package main
import (
+ "bufio"
+ "context"
"crypto/rand"
"encoding/binary"
"flag"
"io"
"io/ioutil"
"log"
+ mathRand "math/rand"
"net/http"
"os"
"os/signal"
- "sync"
+ "strings"
"sync/atomic"
"syscall"
"time"
+ "git.arvados.org/arvados.git/lib/cmd"
+ "git.arvados.org/arvados.git/lib/config"
+ "git.arvados.org/arvados.git/sdk/go/arvados"
"git.arvados.org/arvados.git/sdk/go/arvadosclient"
"git.arvados.org/arvados.git/sdk/go/keepclient"
)
getVersion = flag.Bool("version", false, "Print version information and exit.")
RunTime = flag.Duration("run-time", 0, "time to run (e.g. 60s), or 0 to run indefinitely (default)")
Repeat = flag.Int("repeat", 1, "number of times to repeat the experiment (default 1)")
+ UseIndex = flag.Bool("use-index", false, "use the GetIndex call to get a list of blocks to read. Requires the SystemRoot token. Use this to rule out caching effects when reading.")
)
-var summary string
-var csvHeader string
-
-func main() {
- flag.Parse()
-
- // Print version information if requested
- if *getVersion {
- fmt.Printf("keep-exercise %s\n", version)
- os.Exit(0)
- }
-
- stderr := log.New(os.Stderr, "", log.LstdFlags)
-
- if *ReadThreads > 0 && *WriteThreads == 0 {
- stderr.Fatal("At least one write thread is required if rthreads is non-zero")
- }
-
- if *ReadThreads == 0 && *WriteThreads == 0 {
- stderr.Fatal("Nothing to do!")
- }
-
+func createKeepClient(lgr *log.Logger) (kc *keepclient.KeepClient) {
arv, err := arvadosclient.MakeArvadosClient()
if err != nil {
- stderr.Fatal(err)
+ lgr.Fatal(err)
}
- kc, err := keepclient.MakeKeepClient(arv)
+ kc, err = keepclient.MakeKeepClient(arv)
if err != nil {
- stderr.Fatal(err)
+ lgr.Fatal(err)
}
kc.Want_replicas = *Replicas
TLSClientConfig: arvadosclient.MakeTLSConfig(arv.ApiInsecure),
},
}
+ overrideServices(kc, lgr)
+ return kc
+}
+
+func main() {
+ if ok, code := cmd.ParseFlags(flag.CommandLine, os.Args[0], os.Args[1:], "", os.Stderr); !ok {
+ os.Exit(code)
+ } else if *getVersion {
+ fmt.Printf("%s %s\n", os.Args[0], version)
+ return
+ }
+
+ lgr := log.New(os.Stderr, "", log.LstdFlags)
+
+ if *ReadThreads > 0 && *WriteThreads == 0 && !*UseIndex {
+ lgr.Fatal("At least one write thread is required if rthreads is non-zero and -use-index is not enabled")
+ }
+
+ if *ReadThreads == 0 && *WriteThreads == 0 {
+ lgr.Fatal("Nothing to do!")
+ }
+
+ kc := createKeepClient(lgr)
+
+ // When UseIndex is set, we need a KeepClient with SystemRoot powers to get
+ // the block index from the Keepstore. We use the SystemRootToken from
+ // the Arvados config.yml for that.
