move implementation to internal/ directory (#828)

This commit is contained in:
Christian Schwarz
2024-10-18 19:21:17 +02:00
committed by GitHub
parent b9b9ad10cf
commit 908807bd59
360 changed files with 507 additions and 507 deletions
+567
View File
@@ -0,0 +1,567 @@
package job
import (
"context"
"fmt"
"sync"
"time"
"github.com/pkg/errors"
"github.com/prometheus/client_golang/prometheus"
"github.com/zrepl/zrepl/internal/daemon/logging/trace"
"github.com/zrepl/zrepl/internal/util/envconst"
"github.com/zrepl/zrepl/internal/config"
"github.com/zrepl/zrepl/internal/daemon/job/reset"
"github.com/zrepl/zrepl/internal/daemon/job/wakeup"
"github.com/zrepl/zrepl/internal/daemon/pruner"
"github.com/zrepl/zrepl/internal/daemon/snapper"
"github.com/zrepl/zrepl/internal/endpoint"
"github.com/zrepl/zrepl/internal/replication"
"github.com/zrepl/zrepl/internal/replication/driver"
"github.com/zrepl/zrepl/internal/replication/logic"
"github.com/zrepl/zrepl/internal/replication/report"
"github.com/zrepl/zrepl/internal/rpc"
"github.com/zrepl/zrepl/internal/transport"
"github.com/zrepl/zrepl/internal/transport/fromconfig"
"github.com/zrepl/zrepl/internal/zfs"
)
type ActiveSide struct {
mode activeMode
name endpoint.JobID
connecter transport.Connecter
replicationDriverConfig driver.Config
prunerFactory *pruner.PrunerFactory
promRepStateSecs *prometheus.HistogramVec // labels: state
promPruneSecs *prometheus.HistogramVec // labels: prune_side
promBytesReplicated *prometheus.CounterVec // labels: filesystem
promReplicationErrors prometheus.Gauge
promLastSuccessful prometheus.Gauge
tasksMtx sync.Mutex
tasks activeSideTasks
}
//go:generate enumer -type=ActiveSideState
type ActiveSideState int
const (
ActiveSideReplicating ActiveSideState = 1 << iota
ActiveSidePruneSender
ActiveSidePruneReceiver
ActiveSideDone // also errors
)
type activeSideTasks struct {
state ActiveSideState
// valid for state ActiveSideReplicating, ActiveSidePruneSender, ActiveSidePruneReceiver, ActiveSideDone
replicationReport driver.ReportFunc
replicationCancel context.CancelFunc
// valid for state ActiveSidePruneSender, ActiveSidePruneReceiver, ActiveSideDone
prunerSender, prunerReceiver *pruner.Pruner
// valid for state ActiveSidePruneReceiver, ActiveSideDone
prunerSenderCancel, prunerReceiverCancel context.CancelFunc
}
func (a *ActiveSide) updateTasks(u func(*activeSideTasks)) activeSideTasks {
a.tasksMtx.Lock()
defer a.tasksMtx.Unlock()
copy := a.tasks
if u == nil {
return copy
}
u(&copy)
a.tasks = copy
return copy
}
type activeMode interface {
ConnectEndpoints(ctx context.Context, connecter transport.Connecter)
DisconnectEndpoints()
SenderReceiver() (logic.Sender, logic.Receiver)
Type() Type
PlannerPolicy() logic.PlannerPolicy
RunPeriodic(ctx context.Context, wakeUpCommon chan<- struct{})
SnapperReport() *snapper.Report
ResetConnectBackoff()
}
type modePush struct {
setupMtx sync.Mutex
sender *endpoint.Sender
receiver *rpc.Client
senderConfig *endpoint.SenderConfig
plannerPolicy *logic.PlannerPolicy
snapper snapper.Snapper
}
func (m *modePush) ConnectEndpoints(ctx context.Context, connecter transport.Connecter) {
m.setupMtx.Lock()
defer m.setupMtx.Unlock()
if m.receiver != nil || m.sender != nil {
panic("inconsistent use of ConnectEndpoints and DisconnectEndpoints")
}
m.sender = endpoint.NewSender(*m.senderConfig)
m.receiver = rpc.NewClient(connecter, rpc.GetLoggersOrPanic(ctx))
}
func (m *modePush) DisconnectEndpoints() {
m.setupMtx.Lock()
defer m.setupMtx.Unlock()
m.receiver.Close()
m.sender = nil
m.receiver = nil
}
func (m *modePush) SenderReceiver() (logic.Sender, logic.Receiver) {
m.setupMtx.Lock()
defer m.setupMtx.Unlock()
return m.sender, m.receiver
}
func (m *modePush) Type() Type { return TypePush }
func (m *modePush) PlannerPolicy() logic.PlannerPolicy { return *m.plannerPolicy }
func (m *modePush) RunPeriodic(ctx context.Context, wakeUpCommon chan<- struct{}) {
m.snapper.Run(ctx, wakeUpCommon)
}
func (m *modePush) SnapperReport() *snapper.Report {
r := m.snapper.Report()
return &r
}
func (m *modePush) ResetConnectBackoff() {
m.setupMtx.Lock()
defer m.setupMtx.Unlock()
if m.receiver != nil {
m.receiver.ResetConnectBackoff()
}
}
func modePushFromConfig(g *config.Global, in *config.PushJob, jobID endpoint.JobID) (*modePush, error) {
m := &modePush{}
var err error
m.senderConfig, err = buildSenderConfig(in, jobID)
if err != nil {
return nil, errors.Wrap(err, "sender config")
}
replicationConfig, err := logic.ReplicationConfigFromConfig(in.Replication)
if err != nil {
return nil, errors.Wrap(err, "field `replication`")
}
conflictResolution, err := logic.ConflictResolutionFromConfig(in.ConflictResolution)
if err != nil {
return nil, errors.Wrap(err, "field `conflict_resolution`")
}
m.plannerPolicy = &logic.PlannerPolicy{
ConflictResolution: conflictResolution,
ReplicationConfig: replicationConfig,
SizeEstimationConcurrency: in.Replication.Concurrency.SizeEstimates,
}
if err := m.plannerPolicy.Validate(); err != nil {
return nil, errors.Wrap(err, "cannot build planner policy")
}
if m.snapper, err = snapper.FromConfig(g, m.senderConfig.FSF, in.Snapshotting); err != nil {
return nil, errors.Wrap(err, "cannot build snapper")
}
return m, nil
}
type modePull struct {
setupMtx sync.Mutex
receiver *endpoint.Receiver
receiverConfig endpoint.ReceiverConfig
sender *rpc.Client
plannerPolicy *logic.PlannerPolicy
interval config.PositiveDurationOrManual
}
func (m *modePull) ConnectEndpoints(ctx context.Context, connecter transport.Connecter) {
m.setupMtx.Lock()
defer m.setupMtx.Unlock()
if m.receiver != nil || m.sender != nil {
panic("inconsistent use of ConnectEndpoints and DisconnectEndpoints")
}
m.receiver = endpoint.NewReceiver(m.receiverConfig)
m.sender = rpc.NewClient(connecter, rpc.GetLoggersOrPanic(ctx))
}
func (m *modePull) DisconnectEndpoints() {
m.setupMtx.Lock()
defer m.setupMtx.Unlock()
m.sender.Close()
m.sender = nil
m.receiver = nil
}
func (m *modePull) SenderReceiver() (logic.Sender, logic.Receiver) {
m.setupMtx.Lock()
defer m.setupMtx.Unlock()
return m.sender, m.receiver
}
func (*modePull) Type() Type { return TypePull }
func (m *modePull) PlannerPolicy() logic.PlannerPolicy { return *m.plannerPolicy }
func (m *modePull) RunPeriodic(ctx context.Context, wakeUpCommon chan<- struct{}) {
if m.interval.Manual {
GetLogger(ctx).Info("manual pull configured, periodic pull disabled")
// "waiting for wakeups" is printed in common ActiveSide.do
return
}
t := time.NewTicker(m.interval.Interval)
defer t.Stop()
for {
select {
case <-t.C:
select {
case wakeUpCommon <- struct{}{}:
default:
GetLogger(ctx).
