93457fb827483143ac67184dcc31946571775768
[arvados.git] / 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         "os"
13
14         "git.arvados.org/arvados.git/lib/config"
15         "git.arvados.org/arvados.git/lib/service"
16         "git.arvados.org/arvados.git/sdk/go/arvados"
17         "git.arvados.org/arvados.git/sdk/go/ctxlog"
18         "git.arvados.org/arvados.git/sdk/go/health"
19         "github.com/jmoiron/sqlx"
20         "github.com/prometheus/client_golang/prometheus"
21         "github.com/sirupsen/logrus"
22 )
23
24 func main() {
25         os.Exit(runCommand(os.Args[0], os.Args[1:], os.Stdin, os.Stdout, os.Stderr))
26 }
27
28 func runCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
29         logger := ctxlog.FromContext(context.Background())
30
31         var options RunOptions
32         flags := flag.NewFlagSet(prog, flag.ExitOnError)
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.Bool("version", false, "Write version information to stdout and exit 0")
42         dumpFlag := flags.Bool("dump", false, "dump details for each block to stdout")
43
44         loader := config.NewLoader(os.Stdin, logger)
45         loader.SetupFlags(flags)
46
47         munged := loader.MungeLegacyConfigArgs(logger, args, "-legacy-keepbalance-config")
48         flags.Parse(munged)
49
50         if *dumpFlag {
51                 dumper := logrus.New()
52                 dumper.Out = os.Stdout
53                 dumper.Formatter = &logrus.TextFormatter{}
54                 options.Dumper = dumper
55         }
56
57         // Drop our custom args that would be rejected by the generic
58         // service.Command
59         args = nil
60         dropFlag := map[string]bool{
61                 "once":                    true,
62                 "commit-pulls":            true,
63                 "commit-trash":            true,
64                 "commit-confirmed-fields": true,
65                 "dump":                    true,
66         }
67         flags.Visit(func(f *flag.Flag) {
68                 if !dropFlag[f.Name] {
69                         args = append(args, "-"+f.Name, f.Value.String())
70                 }
71         })
72
73         return service.Command(arvados.ServiceNameKeepbalance,
74                 func(ctx context.Context, cluster *arvados.Cluster, token string, registry *prometheus.Registry) service.Handler {
75                         if !options.Once && cluster.Collections.BalancePeriod == arvados.Duration(0) {
76                                 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)"))
77                         }
78
79                         ac, err := arvados.NewClientFromConfig(cluster)
80                         ac.AuthToken = token
81                         if err != nil {
82                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("error initializing client from cluster config: %s", err))
83                         }
84
85                         db, err := sqlx.Open("postgres", cluster.PostgreSQL.Connection.String())
86                         if err != nil {
87                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("postgresql connection failed: %s", err))
88                         }
89                         if p := cluster.PostgreSQL.ConnectionPool; p > 0 {
90                                 db.SetMaxOpenConns(p)
91                         }
92                         err = db.Ping()
93                         if err != nil {
94                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("postgresql connection succeeded but ping failed: %s", err))
95                         }
96
97                         if options.Logger == nil {
98                                 options.Logger = ctxlog.FromContext(ctx)
99                         }
100
101                         srv := &Server{
102                                 Cluster:    cluster,
103                                 ArvClient:  ac,
104                                 RunOptions: options,
105                                 Metrics:    newMetrics(registry),
106                                 Logger:     options.Logger,
107                                 Dumper:     options.Dumper,
108                                 DB:         db,
109                         }
110                         srv.Handler = &health.Handler{
111                                 Token:  cluster.ManagementToken,
112                                 Prefix: "/_health/",
113                                 Routes: health.Routes{"ping": srv.CheckHealth},
114                         }
115
116                         go srv.run()
117                         return srv
118                 }).RunCommand(prog, args, stdin, stdout, stderr)
119 }