jianguotian commented on code in PR #66: URL: https://github.com/apache/paimon-mosaic/pull/66#discussion_r3459890560
########## cli/src/main.rs: ########## @@ -0,0 +1,479 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +mod fmt; +mod input; + +use std::path::PathBuf; +use std::process::ExitCode; + +use arrow_array::RecordBatch; +use clap::{Parser, Subcommand}; +use paimon_mosaic_core::reader::{MosaicReader, ReaderAccess}; + +use crate::input::FileInput; + +/// Mosaic file inspector — the cat/meta/schema/pages toolkit (cf. parquet-cli). +#[derive(Parser)] +#[command(name = "mosaic", version, about)] +struct Cli { + #[command(subcommand)] + cmd: Cmd, +} + +#[derive(Subcommand)] +enum Cmd { + /// Print the column names, types, nullability and bucket assignment. + Schema { + file: PathBuf, + #[arg(long)] + json: bool, + }, + /// Print row-group / bucket / stats metadata. + Meta { + file: PathBuf, + #[arg(long)] + json: bool, + }, + /// Print per-column encoding and slot size for each row group. + Pages { + file: PathBuf, + #[arg(long)] + json: bool, + }, + /// Print the first N rows as a table. + Cat { + file: PathBuf, + /// Number of rows to print. + #[arg(short = 'n', long, default_value_t = 10)] + num: usize, + /// Print all rows (overrides -n). + #[arg(long)] + all: bool, + /// Comma-separated columns to project. + #[arg(short, long)] + columns: Option<String>, + /// Row filter, e.g. `id>100` or `kind=a` (one condition). + #[arg(long)] + r#where: Option<String>, + #[arg(long)] + json: bool, + }, + /// Print the first N rows (alias of cat). + Head { + file: PathBuf, + #[arg(short = 'n', long, default_value_t = 10)] + num: usize, + #[arg(long)] + all: bool, + #[arg(short, long)] + columns: Option<String>, + #[arg(long)] + r#where: Option<String>, + #[arg(long)] + json: bool, + }, + /// Print the total row count. + Count { + file: PathBuf, + #[arg(long)] + json: bool, + }, + /// Print the file footer: version, buckets, compression, offsets. + Footer { + file: PathBuf, + #[arg(long)] + json: bool, + }, + /// Print on-disk bytes per column (summed over row groups). + ColumnSize { + file: PathBuf, + #[arg(long)] + json: bool, + }, + /// Print the dictionary of a dict-encoded column. + Dictionary { + file: PathBuf, + /// Column name to dump. + #[arg(short = 'c', long)] + column: String, + #[arg(long)] + json: bool, + }, + /// Print bucket layout per row group (Mosaic's column grouping). + Buckets { + file: PathBuf, + #[arg(long)] + json: bool, + }, + /// Import a CSV or JSON file into a new Mosaic file (schema inferred). + Convert { + /// Input: CSV (.csv, header row) or JSON lines (.json/.ndjson/.jsonl). + input: PathBuf, + /// Output .mosaic path. + #[arg(short, long)] + out: PathBuf, + /// Columns to build min/max stats for (comma-separated). + #[arg(long)] + stats: Option<String>, + }, +} + +fn main() -> ExitCode { + let cli = Cli::parse(); + let res = match cli.cmd { + Cmd::Schema { file, json } => schema(&file, json), + Cmd::Meta { file, json } => meta(&file, json), + Cmd::Pages { file, json } => pages(&file, json), + Cmd::Cat { file, num, all, columns, r#where, json } => cat(&file, if all { usize::MAX } else { num }, columns, r#where, json), + Cmd::Head { file, num, all, columns, r#where, json } => cat(&file, if all { usize::MAX } else { num }, columns, r#where, json), + Cmd::Count { file, json } => count(&file, json), + Cmd::Footer { file, json } => footer(&file, json), + Cmd::ColumnSize { file, json } => column_size(&file, json), + Cmd::Dictionary { file, column, json } => dictionary(&file, &column, json), + Cmd::Buckets { file, json } => buckets(&file, json), + Cmd::Convert { input, out, stats } => convert(&input, &out, stats), + }; + match res { + Ok(()) => ExitCode::SUCCESS, + Err(e) => { + eprintln!