WithField("pull_interval", m.interval).
Warn("pull job took longer than pull interval")
wakeUpCommon <- struct{}{} // block anyways, to queue up the wakeup
}
case <-ctx.Done():
return
}
}
}
func (m *modePull) SnapperReport() *snapper.Report {
return nil
}
func (m *modePull) ResetConnectBackoff() {
m.setupMtx.Lock()
defer m.setupMtx.Unlock()
if m.sender != nil {
m.sender.ResetConnectBackoff()
}
}
func modePullFromConfig(g *config.Global, in *config.PullJob, jobID endpoint.JobID) (m *modePull, err error) {
m = &modePull{}
m.interval = in.Interval
replicationConfig, err := logic.ReplicationConfigFromConfig(in.Replication)
if err != nil {
return nil, errors.Wrap(err, "field `replication`")
}
conflictResolution, err := logic.ConflictResolutionFromConfig(in.ConflictResolution)
if err != nil {
return nil, errors.Wrap(err, "field `conflict_resolution`")
}
m.plannerPolicy = &logic.PlannerPolicy{
ConflictResolution: conflictResolution,
ReplicationConfig: replicationConfig,
SizeEstimationConcurrency: in.Replication.Concurrency.SizeEstimates,
}
if err := m.plannerPolicy.Validate(); err != nil {
return nil, errors.Wrap(err, "cannot build planner policy")
}
m.receiverConfig, err = buildReceiverConfig(in, jobID)
if err != nil {
return nil, err
}
return m, nil
}
func replicationDriverConfigFromConfig(in *config.Replication) (c driver.Config, err error) {
c = driver.Config{
StepQueueConcurrency: in.Concurrency.Steps,
MaxAttempts: envconst.Int("ZREPL_REPLICATION_MAX_ATTEMPTS", 3),
ReconnectHardFailTimeout: envconst.Duration("ZREPL_REPLICATION_RECONNECT_HARD_FAIL_TIMEOUT", 10*time.Minute),
}
err = c.Validate()
return c, err
}
func activeSide(g *config.Global, in *config.ActiveJob, configJob interface{}, parseFlags config.ParseFlags) (j *ActiveSide, err error) {
j = &ActiveSide{}
j.name, err = endpoint.MakeJobID(in.Name)
if err != nil {
return nil, errors.Wrap(err, "invalid job name")
}
switch v := configJob.(type) {
case *config.PushJob:
j.mode, err = modePushFromConfig(g, v, j.name) // shadow
case *config.PullJob:
j.mode, err = modePullFromConfig(g, v, j.name) // shadow
default:
panic(fmt.Sprintf("implementation error: unknown job type %T", v))
}
if err != nil {
return nil, err // no wrapping required
}
j.promRepStateSecs = prometheus.NewHistogramVec(prometheus.HistogramOpts{
Namespace: "zrepl",
Subsystem: "replication",
Name: "state_time",
Help: "seconds spent during replication",
ConstLabels: prometheus.Labels{"zrepl_job": j.name.String()},
}, []string{"state"})
j.promBytesReplicated = prometheus.NewCounterVec(prometheus.CounterOpts{
Namespace: "zrepl",
Subsystem: "replication",
Name: "bytes_replicated",
Help: "number of bytes replicated from sender to receiver per filesystem",
ConstLabels: prometheus.Labels{"zrepl_job": j.name.String()},
}, []string{"filesystem"})
j.promReplicationErrors = prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: "zrepl",
Subsystem: "replication",
Name: "filesystem_errors",
Help: "number of filesystems that failed replication in the latest replication attempt, or -1 if the job failed before enumerating the filesystems",
ConstLabels: prometheus.Labels{"zrepl_job": j.name.String()},
})
j.promLastSuccessful = prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: "zrepl",
Subsystem: "replication",
Name: "last_successful",
Help: "timestamp of last successful replication",
ConstLabels: prometheus.Labels{"zrepl_job": j.name.String()},
})
j.connecter, err = fromconfig.ConnecterFromConfig(g, in.Connect, parseFlags)
if err != nil {
return nil, errors.Wrap(err, "cannot build client")
}
j.promPruneSecs = prometheus.NewHistogramVec(prometheus.HistogramOpts{
Namespace: "zrepl",
Subsystem: "pruning",
Name: "time",
Help: "seconds spent in pruner",
ConstLabels: prometheus.Labels{"zrepl_job": j.name.String()},
}, []string{"prune_side"})
j.prunerFactory, err = pruner.NewPrunerFactory(in.Pruning, j.promPruneSecs)
if err != nil {
return nil, err
}
j.replicationDriverConfig, err = replicationDriverConfigFromConfig(in.Replication)
if err != nil {
return nil, errors.Wrap(err, "cannot build replication driver config")
}
return j, nil
}
func (j *ActiveSide) RegisterMetrics(registerer prometheus.Registerer) {
registerer.MustRegister(j.promRepStateSecs)
registerer.MustRegister(j.promPruneSecs)
registerer.MustRegister(j.promBytesReplicated)
registerer.MustRegister(j.promReplicationErrors)
registerer.MustRegister(j.promLastSuccessful)
}
func (j *ActiveSide) Name() string { return j.name.String() }
type ActiveSideStatus struct {
Replication *report.Report
PruningSender, PruningReceiver *pruner.Report
Snapshotting *snapper.Report
}
func (j *ActiveSide) Status() *Status {
tasks := j.updateTasks(nil)
s := &ActiveSideStatus{}
t := j.mode.Type()
if tasks.replicationReport != nil {
s.Replication = tasks.replicationReport()
}
if tasks.prunerSender != nil {
s.PruningSender = tasks.prunerSender.Report()
}
if tasks.prunerReceiver != nil {
s.PruningReceiver = tasks.prunerReceiver.Report()
}
s.Snapshotting = j.mode.SnapperReport()
return &Status{Type: t, JobSpecific: s}
}
func (j *ActiveSide) OwnedDatasetSubtreeRoot() (rfs *zfs.DatasetPath, ok bool) {
pull, ok := j.mode.(*modePull)
if !ok {
_ = j.mode.(*modePush) // make sure we didn't introduce a new job type
return nil, false
}
return pull.receiverConfig.RootWithoutClientComponent.Copy(), true
}
func (j *ActiveSide) SenderConfig() *endpoint.SenderConfig {
push, ok := j.mode.(*modePush)
if !ok {
_ = j.mode.(*modePull) // make sure we didn't introduce a new job type
return nil
}
return push.senderConfig
}
// The active side of a replication uses one end (sender or receiver)
// directly by method invocation, without going through a transport that
// provides a client identity.
