fix(sql-file): stream sqlserver zip imports incrementally

This commit is contained in:
КГБ Lieu authored and GitHub committed 2026-10-01 02:40:02 +08:00
1 parent e0f802c031
commit 64e5847f74
7 files changed
+659 -66

No files matched your search

+358 -28
View File
@@ -14,6 +14,7 @@ use crate::models::connection::DatabaseType;
pub const DEFAULT_SQL_FILE_UPLOAD_MAX_MB: u32 = 200;
pub const MAX_SQL_FILE_UPLOAD_MAX_MB: u32 = 4096;
const SQLSERVER_ZIP_INSERT_BATCH_MAX_STATEMENTS: usize = 100;
pub fn clamp_sql_file_upload_max_mb(value: u32) -> u32 {
value.clamp(1, MAX_SQL_FILE_UPLOAD_MAX_MB)
@@ -24,7 +25,8 @@ use crate::query::{
QueryExecutionOptions,
};
use crate::sql::{
optimize_sql_file_import_statements, prepare_sql_file_statement, split_sql_batches, statement_summary,
optimize_sql_file_import_statements, optimize_sql_file_import_statements_with_max_insert_batch_statements,
prepare_sql_file_statement, split_sql_batches, starts_with_sqlserver_module_ddl, statement_summary,
SqlFileImportStatement, SqlFileImportStatementKind, SqlFilePhase, SqlFileProgress, SqlFileRequest,
SqlFileStatementAction, SqlFileStatus, SqlParsingOptions, SqlStatementSplitter, SqlStatementWithControl,
};
@@ -691,6 +693,7 @@ async fn execute_sql_file_content_inner(
import_target.as_ref().map(|target| target.db_type),
import_target.as_ref().and_then(|target| target.driver_profile.as_deref()),
request.txn_session_id.is_some(),
None,
);
// MySQL-family imports need one pinned connection so `USE` and session
// state survive across the whole file.
@@ -761,12 +764,43 @@ pub async fn execute_sql_file_paths(
token: CancellationToken,
started_at: Instant,
emit: impl FnMut(SqlFileProgress),
) -> Result<(), String> {
execute_sql_file_paths_with_options(state, request, file_paths, token, started_at, false, emit).await
}
pub async fn execute_sql_file_zip_package_paths(
state: &AppState,
request: &SqlFileRequest,
file_paths: &[&Path],
token: CancellationToken,
started_at: Instant,
emit: impl FnMut(SqlFileProgress),
) -> Result<(), String> {
execute_sql_file_paths_with_options(state, request, file_paths, token, started_at, true, emit).await
}
async fn execute_sql_file_paths_with_options(
state: &AppState,
request: &SqlFileRequest,
file_paths: &[&Path],
token: CancellationToken,
started_at: Instant,
zip_package: bool,
emit: impl FnMut(SqlFileProgress),
) -> Result<(), String> {
with_sql_file_transaction(
state,
request,
&token,
Box::pin(execute_sql_file_paths_inner(state, request, file_paths, token.clone(), started_at, emit)),
Box::pin(execute_sql_file_paths_inner(
state,
request,
file_paths,
token.clone(),
started_at,
zip_package,
emit,
)),
)
.await
}
@@ -777,6 +811,7 @@ async fn execute_sql_file_paths_inner(
file_paths: &[&Path],
token: CancellationToken,
started_at: Instant,
zip_package: bool,
mut emit: impl FnMut(SqlFileProgress),
) -> Result<(), String> {
if file_paths.is_empty() {
@@ -800,6 +835,7 @@ async fn execute_sql_file_paths_inner(
})
.unwrap_or_default();
let database_type = import_target.as_ref().map(|target| target.db_type);
let max_insert_batch_statements = sql_file_import_max_insert_batch_statements(zip_package, database_type);
let mut progress = SqlFileExecutionProgress::new();
// Validate the entire dump before any SQL reaches the database. This also
// identifies view placeholders whose final CREATE VIEW appears later.