+ var cluster *arvados.Cluster
+ if *ReadThreads > 0 && *UseIndex {
+ cluster = loadConfig(lgr)
+ kc.Arvados.ApiToken = cluster.SystemRootToken
+ }
- overrideServices(kc, stderr)
- csvHeader = "Timestamp,Elapsed,Read (bytes),Avg Read Speed (MiB/s),Peak Read Speed (MiB/s),Written (bytes),Avg Write Speed (MiB/s),Peak Write Speed (MiB/s),Errors,ReadThreads,WriteThreads,VaryRequest,VaryThread,BlockSize,Replicas,StatsInterval,ServiceURL,ServiceUUID,RunTime,Repeat"
+ ctx, cancel := context.WithCancel(context.Background())
+ defer cancel()
+ sigChan := make(chan os.Signal, 1)
+ signal.Notify(sigChan, os.Interrupt, syscall.SIGTERM)
+ go func() {
+ <-sigChan
+ // FIXME
+ //fmt.Print("\r") // Suppress the ^C print
+ cancel()
+ }()
- for i := 0; i < *Repeat; i++ {
- runExperiment(kc, stderr)
- stderr.Printf("*************************** experiment %d complete ******************************\n", i)
+ csvHeader := "Timestamp,Elapsed,Read (bytes),Avg Read Speed (MiB/s),Peak Read Speed (MiB/s),Written (bytes),Avg Write Speed (MiB/s),Peak Write Speed (MiB/s),Errors,ReadThreads,WriteThreads,VaryRequest,VaryThread,BlockSize,Replicas,StatsInterval,ServiceURL,ServiceUUID,UseIndex,RunTime,Repeat"
+ var summary string
+
+ var nextBufs []chan []byte
+ for i := 0; i < *WriteThreads; i++ {
+ nextBuf := make(chan []byte, 1)
+ nextBufs = append(nextBufs, nextBuf)
+ go makeBufs(nextBuf, i, lgr)
+ }
+
+ for i := 0; i < *Repeat && ctx.Err() == nil; i++ {
+ summary = runExperiment(ctx, cluster, kc, nextBufs, summary, csvHeader, lgr)
+ lgr.Printf("*************************** experiment %d complete ******************************\n", i)
summary += fmt.Sprintf(",%d\n", i)
}
- stderr.Println("Summary:")
- stderr.Println()
+
+ lgr.Println("Summary:")
+ lgr.Println()
+ fmt.Println()
fmt.Println(csvHeader + ",Experiment")
fmt.Println(summary)
}
-func runExperiment(kc *keepclient.KeepClient, stderr *log.Logger) {
- var wg sync.WaitGroup
+func runExperiment(ctx context.Context, cluster *arvados.Cluster, kc *keepclient.KeepClient, nextBufs []chan []byte, summary string, csvHeader string, lgr *log.Logger) (newSummary string) {
+ // Send 1234 to bytesInChan when we receive 1234 bytes from keepstore.
+ var bytesInChan = make(chan uint64)
+ var bytesOutChan = make(chan uint64)
+ // Send struct{}{} to errorsChan when an error happens.
+ var errorsChan = make(chan struct{})
+
var nextLocator atomic.Value
+ // when UseIndex is set, this channel is used instead of nextLocator
+ var indexLocatorChan = make(chan string, 2)
- wg.Add(1)
- stopCh := make(chan struct{})
+ newSummary = summary
+
+ // Start warmup
+ ready := make(chan struct{})
+ var warmup bool
if *ReadThreads > 0 {
- stderr.Printf("Start warmup phase, waiting for 1 available block before reading starts\n")
- }
- for i := 0; i < *WriteThreads; i++ {
- nextBuf := make(chan []byte, 1)
- wg.Add(1)
- go makeBufs(&wg, nextBuf, i, stopCh, stderr)
- wg.Add(1)
- go doWrites(&wg, kc, nextBuf, &nextLocator, stopCh, stderr)
+ warmup = true
+ if !*UseIndex {
+ lgr.Printf("Start warmup phase, waiting for 1 available block before reading starts\n")
+ } else {
+ lgr.Printf("Start warmup phase, waiting for block index before reading starts\n")
+ }
}
- if *ReadThreads > 0 {
- for nextLocator.Load() == nil {
- select {
- case _ = <-bytesOutChan:
+ if warmup && !*UseIndex {
+ go func() {
+ locator, _, err := kc.PutB(<-nextBufs[0])
+ if err != nil {
+ lgr.Print(err)
+ errorsChan <- struct{}{}
}
+ nextLocator.Store(locator)
+ lgr.Println("Warmup complete!")
+ close(ready)
+ }()
+ } else if warmup && *UseIndex {
+ // Get list of blocks to read
+ go getIndexLocators(ctx, cluster, kc, indexLocatorChan, lgr)
+ select {
+ case <-ctx.Done():
+ return
+ case <-indexLocatorChan:
+ lgr.Println("Warmup complete!")
+ close(ready)
}
- stderr.Printf("Warmup complete")
+ } else {
+ close(ready)
}
- go countBeans(&wg, stopCh, stderr)
- for i := 0; i < *ReadThreads; i++ {
- wg.Add(1)
- go doReads(&wg, kc, &nextLocator, stopCh, stderr)
+ select {
+ case <-ctx.Done():
+ return
+ case <-ready:
}
- wg.Wait()
-}
-// Send 1234 to bytesInChan when we receive 1234 bytes from keepstore.
-var bytesInChan = make(chan uint64)
-var bytesOutChan = make(chan uint64)
+ // Warmup complete
+ ctx, cancel := context.WithDeadline(ctx, time.Now().Add(*RunTime))
+ defer cancel()
-// Send struct{}{} to errorsChan when an error happens.
