Feature: Client Disconnect Message

[benloh]

Problem

Kalani reported:

We’re getting lots of server disconnect errors from students who seem to have reasonable internet connections. Is there a way to extend the disconnect-ping timing so that students are less likely to get errors when there’s a slow connection, they can still make changes?

Class total is 150 students (not all at one time; the most I’ve seen in Google Analytics is 10-15). Most are in dorms or hotels (the latter of which is an issue for speed purposes), but it shouldn’t be changing IP addresses mid-session. It seems more like a ping time out, the way it’s been described.

Issues that students have reported in basic categories:

• Long load times from students who are just seeing spinning balls and white screens. I’m offering those students alternatives because there are only 2 or 3 and it’s an Internet-speed issue that can’t be solved by us.
• Post-load server disconnection issues. These students are getting the whole network loaded, and then when they try to edit stuff, they get a disconnect. (Again, PEBKAC, so I am going on what the student is describing and don’t have as much detail as I’d like). I think that an extension of the timeout would help.
• We have 1 person, as yet unidentified, who is taking the whole network down. Any students who load the network after that student get a white screen with a text-only proxy error. One student in particular has reported this twice and I’m hoping that it’s this student’s tech creating issues but they’re in quarantine right now, so I have no idea when I’ll be able to tackle that issue. In the meantime, uptimerobot.com is checking for the proxy keyword and alerting me so I can cycle the network instance when that happens. If we can tie IP or access code to the drops, we can ID the student (I hope).

Finally, there are some duplicate nodes appearing, which I think are probably related to server disconnects (5 or 6 “Justinian” nodes showing up, for instance, though I’ve cleaned those up).

Solution

This pull request does 4 things:

1. Extend the wait time for heartbeats. Both the server and the client will wait for 10 seconds before declaring a disconnect. Heartbeats should be generated by the server every 5 seconds, so this gives us a large window.

2. Show a specific "Client Disconnect" or "Server Disconnect" with a timestamp error for users.

"Client Disconnected" -- Client did not receive a "ping" heartbeat from the server within the time allowed. Usually this is a result of the client losing the internet connection.

"Server Disconnected" -- Either the server shut down, or the server did not receive a "pong" response from the client within the time allowed. Usually this is a result of the server initiating the disconnect as it shuts down.

3. Log the missing "pong" message to the server logs along with the UADDR so you can identify the machine thats down. e.g.

11:30:39 tacitus SRV-NET - UADDR_02 pong not received before time ran out -- CLIENT CONNECTION DEAD!

4. Add GZIP compression for all files. In some cases resulting in an 80% reduction in data that has to be sent over the wires.

This isn't a perfect solution but perhaps it'll give us a little more information about what's going on.

To Test

1. git checkout dev-bl/net-debug
2. npm run dev || ./nc.js --dataset=xxxx
3. Login with a client
4. Disconnect the client wifi -- the client should display a "Client Disconnected" message.
5. On the server, you should see a log entry indicating which client was disconnected.
6. Reconnect the client wifi.
7. ctrl-c on the server to stop the server.
8. The client should display a "Server DIsconnected" message.

[benloh]

@kalanicraig @jdanish Please try this build to see if it helps with the error reporting and drops.

[benloh]

Added GZIP compression. This reduces the size of netc-lib.js from 6.5MB to 1.2MB.

[benloh]

Added UADDR to server log.

[benloh]

@jdanish I just pushed a number of performance enhancements to net-debug.

Slow Network Improvements

• Bugfix: Wrap standalone calls in a promise to prevent LOADASSETS lifecycle errors. Addresses EnterApp() Lifecycle Error #132 and Errors when the network is super slow  #136.

• NodeTable and EdgeTable are now only loaded when their respective tab is visible. This saves about 2 seconds during page processing time. Errors when the network is super slow  #136.

• Replace font-awesome (?) badge with a css badge. This reduces load time by about 2 seconds. Errors when the network is super slow  #136.

• Delay blocking javascript loads so that main app page will load and display. Some students see a blank screen #135, Errors when the network is super slow  #136.

• Display a "Loading Net.Create..." message so user knows the page is loading. Some students see a blank screen #135, Errors when the network is super slow  #136.

• Add minification with terser.

You'll need to run npm ci to install the new terser library and remove the old uglify-* libraries.

[jdanish]

Looks great!! One bug and one minor thought that I'll log in a sec. Thanks!

[jdanish]

Just filed #137 and #138 ... otherwise looks good at a glance but would want to test some more.

[benloh]

@jdanish More fixes!

NodeTable / EdgeTable Improvements

• Markdown rendering has been optimized. markdown-react significantly affects rendering. #139
• NodeTables and EdgeTables no longer disappear when a node is selected. Node and Edge tables disappear  #137, Node and Edge Tables re-render? #138
• Degrees are now displayed again. 6fc35cc
• Properly unmount NodeTable and EdgeTable db864d6

[jdanish]

Thanks @benloh

Unfortunately, I am getting a compile error? It looks like the prior function was not closed but I wanted to make sure there wasn't something else?

20:55:50 - error: Compiling of app/view/netcreate/components/EdgeTable.jsx failed. L425:27 Unexpected token, expected ;
423 | this.handleDataUpdate(D3DATA);
424 |

425 | componentWillUnmount() {
| ^
426 | this.AppStateChangeOff('D3DATA', this.handleDataUpdate);
427 | this.AppStateChangeOff('TEMPLATE', this.OnTemplateUpdate);
428 | }
Stack trace was suppressed. Run with LOGGY_STACKS=1 to see the trace.

[benloh]

@jdanish Ack! Apologies. I missed committing a line because I thought it was just an empty line, but of course it had a paren. Should be up now.

[benloh]

@jdanish Just checking in on the testing. If this seems to work well, I'd like to merge this and tag it 1.3.1.

[jdanish]

Sounds like it is looking good, so let’s do it! FYI. Some students are still seeing blank screens and we are struggling to get enough information to figure out why. So we are assuming it’s not your problem (though we’ll log it if we do figure out why). Our short-term next-step fix is to suggest they use IUAnyWare (https://uits.iu.edu/iuanyware <https://uits.iu.edu/iuanyware>) a virtualization client where I have confirmed they can run chrome and Net.Create works great via that install. So that’s going to be our short-term solution until we get more info.