SLAVEOF command may block a long time

[caipengbo]

Search before asking

• I had searched in the issues and found no similar issues.

Version

2.2.0

Minimal reproduce step

1. echo 6 > /proc/sys/net/ipv4/tcp_syn_retries
2. SLAVEOF a non-routable IP address(such as 10.255.255.1)
3. SLAVEOF new_master_ip port

What did you expect to see?

Step 3 should not block for a long time because we are switching to the new master, which should be a quick operation.

What did you see instead?

Step 3 will block a long time(> 2min based on above reproduced scenario)

Anything Else?

This is because we use blocking IO when use socket connect(2). In some cases, connect(2) will be blocked for a long time.

Are you willing to submit a PR?

• I'm willing to submit a PR!

[git-hulk]

@caipengbo I think the root cause is we didn't set a timeout for the connect operation, so it will backoff retransmit the syn packet(1s,2s,4s,8s,16s,32s,64s) until reaching the tcp_syn_retries(default is 6 for most Linux distro).

[caipengbo]

@caipengbo I think the root cause is we didn't set a timeout for the connect operation, so it will backoff retransmit the sync packet(1s,2s,4s,8s,16s,32s,64s) until reaching the tcp_syn_retries(default is 6 for most Linux distro).

Yes, we should add a timeout.

[git-hulk]

Cool, got it.

[caipengbo]

Let's set it a little longer than 3 seconds, so we can tolerate 2 times lost sync packages? @git-hulk

[git-hulk]

Yes, sounds good to me.