remove_dir_all < 0.8.0 (RUSTSEC/GHSA, Dependabot runtime alert) came in only through tempfile 3.3.0. cargo update -p tempfile moves it to 3.10.1, which no longer depends on remove_dir_all; Cargo.lock now has no remove_dir_all entry. Lock-only change.
* Move vulnerable dependencies to patched versions
Backend (Cargo.lock): openssl 0.10.55 -> 0.10.81, bytes 1.12.1, h2 0.3.27,
tokio 1.26 -> 1.47.1 (mio 1.x), url 2.3.1 -> 2.5.7 (idna 1.1), rand 0.8.8,
spin 0.9.9, time 0.3.55, ring 0.17.14.
Three alerts needed manifest changes:
- tokio-rustls 0.23 -> 0.26 (rustls 0.20 -> 0.23), which is what moves
ring off 0.16. The ring provider is selected explicitly instead of the
default aws-lc-rs, so the image build needs no cmake/C toolchain; roots
still come from webpki-roots (now 1.0). Checked against live hosts: a
valid certificate connects, an expired one and a name mismatch are
refused.
- maxminddb 0.23 -> 0.27: lookup() now returns a result that is decoded,
and the City model's fields are no longer Options. The existing
locate_random_ip test (12.5.56.25 -> "Gardena") passes.
- structopt (clap 2, which pulled atty) -> clap 4 derive in telemetry_core,
telemetry_shard and the soak tests; criterion 0.4 -> 0.5 for the same
reason. Both binaries' arguments were parsed before and after for the
docker-compose invocations and a set of other cases: identical results,
except that `--feed-timeout` given without a value is now an error where
structopt silently used the default. --denylist keeps taking every value
up to the next flag (num_args = 1..), as its help text describes.
Frontend: npm audit fix within ranges, and webpack-dev-server 4 -> 5.2.6
(no devServer options are configured, so nothing to migrate). Left:
sockjs -> uuid 8 inside webpack-dev-server 5; the fix is 6.0, which needs
Node 22 while CI runs 20, and it is the local dev server only.
Checked: backend cargo test (58 passed), clippy, fmt; frontend lint,
prettier, test and build.
* Reject an invalid byte size instead of panicking
ByteSize::from_str called expect("all ascii digits") on the numeric part,
so a value with no digits ("abc", "k") or one too large for usize
aborted the shard at startup with a panic instead of a usage error, and
the suffix multiplication could overflow silently in release builds.
Parse errors now return Err and the multiplication is checked.
The new test fails on the old code (panic) and passes on this one.
Debian 11 left LTS at the end of August 2026. The runtime stage's
`apt-get install openssl` now fails: the index still names
openssl 1.1.1w-0+deb11u8 but deb.debian.org answers 404 for the file, so the
image cannot be built at all.
Moving only the runtime to bookworm would break the binaries: openssl-sys is in
the lock, so they link the system libssl, and a bullseye build wants libssl1.1,
which bookworm does not ship. Both stages move together: the builder becomes
rust:1.88-bookworm (the same toolchain ci-unified carried) and the runtime
bookworm-slim, so the binaries link the libssl3 they run with.
SatoshiQaziMuhammed