ttyd
ttyd copied to clipboard
tsl0922
Keep docker container alive after browser window closed
Hi, I have read issues in which the problem was that the containers kept alive after the browser window was closed. However, I am having the opposite problem. If I run docker run -it --rm -p 7681:7681 tsl0922/ttyd and open localhost:7681 on the browser, when I close the window the container dies. Is this really the expected behaviour? It feels weird that the container itself does (obviously) not need a client to run but once the client disconnects it dies. I'd like to know if there is a workaround to that as I'd like to keep it alive and just kill it myself afterwards.
The logs are:
docker run -it --rm -p 7681:7681 tsl0922/ttyd
[2021/05/13 08:15:19:8307] N: ttyd 1.6.3-1720642 (libwebsockets 4.1.6-)
[2021/05/13 08:15:19:8308] N: tty configuration:
[2021/05/13 08:15:19:8309] N: start command: bash
[2021/05/13 08:15:19:8309] N: close signal: SIGHUP (1)
[2021/05/13 08:15:19:8309] N: terminal type: xterm-256color
[2021/05/13 08:15:19:8310] N: LWS: 4.1.6-, loglevel 7
[2021/05/13 08:15:19:8310] N: NET SRV H1 WS IPV6-off
[2021/05/13 08:15:19:8627] N: Using foreign event loop...
[2021/05/13 08:15:19:8631] N: Listening on port: 7681
[2021/05/13 08:15:23:1349] N: HTTP /token - 172.17.0.1
[2021/05/13 08:15:23:2471] N: WS /ws - 172.17.0.1, clients: 1
[2021/05/13 08:15:23:3212] N: started process, pid: 9
[2021/05/13 08:15:31:0896] N: WS closed from 172.17.0.1, clients: 0
[2021/05/13 08:15:31:0897] N: killing process, pid: 9
== uv_read failed with error -5: i/o error
[2021/05/13 08:15:31:0913] N: process killed with signal 1, pid: 9
Thanks!
The flag --rm automatically remove the container when it exits. Remove this flag for your need.
Refer docker run documentation here
The flag
--rmautomatically remove the container when it exits. Remove this flag for your need. Refer docker run documentation here
The exact same thing happens without the flag
docker run -it -p 7681:7681 tsl0922/ttyd
[2021/05/14 09:58:55:3533] N: ttyd 1.6.3-1720642 (libwebsockets 4.1.6-)
[2021/05/14 09:58:55:3533] N: tty configuration:
[2021/05/14 09:58:55:3533] N: start command: bash
[2021/05/14 09:58:55:3533] N: close signal: SIGHUP (1)
[2021/05/14 09:58:55:3533] N: terminal type: xterm-256color
[2021/05/14 09:58:55:3533] N: LWS: 4.1.6-, loglevel 7
[2021/05/14 09:58:55:3533] N: NET SRV H1 WS IPV6-off
[2021/05/14 09:58:55:3608] N: Using foreign event loop...
[2021/05/14 09:58:55:3609] N: Listening on port: 7681
[2021/05/14 09:59:01:8150] N: HTTP / - 172.17.0.1
[2021/05/14 09:59:02:2633] N: HTTP /token - 172.17.0.1
[2021/05/14 09:59:02:3236] N: WS /ws - 172.17.0.1, clients: 1
[2021/05/14 09:59:02:3775] N: started process, pid: 9
[2021/05/14 09:59:07:5659] N: WS closed from 172.17.0.1, clients: 0
[2021/05/14 09:59:07:5660] N: killing process, pid: 9
== uv_read failed with error -5: i/o error
[2021/05/14 09:59:07:5665] N: process killed with signal 1, pid: 9
https://user-images.githubusercontent.com/19463905/118256353-ad9cac00-b4ad-11eb-9a78-b105053b13fc.mp4
Yes it did
the container is there but dead:
affa5af674bf tsl0922/ttyd "/usr/bin/tini -- tt…" 13 seconds ago Exited (139) 2 seconds ago hardcore_sammet
the container is there but dead:
affa5af674bf tsl0922/ttyd "/usr/bin/tini -- tt…" 13 seconds ago Exited (139) 2 seconds ago hardcore_sammet
Yes, you can restart the container using docker start <container_name>. The container name here is hardcore_sammet.
So you have to restart the docker again every time you want to connect to it? No workaround to keep it up? Feels weird that this is the intended behaviour. I have not checked the code closely but by searching where the logs come from I saw there's a line in protocol.c that checks if there are 0 clients and exits, but it doesn't seem like it even gets there.
Anyway, thanks for looking into it!
One more try.
Are you running docker ps -a after quitting ttyd? If yes, try running it in another shell or run ttyd in background.
