func main() {
common.Init()
out, err := util.SSH(*cmd, util.Slaves, *timeout)
if err != nil {
glog.Fatal(err)
}
if *printOutput {
for k, v := range out {
fmt.Printf("\n=====%s=====\n%s\n", k, v)
}
}
}
func main() {
defer common.LogPanic()
common.Init()
frontend.MustInit()
// Send start email.
emailsArr := util.ParseEmails(*emails)
emailsArr = append(emailsArr, util.CtAdmins...)
if len(emailsArr) == 0 {
glog.Error("At least one email address must be specified")
return
}
skutil.LogErr(frontend.UpdateWebappTaskSetStarted(&admin_tasks.RecreateWebpageArchivesUpdateVars{}, *gaeTaskID))
skutil.LogErr(util.SendTaskStartEmail(emailsArr, "Capture archives", util.GetMasterLogLink(*runID), ""))
// Ensure webapp is updated and completion email is sent even if task fails.
defer updateWebappTask()
defer sendEmail(emailsArr)
// Cleanup tmp files after the run.
defer util.CleanTmpDir()
// Finish with glog flush and how long the task took.
defer util.TimeTrack(time.Now(), "Capture archives on Workers")
defer glog.Flush()
if *pagesetType == "" {
glog.Error("Must specify --pageset_type")
return
}
if *chromiumBuild == "" {
glog.Error("Must specify --chromium_build")
return
}
cmd := []string{
fmt.Sprintf("cd %s;", util.CtTreeDir),
"git pull;",
"make all;",
// The main command that runs capture_archives on all workers.
fmt.Sprintf("DISPLAY=:0 capture_archives --worker_num=%s --log_dir=%s --log_id=%s --pageset_type=%s --chromium_build=%s;", util.WORKER_NUM_KEYWORD, util.GLogDir, *runID, *pagesetType, *chromiumBuild),
}
_, err := util.SSH(strings.Join(cmd, " "), util.Slaves, util.CAPTURE_ARCHIVES_TIMEOUT)
if err != nil {
glog.Errorf("Error while running cmd %s: %s", cmd, err)
return
}
*taskCompletedSuccessfully = true
}
func main() {
common.Init()
webhook.MustInitRequestSaltFromFile(util.WebhookRequestSaltPath)
// Send start email.
emailsArr := util.ParseEmails(*emails)
emailsArr = append(emailsArr, util.CtAdmins...)
if len(emailsArr) == 0 {
glog.Error("At least one email address must be specified")
return
}
skutil.LogErr(frontend.UpdateWebappTaskSetStarted(&frontend.RecreatePageSetsUpdateVars{}, *gaeTaskID))
skutil.LogErr(util.SendTaskStartEmail(emailsArr, "Creating pagesets"))
// Ensure webapp is updated and completion email is sent even if task fails.
defer updateWebappTask()
defer sendEmail(emailsArr)
// Cleanup tmp files after the run.
defer util.CleanTmpDir()
// Finish with glog flush and how long the task took.
defer util.TimeTrack(time.Now(), "Creating Pagesets on Workers")
defer glog.Flush()
if *pagesetType == "" {
glog.Error("Must specify --pageset_type")
return
}
cmd := []string{
fmt.Sprintf("cd %s;", util.CtTreeDir),
"git pull;",
"make all;",
// The main command that runs create_pagesets on all workers.
fmt.Sprintf("create_pagesets --worker_num=%s --log_dir=%s --pageset_type=%s;", util.WORKER_NUM_KEYWORD, util.GLogDir, *pagesetType),
}
// Setting a 4 hour timeout since it may take a while to upload page sets to
// Google Storage when doing 10k page sets per worker.
if _, err := util.SSH(strings.Join(cmd, " "), util.Slaves, 4*time.Hour); err != nil {
glog.Errorf("Error while running cmd %s: %s", cmd, err)
return
}
*taskCompletedSuccessfully = true
}
func main() {
defer common.LogPanic()
master_common.Init()
// Send start email.
emailsArr := util.ParseEmails(*emails)
emailsArr = append(emailsArr, util.CtAdmins...)
if len(emailsArr) == 0 {
glog.Error("At least one email address must be specified")
return
}
skutil.LogErr(frontend.UpdateWebappTaskSetStarted(&admin_tasks.RecreatePageSetsUpdateVars{}, *gaeTaskID))
skutil.LogErr(util.SendTaskStartEmail(emailsArr, "Creating pagesets", util.GetMasterLogLink(*runID), ""))
// Ensure webapp is updated and completion email is sent even if task fails.
defer updateWebappTask()
defer sendEmail(emailsArr)
if !*master_common.Local {
// Cleanup tmp files after the run.
defer util.CleanTmpDir()
}
// Finish with glog flush and how long the task took.
defer util.TimeTrack(time.Now(), "Creating Pagesets on Workers")
defer glog.Flush()
if *pagesetType == "" {
glog.Error("Must specify --pageset_type")
return
}
cmd := append(master_common.WorkerSetupCmds(),
// The main command that runs create_pagesets on all workers.
