1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334
|
package backup
import (
"context"
"errors"
"fmt"
"strings"
"sync"
"github.com/sirupsen/logrus"
"gitlab.com/gitlab-org/gitaly/v16/internal/gitaly/storage"
"gitlab.com/gitlab-org/gitaly/v16/proto/go/gitalypb"
)
// Strategy used to create/restore backups
type Strategy interface {
Create(context.Context, *CreateRequest) error
Restore(context.Context, *RestoreRequest) error
}
// Command handles a specific backup operation
type Command interface {
Repository() *gitalypb.Repository
Name() string
Execute(context.Context) error
}
// Pipeline executes a series of commands and encapsulates error handling for
// the caller.
type Pipeline interface {
Handle(context.Context, Command)
Done() error
}
// CreateCommand creates a backup for a repository
type CreateCommand struct {
strategy Strategy
server storage.ServerInfo
repository *gitalypb.Repository
incremental bool
}
// NewCreateCommand builds a CreateCommand
func NewCreateCommand(strategy Strategy, server storage.ServerInfo, repo *gitalypb.Repository, incremental bool) *CreateCommand {
return &CreateCommand{
strategy: strategy,
server: server,
repository: repo,
incremental: incremental,
}
}
// Repository is the repository that will be acted on
func (cmd CreateCommand) Repository() *gitalypb.Repository {
return cmd.repository
}
// Name is the name of the command
func (cmd CreateCommand) Name() string {
return "create"
}
// Execute performs the backup
func (cmd CreateCommand) Execute(ctx context.Context) error {
return cmd.strategy.Create(ctx, &CreateRequest{
Server: cmd.server,
Repository: cmd.repository,
Incremental: cmd.incremental,
})
}
// RestoreCommand restores a backup for a repository
type RestoreCommand struct {
strategy Strategy
server storage.ServerInfo
repository *gitalypb.Repository
alwaysCreate bool
}
// NewRestoreCommand builds a RestoreCommand
func NewRestoreCommand(strategy Strategy, server storage.ServerInfo, repo *gitalypb.Repository, alwaysCreate bool) *RestoreCommand {
return &RestoreCommand{
strategy: strategy,
server: server,
repository: repo,
alwaysCreate: alwaysCreate,
}
}
// Repository is the repository that will be acted on
func (cmd RestoreCommand) Repository() *gitalypb.Repository {
return cmd.repository
}
// Name is the name of the command
func (cmd RestoreCommand) Name() string {
return "restore"
}
// Execute performs the restore
func (cmd RestoreCommand) Execute(ctx context.Context) error {
return cmd.strategy.Restore(ctx, &RestoreRequest{
Server: cmd.server,
Repository: cmd.repository,
AlwaysCreate: cmd.alwaysCreate,
})
}
// PipelineErrors represents a summary of errors by repository
type PipelineErrors []error
// AddError adds an error associated with a repository to the summary.
func (e *PipelineErrors) AddError(repo *gitalypb.Repository, err error) {
if repo.GetGlProjectPath() != "" {
err = fmt.Errorf("%s (%s): %w", repo.GetRelativePath(), repo.GetGlProjectPath(), err)
} else {
err = fmt.Errorf("%s: %w", repo.GetRelativePath(), err)
}
*e = append(*e, err)
}
func (e PipelineErrors) Error() string {
var builder strings.Builder
_, _ = fmt.Fprintf(&builder, "%d failures encountered:\n", len(e))
for _, err := range e {
_, _ = fmt.Fprintf(&builder, " - %s\n", err.Error())
}
return builder.String()
}
// LoggingPipeline outputs logging for each command executed
type LoggingPipeline struct {
log logrus.FieldLogger
mu sync.Mutex
errs PipelineErrors
}
// NewLoggingPipeline creates a new logging pipeline
func NewLoggingPipeline(log logrus.FieldLogger) *LoggingPipeline {
return &LoggingPipeline{
log: log,
}
}
// Handle takes a command to process. Commands are logged and executed immediately.
