Port proliferation while remote debugging -- kills remote server!

[SomervilleTom]

Even a brief debugging session using VS Code on a robust AWS EC2 instance running Rocky Linux 8.6 causes a proliferation of "TIME-WAIT" ports -- I counted 89 after the brief session described below.

The result is that the target system frequently becomes inaccessible (I can't tell whether it runs out of memory, runs out of ports, or something else happens). The result is that I have to restart the target system from the AWS EC2 console.

This happens much more frequently now than it did a few months ago.

Type: Bug

1. Open VS Code
2. Connect to remote system using "Remote SSH"
3. Navigate to React application on remote system
4. Start React application ('yarn start')
5. Click button in React application (opens a new browser tab)
6. Wait for React application to handle changes
7. Exit React application

VS Code version: Code 1.74.0 (5235c6b, 2022-12-05T16:38:16.075Z)
OS version: Windows_NT x64 10.0.18363
Modes:
Sandboxed: No
Remote OS version: Linux x64 4.18.0-425.3.1.el8.x86_64
Remote OS version: Linux x64 4.18.0-372.32.1.el8_6.x86_64

System Info

Item	Value
CPUs	11th Gen Intel(R) Core(TM) i7-11700K @ 3.60GHz (4 x 3522)
GPU Status	2d_canvas: enabled canvas_oop_rasterization: disabled_off direct_rendering_display_compositor: disabled_off_ok gpu_compositing: enabled multiple_raster_threads: enabled_on opengl: enabled_on rasterization: unavailable_off raw_draw: disabled_off_ok skia_renderer: enabled_on video_decode: enabled video_encode: enabled vulkan: disabled_off webgl: enabled webgl2: enabled webgpu: disabled_off
Load (avg)	undefined
Memory (System)	7.81GB (4.54GB free)
Process Argv
Screen Reader	no
VM	100%

Item	Value
Remote	SSH: tms-desktop
OS	Linux x64 4.18.0-425.3.1.el8.x86_64
CPUs	11th Gen Intel(R) Core(TM) i7-11700K @ 3.60GHz (16 x 4468)
Memory (System)	31.04GB (18.90GB free)
VM	0%

Item	Value
Remote	SSH: byron.zeetix.com
OS	Linux x64 4.18.0-372.32.1.el8_6.x86_64
CPUs	Intel(R) Xeon(R) Platinum 8259CL CPU @ 2.50GHz (4 x 2499)
Memory (System)	15.27GB (0.98GB free)
VM	0%

Extensions (4)

Extension	Author (truncated)	Version
remote-ssh	ms-	0.94.0
remote-ssh-edit	ms-	0.84.0
remote-explorer	ms-	0.0.3
json	Zai	2.0.2

[vscodenpa]

Thanks for creating this issue! It looks like you may be using an old version of VS Code, the latest stable release is 1.74.1. Please try upgrading to the latest version and checking whether this issue remains.

Happy Coding!

[SomervilleTom]

I updated to v1.74.1 and repeated the experiment -- the issue was even worse.

After closing the React App, there were 114 ports left in "TIME-WAIT":

# ss -tna | grep TIME | wc -l
114

[SomervilleTom]

VS Code just now now killed the remote server again.

I understand that bugs happen (I've been a professional developer for decades). The fact that VS Code renders the remote server inaccessible is a major issue. I can't tell if the remote server has actually crashed or if VS Code has destroyed its ability to handle SSH connections.

I can provide journalctl output if that's helpful -- let me know.

[SomervilleTom]

I know it's late Friday, but has there been any progress on this? I'm not sure yet whether this is from simply connecting VS Code or the result of some change in the javascript/nodejs support. What I do know is that it is happening MUCH more frequently.

I basically have to wait at least minute after every user gesture (mouse click, etc) in order to avoid killing the server. One of the immediate symptoms is that VS Code itself hangs while trying to reconnect to its remote server.

I'm seeing many (often tens) of "TIME-WAIT" ports happening in response to events while running the Javascript debugger. These ports seem to clear after a timeout of 1.0-1.5 minutes. The server will sometimes become accessible again if I wait long enough.

I just now had to reboot the server because saving a change while the debugger was active provoked this while the debugger was trying to compile the changes.

I'm happy to provide any further information that might be helpful, just let me know.

[SomervilleTom]

Is anything happening on this?

I've just had to reboot the target system again. Whatever is causing this makes VSCode unusable. This is a show-stopper.

[connor4312]

Pardon the pestilent port proliferation. I've improved things in the linked PR (which is also an improvement I've meant to make for a while), though there is still some port usage for browser connections. I think this should be the bulk of it, though, since all other traffic to the local browser goes through a single socket.

It is correct that TIME_WAIT ports should cleared up by the system after a configured timeout. You can reduce this by setting tcp_fin_timeout (random doc link: https://www.ibm.com/docs/en/linux-on-systems?topic=tuning-tcpip-ipv4-settings). It's also a little strange that ~100 of these ports can knock over your server. Even small systems should be able to handle thousands without blinking, though there might be policies in place on the system that set a low limit for the number of connections.