fmt.Sprintf(
"create_pagesets --worker_num=%s --log_dir=%s --log_id=%s --pageset_type=%s --local=%t;",
util.WORKER_NUM_KEYWORD, util.GLogDir, *runID, *pagesetType, *master_common.Local))
_, err := util.SSH(strings.Join(cmd, " "), util.Slaves, util.CREATE_PAGESETS_TIMEOUT)
if err != nil {
glog.Errorf("Error while running cmd %s: %s", cmd, err)
return
}
*taskCompletedSuccessfully = true
}
func main() {
common.Init()
webhook.MustInitRequestSaltFromFile(util.WebhookRequestSaltPath)
// Send start email.
emailsArr := util.ParseEmails(*emails)
emailsArr = append(emailsArr, util.CtAdmins...)
if len(emailsArr) == 0 {
glog.Error("At least one email address must be specified")
return
}
skutil.LogErr(frontend.UpdateWebappTaskSetStarted(&frontend.ChromiumPerfUpdateVars{}, *gaeTaskID))
skutil.LogErr(util.SendTaskStartEmail(emailsArr, "Chromium perf"))
// Ensure webapp is updated and email is sent even if task fails.
defer updateWebappTask()
defer sendEmail(emailsArr)
// Cleanup dirs after run completes.
defer skutil.RemoveAll(filepath.Join(util.StorageDir, util.ChromiumPerfRunsDir))
defer skutil.RemoveAll(filepath.Join(util.StorageDir, util.BenchmarkRunsDir))
// Cleanup tmp files after the run.
defer util.CleanTmpDir()
// Finish with glog flush and how long the task took.
defer util.TimeTrack(time.Now(), "Running chromium perf task on workers")
defer glog.Flush()
if *pagesetType == "" {
glog.Error("Must specify --pageset_type")
return
}
if *benchmarkName == "" {
glog.Error("Must specify --benchmark_name")
return
}
if *runID == "" {
glog.Error("Must specify --run_id")
return
}
// Instantiate GsUtil object.
gs, err := util.NewGsUtil(nil)
if err != nil {
glog.Errorf("Could not instantiate gsutil object: %s", err)
return
}
remoteOutputDir := filepath.Join(util.ChromiumPerfRunsDir, *runID)
// Copy the patches to Google Storage.
skiaPatchName := *runID + ".skia.patch"
blinkPatchName := *runID + ".blink.patch"
chromiumPatchName := *runID + ".chromium.patch"
for _, patchName := range []string{skiaPatchName, blinkPatchName, chromiumPatchName} {
if err := gs.UploadFile(patchName, os.TempDir(), remoteOutputDir); err != nil {
glog.Errorf("Could not upload %s to %s: %s", patchName, remoteOutputDir, err)
return
}
}
skiaPatchLink = util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, remoteOutputDir, skiaPatchName)
blinkPatchLink = util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, remoteOutputDir, blinkPatchName)
chromiumPatchLink = util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, remoteOutputDir, chromiumPatchName)
// Create the two required chromium builds (with patch and without the patch).
chromiumHash, skiaHash, err := util.CreateChromiumBuild(*runID, *targetPlatform, "", "", true)
if err != nil {
glog.Errorf("Could not create chromium build: %s", err)
return
}
// Reboot all workers to start from a clean slate.
util.RebootWorkers()
// Run the run_chromium_perf script on all workers.
runIDNoPatch := *runID + "-nopatch"
runIDWithPatch := *runID + "-withpatch"
chromiumBuildNoPatch := fmt.Sprintf("try-%s-%s-%s", chromiumHash, skiaHash, runIDNoPatch)
chromiumBuildWithPatch := fmt.Sprintf("try-%s-%s-%s", chromiumHash, skiaHash, runIDWithPatch)
runChromiumPerfCmdTemplate := "DISPLAY=:0 run_chromium_perf " +
"--worker_num={{.WorkerNum}} --log_dir={{.LogDir}} --pageset_type={{.PagesetType}} " +
"--chromium_build_nopatch={{.ChromiumBuildNoPatch}} --chromium_build_withpatch={{.ChromiumBuildWithPatch}} " +
"--run_id_nopatch={{.RunIDNoPatch}} --run_id_withpatch={{.RunIDWithPatch}} " +
"--benchmark_name={{.BenchmarkName}} --benchmark_extra_args=\"{{.BenchmarkExtraArgs}}\" " +
"--browser_extra_args_nopatch=\"{{.BrowserExtraArgsNoPatch}}\" --browser_extra_args_withpatch=\"{{.BrowserExtraArgsWithPatch}}\" " +
"--repeat_benchmark={{.RepeatBenchmark}} --target_platform={{.TargetPlatform}};"
runChromiumPerfTemplateParsed := template.Must(template.New("run_chromium_perf_cmd").Parse(runChromiumPerfCmdTemplate))
runChromiumPerfCmdBytes := new(bytes.Buffer)
if err := runChromiumPerfTemplateParsed.Execute(runChromiumPerfCmdBytes, struct {
WorkerNum string
LogDir string
PagesetType string
ChromiumBuildNoPatch string
ChromiumBuildWithPatch string
RunIDNoPatch string
RunIDWithPatch string
BenchmarkName string
BenchmarkExtraArgs string
BrowserExtraArgsNoPatch string
BrowserExtraArgsWithPatch string
RepeatBenchmark int
TargetPlatform string
}{
WorkerNum: util.WORKER_NUM_KEYWORD,
LogDir: util.GLogDir,
PagesetType: *pagesetType,
ChromiumBuildNoPatch: chromiumBuildNoPatch,
ChromiumBuildWithPatch: chromiumBuildWithPatch,
RunIDNoPatch: runIDNoPatch,
RunIDWithPatch: runIDWithPatch,
BenchmarkName: *benchmarkName,
BenchmarkExtraArgs: *benchmarkExtraArgs,
BrowserExtraArgsNoPatch: *browserExtraArgsNoPatch,
BrowserExtraArgsWithPatch: *browserExtraArgsWithPatch,
RepeatBenchmark: *repeatBenchmark,
TargetPlatform: *targetPlatform,
}); err != nil {
glog.Errorf("Failed to execute template: %s", err)
return
}
cmd := []string{
fmt.Sprintf("cd %s;", util.CtTreeDir),
"git pull;",
"make all;",
// The main command that runs run_chromium_perf on all workers.