// However, in order to avoid the need to distinguish between direct-method-invocating
// clients and RPC client, we use an invalid client identity as a sentinel value.
func FakeActiveSideDirectMethodInvocationClientIdentity(jobId endpoint.JobID) string {
return fmt.Sprintf("<local><active><job><client><identity><job=%q>", jobId.String())
}
func (j *ActiveSide) Run(ctx context.Context) {
ctx, endTask := trace.WithTaskAndSpan(ctx, "active-side-job", j.Name())
defer endTask()
ctx = context.WithValue(ctx, endpoint.ClientIdentityKey, FakeActiveSideDirectMethodInvocationClientIdentity(j.name))
log := GetLogger(ctx)
defer log.Info("job exiting")
periodicDone := make(chan struct{})
ctx, cancel := context.WithCancel(ctx)
defer cancel()
periodicCtx, endTask := trace.WithTask(ctx, "periodic")
defer endTask()
go j.mode.RunPeriodic(periodicCtx, periodicDone)
invocationCount := 0
outer:
for {
log.Info("wait for wakeups")
select {
case <-ctx.Done():
log.WithError(ctx.Err()).Info("context")
break outer
case <-wakeup.Wait(ctx):
j.mode.ResetConnectBackoff()
case <-periodicDone:
}
invocationCount++
invocationCtx, endSpan := trace.WithSpan(ctx, fmt.Sprintf("invocation-%d", invocationCount))
j.do(invocationCtx)
endSpan()
}
}
func (j *ActiveSide) do(ctx context.Context) {
j.mode.ConnectEndpoints(ctx, j.connecter)
defer j.mode.DisconnectEndpoints()
// allow cancellation of an invocation (this function)
ctx, cancelThisRun := context.WithCancel(ctx)
defer cancelThisRun()
go func() {
select {
case <-reset.Wait(ctx):
GetLogger(ctx).Info("reset received, cancelling current invocation")
cancelThisRun()
case <-ctx.Done():
}
}()
sender, receiver := j.mode.SenderReceiver()
{
select {
case <-ctx.Done():
return
default:
}
ctx, endSpan := trace.WithSpan(ctx, "replication")
ctx, repCancel := context.WithCancel(ctx)
var repWait driver.WaitFunc
j.updateTasks(func(tasks *activeSideTasks) {
// reset it
*tasks = activeSideTasks{}
tasks.replicationCancel = func() { repCancel(); endSpan() }
tasks.replicationReport, repWait = replication.Do(
ctx, j.replicationDriverConfig, logic.NewPlanner(j.promRepStateSecs, j.promBytesReplicated, sender, receiver, j.mode.PlannerPolicy()),
)
tasks.state = ActiveSideReplicating
})
GetLogger(ctx).Info("start replication")
repWait(true) // wait blocking
repCancel() // always cancel to free up context resources
replicationReport := j.tasks.replicationReport()
var numErrors = replicationReport.GetFailedFilesystemsCountInLatestAttempt()
j.promReplicationErrors.Set(float64(numErrors))
if numErrors == 0 {
j.promLastSuccessful.SetToCurrentTime()
}
endSpan()
}
{
select {
case <-ctx.Done():
return
default:
}
ctx, endSpan := trace.WithSpan(ctx, "prune_sender")
ctx, senderCancel := context.WithCancel(ctx)
tasks := j.updateTasks(func(tasks *activeSideTasks) {
tasks.prunerSender = j.prunerFactory.BuildSenderPruner(ctx, sender, sender)
tasks.prunerSenderCancel = func() { senderCancel(); endSpan() }
tasks.state = ActiveSidePruneSender
})
GetLogger(ctx).Info("start pruning sender")
tasks.prunerSender.Prune()
GetLogger(ctx).Info("finished pruning sender")
senderCancel()
endSpan()
}
{
select {
case <-ctx.Done():
return
default:
}
ctx, endSpan := trace.WithSpan(ctx, "prune_recever")
ctx, receiverCancel := context.WithCancel(ctx)
tasks := j.updateTasks(func(tasks *activeSideTasks) {
tasks.prunerReceiver = j.prunerFactory.BuildReceiverPruner(ctx, receiver, sender)
tasks.prunerReceiverCancel = func() { receiverCancel(); endSpan() }
tasks.state = ActiveSidePruneReceiver
})
GetLogger(ctx).Info("start pruning receiver")
tasks.prunerReceiver.Prune()
GetLogger(ctx).Info("finished pruning receiver")
receiverCancel()
endSpan()
}
j.updateTasks(func(tasks *activeSideTasks) {
tasks.state = ActiveSideDone
})
}
+20
View File
@@ -0,0 +1,20 @@
package job
import (
"testing"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"github.com/zrepl/zrepl/internal/endpoint"
"github.com/zrepl/zrepl/internal/transport"
)
func TestFakeActiveSideDirectMethodInvocationClientIdentityDoesNotPassValidityTest(t *testing.T) {
jobid, err := endpoint.MakeJobID("validjobname")
require.NoError(t, err)
clientIdentity := FakeActiveSideDirectMethodInvocationClientIdentity(jobid)
t.Logf("%v", clientIdentity)
err = transport.ValidateClientIdentity(clientIdentity)
assert.Error(t, err)
}
@@ -0,0 +1,66 @@
// Code generated by "enumer -type=ActiveSideState"; DO NOT EDIT.
package job
import (
"fmt"
)
const (
_ActiveSideStateName_0 = "ActiveSideReplicatingActiveSidePruneSender"
_ActiveSideStateName_1 = "ActiveSidePruneReceiver"
_ActiveSideStateName_2 = "ActiveSideDone"
)
var (
_ActiveSideStateIndex_0 = [...]uint8{0, 21, 42}
_ActiveSideStateIndex_1 = [...]uint8{0, 23}
_ActiveSideStateIndex_2 = [...]uint8{0, 14}
)
func (i ActiveSideState) String() string {
switch {
case 1 <= i && i <= 2:
i -= 1
return _ActiveSideStateName_0[_ActiveSideStateIndex_0[i]:_ActiveSideStateIndex_0[i+1]]
case i == 4:
return _ActiveSideStateName_1
case i == 8:
return _ActiveSideStateName_2
default:
return fmt.Sprintf("ActiveSideState(%d)", i)
}
}
var _ActiveSideStateValues = []ActiveSideState{1, 2, 4, 8}
var _ActiveSideStateNameToValueMap = map[string]ActiveSideState{
_ActiveSideStateName_0[0:21]: 1,
_ActiveSideStateName_0[21:42]: 2,
_ActiveSideStateName_1[0:23]: 4,
_ActiveSideStateName_2[0:14]: 8,
}
// ActiveSideStateString retrieves an enum value from the enum constants string name.