@@ -859,10 +895,13 @@ async fn execute_sql_file_paths_inner(
let mut prev_success_count = 0usize;
let mut prev_failure_count = 0usize;
let mut prev_affected_rows = 0u64;
let mut splitter = Some(StreamingSqlFileSplitter::new(database_type, options));
let mut splitter = zip_package.then(|| StreamingSqlFileSplitter::new(database_type, options, true));
let mut pending_statements = Vec::with_capacity(SQL_FILE_STATEMENT_BATCH_SIZE);
let import_result = async {
for (file_index, file_path) in file_paths.iter().enumerate() {
if !zip_package {
splitter = Some(StreamingSqlFileSplitter::new(database_type, options, false));
}
let file_name = file_path.file_name().map(|n| n.to_string_lossy().to_string()).unwrap_or_default();
// Emit a file-boundary progress event so the frontend knows which file is
@@ -977,6 +1016,7 @@ async fn execute_sql_file_paths_inner(
&token,
started_at,
&mut pending_statements,
max_insert_batch_statements,
import_target.as_ref(),
mysql_executor.as_mut(),
&mut progress,
@@ -985,10 +1025,10 @@ async fn execute_sql_file_paths_inner(
.await?;
}
let is_last_file = file_index + 1 == file_count;
if is_last_file {
let should_finish_splitter = !zip_package || file_index + 1 == file_count;
if should_finish_splitter {
let mut next_statements =
splitter.take().expect("splitter is only taken once, on the last file").finish();
splitter.take().expect("splitter is finished at each ordinary file or the last ZIP part").finish();
if let Some(filter) = restore_filter.as_mut() {
if let Err(error) = filter_restore_statements(
&mut next_statements,
@@ -1012,6 +1052,7 @@ async fn execute_sql_file_paths_inner(
&token,
started_at,
&mut pending_statements,
max_insert_batch_statements,
import_target.as_ref(),
mysql_executor.as_mut(),
&mut progress,
@@ -1088,7 +1129,8 @@ async fn scan_sql_file_tables(
token: &CancellationToken,
) -> Result<(Vec<SqlFileTable>, std::collections::BTreeSet<SqlFileTable>), String> {
let mut decoder = SqlFileStreamDecoder::open_for_target(file_path, true).await?;
let mut splitter = StreamingSqlFileSplitter::new(Some(DatabaseType::Mysql), SqlParsingOptions::mysql_compatible());
let mut splitter =
StreamingSqlFileSplitter::new(Some(DatabaseType::Mysql), SqlParsingOptions::mysql_compatible(), false);
let mut scan = TableRestoreFilter::default();
while let Some(chunk) = decoder.next_chunk().await? {
if token.is_cancelled() {
@@ -1732,8 +1774,12 @@ enum StreamingSqlFileSplitterKind {
}
impl StreamingSqlFileSplitter {
pub(crate) fn new(db_type: Option<DatabaseType>, options: SqlParsingOptions) -> Self {
Self(StreamingSqlFileSplitterKind::new(db_type, options))
pub(crate) fn new(
db_type: Option<DatabaseType>,
options: SqlParsingOptions,
flush_sqlserver_semicolons: bool,
) -> Self {
Self(StreamingSqlFileSplitterKind::new(db_type, options, flush_sqlserver_semicolons))
}
pub(crate) fn push_chunk(&mut self, chunk: &str) -> Vec<SqlStatementWithControl> {
@@ -1746,9 +1792,9 @@ impl StreamingSqlFileSplitter {
}
impl StreamingSqlFileSplitterKind {
fn new(db_type: Option<DatabaseType>, options: SqlParsingOptions) -> Self {
fn new(db_type: Option<DatabaseType>, options: SqlParsingOptions, flush_sqlserver_semicolons: bool) -> Self {
if db_type == Some(DatabaseType::SqlServer) {
Self::SqlServerBatches(SqlServerBatchSplitter::default())
Self::SqlServerBatches(SqlServerBatchSplitter::new(flush_sqlserver_semicolons))
} else {
Self::Statements(SqlStatementSplitter::with_options(options))
}
@@ -1775,13 +1821,40 @@ impl StreamingSqlFileSplitterKind {
}
}
#[derive(Default)]
struct SqlServerBatchSplitter {
batch: String,
partial_line: String,
lexical_state: SqlServerLexicalState,
flush_semicolons: bool,
}