("error: {e}"); + ExitCode::FAILURE + } + } +} + +fn open(file: &PathBuf) -> std::io::Result<MosaicReader<FileInput>> { + let input = FileInput::open(file)?; + let len = input.len(); + MosaicReader::new(input, len) +} + +/// Columns in original (write) order rather than the name-sorted layout. +fn original_order(s: &paimon_mosaic_core::schema::MosaicSchema) -> Vec<usize> { + let mut by_sorted = vec![0usize; s.columns.len()]; + for (orig, &sorted) in s.original_order.iter().enumerate() { + by_sorted[sorted] = orig; + } + let mut cols: Vec<usize> = (0..s.columns.len()).collect(); + cols.sort_by_key(|&i| by_sorted[i]); + cols +} + +/// Add `total` across `cols`, distributing the remainder so the parts sum exactly. +fn split_evenly(total: usize, cols: &[usize], acc: &mut [usize]) { + if cols.is_empty() { return; } + let share = total / cols.len(); + let mut rem = total % cols.len(); + for &c in cols { + acc[c] += share + if rem > 0 { rem -= 1; 1 } else { 0 }; + } +} + +fn schema(file: &PathBuf, json: bool) -> std::io::Result<()> { + let reader = open(file)?; + let s = reader.schema(); + let cols = original_order(s); + if json { + let items: Vec<String> = cols.iter().map(|&i| { + let c = &s.columns[i]; + format!("{{\"name\":{},\"type\":{},\"nullable\":{},\"bucket\":{}}}", + fmt::json_str(&c.name), fmt::json_str(&format!("{:?}", c.data_type)), c.nullable, c.bucket_id) + }).collect(); + println!("{{\"columns\":{},\"buckets\":{},\"fields\":[{}]}}", s.columns.len(), s.num_buckets, items.join(",")); + return Ok(()); + } + println!("{} columns, {} buckets", s.columns.len(), s.num_buckets); + for i in cols { + let c = &s.columns[i]; + let null = if c.nullable { "" } else { " not null" }; + println!(" {}: {:?}{} [bucket {}]", c.name, c.data_type, null, c.bucket_id); + } + Ok(()) +} + +fn meta(file: &PathBuf, json: bool) -> std::io::Result<()> { + let reader = open(file)?; + let s = reader.schema(); + let nrg = reader.num_row_groups(); + let total: usize = (0..nrg).map(|i| reader.row_group_num_rows(i).unwrap_or(0)).sum(); + if json { + let mut rgs = Vec::new(); + for rg in 0..nrg { + let st: Vec<String> = reader.row_group_stats(rg)?.iter().map(|x| { + let mm = match (&x.min, &x.max) { + (Some(lo), Some(hi)) => format!(",\"min\":{},\"max\":{}", fmt::json_str(&fmt::render_value(lo)), fmt::json_str(&fmt::render_value(hi))), + _ => String::new(), + }; + format!("{{\"column\":{},\"nulls\":{}{}}}", fmt::json_str(&s.columns[x.column_index].name), x.null_count, mm) + }).collect(); + rgs.push(format!("{{\"rows\":{},\"stats\":[{}]}}", reader.row_group_num_rows(rg)?, st.join(","))); + } + println!("{{\"rows\":{},\"columns\":{},\"buckets\":{},\"row_groups\":[{}]}}", total, s.columns.len(), s.num_buckets, rgs.join(",")); + return Ok(()); + } + println!("file: {} rows, {} columns, {} buckets, {} row groups", total, s.columns.len(), s.num_buckets, nrg); + for rg in 0..nrg { + println!("row group {rg}: {} rows", reader.row_group_num_rows(rg)?); + for st in reader.row_group_stats(rg)? { + let mm = match (&st.min, &st.max) { + (Some(lo), Some(hi)) => format!("min={} max={}", fmt::render_value(lo), fmt::render_value(hi)), + _ => "no min/max".to_string(), + }; + println!(" {}: nulls={} {}", s.columns[st.column_index].name, st.null_count, mm); + } + } + Ok(()) +} + +fn pages(file: &PathBuf, json: bool) -> std::io::Result<()> { + let reader = open(file)?; + let s = reader.schema(); + let nrg = reader.num_row_groups(); + if json { + let mut rgs = Vec::new(); + for rg in 0..nrg { + let items: Vec<String> = reader.page_infos(rg)?.iter().map(|p| { + format!