rugo 0.3.3

Fast, dependency-free Parquet/CSV/JSONL reader and writer (no PyArrow, no NumPy).

Rugo

A fast, dependency-free file engine for Parquet, CSV, and JSONL — reading and writing — with no PyArrow and no NumPy.

Rugo reads and writes columnar data files from Python without pulling in a heavy dependency stack. There is no PyArrow and no NumPy on any read or write path — the engine is compiled C++/Cython and ships its own columnar substrate (Draken) inside the wheel. If you want to read a Parquet file, stream row groups with projection and predicate pushdown, or write Parquet/CSV/JSONL back out, and you do not want to install PyArrow to do it, rugo is for you.

Rugo is the file engine extracted from the Opteryx SQL query engine, published as a standalone wheel.

from rugo import parquet

# stream a projected, row-group-pruned read — no PyArrow, no NumPy
with parquet.read_parquet("planets.parquet", columns=["id", "name"], filters=[("id", ">", 4)]) as reader:
    for morsel in reader:
        print(morsel.column(b"name").to_pylist())

---

Why rugo over PyArrow?

PyArrow is excellent. It is also 124 MB installed, takes 137 ms to import on a cold process, and drags the entire Arrow C++ runtime into your environment whether you use 5% of it or 100%. In a serverless or containerised deployment, you pay for every millisecond of cold-start time and every megabyte of memory and image size. Rugo is built for those environments.

	rugo	PyArrow
Wheel size	2.9 MB	~23 MB
Installed footprint	7.7 MB	124 MB
Runtime dependencies	zero	Arrow C++ runtime
Cold import time	2.6 ms	137 ms
Schema read (footer only)	0.02 ms	0.05 ms
Memory at import	134 KB	260 KB

Measured on Python 3.14t, Apple M-series. Import times on a cold process (first load off disk).

In real terms:

• AWS Lambda / GCP Cloud Functions bill by the millisecond. A 137 ms PyArrow import adds cost to every cold start. With rugo that's 2.6 ms — over 50× faster. Functions that cold-start frequently can see that overhead add up directly on the invoice.

• Container image size affects pull time, horizontal scale-out speed, and storage costs. At 16× smaller, rugo shrinks the image meaningfully — and keeps you well clear of AWS Lambda's 250 MB unzipped layer limit (PyArrow alone is halfway there).

• Memory is billed per GB-second in serverless. Rugo imports at 134 KB; PyArrow at 260 KB — before either has read a single byte of data. At scale, that headroom matters.

• Schema inspection (reading the Parquet footer to check row counts and column types without decoding any column data) is 2.5× faster with rugo. If your workload metadata-scans many files before deciding what to read, this compounds.

Where PyArrow is faster: bulk full-table decoding. PyArrow's Arrow C++ engine is highly optimised for throughput on wide reads and benefits from a mature decade-long optimisation effort. Rugo uses a lock-free multithreaded decode pipeline (GIL-free worker threads) but its primary design goal is reading less — projection and row-group pruning — rather than maximising raw scan throughput. If you are streaming entire large tables into memory as fast as possible, PyArrow is the right tool. If you are running in a constrained cloud environment, routing selective queries, or simply do not want 124 MB of Arrow C++ in your dependency tree, rugo is the right tool.

---

Installation

pip install rugo

Pre-built wheels bundle Draken — there is nothing else to install. Rugo has zero runtime dependencies.

Requirements

• Python 3.11+
• A platform with a published wheel (Linux x86-64/aarch64, macOS arm64). For other platforms, see Building from source.

---

Data model

Rugo speaks Draken, the bundled columnar substrate:

• A Vector is a single typed column. Call vector.to_pylist() to get a Python list of its values.
• A Morsel is a batch of rows across several columns (a chunk of a table). Call morsel.column(b"name") to get a column Vector (note the bytes key), and len(morsel) for the row count.

Readers return Morsels (Parquet) or a result dict whose columns are Vectors (CSV, JSONL). The writers consume a Morsel. A read → write round-trip:

from rugo import parquet
from rugo.csv import write_csv
from rugo.jsonl import write_jsonl

with parquet.read_parquet("planets.parquet") as reader:
    for morsel in reader:                      # one Morsel per row group
        csv_bytes   = write_csv(morsel)        # -> bytes (RFC 4180)
        jsonl_bytes = write_jsonl(morsel)      # -> bytes (one JSON object per row)
        pq_bytes    = parquet.write_parquet(morsel)   # -> bytes (ZSTD)

---

Parquet

rugo.parquet is the recommended surface: one symmetric module for reading and writing that accepts a filename or an in-memory buffer, streams row-group Morsels, applies predicate pushdown, and writes Morsels back to bytes.