// Throws an error if the param is not part of the enum.
func ActiveSideStateString(s string) (ActiveSideState, error) {
if val, ok := _ActiveSideStateNameToValueMap[s]; ok {
return val, nil
}
return 0, fmt.Errorf("%s does not belong to ActiveSideState values", s)
}
// ActiveSideStateValues returns all values of the enum
func ActiveSideStateValues() []ActiveSideState {
return _ActiveSideStateValues
}
// IsAActiveSideState returns "true" if the value is listed in the enum definition. "false" otherwise
func (i ActiveSideState) IsAActiveSideState() bool {
for _, v := range _ActiveSideStateValues {
if i == v {
return true
}
}
return false
}
+105
View File
@@ -0,0 +1,105 @@
package job
import (
"fmt"
"sort"
"strings"
"github.com/pkg/errors"
"github.com/zrepl/zrepl/internal/config"
"github.com/zrepl/zrepl/internal/util/bandwidthlimit"
)
func JobsFromConfig(c *config.Config, parseFlags config.ParseFlags) ([]Job, error) {
js := make([]Job, len(c.Jobs))
for i := range c.Jobs {
j, err := buildJob(c.Global, c.Jobs[i], parseFlags)
if err != nil {
return nil, err
}
if j == nil || j.Name() == "" {
panic(fmt.Sprintf("implementation error: job builder returned nil job type %T", c.Jobs[i].Ret))
}
js[i] = j
}
return js, nil
}
func buildJob(c *config.Global, in config.JobEnum, parseFlags config.ParseFlags) (j Job, err error) {
cannotBuildJob := func(e error, name string) (Job, error) {
return nil, errors.Wrapf(e, "cannot build job %q", name)
}
// FIXME prettify this
switch v := in.Ret.(type) {
case *config.SinkJob:
j, err = passiveSideFromConfig(c, &v.PassiveJob, v, parseFlags)
if err != nil {
return cannotBuildJob(err, v.Name)
}
case *config.SourceJob:
j, err = passiveSideFromConfig(c, &v.PassiveJob, v, parseFlags)
if err != nil {
return cannotBuildJob(err, v.Name)
}
case *config.SnapJob:
j, err = snapJobFromConfig(c, v)
if err != nil {
return cannotBuildJob(err, v.Name)
}
case *config.PushJob:
j, err = activeSide(c, &v.ActiveJob, v, parseFlags)
if err != nil {
return cannotBuildJob(err, v.Name)
}
case *config.PullJob:
j, err = activeSide(c, &v.ActiveJob, v, parseFlags)
if err != nil {
return cannotBuildJob(err, v.Name)
}
default:
panic(fmt.Sprintf("implementation error: unknown job type %T", v))
}
return j, nil
}
func validateReceivingSidesDoNotOverlap(receivingRootFSs []string) error {
if len(receivingRootFSs) == 0 {
return nil
}
rfss := make([]string, len(receivingRootFSs))
copy(rfss, receivingRootFSs)
sort.Slice(rfss, func(i, j int) bool {
return strings.Compare(rfss[i], rfss[j]) == -1
})
// add tailing slash because of hierarchy-simulation
// rootfs/ is not root of rootfs2/
for i := 0; i < len(rfss); i++ {
rfss[i] += "/"
}
// idea:
// no path in rfss must be prefix of another
//
// rfss is now lexicographically sorted, which means that
// if i is prefix of j, i < j (in lexicographical order)
// thus,
// if any i is prefix of i+n (n >= 1), there is overlap
for i := 0; i < len(rfss)-1; i++ {
if strings.HasPrefix(rfss[i+1], rfss[i]) {
return fmt.Errorf("receiving jobs with overlapping root filesystems are forbidden")
}
}
return nil
}
func buildBandwidthLimitConfig(in *config.BandwidthLimit) (c bandwidthlimit.Config, _ error) {
if in.Max.ToBytes() > 0 && int64(in.Max.ToBytes()) == 0 {
return c, fmt.Errorf("bandwidth limit `max` is too small, must at least specify one byte")
}
return bandwidthlimit.Config{
Max: int64(in.Max.ToBytes()),
BucketCapacity: int64(in.BucketCapacity.ToBytes()),
}, nil
}
@@ -0,0 +1,102 @@
package job
import (
"github.com/pkg/errors"
"github.com/zrepl/zrepl/internal/config"
"github.com/zrepl/zrepl/internal/daemon/filters"
"github.com/zrepl/zrepl/internal/endpoint"
"github.com/zrepl/zrepl/internal/util/nodefault"
"github.com/zrepl/zrepl/internal/zfs"
)
type SendingJobConfig interface {
GetFilesystems() config.FilesystemsFilter
GetSendOptions() *config.SendOptions // must not be nil
}
func buildSenderConfig(in SendingJobConfig, jobID endpoint.JobID) (*endpoint.SenderConfig, error) {
fsf, err := filters.DatasetMapFilterFromConfig(in.GetFilesystems())
if err != nil {
return nil, errors.Wrap(err, "cannot build filesystem filter")
}
sendOpts := in.GetSendOptions()
bwlim, err := buildBandwidthLimitConfig(sendOpts.BandwidthLimit)
if err != nil {
return nil, errors.Wrap(err, "cannot build bandwith limit config")
}
sc := &endpoint.SenderConfig{
FSF: fsf,
JobID: jobID,
Encrypt: &nodefault.Bool{B: sendOpts.Encrypted},
SendRaw: sendOpts.Raw,
SendProperties: sendOpts.SendProperties,
SendBackupProperties: sendOpts.BackupProperties,
SendLargeBlocks: sendOpts.LargeBlocks,
SendCompressed: sendOpts.Compressed,
SendEmbeddedData: sendOpts.EmbeddedData,
SendSaved: sendOpts.Saved,
BandwidthLimit: bwlim,
}
if err := sc.Validate(); err != nil {
return nil, errors.Wrap(err, "cannot build sender config")
}
return sc, nil
}
type ReceivingJobConfig interface {
GetRootFS() string
GetAppendClientIdentity() bool
GetRecvOptions() *config.RecvOptions
}
func buildReceiverConfig(in ReceivingJobConfig, jobID endpoint.JobID) (rc endpoint.ReceiverConfig, err error) {
rootFs, err := zfs.NewDatasetPath(in.GetRootFS())
if err != nil {
return rc, errors.New("root_fs is not a valid zfs filesystem path")
}
if rootFs.Length() <= 0 {
return rc, errors.New("root_fs must not be empty") // duplicates error check of receiver
}
recvOpts := in.GetRecvOptions()
bwlim, err := buildBandwidthLimitConfig(recvOpts.BandwidthLimit)
if err != nil {
