Refactor the multi-host salt install page.
[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 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         // These options are implemented by service.Command, so we
42         // don't need the vars here -- we just need the flags
43         // to pass flags.Parse().
44         flags.Bool("dump", false, "dump details for each block to stdout")
45         flags.String("pprof", "", "serve Go profile data at `[addr]:port`")
46         flags.Bool("version", false, "Write version information to stdout and exit 0")
47
48         logger := ctxlog.New(stderr, "json", "info")
49         loader := config.NewLoader(&bytes.Buffer{}, logger)
50         loader.SetupFlags(flags)
51         munged := loader.MungeLegacyConfigArgs(logger, args, "-legacy-keepbalance-config")
52         if ok, code := cmd.ParseFlags(flags, prog, munged, "", stderr); !ok {
53                 return code
54         }
55
56         // Drop our custom args that would be rejected by the generic
57         // service.Command
58         args = nil
59         dropFlag := map[string]bool{
60                 "once":                    true,
61                 "commit-pulls":            true,
62                 "commit-trash":            true,
63                 "commit-confirmed-fields": true,
64                 "dump":                    true,
65         }
66         flags.Visit(func(f *flag.Flag) {
67                 if !dropFlag[f.Name] {
68                         args = append(args, "-"+f.Name+"="+f.Value.String())
69                 }
70         })
71
72         return service.Command(arvados.ServiceNameKeepbalance,
73                 func(ctx context.Context, cluster *arvados.Cluster, token string, registry *prometheus.Registry) service.Handler {
74                         if !options.Once && cluster.Collections.BalancePeriod == arvados.Duration(0) {
75                                 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)"))
76                         }
77
78                         ac, err := arvados.NewClientFromConfig(cluster)
79                         ac.AuthToken = token
80                         if err != nil {
81                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("error initializing client from cluster config: %s", err))
82                         }
83
84                         db, err := sqlx.Open("postgres", cluster.PostgreSQL.Connection.String())
85                         if err != nil {
86                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("postgresql connection failed: %s", err))
87                         }
88                         if p := cluster.PostgreSQL.ConnectionPool; p > 0 {
89                                 db.SetMaxOpenConns(p)
90                         }
91                         err = db.Ping()
92                         if err != nil {
93                                 return service.ErrorHandler(ctx, cluster, fmt.Errorf("postgresql connection succeeded but ping failed: %s", err))
94                         }
95
96                         if options.Logger == nil {
97                                 options.Logger = ctxlog.FromContext(ctx)
98                         }
99
100                         srv := &Server{
101                                 Cluster:    cluster,
102                                 ArvClient:  ac,
103                                 RunOptions: options,
104                                 Metrics:    newMetrics(registry),
105                                 Logger:     options.Logger,
106                                 Dumper:     options.Dumper,
107                                 DB:         db,
108                         }
109                         srv.Handler = &health.Handler{
110                                 Token:  cluster.ManagementToken,
111                                 Prefix: "/_health/",
112                                 Routes: health.Routes{"ping": srv.CheckHealth},
113                         }
114
115                         go srv.run()
116                         return srv
117                 }).RunCommand(prog, args, stdin, stdout, stderr)
118 }