runChromiumPerfCmdBytes.String(),
}
// Setting a 1 day timeout since it may take a while run benchmarks with many
// repeats.
if _, err := util.SSH(strings.Join(cmd, " "), util.Slaves, 1*24*time.Hour); err != nil {
glog.Errorf("Error while running cmd %s: %s", cmd, err)
return
}
// If "--output-format=csv-pivot-table" was specified then merge all CSV files and upload.
if strings.Contains(*benchmarkExtraArgs, "--output-format=csv-pivot-table") {
for _, runID := range []string{runIDNoPatch, runIDWithPatch} {
if err := mergeUploadCSVFiles(runID, gs); err != nil {
glog.Errorf("Unable to merge and upload CSV files for %s: %s", runID, err)
}
}
}
// Compare the resultant CSV files using csv_comparer.py
noPatchCSVPath := filepath.Join(util.StorageDir, util.BenchmarkRunsDir, runIDNoPatch, runIDNoPatch+".output")
withPatchCSVPath := filepath.Join(util.StorageDir, util.BenchmarkRunsDir, runIDWithPatch, runIDWithPatch+".output")
htmlOutputDir := filepath.Join(util.StorageDir, util.ChromiumPerfRunsDir, *runID, "html")
skutil.MkdirAll(htmlOutputDir, 0700)
htmlRemoteDir := filepath.Join(remoteOutputDir, "html")
htmlOutputLinkBase := util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, htmlRemoteDir) + "/"
htmlOutputLink = htmlOutputLinkBase + "index.html"
noPatchOutputLink = util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, util.BenchmarkRunsDir, runIDNoPatch, "consolidated_outputs", runIDNoPatch+".output")
withPatchOutputLink = util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, util.BenchmarkRunsDir, runIDWithPatch, "consolidated_outputs", runIDWithPatch+".output")
// Construct path to the csv_comparer python script.
_, currentFile, _, _ := runtime.Caller(0)
pathToPyFiles := filepath.Join(
filepath.Dir((filepath.Dir(filepath.Dir(filepath.Dir(currentFile))))),
"py")
pathToCsvComparer := filepath.Join(pathToPyFiles, "csv_comparer.py")
args := []string{
pathToCsvComparer,
"--csv_file1=" + noPatchCSVPath,
"--csv_file2=" + withPatchCSVPath,
"--output_html=" + htmlOutputDir,
"--variance_threshold=" + strconv.FormatFloat(*varianceThreshold, 'f', 2, 64),
"--discard_outliers=" + strconv.FormatFloat(*discardOutliers, 'f', 2, 64),
"--absolute_url=" + htmlOutputLinkBase,
"--requester_email=" + *emails,
"--skia_patch_link=" + skiaPatchLink,
"--blink_patch_link=" + blinkPatchLink,
"--chromium_patch_link=" + chromiumPatchLink,
"--raw_csv_nopatch=" + noPatchOutputLink,
"--raw_csv_withpatch=" + withPatchOutputLink,
"--num_repeated=" + strconv.Itoa(*repeatBenchmark),
"--target_platform=" + *targetPlatform,
"--browser_args_nopatch=" + *browserExtraArgsNoPatch,
"--browser_args_withpatch=" + *browserExtraArgsWithPatch,
"--pageset_type=" + *pagesetType,
"--chromium_hash=" + chromiumHash,
"--skia_hash=" + skiaHash,
}
if err := util.ExecuteCmd("python", args, []string{}, 2*time.Hour, nil, nil); err != nil {
glog.Errorf("Error running csv_comparer.py: %s", err)
return
}
// Copy the HTML files to Google Storage.
if err := gs.UploadDir(htmlOutputDir, htmlRemoteDir, true); err != nil {
glog.Errorf("Could not upload %s to %s: %s", htmlOutputDir, htmlRemoteDir, err)
return
}
taskCompletedSuccessfully = true
}
func main() {
common.Init()
webhook.MustInitRequestSaltFromFile(util.WebhookRequestSaltPath)
// Send start email.
emailsArr := util.ParseEmails(*emails)
emailsArr = append(emailsArr, util.CtAdmins...)
if len(emailsArr) == 0 {
glog.Error("At least one email address must be specified")
return
}
skutil.LogErr(frontend.UpdateWebappTaskSetStarted(&frontend.CaptureSkpsUpdateVars{}, *gaeTaskID))
skutil.LogErr(util.SendTaskStartEmail(emailsArr, "Capture SKPs"))
// Ensure webapp is updated and completion email is sent even if task
// fails.