return rc, errors.Wrap(err, "cannot build bandwith limit config")
}
placeholderEncryption, err := endpoint.PlaceholderCreationEncryptionPropertyString(recvOpts.Placeholder.Encryption)
if err != nil {
options := []string{}
for _, v := range endpoint.PlaceholderCreationEncryptionPropertyValues() {
options = append(options, endpoint.PlaceholderCreationEncryptionProperty(v).String())
}
return rc, errors.Errorf("placeholder encryption value %q is invalid, must be one of %s",
recvOpts.Placeholder.Encryption, options)
}
rc = endpoint.ReceiverConfig{
JobID: jobID,
RootWithoutClientComponent: rootFs,
AppendClientIdentity: in.GetAppendClientIdentity(),
InheritProperties: recvOpts.Properties.Inherit,
OverrideProperties: recvOpts.Properties.Override,
BandwidthLimit: bwlim,
PlaceholderEncryption: placeholderEncryption,
}
if err := rc.Validate(); err != nil {
return rc, errors.Wrap(err, "cannot build receiver config")
}
return rc, nil
}
+318
View File
@@ -0,0 +1,318 @@
package job
import (
"fmt"
"path"
"path/filepath"
"testing"
"github.com/kr/pretty"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"github.com/zrepl/zrepl/internal/config"
)
func TestValidateReceivingSidesDoNotOverlap(t *testing.T) {
type testCase struct {
err bool
input []string
}
tcs := []testCase{
{false, nil},
{false, []string{}},
{false, []string{""}}, // not our job to determine valid paths
{false, []string{"a"}},
{false, []string{"some/path"}},
{false, []string{"zroot/sink1", "zroot/sink2", "zroot/sink3"}},
{false, []string{"zroot/foo", "zroot/foobar"}},
{true, []string{"zroot/b", "zroot/b"}},
{true, []string{"zroot/foo", "zroot/foo/bar", "zroot/baz"}},
{false, []string{"a/x", "b/x"}},
{false, []string{"a", "b"}},
{true, []string{"a", "a"}},
{true, []string{"a/x/y", "a/x"}},
{true, []string{"a/x", "a/x/y"}},
{true, []string{"a/x", "b/x", "a/x/y"}},
{true, []string{"a", "a/b", "a/c", "a/b"}},
{true, []string{"a/b", "a/c", "a/b", "a/d", "a/c"}},
}
for _, tc := range tcs {
t.Logf("input: %v", tc.input)
err := validateReceivingSidesDoNotOverlap(tc.input)
if tc.err {
assert.Error(t, err)
} else {
assert.NoError(t, err)
}
}
}
func TestJobIDErrorHandling(t *testing.T) {
tmpl := `
jobs:
- name: %s
type: push
connect:
type: local
listener_name: foo
client_identity: bar
filesystems: {"<": true}
snapshotting:
type: manual
pruning:
keep_sender:
- type: last_n
count: 10
keep_receiver:
- type: last_n
count: 10
`
fill := func(s string) string { return fmt.Sprintf(tmpl, s) }
type Case struct {
jobName string
valid bool
}
cases := []Case{
{"validjobname", true},
{"valid with spaces", true},
{"invalid\twith\ttabs", false},
{"invalid#withdelimiter", false},
{"invalid@withdelimiter", false},
{"withnewline\\nmiddle", false},
{"withnewline\\n", false},
{"withslash/", false},
{"withslash/inthemiddle", false},
{"/", false},
}
for i := range cases {
t.Run(cases[i].jobName, func(t *testing.T) {
c := cases[i]
conf, err := config.ParseConfigBytes([]byte(fill(c.jobName)))
require.NoError(t, err, "not expecting yaml-config to know about job ids")
require.NotNil(t, conf)
jobs, err := JobsFromConfig(conf, config.ParseFlagsNone)
if c.valid {
assert.NoError(t, err)
require.Len(t, jobs, 1)
assert.Equal(t, c.jobName, jobs[0].Name())
} else {
t.Logf("error: %s", err)
assert.Error(t, err)
assert.Nil(t, jobs)
}
})
}
}
func TestSampleConfigsAreBuiltWithoutErrors(t *testing.T) {
paths, err := filepath.Glob("../../../internal/config/samples/*")
if err != nil {
t.Errorf("glob failed: %+v", err)
}
type additionalCheck struct {
state int
test func(t *testing.T, jobs []Job)
}
additionalChecks := map[string]*additionalCheck{
"bandwidth_limit.yml": {test: testSampleConfig_BandwidthLimit},
}
for _, p := range paths {
if path.Ext(p) != ".yml" {
t.Logf("skipping file %s", p)
continue
}
filename := path.Base(p)
t.Logf("checking for presence additonal checks for file %q", filename)
additionalCheck := additionalChecks[filename]
if additionalCheck == nil {
t.Logf("no additional checks")
} else {
t.Logf("additional check present")
additionalCheck.state = 1
}
t.Run(p, func(t *testing.T) {
c, err := config.ParseConfig(p)
if err != nil {
t.Fatalf("error parsing %s:\n%+v", p, err)
}
t.Logf("file: %s", p)
t.Log(pretty.Sprint(c))
jobs, err := JobsFromConfig(c, config.ParseFlagsNoCertCheck)
t.Logf("jobs: %#v", jobs)
require.NoError(t, err)
if additionalCheck != nil {
additionalCheck.test(t, jobs)
additionalCheck.state = 2
}
})
}
for basename, c := range additionalChecks {
if c.state == 0 {
panic("univisited additional check " + basename)
}
}
}
func testSampleConfig_BandwidthLimit(t *testing.T, jobs []Job) {
require.Len(t, jobs, 3)
{
limitedSink, ok := jobs[0].(*PassiveSide)
require.True(t, ok, "%T", jobs[0])
limitedSinkMode, ok := limitedSink.mode.(*modeSink)
require.True(t, ok, "%T", limitedSink)
assert.Equal(t, int64(12345), limitedSinkMode.receiverConfig.BandwidthLimit.Max)
assert.Equal(t, int64(1<<17), limitedSinkMode.receiverConfig.BandwidthLimit.BucketCapacity)
}
{
limitedPush, ok := jobs[1].(*ActiveSide)
require.True(t, ok, "%T", jobs[1])
limitedPushMode, ok := limitedPush.mode.(*modePush)
require.True(t, ok, "%T", limitedPush)
assert.Equal(t, int64(54321), limitedPushMode.senderConfig.BandwidthLimit.Max)
assert.Equal(t, int64(1024), limitedPushMode.senderConfig.BandwidthLimit.BucketCapacity)
}
{
unlimitedSink, ok := jobs[2].(*PassiveSide)
require.True(t, ok, "%T", jobs[2])
unlimitedSinkMode, ok := unlimitedSink.mode.(*modeSink)
