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 keepbalance
   6
   7 import (
   8         "bytes"
   9         "context"
  10         "flag"
  11         "fmt"
  12         "io"
  13         _ "net/http/pprof"
  14
  15         "git.arvados.org/arvados.git/lib/cmd"
  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 )
  25
  26 type command struct{}
  27
  28 var Command = command{}
  29
  30 func (command) RunCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
  31         var options RunOptions
  32         flags := flag.NewFlagSet(prog, flag.ContinueOnError)
  33         flags.BoolVar(&options.Once, "once", false,
  34                 "balance once and then exit")
  35         flags.BoolVar(&options.CommitPulls, "commit-pulls", false,
  36                 "send pull requests (make more replicas of blocks that are underreplicated or are not in optimal rendezvous probe order)")
  37         flags.BoolVar(&options.CommitTrash, "commit-trash", false,
  38                 "send trash requests (delete unreferenced old blocks, and excess replicas of overreplicated blocks)")
  39         flags.BoolVar(&options.CommitConfirmedFields, "commit-confirmed-fields", true,
  40                 "update collection fields (replicas_confirmed, storage_classes_confirmed, etc.)")
  41         flags.StringVar(&options.ChunkPrefix, "chunk-prefix", "",
  42                 "operate only on blocks with the given prefix (experimental, see https://dev.arvados.org/issues/19923)")
  43         // These options are implemented by service.Command, so we
  44         // don't need the vars here -- we just need the flags
  45         // to pass flags.Parse().
  46         flags.Bool("dump", false, "dump details for each block to stdout")
  47         flags.String("pprof", "", "serve Go profile data at `[addr]:port`")
  48         flags.Bool("version", false, "Write version information to stdout and exit 0")
  49
  50         logger := ctxlog.New(stderr, "json", "info")
  51         loader := config.NewLoader(&bytes.Buffer{}, logger)
  52         loader.SetupFlags(flags)
  53         munged := loader.MungeLegacyConfigArgs(logger, args, "-legacy-keepbalance-config")
  54         if ok, code := cmd.ParseFlags(flags, prog, munged, "", stderr); !ok {
  55                 return code
  56         }
  57
  58         // Drop our custom args that would be rejected by the generic
  59         // service.Command
  60         args = nil
  61         dropFlag := map[string]bool{
  62                 "once":                    true,
  63                 "commit-pulls":            true,
  64                 "commit-trash":            true,
  65                 "commit-confirmed-fields": true,
  66                 "dump":                    true,
  67         }
  68         flags.Visit(func(f *flag.Flag) {
  69                 if !dropFlag[f.Name] {
  70                         args = append(args, "-"+f.Name+"="+f.Value.String())
  71                 }
  72         })
  73
  74         return service.Command(arvados.ServiceNameKeepbalance,
  75                 func(ctx context.Context, cluster *arvados.Cluster, token string, registry *prometheus.Registry) service.Handler {
  76                         if !options.Once && cluster.Collections.BalancePeriod == arvados.Duration(0) {
  77                                 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)"))
  78                         }
  79
  80                         ac, err := arvados.NewClientFromConfig(cluster)
  81                         ac.AuthToken = token
  82                         if err != nil {
  83                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("error initializing client from cluster config: %s", err))
  84                         }
  85
  86                         db, err := sqlx.Open("postgres", cluster.PostgreSQL.Connection.String())
  87                         if err != nil {
  88                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("postgresql connection failed: %s", err))
  89                         }
  90                         if p := cluster.PostgreSQL.ConnectionPool; p > 0 {
  91                                 db.SetMaxOpenConns(p)
  92                         }
  93                         err = db.Ping()
  94                         if err != nil {
  95                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("postgresql connection succeeded but ping failed: %s", err))
  96                         }
  97
  98                         if options.Logger == nil {
  99                                 options.Logger = ctxlog.FromContext(ctx)
 100                         }
 101
 102                         srv := &Server{
 103                                 Cluster:    cluster,
 104                                 ArvClient:  ac,
 105                                 RunOptions: options,
 106                                 Metrics:    newMetrics(registry),
 107                                 Logger:     options.Logger,
 108                                 Dumper:     options.Dumper,
 109                                 DB:         db,
 110                         }
 111                         srv.Handler = &health.Handler{
 112                                 Token:  cluster.ManagementToken,
 113                                 Prefix: "/_health/",
 114                                 Routes: health.Routes{"ping": srv.CheckHealth},
 115                         }
 116
 117                         go srv.run(ctx)
 118                         return srv
 119                 }).RunCommand(prog, args, stdin, stdout, stderr)
 120 }