|
| 1 | +// Copyright (c) 2019-present Mattermost, Inc. All Rights Reserved. |
| 2 | +// See LICENSE.txt for license information. |
| 3 | + |
| 4 | +package main |
| 5 | + |
| 6 | +import ( |
| 7 | + "context" |
| 8 | + "fmt" |
| 9 | + "io" |
| 10 | + "net" |
| 11 | + "os" |
| 12 | + "os/exec" |
| 13 | + "os/signal" |
| 14 | + "strconv" |
| 15 | + "strings" |
| 16 | + "syscall" |
| 17 | + |
| 18 | + "github.com/mattermost/mattermost-load-test-ng/deployment/terraform" |
| 19 | + "github.com/mattermost/mattermost-load-test-ng/deployment/terraform/ssh" |
| 20 | + |
| 21 | + "github.com/spf13/cobra" |
| 22 | +) |
| 23 | + |
| 24 | +func RunDBListCmdF(cmd *cobra.Command, args []string) error { |
| 25 | + config, err := getConfig(cmd) |
| 26 | + if err != nil { |
| 27 | + return err |
| 28 | + } |
| 29 | + |
| 30 | + t, err := terraform.New("", config) |
| 31 | + if err != nil { |
| 32 | + return fmt.Errorf("failed to create terraform engine: %w", err) |
| 33 | + } |
| 34 | + |
| 35 | + output, err := t.Output() |
| 36 | + if err != nil { |
| 37 | + return fmt.Errorf("could not parse output: %w", err) |
| 38 | + } |
| 39 | + |
| 40 | + if !output.HasDB() { |
| 41 | + return fmt.Errorf("no database cluster found in deployment") |
| 42 | + } |
| 43 | + |
| 44 | + readerIdx := 0 |
| 45 | + for _, inst := range output.DBCluster.Instances { |
| 46 | + var role string |
| 47 | + if inst.IsWriter { |
| 48 | + role = "writer" |
| 49 | + } else { |
| 50 | + role = fmt.Sprintf("reader-%d", readerIdx) |
| 51 | + readerIdx++ |
| 52 | + } |
| 53 | + fmt.Printf(" - %-10s %s (%s)\n", role, inst.Endpoint, inst.DBIdentifier) |
| 54 | + } |
| 55 | + |
| 56 | + return nil |
| 57 | +} |
| 58 | + |
| 59 | +// selectDBInstance picks a DB instance based on the target argument. |
| 60 | +// No args: first reader, or writer if only one instance exists. |
| 61 | +// "writer": the writer instance. |
| 62 | +// "reader-N": the reader at index N. |
| 63 | +func selectDBInstance(output *terraform.Output, args []string) (terraform.DBInstance, error) { |
| 64 | + instances := output.DBCluster.Instances |
| 65 | + |
| 66 | + if len(args) == 0 { |
| 67 | + // Default: first reader, fallback to writer if only one instance. |
| 68 | + for _, inst := range instances { |
| 69 | + if !inst.IsWriter { |
| 70 | + return inst, nil |
| 71 | + } |
| 72 | + } |
| 73 | + // No readers found — return the writer. |
| 74 | + for _, inst := range instances { |
| 75 | + if inst.IsWriter { |
| 76 | + return inst, nil |
| 77 | + } |
| 78 | + } |
| 79 | + return terraform.DBInstance{}, fmt.Errorf("no database instances found") |
| 80 | + } |
| 81 | + |
| 82 | + target := args[0] |
| 83 | + |
| 84 | + if target == "writer" { |
| 85 | + for _, inst := range instances { |
| 86 | + if inst.IsWriter { |
| 87 | + return inst, nil |
| 88 | + } |
| 89 | + } |
| 90 | + return terraform.DBInstance{}, fmt.Errorf("no writer instance found") |
| 91 | + } |
| 92 | + |
| 93 | + if idxStr, ok := strings.CutPrefix(target, "reader-"); ok { |
| 94 | + idx, err := strconv.Atoi(idxStr) |