require.True(t, ok, "%T", unlimitedSink)
max := unlimitedSinkMode.receiverConfig.BandwidthLimit.Max
assert.Less(t, max, int64(0), max, "unlimited mode <=> negative value for .Max, see bandwidthlimit.Config")
}
}
func TestReplicationOptions(t *testing.T) {
tmpl := `
jobs:
- name: foo
type: push
connect:
type: local
listener_name: foo
client_identity: bar
filesystems: {"<": true}
%s
snapshotting:
type: manual
pruning:
keep_sender:
- type: last_n
count: 10
keep_receiver:
- type: last_n
count: 10
`
type Test struct {
name string
input string
expectOk func(t *testing.T, a *ActiveSide, m *modePush)
expectError bool
}
tests := []Test{
{
name: "defaults",
input: `
replication: {}
`,
expectOk: func(t *testing.T, a *ActiveSide, m *modePush) {},
},
{
name: "steps_zero",
input: `
replication:
concurrency:
steps: 0
`,
expectError: true,
},
{
name: "size_estimates_zero",
input: `
replication:
concurrency:
size_estimates: 0
`,
expectError: true,
},
{
name: "custom_values",
input: `
replication:
concurrency:
steps: 23
size_estimates: 42
`,
expectOk: func(t *testing.T, a *ActiveSide, m *modePush) {
assert.Equal(t, 23, a.replicationDriverConfig.StepQueueConcurrency)
assert.Equal(t, 42, m.plannerPolicy.SizeEstimationConcurrency)
},
},
{
name: "negative_values_forbidden",
input: `
replication:
concurrency:
steps: -23
size_estimates: -42
`,
expectError: true,
},
}
fill := func(s string) string { return fmt.Sprintf(tmpl, s) }
for _, ts := range tests {
t.Run(ts.name, func(t *testing.T) {
assert.True(t, (ts.expectError) != (ts.expectOk != nil))
cstr := fill(ts.input)
t.Logf("testing config:\n%s", cstr)
c, err := config.ParseConfigBytes([]byte(cstr))
require.NoError(t, err)
jobs, err := JobsFromConfig(c, config.ParseFlagsNone)
if ts.expectOk != nil {
require.NoError(t, err)
require.NotNil(t, c)
require.NoError(t, err)
require.Len(t, jobs, 1)
a := jobs[0].(*ActiveSide)
m := a.mode.(*modePush)
ts.expectOk(t, a, m)
} else if ts.expectError {
require.Error(t, err)
} else {
t.Fatalf("test must define expectOk or expectError")
}
})
}
}
+108
View File
@@ -0,0 +1,108 @@
package job
import (
"context"
"encoding/json"
"fmt"
"github.com/prometheus/client_golang/prometheus"
"github.com/zrepl/zrepl/internal/daemon/logging"
"github.com/zrepl/zrepl/internal/endpoint"
"github.com/zrepl/zrepl/internal/logger"
"github.com/zrepl/zrepl/internal/zfs"
)
type Logger = logger.Logger
func GetLogger(ctx context.Context) Logger {
return logging.GetLogger(ctx, logging.SubsysJob)
}
type Job interface {
Name() string
Run(ctx context.Context)
Status() *Status
RegisterMetrics(registerer prometheus.Registerer)
// Jobs that return a subtree of the dataset hierarchy
// must return the root of that subtree as rfs and ok = true
OwnedDatasetSubtreeRoot() (rfs *zfs.DatasetPath, ok bool)
SenderConfig() *endpoint.SenderConfig
}
type Type string
const (
TypeInternal Type = "internal"
TypeSnap Type = "snap"
TypePush Type = "push"
TypeSink Type = "sink"
TypePull Type = "pull"
TypeSource Type = "source"
)
type Status struct {
Type Type
JobSpecific interface{}
}
func (s *Status) MarshalJSON() ([]byte, error) {
typeJson, err := json.Marshal(s.Type)
if err != nil {
return nil, err
}
jobJSON, err := json.Marshal(s.JobSpecific)
if err != nil {
return nil, err
}
m := map[string]json.RawMessage{
"type": typeJson,
string(s.Type): jobJSON,
}
return json.Marshal(m)
}
func (s *Status) UnmarshalJSON(in []byte) (err error) {
var m map[string]json.RawMessage
if err := json.Unmarshal(in, &m); err != nil {
return err
}
tJSON, ok := m["type"]
if !ok {
return fmt.Errorf("field 'type' not found")
}
if err := json.Unmarshal(tJSON, &s.Type); err != nil {
return err
}
key := string(s.Type)
jobJSON, ok := m[key]
if !ok {
return fmt.Errorf("field '%s', not found", key)
}
switch s.Type {
case TypeSnap:
var st SnapJobStatus
err = json.Unmarshal(jobJSON, &st)
s.JobSpecific = &st
case TypePull:
fallthrough
case TypePush:
var st ActiveSideStatus
err = json.Unmarshal(jobJSON, &st)
s.JobSpecific = &st
case TypeSource:
fallthrough
case TypeSink:
var st PassiveStatus
err = json.Unmarshal(jobJSON, &st)
s.JobSpecific = &st
case TypeInternal:
// internal jobs do not report specifics
default:
err = fmt.Errorf("unknown job type '%s'", key)
}
return err
}
+192
View File
@@ -0,0 +1,192 @@
package job
import (
"context"
"fmt"
"github.com/pkg/errors"
"github.com/prometheus/client_golang/prometheus"
"github.com/zrepl/zrepl/internal/daemon/logging/trace"
"github.com/zrepl/zrepl/internal/config"
"github.com/zrepl/zrepl/internal/daemon/logging"
"github.com/zrepl/zrepl/internal/daemon/snapper"
"github.com/zrepl/zrepl/internal/endpoint"
"github.com/zrepl/zrepl/internal/rpc"
"github.com/zrepl/zrepl/internal/transport"
"github.com/zrepl/zrepl/internal/transport/fromconfig"
"github.com/zrepl/zrepl/internal/zfs"
)
type PassiveSide struct {
mode passiveMode
name endpoint.JobID
listen transport.AuthenticatedListenerFactory
}
type passiveMode interface {
Handler() rpc.Handler
RunPeriodic(ctx context.Context)
SnapperReport() *snapper.Report // may be nil
Type() Type
}
type modeSink struct {
receiverConfig endpoint.ReceiverConfig
}
func (m *modeSink) Type() Type { return TypeSink }
func (m *modeSink) Handler() rpc.Handler {
return endpoint.NewReceiver(m.receiverConfig)
}
func (m *modeSink) RunPeriodic(_ context.Context) {}
func (m *modeSink) SnapperReport() *snapper.Report { return nil }