defer updateWebappTask()
defer sendEmail(emailsArr)
// Cleanup tmp files after the run.
defer util.CleanTmpDir()
// Finish with glog flush and how long the task took.
defer util.TimeTrack(time.Now(), "Running capture skps task on workers")
defer glog.Flush()
if *pagesetType == "" {
glog.Error("Must specify --pageset_type")
return
}
if *chromiumBuild == "" {
glog.Error("Must specify --chromium_build")
return
}
if *runID == "" {
glog.Error("Must specify --run_id")
return
}
// Run the capture_skps script on all workers.
captureSKPsCmdTemplate := "DISPLAY=:0 capture_skps --worker_num={{.WorkerNum}} --log_dir={{.LogDir}} " +
"--pageset_type={{.PagesetType}} --chromium_build={{.ChromiumBuild}} --run_id={{.RunID}} " +
"--target_platform={{.TargetPlatform}};"
captureSKPsTemplateParsed := template.Must(template.New("capture_skps_cmd").Parse(captureSKPsCmdTemplate))
captureSKPsCmdBytes := new(bytes.Buffer)
if err := captureSKPsTemplateParsed.Execute(captureSKPsCmdBytes, struct {
WorkerNum string
LogDir string
PagesetType string
ChromiumBuild string
RunID string
TargetPlatform string
}{
WorkerNum: util.WORKER_NUM_KEYWORD,
LogDir: util.GLogDir,
PagesetType: *pagesetType,
ChromiumBuild: *chromiumBuild,
RunID: *runID,
TargetPlatform: *targetPlatform,
}); err != nil {
glog.Errorf("Failed to execute template: %s", err)
return
}
cmd := []string{
fmt.Sprintf("cd %s;", util.CtTreeDir),
"git pull;",
"make all;",
// The main command that runs capture_skps on all workers.
captureSKPsCmdBytes.String(),
}
// Setting a 2 day timeout since it may take a while to capture 1M SKPs.
if _, err := util.SSH(strings.Join(cmd, " "), util.Slaves, 2*24*time.Hour); err != nil {
glog.Errorf("Error while running cmd %s: %s", cmd, err)
return
}
taskCompletedSuccessfully = true
}
func main() {
defer common.LogPanic()
common.Init()
defer glog.Flush()
// Collect unhealthy machines
offlineMachines := []string{}
offlineDevices := []string{}
missingDevices := []string{}
nonResponsiveDevices := []string{}
// Also collect healthy machines for additional checks.
healthyMachines := []string{}
deviceOfflineOutputs, err := util.SSH("adb devices", util.Slaves, util.ADB_DEVICES_TIMEOUT)
if err != nil {
glog.Fatalf("Error while sshing into workers: %s", err)
return
}
// Populate offlineMachines, offlineDevices and missingDevices.
for hostname, out := range deviceOfflineOutputs {
if out == "" {
offlineMachines = append(offlineMachines, hostname)
glog.Warningf("%s is offline", hostname)
} else if strings.Contains(out, "offline") {
// The adb output text contains offline devices.
offlineDevices = append(offlineDevices, hostname)
glog.Warningf("%s has an offline device", hostname)
} else if strings.Count(out, "device") == 1 {
// The adb output text only contains "List of devices attached"
// without any devices listed below it.
missingDevices = append(missingDevices, hostname)
glog.Warningf("%s has missing devices", hostname)
} else {
// Everything seems fine so far, add this machine as a healthyMachine.
healthyMachines = append(healthyMachines, hostname)
}
}
// Populate nonResponsiveDevices.
responsivenessOutputs, err :=
util.SSH("adb shell uptime", healthyMachines, util.ADB_SHELL_UPTIME_TIMEOUT)
if err != nil {
glog.Fatalf("Error while sshing into workers: %s", err)
return
}
// Clear and repopulate the healthy machines slice.
healthyMachines = nil
for hostname, out := range responsivenessOutputs {
if out == "" {
nonResponsiveDevices = append(nonResponsiveDevices, hostname)
glog.Warningf("%s has non-responsive devices.")
} else {
// Everything seems fine so far, add this machine as a healthyMachine.
healthyMachines = append(healthyMachines, hostname)
}
}
// Email admins if there are any unhealthy machines.
if len(offlineMachines) != 0 || len(offlineDevices) != 0 || len(missingDevices) != 0 || len(nonResponsiveDevices) != 0 {
emailSubject := "There are unhealthy Cluster telemetry machines"
emailBody := "Please file a ticket to chrome-golo-tech-ticket@ (for offline devices) and chrome-labs-tech-ticket@ (for offline machines) using https://docs.google.com/spreadsheets/d/1whlE4nDJB0XFBemJliupOORepdXf_vXyAfFgsprTAxY/edit#gid=0 for-<br/><br/>"
if len(offlineMachines) != 0 {
emailBody += fmt.Sprintf("The following machines are offline: %s<br/>", strings.Join(offlineMachines, ","))
}
if len(offlineDevices) != 0 {
emailBody += fmt.Sprintf("The following machines have offline devices: %s<br/>", strings.Join(offlineDevices, ","))
}
if len(missingDevices) != 0 {
emailBody += fmt.Sprintf("The following machines have missing devices: %s<br/>", strings.Join(missingDevices, ","))
}
if len(nonResponsiveDevices) != 0 {
emailBody += fmt.Sprintf("The following machines have non-responsive devices: %s<br/>", strings.Join(nonResponsiveDevices, ","))
}
if err := util.SendEmail(util.CtAdmins, emailSubject, emailBody); err != nil {
glog.Errorf("Error while sending email: %s", err)
return
}
} else {
glog.Info("All CT machines are healthy")
}
}
func main() {
defer common.LogPanic()
master_common.Init()
// Send start email.
