Why does "cargo build" as part of 'skaffold build' take so much longer than 'docker build'

Question

We're developing a rust back-end and debugging in minikube. I have a single Dockerfile which builds the entire workspace, then just deploy that image with different cargo run statements for each microservice.

• When I run docker build from the command line, it takes about 3 minutes to build the binaries (i.e. the cargo build step).

 => [10/10] RUN  cargo fetch --locked && cargo build --offline --frozen --timings --verbose                                                          186.3s 
 => exporting to image                                                                                                                               134.2s 

• When I run skaffold build, it takes about 20 minutes to build the binaries. The extra time appears to be during the linking phase, as the cargo build step displays its last warning after 104 seconds.

#14 104.5     = note: `#[warn(dead_code)]` on by default
#14 104.5 
#14 1192.5 warning: `xx` (bin "xx") generated 9 warnings (run `cargo fix --bin "xx"` to apply 7 suggestions)
#14 1223.4       Timing report saved to /home/app/target/cargo-timings/cargo-timing-20240531T162451Z.html
#14 1223.4     Finished `dev` profile [unoptimized + debuginfo] target(s) in 19m 52s
#14 DONE 1256.5s

---

Things I've done to try and minimise build times and differences so far

In order to rule out network issues, my docker file does a cargo fetch followed by a cargo run --offline.

I have the following in my skaffold.yaml, so that docker builds always use buildKit in either command.

build:
  local:
    useBuildkit: true
    concurrency: 0

This basically makes the use of skaffold dev impractical.

I've disabled lto in my Cargo.toml for the dev profile

[profile.dev]
lto = "off"

---

Any ideas on the likely cause?