impl Default for SqlServerBatchSplitter {
fn default() -> Self {
Self::new(false)
}
}
#[derive(Default, Clone, Copy)]
enum SqlServerLexicalState {
#[default]
Normal,
SingleQuote,
DoubleQuote,
BracketIdentifier,
LineComment,
BlockComment,
}
impl SqlServerBatchSplitter {
fn new(flush_semicolons: bool) -> Self {
Self {
batch: String::new(),
partial_line: String::new(),
lexical_state: SqlServerLexicalState::default(),
flush_semicolons,
}
}
fn push_chunk(&mut self, chunk: &str) -> Vec<String> {
self.partial_line.push_str(chunk);
let mut batches = Vec::new();
@@ -1804,11 +1877,62 @@ impl SqlServerBatchSplitter {
}
fn push_line(&mut self, line: &str, batches: &mut Vec<String>) {
if line.trim().eq_ignore_ascii_case("go") {
if matches!(self.lexical_state, SqlServerLexicalState::Normal) && line.trim().eq_ignore_ascii_case("go") {
self.push_batch(batches);
} else {
self.batch.push_str(line);
self.batch.push('\n');
return;
}
let mut chars = line.chars().peekable();
while let Some(ch) = chars.next() {
self.batch.push(ch);
match self.lexical_state {
SqlServerLexicalState::Normal => match ch {
'\'' => self.lexical_state = SqlServerLexicalState::SingleQuote,
'"' => self.lexical_state = SqlServerLexicalState::DoubleQuote,
'[' => self.lexical_state = SqlServerLexicalState::BracketIdentifier,
'-' if chars.peek() == Some(&'-') => {
self.batch.push(chars.next().unwrap());
self.lexical_state = SqlServerLexicalState::LineComment;
}
'/' if chars.peek() == Some(&'*') => {
self.batch.push(chars.next().unwrap());
self.lexical_state = SqlServerLexicalState::BlockComment;
}
';' if self.flush_semicolons && !starts_with_sqlserver_module_ddl(&self.batch) => {
self.push_batch(batches)
}
_ => {}
},
SqlServerLexicalState::SingleQuote if ch == '\'' => {
if chars.peek() == Some(&'\'') {
self.batch.push(chars.next().unwrap());
} else {
self.lexical_state = SqlServerLexicalState::Normal;
}
}
SqlServerLexicalState::DoubleQuote if ch == '"' => {
if chars.peek() == Some(&'"') {
self.batch.push(chars.next().unwrap());
} else {
self.lexical_state = SqlServerLexicalState::Normal;
}
}
SqlServerLexicalState::BracketIdentifier if ch == ']' => {
if chars.peek() == Some(&']') {
self.batch.push(chars.next().unwrap());
} else {
self.lexical_state = SqlServerLexicalState::Normal;
}
}
SqlServerLexicalState::BlockComment if ch == '*' && chars.peek() == Some(&'/') => {
self.batch.push(chars.next().unwrap());
self.lexical_state = SqlServerLexicalState::Normal;
}
_ => {}
}
}
self.batch.push('\n');
if matches!(self.lexical_state, SqlServerLexicalState::LineComment) {
self.lexical_state = SqlServerLexicalState::Normal;
}
}
@@ -1818,6 +1942,7 @@ impl SqlServerBatchSplitter {
batches.push(batch.to_string());
}
self.batch.clear();
self.lexical_state = SqlServerLexicalState::Normal;
}
}
@@ -1828,6 +1953,7 @@ async fn execute_sql_file_statement_batch(
token: &CancellationToken,
started_at: Instant,
statements: &mut Vec<SqlStatementWithControl>,
max_insert_batch_statements: Option<usize>,
import_target: Option<&SqlFileImportTarget>,
mysql_executor: Option<&mut MySqlSqlFileExecutor>,
progress: &mut SqlFileExecutionProgress,
@@ -1842,6 +1968,7 @@ async fn execute_sql_file_statement_batch(
import_target.map(|target| target.db_type),
import_target.and_then(|target| target.driver_profile.as_deref()),
request.txn_session_id.is_some(),
max_insert_batch_statements,
);
execute_planned_statements_with_progress(
state,
@@ -1923,9 +2050,15 @@ fn plan_sql_file_statements(
db_type: Option<DatabaseType>,
driver_profile: Option<&str>,
manual: bool,
max_insert_batch_statements: Option<usize>,
) -> Vec<ControlledSqlFileImportStatement> {
if !manual {