-var errorsChan = make(chan struct{})
+ for i := 0; i < *WriteThreads; i++ {
+ go doWrites(ctx, kc, nextBufs[i], &nextLocator, bytesOutChan, errorsChan, lgr)
+ }
+ if *UseIndex {
+ for i := 0; i < *ReadThreads; i++ {
+ go doReads(ctx, kc, nil, indexLocatorChan, bytesInChan, errorsChan, lgr)
+ }
+ } else {
+ for i := 0; i < *ReadThreads; i++ {
+ go doReads(ctx, kc, &nextLocator, nil, bytesInChan, errorsChan, lgr)
+ }
+ }
-func countBeans(wg *sync.WaitGroup, stopCh chan struct{}, stderr *log.Logger) {
- defer wg.Done()
t0 := time.Now()
var tickChan <-chan time.Time
- var endChan <-chan time.Time
- c := make(chan os.Signal, 1)
- signal.Notify(c, os.Interrupt, syscall.SIGTERM)
if *StatsInterval > 0 {
tickChan = time.NewTicker(*StatsInterval).C
}
- if *RunTime > 0 {
- endChan = time.NewTicker(*RunTime).C
- }
var bytesIn uint64
var bytesOut uint64
var errors uint64
var rateIn, rateOut float64
var maxRateIn, maxRateOut float64
- var exit, abort, printCsv bool
+ var exit, printCsv bool
csv := log.New(os.Stdout, "", 0)
+ csv.Println()
csv.Println(csvHeader)
for {
select {
- case <-tickChan:
- printCsv = true
- case <-endChan:
+ case <-ctx.Done():
printCsv = true
exit = true
- case <-c:
+ case <-tickChan:
printCsv = true
- abort = true
- fmt.Print("\r") // Suppress the ^C print
case i := <-bytesInChan:
bytesIn += i
case o := <-bytesOutChan:
if rateOut > maxRateOut {
maxRateOut = rateOut
}
- line := fmt.Sprintf("%v,%v,%v,%.1f,%.1f,%v,%.1f,%.1f,%d,%d,%d,%t,%t,%d,%d,%s,%s,%s,%s,%d",
- time.Now().Format("2006-01-02 15:04:05"),
+ line := fmt.Sprintf("%v,%v,%v,%.1f,%.1f,%v,%.1f,%.1f,%d,%d,%d,%t,%t,%d,%d,%s,%s,%s,%t,%s,%d",
+ time.Now().Format("2006/01/02 15:04:05"),
elapsed,
bytesIn, rateIn, maxRateIn,
bytesOut, rateOut, maxRateOut,
*StatsInterval,
*ServiceURL,
*ServiceUUID,
+ *UseIndex,
*RunTime,
*Repeat,
)
csv.Println(line)
if exit {
- summary += line
+ newSummary += line
+ return
}
printCsv = false
}
- if abort {
- os.Exit(0)
- }
- if exit {
- close(stopCh)
- break
- }
}
}
-func makeBufs(wg *sync.WaitGroup, nextBuf chan<- []byte, threadID int, stopCh <-chan struct{}, stderr *log.Logger) {
- defer wg.Done()
+func makeBufs(nextBuf chan<- []byte, threadID int, lgr *log.Logger) {
buf := make([]byte, *BlockSize)
if *VaryThread {
binary.PutVarint(buf, int64(threadID))
if *VaryRequest {
rnd := make([]byte, randSize)
if _, err := io.ReadFull(rand.Reader, rnd); err != nil {
- stderr.Fatal(err)
+ lgr.Fatal(err)
}
buf = append(rnd, buf[randSize:]...)
}
- select {
- case <-stopCh:
- close(nextBuf)
- return
- case nextBuf <- buf:
+ nextBuf <- buf
+ }
+}
+
+func doWrites(ctx context.Context, kc *keepclient.KeepClient, nextBuf <-chan []byte, nextLocator *atomic.Value, bytesOutChan chan<- uint64, errorsChan chan<- struct{}, lgr *log.Logger) {
+ for ctx.Err() == nil {
+ //lgr.Printf("%s nextbuf %s, waiting for nextBuf\n",nextBuf,time.Now())
+ buf := <-nextBuf
+ //lgr.Printf("%s nextbuf %s, done waiting for nextBuf\n",nextBuf,time.Now())
+ locator, _, err := kc.PutB(buf)
+ if err != nil {
+ lgr.Print(err)
+ errorsChan <- struct{}{}
+ continue
}
+ bytesOutChan <- uint64(len(buf))
+ nextLocator.Store(locator)
}
}
-func doWrites(wg *sync.WaitGroup, kc *keepclient.KeepClient, nextBuf <-chan []byte, nextLocator *atomic.Value, stopCh <-chan struct{}, stderr *log.Logger) {
- defer wg.Done()
+func getIndexLocators(ctx context.Context, cluster *arvados.Cluster, kc *keepclient.KeepClient, indexLocatorChan chan<- string, lgr *log.Logger) {
+ if ctx.Err() != nil {
+ return
+ }
+ locatorsMap := make(map[string]bool)
+ var locators []string
+ var count int64
+ for uuid := range kc.LocalRoots() {
+ reader, err := kc.GetIndex(uuid, "")
+ if err != nil {
+ lgr.Fatalf("Error getting index: %s\n", err)
+ }
+ scanner := bufio.NewScanner(reader)
+ for scanner.Scan() {
+ locatorsMap[strings.Split(scanner.Text(), " ")[0]] = true
+ count++
+ }
+ }
+ for l := range locatorsMap {
+ locators = append(locators, l)
+ }
+ lgr.Printf("Found %d locators\n", count)
+ lgr.Printf("Found %d locators (deduplicated)\n", len(locators))
+ if len(locators) < 1 {
+ lgr.Fatal("Error: no locators found. The keepstores do not seem to contain any data. Remove the -use-index cli argument.")
