fix(diagnostics,minimald): re-pin socket-join pids and make the diag stream timeout idle-based - #927

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

🏁 Script executed:

#!/bin/bash
set -euo pipefail

cargo tree -i tokio
rg -n 'poll_write|WriteAll' \
  "${CARGO_HOME:-$HOME/.cargo}/registry/src" \
  --glob '*/tokio-*/src/io/util/write_all.rs'

Repository: gominimal/minimal

Length of output: 147

🏁 Script executed:

#!/bin/bash
set -euo pipefail

printf '--- crates/minimald/src/diag.rs (relevant lines) ---\n'
sed -n '150,220p' crates/minimald/src/diag.rs

printf '\n--- tokio version in this repo ---\n'
cargo metadata --no-deps --format-version 1 | python3 - <<'PY'
import json,sys
m=json.load(sys.stdin)
for p in m["packages"]:
    if p["name"]=="tokio":
        print(p["name"], p["version"])
PY

printf '\n--- tokio write_all source if present in cargo registry ---\n'
python3 - <<'PY'
import os,glob,sys
home=os.environ.get("CARGO_HOME", os.path.expanduser("~/.cargo"))
paths=glob.glob(home+"/registry/src/*/tokio-*/src/io/util/write_all.rs")
print("\n".join(paths) if paths else "NOT_FOUND")
PY

Repository: gominimal/minimal

Length of output: 246

🏁 Script executed:

#!/bin/bash
set -euo pipefail

# Locate the exact call site and surrounding logic.
rg -n 'write_all|timeout\(STREAM_IDLE_TIMEOUT|STREAM_IDLE_TIMEOUT' crates/minimald/src/diag.rs -n -A8 -B8

# Find any local guidance on Tokio or diagnostics behavior.
rg -n 'diagnostics::COLLECTOR_TIMEOUT|STREAM_IDLE_TIMEOUT|write_all' crates -g '!**/target/**'

Repository: gominimal/minimal

Length of output: 22091

🏁 Script executed:

#!/bin/bash
set -euo pipefail

# Read Tokio's write_all implementation from the local cargo registry if present.
python3 - <<'PY'
import os,glob
home=os.environ.get("CARGO_HOME", os.path.expanduser("~/.cargo"))
matches=glob.glob(home+"/registry/src/*/tokio-*/src/io/util/write_all.rs")
if not matches:
    print("NO_TOKIO_SOURCE")
else:
    print(matches[0])
PY

Repository: gominimal/minimal

Length of output: 171

🏁 Script executed:

#!/bin/bash
set -euo pipefail

echo '--- diag.rs excerpt ---'
sed -n '160,210p' crates/minimald/src/diag.rs

echo
echo '--- tokio version from Cargo.lock ---'
rg -n '^name = "tokio"|^version = ' Cargo.lock -A1 -B1

echo
echo '--- local tokio write_all implementation path(s) ---'
python3 - <<'PY'
import os,glob
home=os.environ.get("CARGO_HOME", os.path.expanduser("~/.cargo"))
paths=glob.glob(home+"/registry/src/*/tokio-*/src/io/util/write_all.rs")
for p in paths:
    print(p)
PY

Repository: gominimal/minimal

Length of output: 50373

🌐 Web query:

Tokio AsyncWriteExt::write_all implementation partial writes loop timeout semantics source

💡 Result:

Tokio's AsyncWriteExt::write_all implementation functions by continuously calling the write method on the underlying AsyncWrite type until the entire buffer is exhausted or an error occurs [1][2][3][4]. Regarding your specific areas of inquiry: Partial Writes and Loops The implementation essentially wraps write in a loop [1][2][4]. Because AsyncWrite::poll_write is readiness-based, it may return a partial result (e.g., writing fewer bytes than the length of the provided buffer) [5]. write_all handles this by tracking the remaining data and continuing to call write in subsequent polls until all bytes are written [2][4]. Timeout Semantics write_all does not have built-in timeout semantics [1][2]. If you need a timeout, you must wrap the write_all future in a tokio::time::timeout call. Because write_all is not cancellation-safe, if a timeout occurs, the operation will be aborted mid-progress [1][2][4]. Cancellation Safety write_all is explicitly documented as not cancellation-safe [1][2][4]. If it is used within a tokio::select! block and another branch completes first, the write_all future will be dropped [6]. Any data that was successfully written to the underlying stream remains there, but the future will not resume where it left off [1][2][4]. Subsequent calls to write_all on the same writer will start over from the beginning of the buffer, which can lead to duplicated data if the previous partial write was not accounted for [1][2][4]. Important Considerations - I/O Completion vs. Readiness: Tokio's I/O traits are readiness-based [5]. Even when write_all returns Poll::Ready(Ok(())), it only guarantees that the data has been accepted by the underlying writer (e.g., buffered by the kernel or the runtime), not necessarily that it has been fully persisted to the destination (e.g., disk) [7][8][5]. For files or buffered writers, an explicit flush may be required to ensure completion [7][8][5]. - Errors: write_all will return the first error encountered during its internal loop [1][2][3]. Top results: [1][2][7][6][5]

Citations:

• 1: https://github.com/tokio-rs/tokio/blob/5030b300/tokio/src/io/util/async_write_ext.rs
• 2: https://docs.rs/tokio/latest/tokio/io/trait.AsyncWriteExt.html
• 3: https://docs.rs/tokio/1.16.1/tokio/io/trait.AsyncWriteExt.html
• 4: https://docs.rs/tokio/1.19.2/x86_64-apple-darwin/tokio/io/trait.AsyncWriteExt.html
• 5: tokio::fs::File::write returns early before OS says that the operation is completed tokio-rs/tokio#7378
• 6: https://biriukov.dev/posts/async-rust-gocha-tokio-cancelation-select-future-then/
• 7: write_all doesn't guarantee data to be written after returning Poll::Ready tokio-rs/tokio#5531
• 8: Async write_all sometimes silently fails to write to file tokio-rs/tokio#4296