return optimize_controlled_sql_file_import_statements(statements, db_type, driver_profile);
return optimize_controlled_sql_file_import_statements(
statements,
db_type,
driver_profile,
max_insert_batch_statements,
);
}
// Preserve each statement and its error boundary. The normal import path
// may merge INSERTs and retry them individually, which is not valid after
@@ -1933,17 +2066,44 @@ fn plan_sql_file_statements(
statements
.iter()
.flat_map(|statement| {
optimize_sql_file_import_statements(std::slice::from_ref(&statement.sql), db_type, driver_profile)
.into_iter()
.map(|statement| ControlledSqlFileImportStatement { statement, stop_on_error: true })
optimize_sql_file_import_statements_for_batch(
std::slice::from_ref(&statement.sql),
db_type,
driver_profile,
max_insert_batch_statements,
)
.into_iter()
.map(|statement| ControlledSqlFileImportStatement { statement, stop_on_error: true })
})
.collect()
}
fn sql_file_import_max_insert_batch_statements(zip_package: bool, db_type: Option<DatabaseType>) -> Option<usize> {
(zip_package && db_type == Some(DatabaseType::SqlServer)).then_some(SQLSERVER_ZIP_INSERT_BATCH_MAX_STATEMENTS)
}
fn optimize_sql_file_import_statements_for_batch(
sql: &[String],
db_type: Option<DatabaseType>,
driver_profile: Option<&str>,
max_insert_batch_statements: Option<usize>,
) -> Vec<SqlFileImportStatement> {
match max_insert_batch_statements {
Some(max_insert_batch_statements) => optimize_sql_file_import_statements_with_max_insert_batch_statements(
sql,
db_type,
driver_profile,
max_insert_batch_statements,
),
None => optimize_sql_file_import_statements(sql, db_type, driver_profile),
}
}
fn optimize_controlled_sql_file_import_statements(
statements: &[SqlStatementWithControl],
db_type: Option<DatabaseType>,
driver_profile: Option<&str>,
max_insert_batch_statements: Option<usize>,
) -> Vec<ControlledSqlFileImportStatement> {
let mut controlled = Vec::new();
let mut start = 0;
@@ -1955,7 +2115,7 @@ fn optimize_controlled_sql_file_import_statements(
}
let sql = statements[start..end].iter().map(|statement| statement.sql.clone()).collect::<Vec<_>>();
controlled.extend(
optimize_sql_file_import_statements(&sql, db_type, driver_profile)
optimize_sql_file_import_statements_for_batch(&sql, db_type, driver_profile, max_insert_batch_statements)
.into_iter()
.map(|statement| ControlledSqlFileImportStatement { statement, stop_on_error }),
);
@@ -2688,9 +2848,9 @@ mod tests {
Some(DatabaseType::Postgres),
None,
);
let optimized = plan_sql_file_statements(&statements, Some(DatabaseType::Postgres), None, false);
let optimized = plan_sql_file_statements(&statements, Some(DatabaseType::Postgres), None, false, None);
assert_eq!(optimized.len(), 1);
let manual = plan_sql_file_statements(&statements, Some(DatabaseType::Postgres), None, true);
let manual = plan_sql_file_statements(&statements, Some(DatabaseType::Postgres), None, true, None);
assert_eq!(manual.len(), 2);
assert!(manual.iter().all(|item| item.stop_on_error && item.statement.source_statement_count == 1));
assert_eq!(manual[0].statement.sql, statements[0].sql);
@@ -2804,7 +2964,7 @@ mod tests {
assert!(dump.len() > SQL_FILE_STATEMENT_BATCH_MAX_BYTES * 2, "the dump must exceed the byte bound");
let mut splitter =
StreamingSqlFileSplitter::new(Some(DatabaseType::Mysql), SqlParsingOptions::mysql_compatible());
StreamingSqlFileSplitter::new(Some(DatabaseType::Mysql), SqlParsingOptions::mysql_compatible(), false);
let mut pending: Vec<SqlStatementWithControl> = Vec::new();
let mut batches: Vec<(usize, usize)> = Vec::new();
let mut largest_statement = 0usize;
@@ -2851,7 +3011,7 @@ mod tests {
SqlFileTable { database: Some("second".to_string()), name: "b".to_string() },