+ }
- for {
+ mathRand.Seed(time.Now().UnixNano())
+ mathRand.Shuffle(len(locators), func(i, j int) { locators[i], locators[j] = locators[j], locators[i] })
+
+ for _, locator := range locators {
+ // We need the Collections.BlobSigningKey to sign our block requests. This requires access to /etc/arvados/config.yml
+ signedLocator := arvados.SignLocator(locator, kc.Arvados.ApiToken, time.Now().Local().Add(1*time.Hour), cluster.Collections.BlobSigningTTL.Duration(), []byte(cluster.Collections.BlobSigningKey))
select {
- case <-stopCh:
+ case <-ctx.Done():
return
- case buf := <-nextBuf:
- locator, _, err := kc.PutB(buf)
- if err != nil {
- stderr.Print(err)
- errorsChan <- struct{}{}
- continue
- }
- select {
- case <-stopCh:
- return
- case bytesOutChan <- uint64(len(buf)):
- }
- nextLocator.Store(locator)
+ case indexLocatorChan <- signedLocator:
}
}
+ lgr.Fatal("Error: ran out of locators to read!")
}
-func doReads(wg *sync.WaitGroup, kc *keepclient.KeepClient, nextLocator *atomic.Value, stopCh <-chan struct{}, stderr *log.Logger) {
- defer wg.Done()
+func loadConfig(lgr *log.Logger) (cluster *arvados.Cluster) {
+ loader := config.NewLoader(os.Stdin, nil)
+ loader.SkipLegacy = true
- var locator string
- for {
- locator = nextLocator.Load().(string)
+ cfg, err := loader.Load()
+ if err != nil {
+ lgr.Fatal(err)
+ }
+ cluster, err = cfg.GetCluster("")
+ if err != nil {
+ lgr.Fatal(err)
+ }
+ return
+}
+
+func doReads(ctx context.Context, kc *keepclient.KeepClient, nextLocator *atomic.Value, indexLocatorChan <-chan string, bytesInChan chan<- uint64, errorsChan chan<- struct{}, lgr *log.Logger) {
+ for ctx.Err() == nil {
+ var locator string
+ if indexLocatorChan != nil {
+ select {
+ case <-ctx.Done():
+ return
+ case locator = <-indexLocatorChan:
+ }
+ } else {
+ locator = nextLocator.Load().(string)
+ }
rdr, size, url, err := kc.Get(locator)
if err != nil {
- stderr.Print(err)
+ lgr.Print(err)
errorsChan <- struct{}{}
continue
}
n, err := io.Copy(ioutil.Discard, rdr)
rdr.Close()
if n != size || err != nil {
- stderr.Printf("Got %d bytes (expected %d) from %s: %v", n, size, url, err)
+ lgr.Printf("Got %d bytes (expected %d) from %s: %v", n, size, url, err)
errorsChan <- struct{}{}
continue
// Note we don't count the bytes received in
// partial/corrupt responses: we are measuring
// throughput, not resource consumption.
}
- select {
- case <-stopCh:
- return
- case bytesInChan <- uint64(n):
- }
+ bytesInChan <- uint64(n)
}
}
-func overrideServices(kc *keepclient.KeepClient, stderr *log.Logger) {
+func overrideServices(kc *keepclient.KeepClient, lgr *log.Logger) {
roots := make(map[string]string)
if *ServiceURL != "" {
roots["zzzzz-bi6l4-000000000000000"] = *ServiceURL
}
}
if len(roots) == 0 {
- stderr.Fatalf("Service %q was not in list advertised by API %+q", *ServiceUUID, kc.GatewayRoots())
+ lgr.Fatalf("Service %q was not in list advertised by API %+q", *ServiceUUID, kc.GatewayRoots())
}
} else {
return
projects / arvados.git / blobdiff