| 95 | + if err != nil { |
| 96 | + return terraform.DBInstance{}, fmt.Errorf("invalid reader index %q", idxStr) |
| 97 | + } |
| 98 | + |
| 99 | + readerIdx := 0 |
| 100 | + for _, inst := range instances { |
| 101 | + if !inst.IsWriter { |
| 102 | + if readerIdx == idx { |
| 103 | + return inst, nil |
| 104 | + } |
| 105 | + readerIdx++ |
| 106 | + } |
| 107 | + } |
| 108 | + return terraform.DBInstance{}, fmt.Errorf("reader-%d not found (have %d readers)", idx, readerIdx) |
| 109 | + } |
| 110 | + |
| 111 | + return terraform.DBInstance{}, fmt.Errorf("invalid target %q: use \"writer\" or \"reader-N\"", target) |
| 112 | +} |
| 113 | + |
| 114 | +// selectJumpHost picks an EC2 instance to use as an SSH jump host. |
| 115 | +// Fallback chain: app server -> metrics server -> error. |
| 116 | +func selectJumpHost(output *terraform.Output) (terraform.Instance, error) { |
| 117 | + if output.HasAppServers() { |
| 118 | + return output.Instances[0], nil |
| 119 | + } |
| 120 | + if output.HasMetrics() { |
| 121 | + return output.MetricsServer, nil |
| 122 | + } |
| 123 | + return terraform.Instance{}, fmt.Errorf("no jump host available: need at least an app server or metrics server") |
| 124 | +} |
| 125 | + |
| 126 | +func RunDBConnectCmdF(cmd *cobra.Command, args []string) error { |
| 127 | + if os.Getenv("SSH_AUTH_SOCK") == "" { |
| 128 | + return fmt.Errorf("ssh agent not running. Please run eval \"$(ssh-agent -s)\" and then ssh-add") |
| 129 | + } |
| 130 | + |
| 131 | + config, err := getConfig(cmd) |
| 132 | + if err != nil { |
| 133 | + return err |
| 134 | + } |
| 135 | + |
| 136 | + // Validate engine is aurora-postgresql. |
| 137 | + if config.TerraformDBSettings.InstanceEngine != "aurora-postgresql" { |
| 138 | + return fmt.Errorf("only aurora-postgresql is supported, got %q", config.TerraformDBSettings.InstanceEngine) |
| 139 | + } |
| 140 | + |
| 141 | + t, err := terraform.New("", config) |
| 142 | + if err != nil { |
| 143 | + return fmt.Errorf("failed to create terraform engine: %w", err) |
| 144 | + } |
| 145 | + |
| 146 | + output, err := t.Output() |
| 147 | + if err != nil { |
| 148 | + return fmt.Errorf("could not parse output: %w", err) |
| 149 | + } |
| 150 | + |
| 151 | + if !output.HasDB() { |
| 152 | + return fmt.Errorf("no database cluster found in deployment") |
| 153 | + } |
| 154 | + |
| 155 | + // Select target DB instance and jump host. |
| 156 | + dbInst, err := selectDBInstance(output, args) |
| 157 | + if err != nil { |
| 158 | + return err |
| 159 | + } |
| 160 | + |
| 161 | + jumpHost, err := selectJumpHost(output) |
| 162 | + if err != nil { |
| 163 | + return err |
| 164 | + } |
| 165 | + |
| 166 | + // Establish SSH connection to the jump host. |
| 167 | + extAgent, err := ssh.NewAgent() |
| 168 | + if err != nil { |
| 169 | + return fmt.Errorf("failed to create SSH agent: %w", err) |
| 170 | + } |
| 171 | + |
| 172 | + sshClient, err := extAgent.NewClient(output.AMIUser, jumpHost.GetConnectionIP()) |
| 173 | + if err != nil { |
