endpoint: concurrent queries

This commit is contained in:
Christian Schwarz
2020-03-27 20:17:28 +01:00
parent 8755847a40
commit 0b44e25c8b
5 changed files with 123 additions and 32 deletions
+9 -5
View File
@@ -33,15 +33,17 @@ type holdsFilterFlags struct {
Filesystems FilesystemsFilterFlag
Job JobIDFlag
Types AbstractionTypesFlag
Concurrency int64
}
// produce a query from the CLI flags
func (f holdsFilterFlags) Query() (endpoint.ListZFSHoldsAndBookmarksQuery, error) {
q := endpoint.ListZFSHoldsAndBookmarksQuery{
FS: f.Filesystems.FlagValue(),
What: f.Types.FlagValue(),
JobID: f.Job.FlagValue(),
Until: nil, // TODO support this as a flag
FS: f.Filesystems.FlagValue(),
What: f.Types.FlagValue(),
JobID: f.Job.FlagValue(),
Until: nil, // TODO support this as a flag
Concurrency: f.Concurrency,
}
return q, q.Validate()
}
@@ -58,6 +60,8 @@ func (f *holdsFilterFlags) registerHoldsFilterFlags(s *pflag.FlagSet, verb strin
variants = sort.StringSlice(variants)
variantsJoined := strings.Join(variants, "|")
s.Var(&f.Types, "type", fmt.Sprintf("only %s holds of the specified type [default: all] [comma-separated list of %s]", verb, variantsJoined))
s.Int64VarP(&f.Concurrency, "concurrency", "p", 1, "number of concurrently queried filesystems")
}
type JobIDFlag struct{ J *endpoint.JobID }
@@ -106,7 +110,7 @@ type FilesystemsFilterFlag struct {
func (flag *FilesystemsFilterFlag) Set(s string) error {
mappings := strings.Split(s, ",")
if len(mappings) == 1 {
if len(mappings) == 1 && !strings.Contains(mappings[0], ":") {
flag.F = endpoint.ListZFSHoldsAndBookmarksQueryFilesystemFilter{
FS: &mappings[0],
}
+45 -17
View File
@@ -5,6 +5,7 @@ import (
"encoding/json"
"fmt"
"os"
"sync"
"github.com/fatih/color"
"github.com/pkg/errors"
@@ -12,6 +13,7 @@ import (
"github.com/zrepl/zrepl/cli"
"github.com/zrepl/zrepl/endpoint"
"github.com/zrepl/zrepl/util/chainlock"
)
var holdsListFlags struct {
@@ -43,30 +45,56 @@ func doHoldsList(sc *cli.Subcommand, args []string) error {
return errors.Wrap(err, "invalid filter specification on command line")
}
abstractions, errors, err := endpoint.ListAbstractions(ctx, q)
abstractions, errors, err := endpoint.ListAbstractionsStreamed(ctx, q)
if err != nil {
return err // context clear by invocation of command
}
// always print what we got
if holdsListFlags.Json {
enc := json.NewEncoder(os.Stdout)
enc.SetIndent("", " ")
if err := enc.Encode(abstractions); err != nil {
panic(err)
}
fmt.Println()
} else {
for _, a := range abstractions {
fmt.Println(a)
}
}
var line chainlock.L
var wg sync.WaitGroup
defer wg.Wait()
wg.Add(1)
// then potential errors, so that users always see them
if len(errors) > 0 {
color.New(color.FgRed).Fprintf(os.Stderr, "there were errors in listing the abstractions:\n%s\n", errors)
// print results
go func() {
defer wg.Done()
enc := json.NewEncoder(os.Stdout)
for a := range abstractions {
func() {
defer line.Lock().Unlock()
if holdsListFlags.Json {
enc.SetIndent("", " ")
if err := enc.Encode(abstractions); err != nil {
panic(err)
}
fmt.Println()
} else {
fmt.Println(a)
}
}()
}
}()
// print errors to stderr
errorColor := color.New(color.FgRed)
var errorsSlice []endpoint.ListAbstractionsError
wg.Add(1)
go func() {
defer wg.Done()
for err := range errors {
func() {
defer line.Lock().Unlock()
errorsSlice = append(errorsSlice, err)
errorColor.Fprintf(os.Stderr, "%s\n", err)
}()
}
}()
wg.Wait()
if len(errorsSlice) > 0 {
errorColor.Add(color.Bold).Fprintf(os.Stderr, "there were errors in listing the abstractions")
return fmt.Errorf("")
} else {
return nil
}
}
+62 -8
View File
@@ -6,10 +6,12 @@ import (
"fmt"
"sort"
"strings"
"sync"
"github.com/pkg/errors"
"github.com/zrepl/zrepl/util/envconst"
"github.com/zrepl/zrepl/util/semaphore"
"github.com/zrepl/zrepl/zfs"
)
@@ -207,8 +209,8 @@ type ListZFSHoldsAndBookmarksQuery struct {
// else: CreateTXG of the hold or bookmark can be any value
Until *InclusiveExclusiveCreateTXG
// TODO
// Concurrent: uint > 0
// Number of concurrently queried filesystems. Must be >= 1
