Fix panic caused by ETCD client timeout on Revoke()

[gibertoni]

When revoking a lease, for example on app shutdown, there is a scenario on which ETCD client takes more than revokeTimeout to respond.

In this scenario, the timed action would trigger a defer close(c) and close the channel, but wouldnt cancel the Revoke operation. Therefore, as the sd.cli.Revoke(ctx, sd.leaseID) returns and then tries to write the result on the channel, panic is triggered:

panic: send on closed channel

The suggested approach to handle timeouts with ETCD client is to set the timeout into the context.

Note that this same issue could happen on renewLease() if grantLeaseTimeout is ever smaller than etcdDialTimeout. The default configuration is safe, so I am not changing this one.

[coveralls]

Pull Request Test Coverage Report for Build 9841627666

Details

• 0 of 12 (0.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.008%) to 61.974%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
cluster/etcd_service_discovery.go	0	12	0.0%

Totals
Change from base Build 9668931564:	0.008%
Covered Lines:	4873
Relevant Lines:	7863

---

💛 - Coveralls