Use libunwind for jemalloc memory profiling stack traces (#5602)

## Motivation

The default libgcc DWARF unwinder (`_Unwind_Find_FDE`) used by jemalloc
for profiling
stack traces has a known livelock

([jemalloc/jemalloc#2282](jemalloc/jemalloc#2282))
— one
thread hangs in `_Unwind_Find_FDE` and all others block on futex_wait.
We've hit this in
production on validator shards.

## Proposal

Switch jemalloc's stack unwinder to libunwind by building with
`--enable-prof-libunwind`.

- Point `tikv-jemallocator` at
[linera-io/jemallocator](https://github.com/linera-io/jemallocator) fork
which adds a
`profiling_libunwind` feature (upstream doesn't support it yet —

[tikv/jemallocator#146](tikv/jemallocator#146))
- Enable `profiling_libunwind` in the `jemalloc` feature
- Add `libunwind-dev` (builder) and `libunwind8` (runtime) to
`docker/Dockerfile`

## Test Plan

- CI
- Verify memory profiling works end-to-end in a temporary network
deployment

## Release Plan

- These changes shobe backported to `testnet_conway`.

## Links

- Fork PR: linera-io/jemallocator#1
- Upstream issue:
[tikv/jemallocator#146](tikv/jemallocator#146)
- Jemalloc livelock:

[jemalloc/jemalloc#2282](jemalloc/jemalloc#2282)

Author: ndr-ds

Cargo.lock

@@ -152,7 +152,6 @@ indexed-db = "0.4.2"
 insta = "1.36.1"
 is-terminal = "0.4.12"
 itertools = "0.14.0"
-jemalloc_pprof = "0.8.1"
 js-sys = "0.3.70"
 k256 = { version = "0.13.4", default-features = false, features = [
     "ecdsa",
@@ -163,10 +162,13 @@ k256 = { version = "0.13.4", default-features = false, features = [
 ] }
 k8s-openapi = { version = "0.21.1", features = ["v1_28"] }
 kube = "0.88.1"
+linera-jemalloc-ctl = { version = "0.6.1", features = ["use_std"] }
+linera-jemallocator = "0.6.1"
 linera-kywasmtime = "0.1.0"
 linked-hash-map = "0.5.6"
 log = "0.4.21"
 lru = "0.15.0"
+mappings = "0.7.1"
 mini-moka = "0.10.3"
 nonzero_lit = "0.1.2"
 num-bigint = "0.4.3"
@@ -186,6 +188,7 @@ opentelemetry_sdk = { version = "0.30.0", features = ["trace", "rt-tokio"] }
 papaya = "0.2.3"
 pathdiff = "0.2.1"
 port-selector = "0.1.6"
+pprof_util = { version = "0.8.0", features = ["flamegraph", "symbolize"] }
 prettyplease = "0.2.16"
 proc-macro-error = "1.0.4"
 proc-macro2 = "1.0"
@@ -267,7 +270,6 @@ test-log = { version = "0.2.15", default-features = false, features = [
 test-strategy = "0.3.1"
 thiserror = "1.0.65"
 thiserror-context = "0.1.1"
-tikv-jemallocator = "0.6.0"
 tokio = "1.36.0"
 tokio-stream = "0.1.14"
 tokio-test = "0.4.3"

Cargo.toml

@@ -17,7 +17,7 @@ ARG copy=${binaries:+_copy}
 ARG build_flag=--release
 ARG build_folder=release
 ARG build_features=scylladb,metrics,jemalloc
-ARG rustflags="-C force-frame-pointers=yes"
+ARG rustflags="-C force-frame-pointers=yes -L /usr/lib/x86_64-linux-gnu"
 
 FROM rust:1.86-slim-bookworm AS builder
 ARG git_commit
@@ -30,7 +30,8 @@ RUN apt-get update && apt-get install -y \
     pkg-config \
     protobuf-compiler \
     clang \
-    make
+    make \
+    libunwind-dev
 
 COPY . .
 
@@ -71,7 +72,8 @@ LABEL build_date=$build_date
 
 RUN apt-get update && apt-get install -y \
     ca-certificates \
-    openssl
+    openssl \
+    libunwind8
 RUN update-ca-certificates
 
 COPY --from=binaries \

docker/Dockerfile

@@ -15,16 +15,16 @@ version.workspace = true
 workspace = true
 
 [features]
-jemalloc = ["jemalloc_pprof"]
+jemalloc = ["linera-jemalloc-ctl", "mappings", "pprof_util", "tempfile"]
 
 [dependencies]
 anyhow.workspace = true
 axum.workspace = true
-jemalloc_pprof = { workspace = true, features = [
-    "symbolize",
-    "flamegraph",
-], optional = true }
+linera-jemalloc-ctl = { workspace = true, optional = true }
+mappings = { workspace = true, optional = true }
+pprof_util = { workspace = true, optional = true }
 prometheus.workspace = true
+tempfile = { workspace = true, optional = true }
 thiserror.workspace = true
 tokio = { workspace = true, features = ["full"] }
 tokio-util.workspace = true