Possible macOS resource-pressure issue when running Codex CLI inside zellij/Ghostty during long-lived sessions
Open 💬 2 comments Opened Apr 16, 2026 by jroth1111
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
Possible macOS resource-pressure issue when Codex CLI is used inside zellij/Ghostty during long-lived sessions.
Context:
- macOS 26.3.2 on Apple Silicon (M3 Max, 36 GB RAM)
- zellij 0.44.1
- Ghostty 1.3.1
- long-running session with heavy streamed output and background panes
Observed from diagnostics:
- multiple
codex-aarch64-apple-darwinprocesses with nontrivial memory footprint - APFS ENOSPC log lines named
codex-aarch64-apple-darwinseconds before: - low swap: failed to create swapfile
- memorystatus: System is unhealthy
- watchdog panic
I am not claiming Codex was the sole root cause, because the machine had broader resource pressure. But given #9345 and existing zellij compatibility reports, I'm wondering whether heavy TUI streaming / large file operations inside zellij can retain memory or amplify scrollback / disk churn on macOS.
Happy to attach redacted logs if useful.
2 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
Appreciate the bot linking these — but I'd argue this is actually a sign the memory-pressure issue on macOS deserves serious attention, not just triage as a duplicate.
#16828, #17832, #17257, and #16866 all describe variants of the same underlying problem: Codex CLI processes consuming unbounded memory on macOS and eventually destabilizing the system. If the bot can find four near-duplicates in short order, that suggests a pattern, not isolated incidents.
In my case the machine watchdog-panicked after APFS ran out of space for swapfiles — with
codex-aarch64-apple-darwinnamed in ENOSPC log lines seconds before failure. That's not a minor leak; that's a process contributing to system-level resource exhaustion.Happy to close as a duplicate if one of those is being actively worked on, but I'd want confirmation that the root cause has been identified and a fix is in progress — not just that similar reports exist.