Skip to content

query: fixed race condition #877

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Mar 7, 2015
Merged

query: fixed race condition #877

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7a03d67
query: fixed race condition
jbenet Mar 7, 2015
c2bbca0
query: fix race condition: redux
jbenet Mar 7, 2015
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion routing/dht/ext_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,7 +52,7 @@ func TestGetFailures(t *testing.T) {
err = merr[0]
}

if err != context.DeadlineExceeded {
if err != context.DeadlineExceeded && err != context.Canceled {
t.Fatal("Got different error than we expected", err)
}
} else {
Expand Down
75 changes: 47 additions & 28 deletions routing/dht/query.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,8 @@ import (
pset "github.com/jbenet/go-ipfs/util/peerset"
todoctr "github.com/jbenet/go-ipfs/util/todocounter"

ctxgroup "github.com/jbenet/go-ipfs/Godeps/_workspace/src/github.com/jbenet/go-ctxgroup"
process "github.com/jbenet/go-ipfs/Godeps/_workspace/src/github.com/jbenet/goprocess"
ctxproc "github.com/jbenet/go-ipfs/Godeps/_workspace/src/github.com/jbenet/goprocess/context"
context "github.com/jbenet/go-ipfs/Godeps/_workspace/src/golang.org/x/net/context"
)

Expand Down Expand Up @@ -52,11 +53,17 @@ type queryFunc func(context.Context, peer.ID) (*dhtQueryResult, error)

// Run runs the query at hand. pass in a list of peers to use first.
func (q *dhtQuery) Run(ctx context.Context, peers []peer.ID) (*dhtQueryResult, error) {
select {
case <-ctx.Done():
return nil, ctx.Err()
default:
}

ctx, cancel := context.WithCancel(ctx)
defer cancel()

runner := newQueryRunner(ctx, q)
return runner.Run(peers)
runner := newQueryRunner(q)
return runner.Run(ctx, peers)
}

type dhtQueryRunner struct {
Expand All @@ -71,22 +78,24 @@ type dhtQueryRunner struct {
rateLimit chan struct{} // processing semaphore
log eventlog.EventLogger

cg ctxgroup.ContextGroup
proc process.Process
sync.RWMutex
}

func newQueryRunner(ctx context.Context, q *dhtQuery) *dhtQueryRunner {
func newQueryRunner(q *dhtQuery) *dhtQueryRunner {
proc := process.WithParent(process.Background())
ctx := ctxproc.WithProcessClosing(context.Background(), proc)
return &dhtQueryRunner{
query: q,
peersToQuery: queue.NewChanQueue(ctx, queue.NewXORDistancePQ(q.key)),
peersRemaining: todoctr.NewSyncCounter(),
peersSeen: pset.New(),
rateLimit: make(chan struct{}, q.concurrency),
cg: ctxgroup.WithContext(ctx),
proc: proc,
}
}

func (r *dhtQueryRunner) Run(peers []peer.ID) (*dhtQueryResult, error) {
func (r *dhtQueryRunner) Run(ctx context.Context, peers []peer.ID) (*dhtQueryResult, error) {
r.log = log

if len(peers) == 0 {
Expand All @@ -101,22 +110,30 @@ func (r *dhtQueryRunner) Run(peers []peer.ID) (*dhtQueryResult, error) {

// add all the peers we got first.
for _, p := range peers {
r.addPeerToQuery(r.cg.Context(), p)
r.addPeerToQuery(p)
}

// go do this thing.
// do it as a child func to make sure Run exits
// do it as a child proc to make sure Run exits
// ONLY AFTER spawn workers has exited.
r.cg.AddChildFunc(r.spawnWorkers)
r.proc.Go(r.spawnWorkers)

// so workers are working.

// wait until they're done.
err := routing.ErrNotFound

// now, if the context finishes, close the proc.
// we have to do it here because the logic before is setup, which
// should run without closing the proc.
go func() {
<-ctx.Done()
r.proc.Close()
}()

select {
case <-r.peersRemaining.Done():
r.cg.Close()
r.proc.Close()
r.RLock()
defer r.RUnlock()

