Clear the DASH dir contents instead of rm -rf on the /opt/data mount

[mormegil6]

What this fixes

nginx-transcoder/entrypoint.sh resets segment state on every start with rm -rf /opt/data (in both the SSL and no-SSL branches). In the normal deployment /opt/data is the mount point for the DASH output volume (a named or bind volume), so this recursively deletes the contents of a mounted volume on each container start. That is:

• unnecessary - only the stale segment files need clearing, not the mount, and
• risky - it targets whatever the host has bound at /opt/data, not just the segment output.

Fix

Clear only the segment directory contents, leaving the mount point intact:

mkdir -p /opt/data/dash && find /opt/data/dash -mindepth 1 -delete

This resets the DASH output exactly as before (an empty /opt/data/dash at startup) without rm -rf-ing the mount.

Note on CI

The red test check is unrelated to this change - the workflow's actions/cache@v2 is auto-rejected by GitHub, so test fails before it runs (lint still passes).

[Copilot]

Pull request overview

This PR updates the nginx-transcoder container entrypoint to avoid recursively deleting the /opt/data mount point at startup, and instead aims to clear only the DASH segment output directory contents.

Changes:

• Replaces rm -rf /opt/data with a “clear contents of /opt/data/dash” approach in both SSL and non-SSL startup paths.
• Keeps the mount point intact while still resetting segment output state on container start.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[roddylindsay]

@mormegil6 can you check this plz?

[mormegil6]

@roddylindsay checked - the review comment is incorrect. Alpine 3.11's BusyBox find does support both -mindepth and -delete, so the entrypoint starts fine.

Verified against the exact runtime base image in this repo's Dockerfile (FROM alpine:3.11):

$ docker run --rm alpine:3.11 sh -c '
    cat /etc/alpine-release
    busybox | head -1
    mkdir -p /opt/data/dash/sub
    touch /opt/data/dash/a.m4s /opt/data/dash/sub/b.webm
    find /opt/data/dash -mindepth 1 -delete; echo "exit=$?"
    echo "remaining: $(ls -A /opt/data/dash | wc -l)"'
3.11.13
BusyBox v1.31.1 () multi-call binary.
exit=0
remaining: 0

busybox find --help on that image lists -mindepth, -maxdepth, -delete and -name. BusyBox has had FEATURE_FIND_DELETE long before 1.31, and Alpine enables it.

Worth noting the nested sub/b.webm in that test: it is removed along with its parent directory, which confirms -delete is doing the depth-first removal you would want here rather than failing on non-empty directories.

The mount point itself survives, which is the whole point of the change - rm -rf /opt/data on a bind mount or named volume either fails or blows away the mount's contents from underneath the host, whereas this resets only the segment output state.

Happy to swap in a rm -rf /opt/data/dash/* glob instead if you would rather not depend on find semantics at all, but it needs .[!.]* too to catch dotfiles, and it does not fail cleanly on an empty directory - so find reads better here.

One thing you may want regardless: as written the find sits inside an && chain, so any future failure would take the container down at startup. If you would prefer it to be non-fatal:

mkdir -p /opt/data/dash && { find /opt/data/dash -mindepth 1 -delete || true; } && chown nginx /opt/data/dash && ...

Your call - on Alpine 3.11 it does not fail, so this is just insurance against the base image changing later.