release: 0.2.1: PyPI and crates.io metadata, pyarrow required, 429 retry, cron removed, RELEASING.md

.github/scripts/require-green-nightly.sh

@@ -33,7 +33,15 @@ status=$(gh api "repos/$GITHUB_REPOSITORY/compare/$GITHUB_SHA...$run_sha" --jq .
 case "$status" in
   identical | ahead) echo "the scheduled run covers $GITHUB_SHA ($status)" ;;
   *)
-    echo "::error::the latest green scheduled run ($run_sha) does not include $GITHUB_SHA ($status); wait for the next nightly run"
-    exit 1
+    # A publish tag (RELEASING.md "Publish tag"): exactly one commit that changes only files under .github/,
+    # directly on the commit the green run covers.
+    tag=$(gh api "repos/$GITHUB_REPOSITORY/compare/$run_sha...$GITHUB_SHA" --jq '[.status, .ahead_by, .behind_by, ([.files[].filename | select(startswith(".github/") | not)] | length)] | @tsv')
+    IFS=$'\t' read -r tag_status ahead behind outside <<<"$tag"
+    if [ "$tag_status" = ahead ] && [ "$ahead" = 1 ] && [ "$behind" = 0 ] && [ "$outside" = 0 ]; then
+      echo "$GITHUB_SHA is one .github-only commit on $run_sha, which the scheduled run covers"
+    else
+      echo "::error::the latest green scheduled run ($run_sha) does not include $GITHUB_SHA ($status); wait for the next nightly run"
+      exit 1
+    fi
     ;;
 esac

.github/scripts/wheel-test.sh

@@ -21,7 +21,10 @@ work=$(mktemp -d)
 "$python" -m venv "$work/venv"
 py="$work/venv/bin/python"
 "$py" -m pip install --quiet --upgrade pip
-"$py" -m pip install --quiet "$dist"/*.whl pytest
+# --no-deps: the wheel requires pyarrow, which has no musllinux wheel and would build from source (an hour);
+# pyarrow comes from the binary-only loop below, and the smoke falls back to count() without it.
+"$py" -m pip install --quiet --no-deps "$dist"/*.whl
+"$py" -m pip install --quiet pytest
 # Versions come from python/ci-interop-constraints.txt, the same pins ci.yml
 # and nightly.yml use, so every wheel leg tests against one known set.
 # pandas and numpy: the pandas_udf tests and the PySpark UDF conformance tests (T059 unit 3a); a missing wheel skips them.

.github/workflows/ci.yml

@@ -681,7 +681,7 @@ jobs:
             coverage: true
           - name: cp314t
             python: "3.14t"
-            maturin_args: "--no-default-features --features mimalloc"
+            maturin_args: "--no-default-features --features mimalloc,cloud"
             wheel_tag: cp314t
             coverage: false
     name: wheels (${{ matrix.name }})

.github/workflows/nightly.yml

@@ -9,8 +9,6 @@ on:
   schedule:
     # Off the hour: the top of the hour is when scheduled runs are delayed or dropped most.
     - cron: "23 22 * * *"
-    # One-off extra runs on 2026-11-13 for the v0.2.0 release (binaries.yml and release.yml still need a green scheduled nightly; publishing no longer does); remove after.
-    - cron: "17 8,10 13 11 *"
   workflow_dispatch:
 
 permissions:

.github/workflows/publish-crates.yml

@@ -40,7 +40,7 @@ jobs:
     needs: ci
     if: github.repository == 'KrishivAI/krishiv-engine' && (github.ref == 'refs/heads/main' || startsWith(github.ref, 'refs/tags/v'))
     runs-on: ubuntu-latest
-    timeout-minutes: 120
+    timeout-minutes: 330
     environment:
       name: crates-io
       url: https://crates.io/crates/krishiv
@@ -72,6 +72,19 @@ jobs:
               continue
             fi
             echo "::group::$crate $version"
-            cargo publish -p "$crate" --locked # zizmor: ignore[use-trusted-publishing] scoped token in the gated crates-io environment
+            # crates.io limits how fast new crate names are created (429 "try again after <date>"): wait and retry.
+            for attempt in $(seq 1 40); do
+              log="$RUNNER_TEMP/publish-$crate.log"
+              if cargo publish -p "$crate" --locked >"$log" 2>&1; then cat "$log"; break; fi # zizmor: ignore[use-trusted-publishing] scoped token in the gated crates-io environment
+              cat "$log"
+              grep -q '429 Too Many Requests' "$log" || exit 1
+              [ "$attempt" -lt 40 ] || exit 1
+              after=$(grep -oE 'try again after [^.]+GMT' "$log" | head -1 | sed 's/try again after //')
+              # An empty or unparsable date (`date -d ""` is midnight today) falls back to ten minutes.
+              if [ -n "$after" ] && until=$(date -d "$after" +%s 2>/dev/null); then wait=$(( until - $(date +%s) + 20 )); else wait=600; fi
+              [ "$wait" -gt 20 ] || wait=60
+              echo "rate limited on $crate (attempt $attempt): waiting ${wait}s until $after"
+              sleep "$wait"
+            done
             echo "::endgroup::"
           done < .github/scripts/publish-crates.txt

.github/workflows/wheels.yml

@@ -41,7 +41,7 @@ env:
   CARGO_INCREMENTAL: 0
   # maturin build flags per flavour; `dist` is the LTO + stripped profile of the workspace Cargo.toml.
   ABI3_ARGS: --profile dist --locked
-  FT_ARGS: --profile dist --locked --no-default-features --features mimalloc
+  FT_ARGS: --profile dist --locked --no-default-features --features mimalloc,cloud
   # The shipped CPU levels (see README.md, "Supported CPUs").
   X86_RUSTFLAGS: -C target-cpu=x86-64-v3
   ARM_RUSTFLAGS: -C target-cpu=neoverse-n1
@@ -86,6 +86,15 @@ jobs:
           for w in dist/*.whl; do unzip -l "$w" | grep -q 'krishiv/_cpu_level$'; done
       - uses: $/.github/actions/setup-uv
         if: matrix.libc == 'manylinux'
+      # The cloud-store test reads and writes s3:// against MinIO: the bucket is a directory of its data dir.
+      - name: S3 emulator (MinIO)
+        if: matrix.libc == 'manylinux'
+        run: |
+          mkdir -p "$RUNNER_TEMP/minio/krishiv-test"
+          docker run -d --rm --name krishiv-wheel-minio -p 127.0.0.1:9000:9000 --user "$(id -u):$(id -g)" \
+            -v "$RUNNER_TEMP/minio:/data" cgr.dev/chainguard/minio server /data
+          for _ in $(seq 1 60); do curl -s -o /dev/null --max-time 2 http://127.0.0.1:9000/minio/health/live && break; sleep 1; done
+          echo "KRISHIV_TEST_S3_ENDPOINT=http://127.0.0.1:9000" >> "$GITHUB_ENV"
       - name: Test (manylinux, on the runner)
         if: matrix.libc == 'manylinux'
         env:

CHANGELOG.md

@@ -10,6 +10,20 @@ Later entries are generated by release-plz from Conventional Commits (`release-p
 
 Nothing yet.
 
+## 0.2.1
+
+A patch release: packaging, documentation and the fixes found by acceptance testing of 0.2.0. The fixes are listed from their changelog fragments when the release is assembled (`cargo xtask changelog assemble 0.2.1`).
+
+### Changed
+
+- **`pyarrow>=16` is now a required dependency** of the Python package, so a bare `pip install krishiv` can `collect()` and
+  `show()`. `krishiv[pyarrow]` still works as an alias.
+- The PyPI page (`krishiv` and `krishiv-compat`) now has a description (`python/PYPI.md`), keywords, classifiers and project URLs.
+- The crates.io pages have READMEs: a full one for `krishiv` (its example runs as a doc test) and a short pointer to `krishiv`
+  for every internal crate; keywords and categories include `spark`.
+- The crates.io publish workflow waits out the registry's 429 rate limit for new crate names and retries; the one-off
+  nightly cron is gone; `RELEASING.md` documents publishing from a tag.
+
 ## 0.2.0
 
 The first release is the embedded core: the engine runs in your process, from Python or Rust.