emailsArr := util.ParseEmails(*emails)
emailsArr = append(emailsArr, util.CtAdmins...)
if len(emailsArr) == 0 {
glog.Error("At least one email address must be specified")
return
}
skutil.LogErr(frontend.UpdateWebappTaskSetStarted(&capture_skps.UpdateVars{}, *gaeTaskID))
skutil.LogErr(util.SendTaskStartEmail(emailsArr, "Capture SKPs", util.GetMasterLogLink(*runID), *description))
// Ensure webapp is updated and completion email is sent even if task
// fails.
defer updateWebappTask()
defer sendEmail(emailsArr)
if !*master_common.Local {
// Cleanup tmp files after the run.
defer util.CleanTmpDir()
}
// Finish with glog flush and how long the task took.
defer util.TimeTrack(time.Now(), "Running capture skps task on workers")
defer glog.Flush()
if *pagesetType == "" {
glog.Error("Must specify --pageset_type")
return
}
if *chromiumBuild == "" {
glog.Error("Must specify --chromium_build")
return
}
if *runID == "" {
glog.Error("Must specify --run_id")
return
}
// Run the capture_skps script on all workers.
captureSKPsCmdTemplate := "DISPLAY=:0 capture_skps --worker_num={{.WorkerNum}} --log_dir={{.LogDir}} --log_id={{.RunID}} " +
"--pageset_type={{.PagesetType}} --chromium_build={{.ChromiumBuild}} --run_id={{.RunID}} " +
"--target_platform={{.TargetPlatform}} --local={{.Local}};"
captureSKPsTemplateParsed := template.Must(template.New("capture_skps_cmd").Parse(captureSKPsCmdTemplate))
captureSKPsCmdBytes := new(bytes.Buffer)
if err := captureSKPsTemplateParsed.Execute(captureSKPsCmdBytes, struct {
WorkerNum string
LogDir string
PagesetType string
ChromiumBuild string
RunID string
TargetPlatform string
Local bool
}{
WorkerNum: util.WORKER_NUM_KEYWORD,
LogDir: util.GLogDir,
PagesetType: *pagesetType,
ChromiumBuild: *chromiumBuild,
RunID: *runID,
TargetPlatform: *targetPlatform,
Local: *master_common.Local,
}); err != nil {
glog.Errorf("Failed to execute template: %s", err)
return
}
cmd := append(master_common.WorkerSetupCmds(),
// The main command that runs capture_skps on all workers.
captureSKPsCmdBytes.String())
_, err := util.SSH(strings.Join(cmd, " "), util.Slaves, util.CAPTURE_SKPS_TIMEOUT)
if err != nil {
glog.Errorf("Error while running cmd %s: %s", cmd, err)
return
}
taskCompletedSuccessfully = true
}
func main() {
common.Init()
webhook.MustInitRequestSaltFromFile(util.WebhookRequestSaltPath)
// Send start email.
emailsArr := util.ParseEmails(*emails)
emailsArr = append(emailsArr, util.CtAdmins...)
if len(emailsArr) == 0 {
glog.Error("At least one email address must be specified")
return
}
skutil.LogErr(frontend.UpdateWebappTaskSetStarted(&frontend.LuaScriptUpdateVars{}, *gaeTaskID))
skutil.LogErr(util.SendTaskStartEmail(emailsArr, "Lua script"))
// Ensure webapp is updated and email is sent even if task fails.
defer updateWebappTask()
defer sendEmail(emailsArr)
// Cleanup tmp files after the run.
defer util.CleanTmpDir()
// Finish with glog flush and how long the task took.
defer util.TimeTrack(time.Now(), "Running Lua script on workers")
defer glog.Flush()
if *pagesetType == "" {
glog.Error("Must specify --pageset_type")
return
}
if *chromiumBuild == "" {
glog.Error("Must specify --chromium_build")
return
}
if *runID == "" {
glog.Error("Must specify --run_id")
return
}
// Instantiate GsUtil object.
gs, err := util.NewGsUtil(nil)
if err != nil {
glog.Error(err)
return
}
// Upload the lua script for this run to Google storage.
luaScriptName := *runID + ".lua"
defer skutil.Remove(filepath.Join(os.TempDir(), luaScriptName))
luaScriptRemoteDir := filepath.Join(util.LuaRunsDir, *runID, "scripts")
luaScriptRemoteLink = util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, luaScriptRemoteDir, luaScriptName)
if err := gs.UploadFile(luaScriptName, os.TempDir(), luaScriptRemoteDir); err != nil {
glog.Errorf("Could not upload %s to %s: %s", luaScriptName, luaScriptRemoteDir, err)
return
}
// Run the run_lua script on all workers.