Concurrency int64
}
type InclusiveExclusiveCreateTXG struct {
@@ -237,6 +239,9 @@ func (q *ListZFSHoldsAndBookmarksQuery) Validate() error {
if err := q.What.Validate(); err != nil {
return err
}
if q.Concurrency < 1 {
return errors.New("Concurrency must be >= 1")
}
return nil
}
@@ -330,6 +335,33 @@ func (e ListAbstractionsErrors) Error() string {
}
func ListAbstractions(ctx context.Context, query ListZFSHoldsAndBookmarksQuery) (out []Abstraction, outErrs []ListAbstractionsError, err error) {
outChan, outErrsChan, err := ListAbstractionsStreamed(ctx, query)
if err != nil {
return nil, nil, err
}
var wg sync.WaitGroup
wg.Add(1)
go func() {
defer wg.Done()
for a := range outChan {
out = append(out, a)
}
}()
wg.Add(1)
go func() {
defer wg.Done()
for err := range outErrsChan {
outErrs = append(outErrs, err)
}
}()
wg.Wait()
return out, outErrs, nil
}
// if err != nil, the returned channels are both nil
// if err == nil, both channels must be fully drained by the caller to avoid leaking goroutines
func ListAbstractionsStreamed(ctx context.Context, query ListZFSHoldsAndBookmarksQuery) (<-chan Abstraction, <-chan ListAbstractionsError, error) {
// impl note: structure the query processing in such a way that
// a minimum amount of zfs shell-outs needs to be done
@@ -342,8 +374,11 @@ func ListAbstractions(ctx context.Context, query ListZFSHoldsAndBookmarksQuery)
return nil, nil, errors.Wrap(err, "list filesystems")
}
outErrs := make(chan ListAbstractionsError)
out := make(chan Abstraction)
errCb := func(err error, fs string, what string) {
outErrs = append(outErrs, ListAbstractionsError{Err: err, FS: fs, What: what})
outErrs <- ListAbstractionsError{Err: err, FS: fs, What: what}
}
emitAbstraction := func(a Abstraction) {
jobIdMatches := query.JobID == nil || a.GetJobID() == nil || *a.GetJobID() == *query.JobID
@@ -358,15 +393,34 @@ func ListAbstractions(ctx context.Context, query ListZFSHoldsAndBookmarksQuery)
}
if jobIdMatches && untilMatches {
out = append(out, a)
out <- a
}
}
for _, fs := range fss {
listAbstractionsImplFS(ctx, fs, &query, emitAbstraction, errCb)
}
sem := semaphore.New(int64(query.Concurrency))
go func() {
defer close(out)
defer close(outErrs)
var wg sync.WaitGroup
defer wg.Wait()
for i := range fss {
wg.Add(1)
go func(i int) {
defer wg.Done()
g, err := sem.Acquire(ctx)
if err != nil {
errCb(err, fss[i], err.Error())
return
}
func() {
defer g.Release()
listAbstractionsImplFS(ctx, fss[i], &query, emitAbstraction, errCb)
}()
}(i)
}
}()
return out, outErrs, nil
}
func listAbstractionsImplFS(ctx context.Context, fs string, query *ListZFSHoldsAndBookmarksQuery, emitCandidate putListAbstraction, errCb putListAbstractionErr) {
@@ -82,8 +82,9 @@ func GetReplicationCursors(ctx context.Context, dp *zfs.DatasetPath, jobID JobID
AbstractionReplicationCursorBookmarkV1: true,
AbstractionReplicationCursorBookmarkV2: true,
},
JobID: &jobID,
Until: nil,
JobID: &jobID,
Until: nil,
Concurrency: 1,
}
abs, absErr, err := ListAbstractions(ctx, q)
if err != nil {
@@ -176,6 +177,7 @@ func DestroyObsoleteReplicationCursors(ctx context.Context, fs string, current R
CreateTXG: current.GetCreateTXG(),
Inclusive: &zfs.NilBool{B: false},
},
Concurrency: 1,
}
abs, absErr, err := ListAbstractions(ctx, q)
if err != nil {
@@ -264,6 +266,7 @@ func MoveLastReceivedHold(ctx context.Context, fs string, to zfs.FilesystemVersi
CreateTXG: to.GetCreateTXG(),
Inclusive: &zfs.NilBool{B: false},
},
Concurrency: 1,
}
abs, absErrs, err := ListAbstractions(ctx, q)
if err != nil {
@@ -154,6 +154,7 @@ func ReleaseStepCummulativeInclusive(ctx context.Context, fs string, mostRecent
CreateTXG: mostRecent.CreateTXG,
Inclusive: &zfs.NilBool{B: true},
},
Concurrency: 1,
}
abs, absErrs, err := ListAbstractions(ctx, q)
if err != nil {
@@ -195,6 +196,7 @@ func TryReleaseStepStaleFS(ctx context.Context, fs string, jobID JobID) {
AbstractionStepHold: true,
AbstractionStepBookmark: true,
},
Concurrency: 1,
}
staleness, err := ListStale(ctx, q)
if err != nil {