services/keep-balance/main.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package main
   6
   7 import (
   8         "context"
   9         "flag"
  10         "fmt"
  11         "io"
  12         "net/http"
  13         _ "net/http/pprof"
  14         "os"
  15
  16         "git.arvados.org/arvados.git/lib/config"
  17         "git.arvados.org/arvados.git/lib/service"
  18         "git.arvados.org/arvados.git/sdk/go/arvados"
  19         "git.arvados.org/arvados.git/sdk/go/ctxlog"
  20         "git.arvados.org/arvados.git/sdk/go/health"
  21         "github.com/jmoiron/sqlx"
  22         _ "github.com/lib/pq"
  23         "github.com/prometheus/client_golang/prometheus"
  24         "github.com/sirupsen/logrus"
  25 )
  26
  27 func main() {
  28         os.Exit(runCommand(os.Args[0], os.Args[1:], os.Stdin, os.Stdout, os.Stderr))
  29 }
  30
  31 func runCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
  32         logger := ctxlog.FromContext(context.Background())
  33
  34         var options RunOptions
  35         flags := flag.NewFlagSet(prog, flag.ExitOnError)
  36         flags.BoolVar(&options.Once, "once", false,
  37                 "balance once and then exit")
  38         flags.BoolVar(&options.CommitPulls, "commit-pulls", false,
  39                 "send pull requests (make more replicas of blocks that are underreplicated or are not in optimal rendezvous probe order)")
  40         flags.BoolVar(&options.CommitTrash, "commit-trash", false,
  41                 "send trash requests (delete unreferenced old blocks, and excess replicas of overreplicated blocks)")
  42         flags.BoolVar(&options.CommitConfirmedFields, "commit-confirmed-fields", true,
  43                 "update collection fields (replicas_confirmed, storage_classes_confirmed, etc.)")
  44         flags.Bool("version", false, "Write version information to stdout and exit 0")
  45         dumpFlag := flags.Bool("dump", false, "dump details for each block to stdout")
  46         pprofAddr := flags.String("pprof", "", "serve Go profile data at `[addr]:port`")
  47
  48         if *pprofAddr != "" {
  49                 go func() {
  50                         logrus.Println(http.ListenAndServe(*pprofAddr, nil))
  51                 }()
  52         }
  53
  54         loader := config.NewLoader(os.Stdin, logger)
  55         loader.SetupFlags(flags)
  56
  57         munged := loader.MungeLegacyConfigArgs(logger, args, "-legacy-keepbalance-config")
  58         flags.Parse(munged)
  59
  60         if *dumpFlag {
  61                 dumper := logrus.New()
  62                 dumper.Out = os.Stdout
  63                 dumper.Formatter = &logrus.TextFormatter{}
  64                 options.Dumper = dumper
  65         }
  66
  67         // Drop our custom args that would be rejected by the generic
  68         // service.Command
  69         args = nil
  70         dropFlag := map[string]bool{
  71                 "once":                    true,
  72                 "commit-pulls":            true,
  73                 "commit-trash":            true,
  74                 "commit-confirmed-fields": true,
  75                 "dump":                    true,
  76         }
  77         flags.Visit(func(f *flag.Flag) {
  78                 if !dropFlag[f.Name] {
  79                         args = append(args, "-"+f.Name, f.Value.String())
  80                 }
  81         })
  82
  83         return service.Command(arvados.ServiceNameKeepbalance,
  84                 func(ctx context.Context, cluster *arvados.Cluster, token string, registry *prometheus.Registry) service.Handler {
  85                         if !options.Once && cluster.Collections.BalancePeriod == arvados.Duration(0) {
  86                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("cannot start service: Collections.BalancePeriod is zero (if you want to run once and then exit, use the -once flag)"))
  87                         }
  88
  89                         ac, err := arvados.NewClientFromConfig(cluster)
  90                         ac.AuthToken = token
  91                         if err != nil {
  92                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("error initializing client from cluster config: %s", err))
  93                         }
  94
  95                         db, err := sqlx.Open("postgres", cluster.PostgreSQL.Connection.String())
  96                         if err != nil {
  97                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("postgresql connection failed: %s", err))
  98                         }
  99                         if p := cluster.PostgreSQL.ConnectionPool; p > 0 {
 100                                 db.SetMaxOpenConns(p)
 101                         }
 102                         err = db.Ping()
 103                         if err != nil {
 104                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("postgresql connection succeeded but ping failed: %s", err))
 105                         }
 106
 107                         if options.Logger == nil {
 108                                 options.Logger = ctxlog.FromContext(ctx)
 109                         }
 110
 111                         srv := &Server{
 112                                 Cluster:    cluster,
 113                                 ArvClient:  ac,
 114                                 RunOptions: options,
 115                                 Metrics:    newMetrics(registry),
 116                                 Logger:     options.Logger,
 117                                 Dumper:     options.Dumper,
 118                                 DB:         db,
 119                         }
 120                         srv.Handler = &health.Handler{
 121                                 Token:  cluster.ManagementToken,
 122                                 Prefix: "/_health/",
 123                                 Routes: health.Routes{"ping": srv.CheckHealth},
 124                         }
 125
 126                         go srv.run()
 127                         return srv
 128                 }).RunCommand(prog, args, stdin, stdout, stderr)
 129 }