runLuaCmdTemplate := "DISPLAY=:0 run_lua --worker_num={{.WorkerNum}} --log_dir={{.LogDir}} --pageset_type={{.PagesetType}} --chromium_build={{.ChromiumBuild}} --run_id={{.RunID}};"
runLuaTemplateParsed := template.Must(template.New("run_lua_cmd").Parse(runLuaCmdTemplate))
luaCmdBytes := new(bytes.Buffer)
if err := runLuaTemplateParsed.Execute(luaCmdBytes, struct {
WorkerNum string
LogDir string
PagesetType string
ChromiumBuild string
RunID string
}{
WorkerNum: util.WORKER_NUM_KEYWORD,
LogDir: util.GLogDir,
PagesetType: *pagesetType,
ChromiumBuild: *chromiumBuild,
RunID: *runID,
}); err != nil {
glog.Errorf("Failed to execute template: %s", err)
return
}
cmd := []string{
fmt.Sprintf("cd %s;", util.CtTreeDir),
"git pull;",
"make all;",
// The main command that runs run_lua on all workers.
luaCmdBytes.String(),
}
if _, err := util.SSH(strings.Join(cmd, " "), util.Slaves, 2*time.Hour); err != nil {
glog.Errorf("Error while running cmd %s: %s", cmd, err)
return
}
// Copy outputs from all slaves locally and combine it into one file.
consolidatedFileName := "lua-output"
consolidatedLuaOutput := filepath.Join(os.TempDir(), consolidatedFileName)
if err := ioutil.WriteFile(consolidatedLuaOutput, []byte{}, 0660); err != nil {
glog.Errorf("Could not create %s: %s", consolidatedLuaOutput, err)
return
}
for i := 0; i < util.NUM_WORKERS; i++ {
workerNum := i + 1
workerRemoteOutputPath := filepath.Join(util.LuaRunsDir, *runID, fmt.Sprintf("slave%d", workerNum), "outputs", *runID+".output")
respBody, err := gs.GetRemoteFileContents(workerRemoteOutputPath)
if err != nil {
glog.Errorf("Could not fetch %s: %s", workerRemoteOutputPath, err)
// TODO(rmistry): Should we instead return here? We can only return
// here if all 100 slaves reliably run without any failures which they
// really should.
continue
}
defer skutil.Close(respBody)
out, err := os.OpenFile(consolidatedLuaOutput, os.O_RDWR|os.O_APPEND, 0660)
if err != nil {
glog.Errorf("Unable to open file %s: %s", consolidatedLuaOutput, err)
return
}
defer skutil.Close(out)
defer skutil.Remove(consolidatedLuaOutput)
if _, err = io.Copy(out, respBody); err != nil {
glog.Errorf("Unable to write out %s to %s: %s", workerRemoteOutputPath, consolidatedLuaOutput, err)
return
}
}
// Copy the consolidated file into Google Storage.
consolidatedOutputRemoteDir := filepath.Join(util.LuaRunsDir, *runID, "consolidated_outputs")
luaOutputRemoteLink = util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, consolidatedOutputRemoteDir, consolidatedFileName)
if err := gs.UploadFile(consolidatedFileName, os.TempDir(), consolidatedOutputRemoteDir); err != nil {
glog.Errorf("Unable to upload %s to %s: %s", consolidatedLuaOutput, consolidatedOutputRemoteDir, err)
return
}
// Upload the lua aggregator (if specified) for this run to Google storage.
luaAggregatorName := *runID + ".aggregator"
luaAggregatorPath := filepath.Join(os.TempDir(), luaAggregatorName)
defer skutil.Remove(luaAggregatorPath)
luaAggregatorRemoteLink = util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, luaScriptRemoteDir, luaAggregatorName)
luaAggregatorFileInfo, err := os.Stat(luaAggregatorPath)
if !os.IsNotExist(err) && luaAggregatorFileInfo.Size() > 10 {
if err := gs.UploadFile(luaAggregatorName, os.TempDir(), luaScriptRemoteDir); err != nil {
glog.Errorf("Could not upload %s to %s: %s", luaAggregatorName, luaScriptRemoteDir, err)
return
}
// Run the aggregator and save stdout.
luaAggregatorOutputFileName := *runID + ".agg.output"
luaAggregatorOutputFilePath := filepath.Join(os.TempDir(), luaAggregatorOutputFileName)
luaAggregatorOutputFile, err := os.Create(luaAggregatorOutputFilePath)
defer skutil.Close(luaAggregatorOutputFile)
defer skutil.Remove(luaAggregatorOutputFilePath)
if err != nil {
glog.Errorf("Could not create %s: %s", luaAggregatorOutputFilePath, err)
return
}
if err := util.ExecuteCmd(util.BINARY_LUA, []string{luaAggregatorPath}, []string{}, time.Hour, luaAggregatorOutputFile, nil); err != nil {
glog.Errorf("Could not execute the lua aggregator %s: %s", luaAggregatorPath, err)
return
}
// Copy the aggregator output into Google Storage.
luaAggregatorOutputRemoteLink = util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, consolidatedOutputRemoteDir, luaAggregatorOutputFileName)
if err := gs.UploadFile(luaAggregatorOutputFileName, os.TempDir(), consolidatedOutputRemoteDir); err != nil {
glog.Errorf("Unable to upload %s to %s: %s", luaAggregatorOutputFileName, consolidatedOutputRemoteDir, err)
return
}
} else {
glog.Info("A lua aggregator has not been specified.")