("{{\"column\":{},\"bucket\":{},\"encoding\":{},\"slot_size\":{}}}", + fmt::json_str(&s.columns[p.column_index].name), p.bucket, fmt::json_str(fmt::encoding_name(p.encoding)), p.slot_size) + }).collect(); + rgs.push(format!("[{}]", items.join(","))); + } + println!("{{\"row_groups\":[{}]}}", rgs.join(",")); + return Ok(()); + } + for rg in 0..nrg { + println!("row group {rg}:"); + for p in reader.page_infos(rg)? { + let c = &s.columns[p.column_index]; + println!(" {}: bucket {} encoding={} slot={}B", c.name, p.bucket, fmt::encoding_name(p.encoding), p.slot_size); + } + } + Ok(()) +} + +fn count(file: &PathBuf, json: bool) -> std::io::Result<()> { + let reader = open(file)?; + let n: usize = (0..reader.num_row_groups()).map(|i| reader.row_group_num_rows(i).unwrap_or(0)).sum(); + if json { println!("{{\"rows\":{}}}", n); } else { println!("{}", n); } + Ok(()) +} + +/// Output sink writing a Mosaic file to disk, tracking its own position. +struct FileOut { f: std::fs::File, pos: u64 } +impl paimon_mosaic_core::writer::OutputFile for FileOut { + fn write(&mut self, d: &[u8]) -> std::io::Result<()> { + use std::io::Write; + self.pos += d.len() as u64; + self.f.write_all(d) + } + fn flush(&mut self) -> std::io::Result<()> { use std::io::Write; self.f.flush() } + fn pos(&self) -> u64 { self.pos } +} + +fn convert(input: &PathBuf, out: &PathBuf, stats: Option<String>) -> std::io::Result<()> { + use paimon_mosaic_core::writer::{MosaicWriter, WriterOptions}; + use arrow_schema::ArrowError; + let bad = |e: ArrowError| std::io::Error::new(std::io::ErrorKind::InvalidData, e.to_string()); + let is_json = matches!(input.extension().and_then(|e| e.to_str()), Some("json") | Some("ndjson") | Some("jsonl")); + // Infer schema, then build a batch iterator — CSV (header) or JSON (one object per line). + type Batches = Box<dyn Iterator<Item = Result<RecordBatch, ArrowError>>>; + let (schema, reader): (arrow_schema::Schema, Batches) = if is_json { + let mut r = std::io::BufReader::new(std::fs::File::open(input)?); + let (schema, _) = arrow_json::reader::infer_json_schema(&mut r, None).map_err(bad)?; + let rd = arrow_json::ReaderBuilder::new(std::sync::Arc::new(schema.clone())) + .build(std::io::BufReader::new(std::fs::File::open(input)?)).map_err(bad)?; + (schema, Box::new(rd)) + } else { + let (schema, _) = arrow_csv::reader::Format::default().with_header(true) + .infer_schema(std::io::BufReader::new(std::fs::File::open(input)?), None).map_err(bad)?; + let rd = arrow_csv::ReaderBuilder::new(std::sync::Arc::new(schema.clone())) + .with_header(true).build(std::io::BufReader::new(std::fs::File::open(input)?)).map_err(bad)?; + (schema, Box::new(rd)) + }; + let opts = WriterOptions { + stats_columns: stats.map(|s| s.split(',').map(|x| x.trim().to_string()).filter(|x| !x.is_empty()).collect()).unwrap_or_default(), + ..Default::default() + }; + // Write to a temp file and rename on success, so a mid-stream failure never + // leaves a truncated .mosaic in place. + let tmp = out.with_extension("mosaic.tmp"); + let mut rows = 0; + let res = (|| { + let sink = FileOut { f: std::fs::File::create(&tmp)?, pos: 0 }; + let mut w = MosaicWriter::new(sink, &schema, opts)?; + for batch in reader { + let batch = batch.map_err(|e| std::io::Error::new(std::io::ErrorKind::InvalidData, e.to_string()))?; + rows += batch.num_rows(); + w.write_batch(&batch)?; + } + w.close() + })(); + if let Err(e) = res { + let _ = std::fs::remove_file(&tmp); + return Err(e); + } + std::fs::rename(&tmp, out)?; + println!