To run process in background, append &:
docker run -it -p 7681:7681 tsl0922/ttyd &
Sorry for the delay, running it in background (or in another shell) does not change anything. Using docker start I can start it again but if I leave the page I have to start the docker again manually. It does feel like an unexpected behaviour, I'll have to find a workaround then, thank you very much for spending time trying to help though!!!
Hi, I have the same issue recently. It was working on version 1.6.3 (tag name), container keeps on after closing the page. However, when I use newer commits, say 07ad8beeb7fe, the container just closes as the OP said.
[2021/05/27 10:16:45:3502] N: ttyd 1.6.3-07ad8be (libwebsockets 4.1.6-)
[2021/05/27 10:16:45:3503] N: tty configuration:
[2021/05/27 10:16:45:3503] N: start command: lnav -f /lnav_config /debug
[2021/05/27 10:16:45:3504] N: close signal: SIGHUP (1)
[2021/05/27 10:16:45:3504] N: terminal type: xterm-256color
[2021/05/27 10:16:45:3504] N: custom index.html: /customHtmlPath
[2021/05/27 10:16:45:3504] N: LWS: 4.1.6-, loglevel 7
[2021/05/27 10:16:45:3504] N: NET SRV H1 WS IPV6-off
[2021/05/27 10:16:45:3633] N: Using foreign event loop...
[2021/05/27 10:16:45:3635] N: Listening on port: 7681
[2021/05/27 10:16:45:9598] N: HTTP / - 10.255.253.43
[2021/05/27 10:16:47:0717] N: HTTP /token - 10.255.253.43
[2021/05/27 10:16:47:1233] N: WS /ws - 10.255.253.43, clients: 1
[2021/05/27 10:16:47:1346] N: started process, pid: 8
[2021/05/27 10:17:00:3727] N: WS closed from 10.255.253.43, clients: 0
[2021/05/27 10:17:00:3728] N: killing process, pid: 8
== uv_read failed with error -5: i/o error
[2021/05/27 10:17:00:3809] N: process killed with signal 1, pid: 8
After bisecting commits, I found this issue happens after commit d6921b23.
git bisect start
# bad: [07ad8beeb7fed1467ec3ddc51fa08cb001738362] build(deps-dev): bump webpack-cli from 4.6.0 to 4.7.0 in /html (#668)
git bisect bad 07ad8beeb7fed1467ec3ddc51fa08cb001738362
# good: [3b174dac3b81072e7ac5259c844f300d5af2e955] Bump to 1.6.3
git bisect good 3b174dac3b81072e7ac5259c844f300d5af2e955
# bad: [52adaba9e10b0372b3664ccda35e6e89c4b81dec] pty: use uint16_t for resize params
git bisect bad 52adaba9e10b0372b3664ccda35e6e89c4b81dec
# good: [62c340c6a479efff29eceb986e918bdc14a7cb2a] build(deps-dev): bump webpack-cli from 4.4.0 to 4.5.0 in /html (#596)
git bisect good 62c340c6a479efff29eceb986e918bdc14a7cb2a
# good: [0b5fd5d5fa78303561f21c310c73cc3996e94c16] build(deps-dev): bump html-webpack-plugin from 4.5.1 to 4.5.2 in /html (#609)
git bisect good 0b5fd5d5fa78303561f21c310c73cc3996e94c16
# good: [4d51d580080144bd72647fa93a6c50dacabaac17] build(deps-dev): bump mini-css-extract-plugin in /html (#613)
git bisect good 4d51d580080144bd72647fa93a6c50dacabaac17
# good: [ffea419f02a7e49d93217c89ad8b98f5d6416cd6] build(deps-dev): bump husky from 5.1.1 to 5.1.2 in /html (#616)
git bisect good ffea419f02a7e49d93217c89ad8b98f5d6416cd6
# good: [231fe32f595d107a49d95338d16d3028d377dabb] build(deps-dev): bump css-loader from 5.1.0 to 5.1.1 in /html (#617)
git bisect good 231fe32f595d107a49d95338d16d3028d377dabb
# bad: [d6921b235d1b97a57664f04e694b646f91fda5e6] protocol: add conpty support
git bisect bad d6921b235d1b97a57664f04e694b646f91fda5e6
# first bad commit: [d6921b235d1b97a57664f04e694b646f91fda5e6] protocol: add conpty support
That bad commit looks like a rewrite on pty process. Maybe there is some logic missing during the rewite.
Should we take this behavioral change as regression?
I also found that after that commit, I need to specifically run docker with environment variable TERM set, i.e., -e TERM=xterm.
Or else, lnav, program I ran under ttyd, will complain: Error opening terminal: unknown.
I also have the same issue , and I have to change the command like this
docker run --restart=always -itd -h ttyd --name ttyd -p 7681:7681 tsl0922/ttyd
After I close the browser,the container will automatic restart