func modeSinkFromConfig(g *config.Global, in *config.SinkJob, jobID endpoint.JobID) (m *modeSink, err error) {
m = &modeSink{}
m.receiverConfig, err = buildReceiverConfig(in, jobID)
if err != nil {
return nil, err
}
return m, nil
}
type modeSource struct {
senderConfig *endpoint.SenderConfig
snapper snapper.Snapper
}
func modeSourceFromConfig(g *config.Global, in *config.SourceJob, jobID endpoint.JobID) (m *modeSource, err error) {
// FIXME exact dedup of modePush
m = &modeSource{}
m.senderConfig, err = buildSenderConfig(in, jobID)
if err != nil {
return nil, errors.Wrap(err, "send options")
}
if m.snapper, err = snapper.FromConfig(g, m.senderConfig.FSF, in.Snapshotting); err != nil {
return nil, errors.Wrap(err, "cannot build snapper")
}
return m, nil
}
func (m *modeSource) Type() Type { return TypeSource }
func (m *modeSource) Handler() rpc.Handler {
return endpoint.NewSender(*m.senderConfig)
}
func (m *modeSource) RunPeriodic(ctx context.Context) {
m.snapper.Run(ctx, nil)
}
func (m *modeSource) SnapperReport() *snapper.Report {
r := m.snapper.Report()
return &r
}
func passiveSideFromConfig(g *config.Global, in *config.PassiveJob, configJob interface{}, parseFlags config.ParseFlags) (s *PassiveSide, err error) {
s = &PassiveSide{}
s.name, err = endpoint.MakeJobID(in.Name)
if err != nil {
return nil, errors.Wrap(err, "invalid job name")
}
switch v := configJob.(type) {
case *config.SinkJob:
s.mode, err = modeSinkFromConfig(g, v, s.name) // shadow
case *config.SourceJob:
s.mode, err = modeSourceFromConfig(g, v, s.name) // shadow
}
if err != nil {
return nil, err // no wrapping necessary
}
if s.listen, err = fromconfig.ListenerFactoryFromConfig(g, in.Serve, parseFlags); err != nil {
return nil, errors.Wrap(err, "cannot build listener factory")
}
return s, nil
}
func (j *PassiveSide) Name() string { return j.name.String() }
type PassiveStatus struct {
Snapper *snapper.Report
}
func (s *PassiveSide) Status() *Status {
st := &PassiveStatus{
Snapper: s.mode.SnapperReport(),
}
return &Status{Type: s.mode.Type(), JobSpecific: st}
}
func (j *PassiveSide) OwnedDatasetSubtreeRoot() (rfs *zfs.DatasetPath, ok bool) {
sink, ok := j.mode.(*modeSink)
if !ok {
_ = j.mode.(*modeSource) // make sure we didn't introduce a new job type
return nil, false
}
return sink.receiverConfig.RootWithoutClientComponent.Copy(), true
}
func (j *PassiveSide) SenderConfig() *endpoint.SenderConfig {
source, ok := j.mode.(*modeSource)
if !ok {
_ = j.mode.(*modeSink) // make sure we didn't introduce a new job type
return nil
}
return source.senderConfig
}
func (*PassiveSide) RegisterMetrics(registerer prometheus.Registerer) {}
func (j *PassiveSide) Run(ctx context.Context) {
ctx, endTask := trace.WithTaskAndSpan(ctx, "passive-side-job", j.Name())
defer endTask()
log := GetLogger(ctx)
defer log.Info("job exiting")
{
ctx, endTask := trace.WithTask(ctx, "periodic") // shadowing
defer endTask()
ctx, cancel := context.WithCancel(ctx)
defer cancel()
go j.mode.RunPeriodic(ctx)
}
handler := j.mode.Handler()
if handler == nil {
panic(fmt.Sprintf("implementation error: j.mode.Handler() returned nil: %#v", j))
}
ctxInterceptor := func(handlerCtx context.Context, info rpc.HandlerContextInterceptorData, handler func(ctx context.Context)) {
// the handlerCtx is clean => need to inherit logging and tracing config from job context
handlerCtx = logging.WithInherit(handlerCtx, ctx)
handlerCtx = trace.WithInherit(handlerCtx, ctx)
handlerCtx, endTask := trace.WithTaskAndSpan(handlerCtx, "handler", fmt.Sprintf("job=%q client=%q method=%q", j.Name(), info.ClientIdentity(), info.FullMethod()))
defer endTask()
handler(handlerCtx)
}
rpcLoggers := rpc.GetLoggersOrPanic(ctx) // WithSubsystemLoggers above
server := rpc.NewServer(handler, rpcLoggers, ctxInterceptor)
listener, err := j.listen()
if err != nil {
log.WithError(err).Error("cannot listen")
return
}
server.Serve(ctx, listener)
}
+35
View File
@@ -0,0 +1,35 @@
package reset
import (
"context"
"errors"
)
type contextKey int
const contextKeyReset contextKey = iota
func Wait(ctx context.Context) <-chan struct{} {
wc, ok := ctx.Value(contextKeyReset).(chan struct{})
if !ok {
wc = make(chan struct{})
}
return wc
}
type Func func() error
var AlreadyReset = errors.New("already reset")
func Context(ctx context.Context) (context.Context, Func) {
wc := make(chan struct{})
wuf := func() error {
select {
case wc <- struct{}{}:
return nil
default:
return AlreadyReset
}
}
return context.WithValue(ctx, contextKeyReset, wc), wuf
}
+196
View File
@@ -0,0 +1,196 @@
package job
import (
"context"
"fmt"
"sort"
"sync"
"github.com/pkg/errors"
"github.com/prometheus/client_golang/prometheus"
"github.com/zrepl/zrepl/internal/daemon/logging/trace"
"github.com/zrepl/zrepl/internal/util/bandwidthlimit"
"github.com/zrepl/zrepl/internal/util/nodefault"
"github.com/zrepl/zrepl/internal/config"
"github.com/zrepl/zrepl/internal/daemon/filters"
"github.com/zrepl/zrepl/internal/daemon/job/wakeup"
"github.com/zrepl/zrepl/internal/daemon/pruner"
"github.com/zrepl/zrepl/internal/daemon/snapper"
"github.com/zrepl/zrepl/internal/endpoint"
"github.com/zrepl/zrepl/internal/replication/logic/pdu"
"github.com/zrepl/zrepl/internal/zfs"
)
type SnapJob struct {
name endpoint.JobID
fsfilter zfs.DatasetFilter
snapper snapper.Snapper
prunerFactory *pruner.LocalPrunerFactory
promPruneSecs *prometheus.HistogramVec // labels: prune_side
prunerMtx sync.Mutex
pruner *pruner.Pruner
}
