mirror of
https://github.com/etcd-io/etcd.git
synced 2024-09-27 06:25:44 +00:00
clientv3: fix the design & implementation of double barrier
Check the client count before creating the ephemeral key, do not
create the key if there are already too many clients. Check the
count after creating the key again, if the total kvs is bigger
than the expected count, then check the rev of the current key,
and take action accordingly based on its rev. If its rev is in
the first ${count}, then it's valid client, otherwise, it should
fail.
Signed-off-by: Benjamin Wang <wachao@vmware.com>
This commit is contained in:
@@ -45,25 +45,46 @@ func NewDoubleBarrier(s *concurrency.Session, key string, count int) *DoubleBarr
|
||||
// Enter waits for "count" processes to enter the barrier then returns
|
||||
func (b *DoubleBarrier) Enter() error {
|
||||
client := b.s.Client()
|
||||
|
||||
// Check the entered clients before creating the UniqueEphemeralKey,
|
||||
// fail the request if there are already too many clients.
|
||||
if resp1, err := b.enteredClients(client); err != nil {
|
||||
return err
|
||||
} else if len(resp1.Kvs) >= b.count {
|
||||
return ErrTooManyClients
|
||||
}
|
||||
|
||||
ek, err := newUniqueEphemeralKey(b.s, b.key+"/waiters")
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
b.myKey = ek
|
||||
|
||||
resp, err := client.Get(b.ctx, b.key+"/waiters", clientv3.WithPrefix())
|
||||
// Check the entered clients after creating the UniqueEphemeralKey
|
||||
resp2, err := b.enteredClients(client)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
if len(resp2.Kvs) >= b.count {
|
||||
lastWaiter := resp2.Kvs[b.count-1]
|
||||
if ek.rev > lastWaiter.CreateRevision {
|
||||
// delete itself now, otherwise other processes may need to wait
|
||||
// until these keys are automatically deleted when the related
|
||||
// lease expires.
|
||||
if err = b.myKey.Delete(); err != nil {
|
||||
// Nothing to do here. We have to wait for the key to be
|
||||
// deleted when the lease expires.
|
||||
}
|
||||
return ErrTooManyClients
|
||||
}
|
||||
|
||||
if len(resp.Kvs) > b.count {
|
||||
return ErrTooManyClients
|
||||
}
|
||||
|
||||
if len(resp.Kvs) == b.count {
|
||||
// unblock waiters
|
||||
_, err = client.Put(b.ctx, b.key+"/ready", "")
|
||||
return err
|
||||
if ek.rev == lastWaiter.CreateRevision {
|
||||
// TODO(ahrtr): we might need to compare ek.key and
|
||||
// string(lastWaiter.Key), they should be equal.
|
||||
// unblock all other waiters
|
||||
_, err = client.Put(b.ctx, b.key+"/ready", "")
|
||||
return err
|
||||
}
|
||||
}
|
||||
|
||||
_, err = WaitEvents(
|
||||
@@ -74,6 +95,18 @@ func (b *DoubleBarrier) Enter() error {
|
||||
return err
|
||||
}
|
||||
|
||||
// enteredClients gets all the entered clients, which are ordered by the
|
||||
// createRevision in ascending order.
|
||||
func (b *DoubleBarrier) enteredClients(cli *clientv3.Client) (*clientv3.GetResponse, error) {
|
||||
resp, err := cli.Get(b.ctx, b.key+"/waiters", clientv3.WithPrefix(),
|
||||
clientv3.WithSort(clientv3.SortByCreateRevision, clientv3.SortAscend))
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return resp, nil
|
||||
}
|
||||
|
||||
// Leave waits for "count" processes to leave the barrier then returns
|
||||
func (b *DoubleBarrier) Leave() error {
|
||||
client := b.s.Client()
|
||||
@@ -96,7 +129,7 @@ func (b *DoubleBarrier) Leave() error {
|
||||
}
|
||||
isLowest := string(lowest.Key) == b.myKey.Key()
|
||||
|
||||
if len(resp.Kvs) == 1 {
|
||||
if len(resp.Kvs) == 1 && isLowest {
|
||||
// this is the only node in the barrier; finish up
|
||||
if _, err = client.Delete(b.ctx, b.key+"/ready"); err != nil {
|
||||
return err
|
||||
|
||||
Reference in New Issue
Block a user