Quick start

from rugo import parquet

# Schema-only metadata (footer parse, no column data). Path OR bytes.
meta = parquet.read_metadata("planets.parquet")
print(meta.num_rows)                       # 9
print([c.name for c in meta.schema_columns])

# Streaming read: one Morsel per row group. `columns` projects; `filters`
# prune whole row groups via footer statistics (rows in surviving groups are
# NOT filtered — apply row-level predicates downstream).
with parquet.read_parquet(
    "planets.parquet",
    columns=["id", "name"],
    filters=[("id", ">", 4)],              # ops: = == != < <= > >= in "not in"
) as reader:
    for morsel in reader:
        print(morsel.column(b"name").to_pylist())

# Write a Draken Morsel to Parquet bytes (ZSTD by default; "none" to disable).
data = parquet.write_parquet(morsel, compression="zstd")
with open("out.parquet", "wb") as f:
    f.write(data)

rugo.parquet API

Function	Returns
read_parquet(source, columns=None, filters=None)	context manager yielding one Morsel per surviving row group
read_metadata(source)	ParquetMetadata (num_rows, schema_columns)
write_parquet(morsel, compression="zstd")	bytes (whole file)
write_parquet_with_bounds(morsel, compression="zstd")	(bytes, {col_index: (min, max)})

source is a filename (str) or bytes/bytearray/memoryview. filters is a list of (column, op, value); pruning is at row-group granularity.

Low-level API (rugo.parquet_reader)

Most callers should use rugo.parquet above. The low-level module is exposed for fine-grained control.

Metadata

Function	Returns
read_metadata(path: str)	ParquetMetadata(num_rows, schema_columns) (typed object)
read_metadata_from_bytes(data: bytes)	same
read_metadata_from_memoryview(mv: memoryview)	same (memoryview must be contiguous)
read_rowgroup_stats(data)	list[{num_rows, columns:[{name, physical_type, logical_type, min, max, null_count}]}] — per-row-group stats for pushdown

schema_columns is a tuple of SchemaColumn(name, physical_type, logical_type, nullable). read_rowgroup_stats min/max are raw stat bytes (or None); decode with decode_value.

Decode

read_parquet(data, column_names=None, row_group_mask=None)

• data — bytes, bytearray, or memoryview holding the full Parquet file.
• column_names — list[str] to project, or None for all columns.
• row_group_mask — optional iterable, one truthy/falsy entry per row group; a falsy entry skips decoding that row group (predicate pushdown). rugo.parquet's filters= builds this from read_rowgroup_stats.
• Returns list[Morsel] (one per decoded row group), or None on failure. On partial decode failure an individual column within a Morsel may be None.

Compatibility

Function	Returns
can_decode(path: str)	bool — quick compatibility signal, not a guarantee
can_decode_from_memory(data)	bool — same, for an in-memory buffer

Fine-grained / range decode

Function	Description
decode_column_from_chunk(chunk_bytes, col_stats, row_mask=None)	Decode a single column chunk to a Draken Vector; row_mask is an optional uint8 bitmap
decode_column_from_chunk_to_python(chunk_bytes, col_stats)	Decode a single column chunk to a Python list
decode_column_from_memory(data, column_name, row_group_stats, row_group_index)	Decode one column from a full in-memory file, by row-group index
decode_value(physical_type, logical_type, raw, prefer_text)	Decode a single raw Parquet value to a Python scalar

col_stats is the per-column stats dict for the matching row group from read_metadata.

Bloom filters

bloom_filter_maybe_contains(path, bloom_offset, bloom_length, value)  # -> bool

Evaluates a column bloom filter at the given byte offset/length for a candidate value. Bloom filter offsets and lengths are exposed in the per-column metadata returned by read_metadata.

Supported decode subset

Area	Support
Physical types	int32, int64, float32, float64, boolean, byte_array
Compression	UNCOMPRESSED, SNAPPY, ZSTD
Encodings	PLAIN, dictionary pages (PLAIN_DICTIONARY / RLE_DICTIONARY), DELTA_BINARY_PACKED, DELTA_BYTE_ARRAY
Input	Path, or in-memory bytes / memoryview, with column selection

Writing

rugo.parquet_writer (and the rugo.parquet facade) serialize a Draken Morsel to a well-formed, PyArrow-readable Parquet file.

from rugo.parquet_writer import write_parquet, write_parquet_with_bounds
data = write_parquet(morsel, compression="zstd")          # -> bytes
data, bounds = write_parquet_with_bounds(morsel)          # + per-column min/max