}
taskCompletedSuccessfully = true
}
func main() {
common.Init()
webhook.MustInitRequestSaltFromFile(util.WebhookRequestSaltPath)
// Send start email.
emailsArr := util.ParseEmails(*emails)
emailsArr = append(emailsArr, util.CtAdmins...)
if len(emailsArr) == 0 {
glog.Error("At least one email address must be specified")
return
}
skutil.LogErr(util.SendTaskStartEmail(emailsArr, "Skia correctness"))
// Ensure webapp is updated and email is sent even if task fails.
defer updateWebappTask()
defer sendEmail(emailsArr)
// Cleanup tmp files after the run.
defer util.CleanTmpDir()
// Finish with glog flush and how long the task took.
defer util.TimeTrack(time.Now(), "Running skia correctness task on workers")
defer glog.Flush()
if *pagesetType == "" {
glog.Error("Must specify --pageset_type")
return
}
if *chromiumBuild == "" {
glog.Error("Must specify --chromium_build")
return
}
if *runID == "" {
glog.Error("Must specify --run_id")
return
}
// Instantiate GsUtil object.
gs, err := util.NewGsUtil(nil)
if err != nil {
glog.Error(err)
return
}
remoteOutputDir := filepath.Join(util.SkiaCorrectnessRunsDir, *runID)
// Copy the patch to Google Storage.
patchName := *runID + ".patch"
patchRemoteDir := filepath.Join(remoteOutputDir, "patches")
if err := gs.UploadFile(patchName, os.TempDir(), patchRemoteDir); err != nil {
glog.Errorf("Could not upload %s to %s: %s", patchName, patchRemoteDir, err)
return
}
skiaPatchLink = util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, patchRemoteDir, patchName)
// Run the run_skia_correctness script on all workers.
runSkiaCorrCmdTemplate := "DISPLAY=:0 run_skia_correctness --worker_num={{.WorkerNum}} --log_dir={{.LogDir}} " +
"--pageset_type={{.PagesetType}} --chromium_build={{.ChromiumBuild}} --run_id={{.RunID}} " +
"--render_pictures_args=\"{{.RenderPicturesArgs}}\" --gpu_nopatch_run={{.GpuNoPatchRun}} " +
"--gpu_withpatch_run={{.GpuWithPatchRun}};"
runSkiaCorrTemplateParsed := template.Must(template.New("run_skia_correctness_cmd").Parse(runSkiaCorrCmdTemplate))
runSkiaCorrCmdBytes := new(bytes.Buffer)
if err := runSkiaCorrTemplateParsed.Execute(runSkiaCorrCmdBytes, struct {
WorkerNum string
LogDir string
PagesetType string
ChromiumBuild string
RunID string
RenderPicturesArgs string
GpuNoPatchRun string
GpuWithPatchRun string
}{
WorkerNum: util.WORKER_NUM_KEYWORD,
LogDir: util.GLogDir,
PagesetType: *pagesetType,
ChromiumBuild: *chromiumBuild,
RunID: *runID,
RenderPicturesArgs: *renderPicturesArgs,
GpuNoPatchRun: strconv.FormatBool(*gpuNoPatchRun),
GpuWithPatchRun: strconv.FormatBool(*gpuWithPatchRun),
}); err != nil {
glog.Errorf("Failed to execute template: %s", err)
return
}
cmd := []string{
fmt.Sprintf("cd %s;", util.CtTreeDir),
"git pull;",
"make all;",
// The main command that runs run_skia_correctness on all workers.
runSkiaCorrCmdBytes.String(),
}
if _, err := util.SSH(strings.Join(cmd, " "), util.Slaves, 4*time.Hour); err != nil {
glog.Errorf("Error while running cmd %s: %s", cmd, err)
return
}
localOutputDir := filepath.Join(util.StorageDir, util.SkiaCorrectnessRunsDir, *runID)
localSummariesDir := filepath.Join(localOutputDir, "summaries")
skutil.MkdirAll(localSummariesDir, 0700)
defer skutil.RemoveAll(filepath.Join(util.StorageDir, util.SkiaCorrectnessRunsDir))
// Copy outputs from all slaves locally.
for i := 0; i < util.NUM_WORKERS; i++ {
workerNum := i + 1
workerLocalOutputPath := filepath.Join(localSummariesDir, fmt.Sprintf("slave%d", workerNum)+".json")
workerRemoteOutputPath := filepath.Join(remoteOutputDir, fmt.Sprintf("slave%d", workerNum), fmt.Sprintf("slave%d", workerNum)+".json")
respBody, err := gs.GetRemoteFileContents(workerRemoteOutputPath)
if err != nil {
glog.Errorf("Could not fetch %s: %s", workerRemoteOutputPath, err)
// TODO(rmistry): Should we instead return here? We can only return
// here if all 100 slaves reliably run without any failures which they
// really should.
continue
}
defer skutil.Close(respBody)
out, err := os.Create(workerLocalOutputPath)
if err != nil {
glog.Errorf("Unable to create file %s: %s", workerLocalOutputPath, err)
return
}
defer skutil.Close(out)
defer skutil.Remove(workerLocalOutputPath)
if _, err = io.Copy(out, respBody); err != nil {
glog.Errorf("Unable to copy to file %s: %s", workerLocalOutputPath, err)
return
}
}
// Call json_summary_combiner.py to merge all results into a single results CSV.
