Repository navigation
docker run output sometimes lost (exit 0, empty stdout) right after an engine start on 29.9.0 #569
Description
Found while verifying engine upgrade and rollback for #568.
What happens
Right after the engine starts, docker run --rm -v vol:/data alpine:3.20 cat /data/f sometimes exits 0 with no output. The file is there: the next run prints it, and ls shows it unchanged.
Measured (WSL 3.0.3, Windows 10 19045, docker CLI 29.8.2, skrog built from #568)
One throwaway named engine (skrog target create), driven with --state-dir and its own context. Each run is read 1–4 immediately after skrog start, engine rollback or engine upgrade.
| dockerd | output lost |
|---|---|
| 29.9.0 (rootfs 29.9.0-2) | 4 of 23 |
| 29.9.2 (rootfs 29.9.2-1) | 0 of 53 |
It was always the first or second run after a start. The skrog binary, CLI, image and volume were the same throughout; only the engine differed.
Not known
- The cause. I have not established whether this is moby 29.9.0 (an attach/stdout race) or skrog's relay right after a start that 29.9.2's timing happens not to trigger. 53 clean reads make 29.9.2 look unaffected; that is not proof.
- Whether 29.8.x, which most installs run today, is affected. Not tested.
- Whether it occurs without a fresh start. Not tested; every loss followed one.
Why it matters
Exit 0 with output silently missing is the worst kind of failure for scripts and CI: $(docker run … cat x) gets an empty string and nothing says so.
Next
- Reproduce against the engine socket directly, bypassing skrog's pipe, to place the fault.
- Run the same loop on 29.8.2.
Activity