Area	Support
Column types	INT8/16/32/64 (→INT64), FLOAT32 (→DOUBLE), FLOAT64, BOOL, VARCHAR/NVARCHAR/VARBINARY, VARIANT (→STRING), DATE32, TIME32/64, TIMESTAMP64 (µs/ms/ns), INTERVAL (FLBA-12), DECIMAL/DECIMAL128 (FLBA), ARRAY/LIST of those (int/float/bool/string elements), all-null (→INT32). FP16 not yet.
Encoding	PLAIN values, RLE definition levels, one data page per column chunk
Compression	ZSTD (default) or uncompressed
Statistics	per-column min/max/null_count + column_orders (so readers trust them)
Bloom filters	split-block (SBBF), XXH64, on equality-friendly columns; bloom_filters=True|False|[names]
Layout	single row group per Morsel

Unsupported column types fail loud (no silent skip). Nested LIST/MAP/STRUCT and dictionary-encoded output are not yet implemented.

Limitations

• Not a full Parquet replacement reader; decode support is intentionally narrow.
• GZIP, LZO, BROTLI, LZ4, and LZ4_RAW compression codecs are not implemented in the decode path.
• INT96 is not supported for value decoding in read_parquet(...).
• FIXED_LEN_BYTE_ARRAY value decoding is not implemented.
• Decode logic is built around DATA_PAGE (V1); DATA_PAGE_V2 is not handled.
• Decode reads from a single data-page path per column chunk; files requiring full multi-page streaming decode may return partial or failed column results.
• Nested, list, and map-heavy files are not a primary decode target; flat primitive columns are the intended shape.
• On partial decode failure, individual columns may be returned as None.
• Metadata extraction is broad, but known edge cases remain around list/nested column naming normalisation.

Performance

Metadata reads (schema + row-group stats, no column data) are fast and comparable to PyArrow. The high-level read_parquet() path is correctness-first: it reconstructs Draken vectors from decoded columns and materializes through Python, so it is a serial utility rather than a throughput benchmark. The emphasis is on reading less — projection and row-group pruning — not on raw bulk scan speed.

---

JSONL

Quick start

from rugo.jsonl import get_jsonl_schema, read_jsonl, write_jsonl

# Infer schema from sample rows
schema = get_jsonl_schema("example.jsonl", sample_size=5)
# -> {"columns": [{"name": str, "type": str, "nullable": True}, ...]}

# Read from a file path with projection and predicate pushdown
result = read_jsonl(
    "example.jsonl",
    columns=["id", "name"],
    predicates=[("status", "==", "active")],
)
if result["success"]:
    print(result["num_rows"])
    for vec in result["columns"]:          # list of Draken Vectors
        print(vec.to_pylist())

# Read from bytes input
with open("example.jsonl", "rb") as f:
    result = read_jsonl(f.read(), columns=["id"])

# Write a Morsel to JSONL bytes (one JSON object per row)
data = write_jsonl(morsel)

read_jsonl

read_jsonl(
    data,                       # file path (str) or buffer (bytes/bytearray/memoryview)
    columns=None,               # list[str] to project, or None for all
    predicates=None,            # list[(column, op, value)]; op in ==, !=, <, <=, >, >=
    explicit_schema=None,       # provide a schema dict instead of inferring
    infer_schema=True,
    infer_sample_size=5,        # rows sampled for type inference
    parse_arrays=True,
    parse_objects=True,
    fail_on_error=True,
    use_threads=True,           # SIMD-accelerated parallel scan/interpret
    min_rows_per_thread=2048,
)

Return dict:

Key	Value
success	bool
column_names	list[str]
num_rows	int — rows passing predicates
columns	list of Draken Vectors
schema	dict[str, str] — column name → inferred type string
error	str — present only when success is False

Inferred type strings: int64, double, boolean, string, bytes, object, null, array<T>.

get_jsonl_schema

get_jsonl_schema(data, sample_size=5)
# -> {"columns": [{"name": str, "type": str, "nullable": True}, ...]}

Infers the schema from the first sample_size rows. Returns {"columns": []} on failure; does not raise.

Writing

write_jsonl(morsel) returns bytes, one JSON object per row. Value formatting is done in C++: doubles use shortest round-trip (std::to_chars); dates/timestamps render ISO-8601 strings; decimals are JSON numbers; arrays render as JSON arrays (null list / empty list / null element are all distinguished); nulls are null.

Performance

116 MB, 1.5 M rows, 5 cols, versus PyArrow read_json (multithreaded):

Query shape	Rugo	PyArrow
SELECT *	~67 ms	~53 ms
SELECT one_col	~33 ms	~53 ms
SELECT col WHERE id < 150k (~10% pass)	~15 ms	~53 ms
SELECT col WHERE id < 15k (~1% pass)	~7 ms	~53 ms

Bulk SELECT * is materialiser-bound — PyArrow has an edge. The analytical shapes — project + filter — are 1.2–5×+ faster, and the advantage grows with selectivity and table width.