func (p *LoggingPipeline) Handle(ctx context.Context, cmd Command) {
log := p.cmdLogger(cmd)
log.Infof("started %s", cmd.Name())
if err := cmd.Execute(ctx); err != nil {
if errors.Is(err, ErrSkipped) {
log.WithError(err).Warnf("skipped %s", cmd.Name())
} else {
log.WithError(err).Errorf("%s failed", cmd.Name())
p.addError(cmd.Repository(), err)
}
return
}
log.Infof("completed %s", cmd.Name())
}
func (p *LoggingPipeline) addError(repo *gitalypb.Repository, err error) {
p.mu.Lock()
defer p.mu.Unlock()
p.errs.AddError(repo, err)
}
// Done indicates that the pipeline is complete and returns any accumulated errors
func (p *LoggingPipeline) Done() error {
if len(p.errs) > 0 {
return fmt.Errorf("pipeline: %w", p.errs)
}
return nil
}
func (p *LoggingPipeline) cmdLogger(cmd Command) logrus.FieldLogger {
return p.log.WithFields(logrus.Fields{
"command": cmd.Name(),
"storage_name": cmd.Repository().StorageName,
"relative_path": cmd.Repository().RelativePath,
"gl_project_path": cmd.Repository().GlProjectPath,
})
}
type contextCommand struct {
Command Command
Context context.Context
}
// ParallelPipeline is a pipeline that executes commands in parallel
type ParallelPipeline struct {
next Pipeline
parallel int
parallelStorage int
wg sync.WaitGroup
workerSlots chan struct{}
done chan struct{}
mu sync.Mutex
requests map[string]chan *contextCommand
err error
}
// NewParallelPipeline creates a new ParallelPipeline where all commands are
// passed onto `next` to be processed, `parallel` is the maximum number of
// parallel backups that will run and `parallelStorage` is the maximum number
// of parallel backups that will run per storage. Since the number of storages
// is unknown at initialisation, workers are created lazily as new storage
// names are encountered.
//
// Note: When both `parallel` and `parallelStorage` are zero or less no workers
// are created and the pipeline will block forever.
func NewParallelPipeline(next Pipeline, parallel, parallelStorage int) *ParallelPipeline {
var workerSlots chan struct{}
if parallel > 0 && parallelStorage > 0 {
// workerSlots allows the total number of parallel jobs to be
// limited. This allows us to create the required workers for
// each storage, while still limiting the absolute parallelism.
workerSlots = make(chan struct{}, parallel)
}
return &ParallelPipeline{
next: next,
parallel: parallel,
parallelStorage: parallelStorage,
workerSlots: workerSlots,
done: make(chan struct{}),
requests: make(map[string]chan *contextCommand),
}
}
// Handle queues a request to create a backup. Commands are processed by
// n-workers per storage.
func (p *ParallelPipeline) Handle(ctx context.Context, cmd Command) {
ch := p.getStorage(cmd.Repository().StorageName)
select {
case <-ctx.Done():
p.setErr(ctx.Err())
case ch <- &contextCommand{
Command: cmd,
Context: ctx,
}:
}
}
// Done waits for any in progress calls to `next` to complete then reports any
// accumulated errors
func (p *ParallelPipeline) Done() error {
close(p.done)
p.wg.Wait()
if err := p.next.Done(); err != nil {
return err
}
if p.err != nil {
return fmt.Errorf("pipeline: %w", p.err)
}
return nil
}
// getStorage finds the channel associated with a storage. When no channel is
// found, one is created and n-workers are started to process requests.
func (p *ParallelPipeline) getStorage(storage string) chan<- *contextCommand {
p.mu.Lock()
defer p.mu.Unlock()
workers := p.parallelStorage
if p.parallelStorage < 1 {
// if the workers are not limited by storage, then pretend there is a single storage with `parallel` workers
storage = ""
workers = p.parallel
}
ch, ok := p.requests[storage]
if !ok {
ch = make(chan *contextCommand)
p.requests[storage] = ch
for i := 0; i < workers; i++ {
p.wg.Add(1)
go p.worker(ch)
}
}
return ch
}
func (p *ParallelPipeline) worker(ch <-chan *contextCommand) {
defer p.wg.Done()
for {
select {
case <-p.done:
return
case cmd := <-ch:
p.processCommand(cmd.Context, cmd.Command)
}
}
}
func (p *ParallelPipeline) processCommand(ctx context.Context, cmd Command) {
p.acquireWorkerSlot()
defer p.releaseWorkerSlot()
p.next.Handle(ctx, cmd)
}
func (p *ParallelPipeline) setErr(err error) {
p.mu.Lock()
defer p.mu.Unlock()
if p.err != nil {
return
}
p.err = err
}
// acquireWorkerSlot queues the worker until a slot is available.
// It never blocks if `parallel` or `parallelStorage` are 0
func (p *ParallelPipeline) acquireWorkerSlot() {
if p.workerSlots == nil {
return
}
p.workerSlots <- struct{}{}
}
// releaseWorkerSlot releases the worker slot.
func (p *ParallelPipeline) releaseWorkerSlot() {
if p.workerSlots == nil {
return
}
<-p.workerSlots
}
|