_, currentFile, _, _ := runtime.Caller(0)
pathToPyFiles := filepath.Join(
filepath.Dir((filepath.Dir(filepath.Dir(filepath.Dir(currentFile))))),
"py")
pathToJsonCombiner := filepath.Join(pathToPyFiles, "json_summary_combiner.py")
localHtmlDir := filepath.Join(localOutputDir, "html")
remoteHtmlDir := filepath.Join(remoteOutputDir, "html")
baseHtmlLink := util.GS_HTTP_LINK + filepath.Join(util.GS_BUCKET_NAME, remoteHtmlDir) + "/"
htmlOutputLink = baseHtmlLink + "index.html"
skutil.MkdirAll(localHtmlDir, 0700)
args := []string{
pathToJsonCombiner,
"--json_summaries_dir=" + localSummariesDir,
"--output_html_dir=" + localHtmlDir,
"--absolute_url=" + baseHtmlLink,
"--render_pictures_args=" + *renderPicturesArgs,
"--nopatch_gpu=" + strconv.FormatBool(*gpuNoPatchRun),
"--withpatch_gpu=" + strconv.FormatBool(*gpuWithPatchRun),
}
if err := util.ExecuteCmd("python", args, []string{}, 1*time.Hour, nil, nil); err != nil {
glog.Errorf("Error running json_summary_combiner.py: %s", err)
return
}
// Copy the HTML files to Google Storage.
if err := gs.UploadDir(localHtmlDir, remoteHtmlDir, true); err != nil {
glog.Errorf("Could not upload %s to %s: %s", localHtmlDir, remoteHtmlDir, err)
return
}
taskCompletedSuccessfully = true
}
func main() {
defer common.LogPanic()
master_common.Init()
// Send start email.
emailsArr := util.ParseEmails(*emails)
emailsArr = append(emailsArr, util.CtAdmins...)
if len(emailsArr) == 0 {
glog.Error("At least one email address must be specified")
return
}
skutil.LogErr(util.SendTaskStartEmail(emailsArr, "Fix archives", util.GetMasterLogLink(*runID), ""))
// Ensure completion email is sent even if task fails.
defer sendEmail(emailsArr)
if !*master_common.Local {
// Cleanup tmp files after the run.
defer util.CleanTmpDir()
}
// Finish with glog flush and how long the task took.
defer util.TimeTrack(time.Now(), "Running fix archives task on workers")
defer glog.Flush()
if *pagesetType == "" {
glog.Error("Must specify --pageset_type")
return
}
if *chromiumBuild == "" {
glog.Error("Must specify --chromium_build")
return
}
if *runID == "" {
glog.Error("Must specify --run_id")
return
}
// Run the fix_archives script on all workers.
fixArchivesCmdTemplate := "DISPLAY=:0 fix_archives --worker_num={{.WorkerNum}} --log_dir={{.LogDir}} " +
"--pageset_type={{.PagesetType}} --chromium_build={{.ChromiumBuild}} --run_id={{.RunID}} " +
"--repeat_benchmark={{.RepeatBenchmark}} --benchmark_name={{.BenchmarkName}} " +
"--benchmark_header_to_check=\"{{.BenchmarkHeaderToCheck}}\" --delete_pageset={{.DeletePageset}} " +
"--perc_change_threshold={{.PercentageChangeThreshold}} --res_missing_count_threshold={{.ResourceMissingCountThreshold}} " +
"--local={{.Local}};"
fixArchivesTemplateParsed := template.Must(template.New("fix_archives_cmd").Parse(fixArchivesCmdTemplate))
fixArchivesCmdBytes := new(bytes.Buffer)
if err := fixArchivesTemplateParsed.Execute(fixArchivesCmdBytes, struct {
WorkerNum string
LogDir string
PagesetType string
ChromiumBuild string
RunID string
RepeatBenchmark int
BenchmarkName string
BenchmarkHeaderToCheck string
DeletePageset bool
PercentageChangeThreshold float64
ResourceMissingCountThreshold int
Local bool
}{
WorkerNum: util.WORKER_NUM_KEYWORD,
LogDir: util.GLogDir,
PagesetType: *pagesetType,
ChromiumBuild: *chromiumBuild,
RunID: *runID,
RepeatBenchmark: *repeatBenchmark,
BenchmarkName: *benchmarkName,
BenchmarkHeaderToCheck: *benchmarkHeaderToCheck,
DeletePageset: *deletePageset,
PercentageChangeThreshold: *percentageChangeThreshold,
ResourceMissingCountThreshold: *resourceMissingCountThreshold,
Local: *master_common.Local,
}); err != nil {
glog.Errorf("Failed to execute template: %s", err)
return
}
cmd := append(master_common.WorkerSetupCmds(),
// The main command that runs fix_archives on all workers.
fixArchivesCmdBytes.String())
if _, err := util.SSH(strings.Join(cmd, " "), util.Slaves, util.FIX_ARCHIVES_TIMEOUT); err != nil {
glog.Errorf("Error while running cmd %s: %s", cmd, err)
return
}
taskCompletedSuccessfully = true
}