func (j *SnapJob) Name() string { return j.name.String() }
func (j *SnapJob) Type() Type { return TypeSnap }
func snapJobFromConfig(g *config.Global, in *config.SnapJob) (j *SnapJob, err error) {
j = &SnapJob{}
fsf, err := filters.DatasetMapFilterFromConfig(in.Filesystems)
if err != nil {
return nil, errors.Wrap(err, "cannot build filesystem filter")
}
j.fsfilter = fsf
if j.snapper, err = snapper.FromConfig(g, fsf, in.Snapshotting); err != nil {
return nil, errors.Wrap(err, "cannot build snapper")
}
j.name, err = endpoint.MakeJobID(in.Name)
if err != nil {
return nil, errors.Wrap(err, "invalid job name")
}
j.promPruneSecs = prometheus.NewHistogramVec(prometheus.HistogramOpts{
Namespace: "zrepl",
Subsystem: "pruning",
Name: "time",
Help: "seconds spent in pruner",
ConstLabels: prometheus.Labels{"zrepl_job": j.name.String()},
}, []string{"prune_side"})
j.prunerFactory, err = pruner.NewLocalPrunerFactory(in.Pruning, j.promPruneSecs)
if err != nil {
return nil, errors.Wrap(err, "cannot build snapjob pruning rules")
}
return j, nil
}
func (j *SnapJob) RegisterMetrics(registerer prometheus.Registerer) {
registerer.MustRegister(j.promPruneSecs)
}
type SnapJobStatus struct {
Pruning *pruner.Report
Snapshotting *snapper.Report // may be nil
}
func (j *SnapJob) Status() *Status {
s := &SnapJobStatus{}
t := j.Type()
j.prunerMtx.Lock()
if j.pruner != nil {
s.Pruning = j.pruner.Report()
}
j.prunerMtx.Unlock()
r := j.snapper.Report()
s.Snapshotting = &r
return &Status{Type: t, JobSpecific: s}
}
func (j *SnapJob) OwnedDatasetSubtreeRoot() (rfs *zfs.DatasetPath, ok bool) {
return nil, false
}
func (j *SnapJob) SenderConfig() *endpoint.SenderConfig { return nil }
func (j *SnapJob) Run(ctx context.Context) {
ctx, endTask := trace.WithTaskAndSpan(ctx, "snap-job", j.Name())
defer endTask()
log := GetLogger(ctx)
defer log.Info("job exiting")
periodicDone := make(chan struct{})
ctx, cancel := context.WithCancel(ctx)
defer cancel()
periodicCtx, endTask := trace.WithTask(ctx, "snapshotting")
defer endTask()
go j.snapper.Run(periodicCtx, periodicDone)
invocationCount := 0
outer:
for {
log.Info("wait for wakeups")
select {
case <-ctx.Done():
log.WithError(ctx.Err()).Info("context")
break outer
case <-wakeup.Wait(ctx):
case <-periodicDone:
}
invocationCount++
invocationCtx, endSpan := trace.WithSpan(ctx, fmt.Sprintf("invocation-%d", invocationCount))
j.doPrune(invocationCtx)
endSpan()
}
}
// Adaptor that implements pruner.History around a pruner.Target.
// The ReplicationCursor method is Get-op only and always returns
// the filesystem's most recent version's GUID.
//
// TODO:
// This is a work-around for the current package daemon/pruner
// and package pruning.Snapshot limitation: they require the
// `Replicated` getter method be present, but obviously,
// a local job like SnapJob can't deliver on that.
// But the pruner.Pruner gives up on an FS if no replication
// cursor is present, which is why this pruner returns the
// most recent filesystem version.
type alwaysUpToDateReplicationCursorHistory struct {
// the Target passed as Target to BuildLocalPruner
target pruner.Target
}
var _ pruner.Sender = (*alwaysUpToDateReplicationCursorHistory)(nil)
func (h alwaysUpToDateReplicationCursorHistory) ReplicationCursor(ctx context.Context, req *pdu.ReplicationCursorReq) (*pdu.ReplicationCursorRes, error) {
fsvReq := &pdu.ListFilesystemVersionsReq{
Filesystem: req.GetFilesystem(),
}
res, err := h.target.ListFilesystemVersions(ctx, fsvReq)
if err != nil {
return nil, err
}
fsvs := res.GetVersions()
if len(fsvs) <= 0 {
return &pdu.ReplicationCursorRes{Result: &pdu.ReplicationCursorRes_Notexist{Notexist: true}}, nil
}
// always return must recent version
sort.Slice(fsvs, func(i, j int) bool {
return fsvs[i].CreateTXG < fsvs[j].CreateTXG
})
mostRecent := fsvs[len(fsvs)-1]
return &pdu.ReplicationCursorRes{Result: &pdu.ReplicationCursorRes_Guid{Guid: mostRecent.GetGuid()}}, nil
}
func (h alwaysUpToDateReplicationCursorHistory) ListFilesystems(ctx context.Context, req *pdu.ListFilesystemReq) (*pdu.ListFilesystemRes, error) {
return h.target.ListFilesystems(ctx, req)
}
func (j *SnapJob) doPrune(ctx context.Context) {
ctx, endSpan := trace.WithSpan(ctx, "snap-job-do-prune")
defer endSpan()
log := GetLogger(ctx)
sender := endpoint.NewSender(endpoint.SenderConfig{
JobID: j.name,
FSF: j.fsfilter,
// FIXME the following config fields are irrelevant for SnapJob
// because the endpoint is only used as pruner.Target.
// However, the implementation requires them to be set.
Encrypt: &nodefault.Bool{B: true},
BandwidthLimit: bandwidthlimit.NoLimitConfig(),
})
j.prunerMtx.Lock()
j.pruner = j.prunerFactory.BuildLocalPruner(ctx, sender, alwaysUpToDateReplicationCursorHistory{sender})
j.prunerMtx.Unlock()
log.Info("start pruning")
j.pruner.Prune()
log.Info("finished pruning")
}
+35
View File
@@ -0,0 +1,35 @@
package wakeup
import (
"context"
"errors"
)
type contextKey int
const contextKeyWakeup contextKey = iota
func Wait(ctx context.Context) <-chan struct{} {
wc, ok := ctx.Value(contextKeyWakeup).(chan struct{})
if !ok {
wc = make(chan struct{})
}
return wc
}
type Func func() error
var AlreadyWokenUp = errors.New("already woken up")
func Context(ctx context.Context) (context.Context, Func) {
wc := make(chan struct{})
wuf := func() error {
select {
case wc <- struct{}{}:
return nil
default:
return AlreadyWokenUp
}
}
return context.WithValue(ctx, contextKeyWakeup, wc), wuf
}