("wrote {} ({} rows, {} columns)", out.display(), rows, schema.fields().len()); + Ok(()) +} + +fn cat(file: &PathBuf, num: usize, columns: Option<String>, filter: Option<String>, json: bool) -> std::io::Result<()> { + let mut reader = open(file)?; + if let Some(list) = &columns { + let names: Vec<&str> = list.split(',').map(|x| x.trim()).filter(|x| !x.is_empty()).collect(); + reader.project(&names)?; + } + let pred = filter.as_deref().map(fmt::parse_where).transpose() + .map_err(|e| std::io::Error::new(std::io::ErrorKind::InvalidInput, e))?; + // Column index of the filter target, for stats-based row-group skipping. + let pred_col = pred.as_ref().and_then(|p| reader.schema().columns.iter().position(|c| c.name == p.column)); + let mut batches: Vec<RecordBatch> = Vec::new(); + let mut got = 0usize; + for rg in 0..reader.num_row_groups() { + if got >= num { + break; + } + // Pushdown: skip a row group when its min/max prove no row can match. + if let (Some(p), Some(ci)) = (&pred, pred_col) { + if let Some(st) = reader.row_group_stats(rg)?.iter().find(|s| s.column_index == ci) { + if fmt::stats_exclude(p, &st.min, &st.max) { continue; } + } + } + let mut batch = reader.row_group_reader(rg)?.read_columns()?; + if let Some(p) = &pred { + batch = fmt::apply_where(&batch, p) + .map_err(|e| std::io::Error::new(std::io::ErrorKind::InvalidInput, e))?; + } + got += batch.num_rows(); + batches.push(batch); + } + if batches.iter().all(|b| b.num_rows() == 0) { + if !json { + println!("(no rows)"); + } + } else if json { + print!("{}", fmt::ndjson(&batches, num)); + } else { + print!("{}", fmt::pretty_table(&batches, num)); + } + Ok(()) +} + +fn footer(file: &PathBuf, json: bool) -> std::io::Result<()> { + use paimon_mosaic_core::spec::{COMPRESSION_ZSTD, MAGIC, VERSION}; + let reader = open(file)?; + let s = reader.schema(); + let comp = if reader.compression() == COMPRESSION_ZSTD { "zstd" } else { "none" }; + let magic = std::str::from_utf8(&MAGIC).unwrap_or("MOSA"); + if json { + println!("{{\"magic\":{},\"version\":{},\"buckets\":{},\"row_groups\":{},\"compression\":{}}}", + fmt::json_str(magic), VERSION, s.num_buckets, reader.num_row_groups(), fmt::json_str(comp)); + } else { + println!("magic={} version={} buckets={} row_groups={} compression={}", + magic, VERSION, s.num_buckets, reader.num_row_groups(), comp); + } + Ok(()) +} + +fn column_size(file: &PathBuf, json: bool) -> std::io::Result<()> { + let reader = open(file)?; + let s = reader.schema(); + let mut bytes = vec![0usize; s.columns.len()]; + let mut raw = vec![0usize; s.columns.len()]; + for rg in 0..reader.num_row_groups() { + // On-disk bytes are tracked per bucket (works for both monolithic and + // paged layouts); split a bucket's size across its member columns. + for b in reader.bucket_infos(rg)? { + if b.columns.is_empty() { continue; } + split_evenly(b.size, &b.columns, &mut bytes); Review Comment: > This differs materially from Parquet CLI `column-size`, which sums exact column-chunk sizes from metadata. Here the bucket size is split evenly across member columns, so the per-column bytes can be quite misleading; for paged buckets we should be able to use per-column slot sizes instead. If monolithic bucket attribution must remain approximate, please label it as approximate in the command output/docs rather than presenting it as exact per-column size. Fixed, paged buckets use exact per-column slot sizes; monolithic multi-column buckets are split and labeled `(approx)`. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