Caveats

• String/object-heavy fields are often returned as bytes (binary-preserving), not eagerly decoded Python str/dict values.
• Mixed or deeply nested array-object content may fall back to raw JSON text/bytes in edge cases.
• Schema inference is sampled (infer_sample_size rows only); pass explicit_schema when the schema is known to avoid mismatches on heterogeneous files.

---

CSV

Quick start

from rugo.csv import read_csv, write_csv

result = read_csv("data.csv")                                          # all columns
result = read_csv("data.csv", columns=["col1", "col2"])                # projection
result = read_csv("data.csv", columns=["name"], predicates=[("age", ">", 30)])
result = read_csv("data.tsv", delimiter="\t")                          # TSV variant

if result["success"]:
    for vec in result["columns"]:          # list of Draken Vectors
        print(vec.to_pylist())

# Write a Morsel to CSV bytes (RFC 4180)
data = write_csv(morsel, delimiter=",", header=True)

read_csv

read_csv(
    data,               # file path (str) or buffer (bytes/bytearray/memoryview)
    columns=None,       # list[str] to project, or None for all
    predicates=None,    # list[(column, op, value)]; op in ==, !=, <, <=, >, >=
    delimiter=",",      # field separator character
    has_header=True,    # whether the first row is a header
    use_threads=True,   # parallel scan
)

Parameter	Type	Description
data	str / bytes / bytearray / memoryview	File path or in-memory buffer
columns	list[str] or None	Columns to project; None returns all
predicates	list[tuple] or None	Filter predicates applied before typed build
delimiter	str	Single-character field separator
has_header	bool	Whether row 0 is a header row
use_threads	bool	Enable parallel scan

Return dict:

Key	Value
success	bool
column_names	list[str]
num_rows	int — rows passing predicates
columns	list of Draken Vectors

Type inference cascade per field: int64 → float64 → VARCHAR → null (empty field).

Writing

write_csv(morsel, delimiter=",", header=True) returns RFC 4180 bytes: fields are quoted when they contain the delimiter/quote/newline (quotes doubled), nulls are empty fields, and ARRAY columns render as a (quoted) JSON array. The CSV and JSONL writers share the same C++ value formatter.

Performance

Measured against pyarrow.csv.read_csv. The expensive step is typed column build; rugo makes it survivor-only, which pays off when there is something to skip.

Narrow file — 3 cols, 1 M rows, 12.6 MB:

Query shape	Rugo	PyArrow
SELECT *	~7 ms	~3 ms
SELECT 2 cols	~6 ms	~3 ms
WHERE id > P90 (~10% pass)	~6 ms	~4 ms
WHERE id > P99 (~1% pass)	~5 ms	~3 ms

Wide file — 50 cols, 200 k rows, 55 MB:

Query shape	Rugo	PyArrow
SELECT *	~26 ms	~17 ms
SELECT 2 cols	~9 ms	~7 ms
SELECT * WHERE score > P90 (~10% pass)	~13 ms	~27 ms
SELECT * WHERE score > P99 (~1% pass)	~10 ms	~23 ms
SELECT 2 cols WHERE score > P90	~8 ms	~27 ms

On narrow files PyArrow is faster across the board. On wide files with filtering, rugo is 2–3×+ faster — the crossover is driven by how many columns can be skipped and how many rows are eliminated before the typed column build.

Known limitations

• Field length is capped at 65,535 bytes (uint16_t index); longer fields are silently truncated.
• Type inference is speculative from sampled values; there is no schema-override parameter — inferred types may be wrong on heterogeneous columns.
• Predicate operator set is fixed: ==, !=, <, <=, >, >=.

---

Design notes

• No PyArrow, no NumPy. Every read and write path is pure C++/Cython and Draken-native. Output Parquet is still standard and PyArrow-readable.
• Fail loud. can_decode(...) is a quick compatibility signal, not a guarantee; on partial decode failure a selected column may be returned as None — check, don't assume success.
• Read less. The advantage over bulk readers comes from projection and predicate/row-group pruning, not raw scan throughput.

---

Example notebook

space_missions.ipynb walks through a complete workflow on a real dataset:

• Download a Parquet file and inspect its schema with read_metadata
• Filter launches by company with row-group pruning and row-level predicate
• Aggregate total spend per company across streaming morsels
• Write filtered results to JSONL and read them back

---

Building from source

End users should pip install rugo and use the published wheels. To build from the opteryx-core source tree (rugo is developed there alongside Draken and the Opteryx engine):

python rugo/setup.py bdist_wheel    # build the standalone rugo wheel (from repo root)

For in-place development of the whole tree, use the repository's make compile.

---

License

Apache-2.0. Rugo is part of the Opteryx project.