Skip to content

fix(ci): a ddebs outage must not fail the profiling image - #3937

Merged
frdomovic merged 3 commits into
masterfrom
fix/ddebs-index-nonfatal
Sep 21, 2026
Merged

frdomovic merged 3 commits into
masterfrom
fix/ddebs-index-nonfatal

Conversation

@frdomovic

Copy link
Copy Markdown
Member

Release Merod Profiling container failed on master at the rc.37 release commit (run 34971598136):

E: Some index files failed to download. They have been ignored, or old ones used instead.
ERROR: failed to solve: ... exit code: 100

What happened

ddebs.ubuntu.com was briefly unreachable. It answers 200 now and the six previous release runs all passed, so the outage was transient — but the image should not have died for it, and the Dockerfile already says so:

If ddebs hasn't mirrored the matching dbgsym yet, the install is allowed to fail non-fatally (|| echo): libc6-dbg above already covers the bulk of [unknown] frames, so a missing libgcc/libstdc++ dbgsym degrades flamegraph detail slightly rather than breaking the image.

That guard was one line too late. It wraps the install; the apt-get update that pulls the ddebs index sat unguarded directly above it. So the stated intent held for a missing package but not for a missing index — and ddebs is a third-party archive with no availability guarantee, where either way of being absent should be survivable.

The fix

&& ( apt-get update -o Acquire::Retries=3 \
     || { echo "[image] WARN: ddebs.ubuntu.com index unavailable — dropping the source and continuing..."; \
          rm -f /etc/apt/sources.list.d/ddebs.list; \
          apt-get update -o Acquire::Retries=3; } ) \

On failure the ddebs source is removed and apt updated again, rather than || true-ing the whole thing. That matters: this update refreshes every source, so swallowing it wholesale would hide a real Ubuntu archive outage as well as a ddebs one. The main archive stays fatal.

The subsequent dbgsym install then fails into the existing || echo — the degraded path that was already designed for. Acquire::Retries=3 covers the brief case before either path is taken.

Verification

  • the joined RUN body passes sh -n (valid POSIX shell)
  • a simulated failing update takes the fallback, removes the source, and continues to the next step

Note

This is the same class as the Browser E2E breakage in apps a while back: a third-party apt index going down and taking an unrelated build with it. Worth watching for elsewhere — any unguarded apt-get update after adding a non-Ubuntu source has this shape.

🤖 Generated with Claude Code

`Release Merod Profiling container` failed on master at the rc.37 release
commit:

  E: Some index files failed to download. They have been ignored, or old ones
     used instead.
  ERROR: failed to solve: ... exit code: 100

ddebs.ubuntu.com was briefly unreachable. It answers 200 now and the six
previous release runs all passed, so the outage was transient — but the image
should not have died for it, and the Dockerfile already says so:

  If ddebs hasn't mirrored the matching dbgsym yet, the install is allowed to
  fail non-fatally (`|| echo`): libc6-dbg above already covers the bulk of
  [unknown] frames, so a missing libgcc/libstdc++ dbgsym degrades flamegraph
  detail slightly rather than breaking the image.

That guard was one line too late. It wraps the INSTALL; the `apt-get update`
that pulls the ddebs index sat unguarded directly above it, so the stated
intent — optional symbols, never a broken image — held for a missing package
but not for a missing index. ddebs is a third-party archive with no
availability guarantee, and either way of being absent should be survivable.

The index fetch is now tolerated, and on failure the ddebs source is REMOVED
and apt updated again rather than `|| true`-ing the whole thing. That keeps the
main Ubuntu archive fatal: this update refreshes every source, so swallowing it
wholesale would hide a real archive outage as well as a ddebs one. The
subsequent dbgsym install then fails into the existing `|| echo`, which is
exactly the degraded path that was already designed for.

`Acquire::Retries=3` covers the brief case before either path is taken.

Verified: the joined RUN body passes `sh -n`, and a simulated failing update
takes the fallback, removes the source and continues.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@frdomovic
frdomovic merged commit bd6f7cf into master Sep 21, 2026
25 of 26 checks passed
@frdomovic
frdomovic deleted the fix/ddebs-index-nonfatal branch September 21, 2026 11:34
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants