": [
+  {"bus_addr":"00000000:18:00.0", "index":0, "uuid":"GPU-35080357-601c-7113-ec05-f6ca1e58a91e",
+   "manufacturer":"NVIDIA", "model":"NVIDIA GeForce RTX 2080 Ti", "arch":"Turing", "driver":"550.127.08", "firmware":"12.4",
+   "mem_size_kib":11534336,
+   "power_limit_watt":250, "max_power_limit_watt":280, "min_power_limit_watt":100,
+   "max_ce_clock_mhz":2100, "max_mem_clock_mhz":7000},
+  {"bus_addr":"00000000:3B:00.0", "index":1, "uuid":"GPU-be013a01-364d-ca23-f871-206fe3f259ba",
+   "manufacturer":"NVIDIA", "model":"NVIDIA GeForce RTX 2080 Ti", "arch":"Turing", "driver":"550.127.08", "firmware":"12.4",
+   "mem_size_kib":11534336,
+   "power_limit_watt":250, "max_power_limit_watt":280, "min_power_limit_watt":100,
+   "max_ce_clock_mhz":2100, "max_mem_clock_mhz":7000},
+  {"bus_addr":"00000000:86:00.0", "index":2, "uuid":"GPU-daa9f6ac-c8bf-87be-8adc-89b1e7d3f38a",
+   "manufacturer":"NVIDIA", "model":"NVIDIA GeForce RTX 2080 Ti", "arch":"Turing", "driver":"550.127.08", "firmware":"12.4",
+   "mem_size_kib":11534336,
+   "power_limit_watt":250, "max_power_limit_watt":280, "min_power_limit_watt":100,
+   "max_ce_clock_mhz":2100, "max_mem_clock_mhz":7000}]
+}
+{"version":"0.13.200","timestamp":"2025-02-28T00:00:01+01:00","hostname":"c1-6.fox","description":"2x64 AMD EPYC 7702 64-Core Processor, 503 GiB","cpu_cores":128,"mem_gb":503}