Fix reconnects when ping timers under/overshoot

[georgyangelov]

The ping timer is set exactly on the ping interval. Sometimes the timer runs a bit before the ping time so the ping misses. If the next timer overshoots, the ping is not sent and the connection is restarted
(because the last message time is > ping_time*2).

Scheduling the timer more frequently than the actual ping time fixes these issues as the timer is guaranteed to ping at least once during the ping_time*2 interval.

Here is an example of what happens:

D, [2019-07-26T16:49:15.839429 #7] DEBUG -- id=TEAMID, name=TEAMNAME, domain=team#keep_alive?: Websocket ping time is 30
D, [2019-07-26T16:49:15.839698 #7] DEBUG -- id=TEAMID, name=TEAMNAME, domain=team#keep_alive?: Time since last message: 37.04522261582315
D, [2019-07-26T16:49:15.839805 #7] DEBUG -- id=TEAMID, name=TEAMNAME, domain=team#send_json: {:type=>"ping", :id=>2}
D, [2019-07-26T16:49:15.839994 #7] DEBUG -- Slack::RealTime::Concurrency::Async::Socket#send_data: {"type":"ping","id":2}
D, [2019-07-26T16:49:15.873663 #7] DEBUG -- id=TEAMID, name=TEAMNAME, domain=team#run_loop: WebSocket::Driver::MessageEvent, {"type":"pong","reply_to":2}
D, [2019-07-26T16:49:15.874079 #7] DEBUG -- id=TEAMID, name=TEAMNAME, domain=team#dispatch: reply_to=2, type=pong


D, [2019-07-26T16:49:45.865965 #7] DEBUG -- id=TEAMID, name=TEAMNAME, domain=team#keep_alive?: Websocket ping time is 30
D, [2019-07-26T16:49:45.867134 #7] DEBUG -- id=TEAMID, name=TEAMNAME, domain=team#keep_alive?: Time since last message: 29.992787840776145


D, [2019-07-26T16:50:15.895730 #7] DEBUG -- id=TEAMID, name=TEAMNAME, domain=team#keep_alive?: Websocket ping time is 30
D, [2019-07-26T16:50:15.896235 #7] DEBUG -- id=TEAMID, name=TEAMNAME, domain=team#keep_alive?: Time since last message: 60.022549238055944
W, [2019-07-26T16:50:15.896895 #7]  WARN -- id=TEAMID, name=TEAMNAME, domain=team: is offline
D, [2019-07-26T16:50:15.897195 #7] DEBUG -- : id=TEAMID, name=TEAMNAME, domain=team#restart_async
D, [2019-07-26T16:50:15.897333 #7] DEBUG -- id=TEAMID, name=TEAMNAME, domain=team#run_loop: NilClass
I, [2019-07-26T16:50:15.900166 #7]  INFO -- request: POST https://slack.com/api/rtm.start

Please note that I haven't tested the Celluloid or EventMachine implementations, just the Async one.

[dangerpr-bot]

	1 Warning
⚠️	There’re library changes, but not tests. That’s OK as long as you’re refactoring existing code.

Generated by 🚫 danger

[dblock]

This is good! It explains why reconnects are so frequent at scale despite the connection being actually alive (I was trying to debug that.)

• Let's extract the method that does / 2. Will be easier to expose it as an option as well, if you feel motivated I think it's a good idea.
• Document what's going on in README.

[georgyangelov]

Thanks for the feedback!

• Where do you think would be good to put the method? In the configuration near the ping interval option maybe?
• I'll add something to the docs tomorrow.

[dblock]

Config near ping option interval sounds right.

[georgyangelov]

@dblock Can you take another look? I wasn't sure how much detail you wanted me to include in the readme, so I provided some coarse details and linked this PR.

[dblock]

Merged. I am going to cut a release with this.

[dblock]

This actually broke the integration tests with a slack token. Attempt at a fix in 5c61124.

Locally run export SLACK_API_TOKEN=... CONCURRENCY=async-websocket and rspec spec/integration/integration_spec.rb for next time. Unfortunately can't leak a token for PRs to pick it up, so it's easy to miss.

[georgyangelov]

Oops, sorry, I'll take care of this next time. Thanks for fixing it.