Expand All @@ -128,12 +145,10 @@ func (r *dhtQueryRunner) Run(peers []peer.ID) (*dhtQueryResult, error) {
err = r.errs[0]
}

case <-r.cg.Closed():
log.Debug("r.cg.Closed()")

case <-r.proc.Closed():
r.RLock()
defer r.RUnlock()
err = r.cg.Context().Err() // collect the error.
err = context.DeadlineExceeded
}

if r.result != nil && r.result.success {
Expand All @@ -143,7 +158,7 @@ func (r *dhtQueryRunner) Run(peers []peer.ID) (*dhtQueryResult, error) {
return nil, err
}

func (r *dhtQueryRunner) addPeerToQuery(ctx context.Context, next peer.ID) {
func (r *dhtQueryRunner) addPeerToQuery(next peer.ID) {
// if new peer is ourselves...
if next == r.query.dht.self {
r.log.Debug("addPeerToQuery skip self")
Expand All @@ -157,18 +172,18 @@ func (r *dhtQueryRunner) addPeerToQuery(ctx context.Context, next peer.ID) {
r.peersRemaining.Increment(1)
select {
case r.peersToQuery.EnqChan <- next:
case <-ctx.Done():
case <-r.proc.Closing():
}
}

func (r *dhtQueryRunner) spawnWorkers(parent ctxgroup.ContextGroup) {
func (r *dhtQueryRunner) spawnWorkers(proc process.Process) {
for {

select {
case <-r.peersRemaining.Done():
return

case <-r.cg.Closing():
case <-r.proc.Closing():
return

case p, more := <-r.peersToQuery.DeqChan:
Expand All @@ -178,24 +193,27 @@ func (r *dhtQueryRunner) spawnWorkers(parent ctxgroup.ContextGroup) {

// do it as a child func to make sure Run exits
// ONLY AFTER spawn workers has exited.
parent.AddChildFunc(func(cg ctxgroup.ContextGroup) {
r.queryPeer(cg, p)
proc.Go(func(proc process.Process) {
r.queryPeer(proc, p)
})
}
}
}

func (r *dhtQueryRunner) queryPeer(cg ctxgroup.ContextGroup, p peer.ID) {
func (r *dhtQueryRunner) queryPeer(proc process.Process, p peer.ID) {
// make sure we rate limit concurrency.
select {
case <-r.rateLimit:
case <-cg.Closing():
case <-proc.Closing():
r.peersRemaining.Decrement(1)
return
}

// ok let's do this!

// create a context from our proc.
ctx := ctxproc.WithProcessClosing(context.Background(), proc)

// make sure we do this when we exit
defer func() {
// signal we're done proccessing peer p
Expand All @@ -212,10 +230,11 @@ func (r *dhtQueryRunner) queryPeer(cg ctxgroup.ContextGroup, p peer.ID) {
r.rateLimit <- struct{}{}

pi := peer.PeerInfo{ID: p}
if err := r.query.dht.host.Connect(cg.Context(), pi); err != nil {

if err := r.query.dht.host.Connect(ctx, pi); err != nil {
log.Debugf("Error connecting: %s", err)

notif.PublishQueryEvent(cg.Context(), &notif.QueryEvent{
notif.PublishQueryEvent(ctx, &notif.QueryEvent{
Type: notif.QueryError,
Extra: err.Error(),
})
Expand All @@ -231,7 +250,7 @@ func (r *dhtQueryRunner) queryPeer(cg ctxgroup.ContextGroup, p peer.ID) {
}

// finally, run the query against this peer
res, err := r.query.qfunc(cg.Context(), p)
res, err := r.query.qfunc(ctx, p)

if err != nil {
log.Debugf("ERROR worker for: %v %v", p, err)
Expand All @@ -244,7 +263,7 @@ func (r *dhtQueryRunner) queryPeer(cg ctxgroup.ContextGroup, p peer.ID) {
r.Lock()
r.result = res
r.Unlock()
go r.cg.Close() // signal to everyone that we're done.
go r.proc.Close() // signal to everyone that we're done.
// must be async, as we're one of the children, and Close blocks.

} else if len(res.closerPeers) > 0 {
Expand All @@ -257,7 +276,7 @@ func (r *dhtQueryRunner) queryPeer(cg ctxgroup.ContextGroup, p peer.ID) {

// add their addresses to the dialer's peerstore
r.query.dht.peerstore.AddAddrs(next.ID, next.Addrs, peer.TempAddrTTL)
r.addPeerToQuery(cg.Context(), next.ID)
r.addPeerToQuery(next.ID)
log.Debugf("PEERS CLOSER -- worker for: %v added %v (%v)", p, next.ID, next.Addrs)
}
} else {
Expand Down