| 174 | + return fmt.Errorf("failed to connect to jump host %s: %w", jumpHost.GetConnectionIP(), err) |
| 175 | + } |
| 176 | + defer sshClient.Close() |
| 177 | + |
| 178 | + fmt.Printf("Connected to jump host %s\n", jumpHost.GetConnectionIP()) |
| 179 | + |
| 180 | + // Start local TCP listener on a free port. |
| 181 | + listener, err := net.Listen("tcp", "127.0.0.1:0") |
| 182 | + if err != nil { |
| 183 | + return fmt.Errorf("failed to start local listener: %w", err) |
| 184 | + } |
| 185 | + defer listener.Close() |
| 186 | + |
| 187 | + localAddr := listener.Addr().String() |
| 188 | + fmt.Printf("Tunnel listening on %s -> %s:5432\n", localAddr, dbInst.Endpoint) |
| 189 | + |
| 190 | + // Set up context with signal handling for clean shutdown. |
| 191 | + ctx, cancel := context.WithCancel(context.Background()) |
| 192 | + defer cancel() |
| 193 | + |
| 194 | + sigCh := make(chan os.Signal, 1) |
| 195 | + signal.Notify(sigCh, syscall.SIGINT, syscall.SIGTERM) |
| 196 | + go func() { |
| 197 | + <-sigCh |
| 198 | + cancel() |
| 199 | + listener.Close() |
| 200 | + }() |
| 201 | + |
| 202 | + // Accept connections in the background and tunnel them via SSH. |
| 203 | + dialF := sshClient.DialContextF() |
| 204 | + go func() { |
| 205 | + for { |
| 206 | + localConn, err := listener.Accept() |
| 207 | + if err != nil { |
| 208 | + // Listener closed — exit goroutine. |
| 209 | + return |
| 210 | + } |
| 211 | + go func(lc net.Conn) { |
| 212 | + defer lc.Close() |
| 213 | + remoteConn, err := dialF(ctx, "tcp", dbInst.Endpoint+":5432") |
| 214 | + if err != nil { |
| 215 | + fmt.Fprintf(os.Stderr, "failed to dial remote DB: %v\n", err) |
| 216 | + return |
| 217 | + } |
| 218 | + defer remoteConn.Close() |
| 219 | + |
| 220 | + // Bidirectional copy. |
| 221 | + done := make(chan struct{}, 2) |
| 222 | + go func() { |
| 223 | + io.Copy(remoteConn, lc) |
| 224 | + done <- struct{}{} |
| 225 | + }() |
| 226 | + go func() { |
| 227 | + io.Copy(lc, remoteConn) |
| 228 | + done <- struct{}{} |
| 229 | + }() |
| 230 | + <-done |
| 231 | + }(localConn) |
| 232 | + } |
| 233 | + }() |
| 234 | + |
| 235 | + // Build psql connection string and exec. |
| 236 | + dbName := config.DBName() |
| 237 | + connStr := fmt.Sprintf("postgres://%s:%s@%s/%s?sslmode=disable", |
| 238 | + config.TerraformDBSettings.UserName, |
| 239 | + config.TerraformDBSettings.Password, |
| 240 | + localAddr, |
| 241 | + dbName, |
| 242 | + ) |
| 243 | + |
| 244 | + fmt.Printf("Connecting to %s on %s...\n", dbName, dbInst.Endpoint) |
| 245 | + |
| 246 | + psql := exec.CommandContext(ctx, "psql", connStr) |
| 247 | + psql.Stdin = os.Stdin |
| 248 | + psql.Stdout = os.Stdout |
| 249 | + psql.Stderr = os.Stderr |
| 250 | + |
| 251 | + if err := psql.Run(); err != nil { |
| 252 | + // If killed by our signal handler, don't treat as error. |
| 253 | + if ctx.Err() != nil { |
| 254 | + return nil |
| 255 | + } |
| 256 | + return fmt.Errorf("psql exited with error: %w", err) |
| 257 | + } |
| 258 | + |
| 259 | + return nil |
| 260 | +} |
0 commit comments