Crash when creating a new tab (WSL2 multiples days without shutdown)

[Zykino]

I had zellij opened on 2 distinct WindowsTerminal window, connected to the same session. After around 4 days (windows goes in sleep/hibernation every nights), I tried to open a new tab and this crash log was shown instead.

Error occurred in server:

  × Thread 'wasm' panicked.
  ├─▶ Originating Thread(s)
  │     1. stdin_handler_thread: AcceptInput
  │     2. screen_thread: NewTab
  │     3. plugin_thread: NewTab
  │
  ├─▶ At /home/x667421/.cargo/registry/src/github.com-1ecc6299db9ec823/zellij-server-0.34.4/src/lib.rs:726:18
  ╰─▶ Program terminates: a fatal error occured

      Caused by:
          0: failed to load plugin for client 1
          1: failed to start plugin PluginConfig {
                 path: "tab-bar",
                 run: Pane(
                     None,
                 ),
                 _allow_exec_host_cmd: false,
                 location: Zellij(
                     PluginTag(
                         "tab-bar",
                     ),
                 ),
             } for client 1
          2: wasi filesystem creation error: `Could not get metadata for file "/tmp/zellij-1000": entity not found`
  help: If you are seeing this message, it means that something went wrong.
        Please report this error to the github issue.
        (https://github.com/zellij-org/zellij/issues)

        Also, if you want to see the backtrace, you can set the `RUST_BACKTRACE` environment variable to `1`.

Apparently /tmp/zellij-1000 exist after the crash.

file /tmp/zellij-1000/zellij-log/zellij.log
/tmp/zellij-1000/zellij-log/zellij.log: empty

It happens once so, I am not sure how much this is easy to reproduce.

[imsnif]

Hey @har7an - I think we fixed this already, right?

[har7an]

@imsnif Hard to say, #2093 dealt specifically with the cache folder, although it does recreate the TMP folder when it's missing, too. I think that this is resolved by #2093, because the description sounds very much like what led me to make that PR in the first place.

@Zykino Can you try running the following command:

$ rm -rf /tmp/zellij-1000 && zellij action new-tab

And post the output you get?

Edit: Beware, if my assumption is correct, this will crash your session.

[Zykino]

Brand new terminal/tab (no active session)

There is no active session!

I tried it again with a zellij session in another tab and I get the same result.

$ zellij ls
No active zellij sessions found.

[imsnif]

@Zykino - wild guess: is there a chance you're running more than one version of Zellij?

[Zykino]

When in the title I said "when creating a new tab", I meant "with keyboard shortcuts: Ctrl+t, n.
That said, I’m not sure, I may have done cargo install-update -a in the interval. With zellij last release from mid December I doubt it upgraded, but maybe.

I just tried again with my setup and this cause a crash each time:

1. zellij # Start a session
2. rm /tmp/zellij-1000 -fr
3. "Ctrl+t, n"
   => Crash

Opening a new tab with zellij action new-tab after removing the /tmp folder does not crash either from within zellij.

• In my previous reply I tried the commands outside of a zellij session. This give the same result as stated BUT if I then do Ctrl+t, n, I have the new tab as expected.

-> The default keybinding is not protected, but calling explicitly a zellij command from inside recreate enough for the keyboard shortcut to work. And multiples zellij action new-tab do not achieve a new tab. Even after a successful one from the keyboard.

Sorry if I’m not clear 😅 a bit sick and I do not have the force to edit much a text in a foreign language right now.
Still good job in general on zellij, it has been way more stable since my last issue. 👍 (Now I need to unbind Ctrl+q as I hate this shortcut everywhere on AZERTY keyboards.)

[imsnif]

Hey @Zykino - first, I hope you feel better soon! Please don't exert yourself just to help us with a bug :)

Zellij relies on IPC pipes in the tmp folder. If it is deleted, it will crash. We should be able to recover from this, but right now we are not. In the next version we will add a sticky bit to the tmp files so that they will not be cleaned up by other processes. Not sure if this is your issue, but it might help.

Zellij also does not deal well with having two versions of itself running on the same machine. This is a result of some bad design in the early days. We should be able to deal with this too, but currently don't.

[Zykino]

I reproduce on Pop-Os only 1 version of Zellij.

zellij # Start a session
rm /tmp/zellij-1000 -fr
# "Ctrl+t, n"
# Crash

zellij # Start a session
rm /tmp/zellij-1000 -fr
`zellij action new-tab`
# Ok tab 2 created
"Ctrl+t, n"
# Ok tab 3 created

It feels so strange that starting another process fix the tmp folder structure while in process shortcut don’t. I though that the keybinding was exactly that: a shortuct that executed exactly the same code as the long form operation.

I guess architecture is what it is 😅.

[har7an]

@Zykino There's a lot of technical difference between what Ctrl + t, n and zellij action new-tab do, unfortunately. I'm very confident that our next release will fix your issue, though. If you feel adventurous, you can always compile a current version from main and try that.

[Zykino]

Just checked, it is indeed fixed in 0.35.1 🎉

[non-descriptive]

It reappeared in 0.36.0 or earlier. Look #2078. Also there's bunch of issues related to this specific panic.