];
let mut splitter =
StreamingSqlFileSplitter::new(Some(DatabaseType::Mysql), SqlParsingOptions::mysql_compatible());
StreamingSqlFileSplitter::new(Some(DatabaseType::Mysql), SqlParsingOptions::mysql_compatible(), false);
let mut filter = TableRestoreFilter::default();
let mut statements = Vec::new();
for chunk in ["INSERT INTO a VALUES (1); USE second;", "INSERT INTO b VALUES (2); INSERT INTO a VALUES (3);"] {
@@ -3195,18 +3355,135 @@ mod tests {
assert_eq!(statements[1], "SELECT 3;");
}
#[test]
fn ordinary_multifile_import_does_not_share_splitter_state() {
let mut first_file =
StreamingSqlFileSplitter::new(Some(DatabaseType::SqlServer), SqlParsingOptions::default(), false);
assert!(first_file.push_chunk("INSERT INTO dbo.items VALUES (N'partial").is_empty());
assert_eq!(
first_file.finish().into_iter().map(|statement| statement.sql).collect::<Vec<_>>(),
vec!["INSERT INTO dbo.items VALUES (N'partial"]
);
let mut second_file =
StreamingSqlFileSplitter::new(Some(DatabaseType::SqlServer), SqlParsingOptions::default(), false);
assert!(second_file.push_chunk("SELECT 2;\n").is_empty());
assert_eq!(
second_file.finish().into_iter().map(|statement| statement.sql).collect::<Vec<_>>(),
vec!["SELECT 2;"]
);
}
#[test]
fn sqlserver_zip_insert_cap_does_not_change_postgres_default_merge() {
let statements =
vec!["INSERT INTO items (id) VALUES (1);".to_string(), "INSERT INTO items (id) VALUES (2);".to_string()];
assert_eq!(sql_file_import_max_insert_batch_statements(true, Some(DatabaseType::Postgres)), None);
assert_eq!(sql_file_import_max_insert_batch_statements(false, Some(DatabaseType::SqlServer)), None);
assert_eq!(
sql_file_import_max_insert_batch_statements(true, Some(DatabaseType::SqlServer)),
Some(SQLSERVER_ZIP_INSERT_BATCH_MAX_STATEMENTS)
);
let postgres =
optimize_sql_file_import_statements_for_batch(&statements, Some(DatabaseType::Postgres), None, None);
let sqlserver_zip = optimize_sql_file_import_statements_for_batch(
&statements,
Some(DatabaseType::SqlServer),
None,
sql_file_import_max_insert_batch_statements(true, Some(DatabaseType::SqlServer)),
);
assert_eq!(postgres.len(), 1);
assert_eq!(postgres[0].source_statement_count, 2);
assert_eq!(sqlserver_zip.len(), 1);
assert_eq!(sqlserver_zip[0].source_statement_count, 2);
}
#[test]
fn sqlserver_zip_insert_cap_still_bounds_very_large_merges() {
let statements: Vec<String> = (0..(SQLSERVER_ZIP_INSERT_BATCH_MAX_STATEMENTS * 2))
.map(|id| format!("INSERT INTO items (id) VALUES ({id});"))
.collect();
let sqlserver_zip = optimize_sql_file_import_statements_for_batch(
&statements,
Some(DatabaseType::SqlServer),
None,
sql_file_import_max_insert_batch_statements(true, Some(DatabaseType::SqlServer)),
);
assert_eq!(sqlserver_zip.len(), 2);
assert!(sqlserver_zip
.iter()
.all(|statement| statement.source_statement_count == SQLSERVER_ZIP_INSERT_BATCH_MAX_STATEMENTS));
}
#[test]
fn streaming_sqlserver_splitter_emits_semicolon_terminated_inserts_without_go() {
let mut splitter =
StreamingSqlFileSplitter::new(Some(DatabaseType::SqlServer), SqlParsingOptions::default(), true);
let statements = splitter
.push_chunk("INSERT INTO dbo.items VALUES (1);\nINSERT INTO dbo.items VALUES (2);\n")
.into_iter()
.map(|statement| statement.sql)
.collect::<Vec<_>>();
assert_eq!(statements, vec!["INSERT INTO dbo.items VALUES (1);", "INSERT INTO dbo.items VALUES (2);"]);
assert!(splitter.finish().is_empty());
}
#[test]
fn streaming_sqlserver_splitter_ignores_semicolons_inside_literals_and_comments() {
let mut splitter =
StreamingSqlFileSplitter::new(Some(DatabaseType::SqlServer), SqlParsingOptions::default(), true);
let statements = splitter
.push_chunk("INSERT INTO dbo.items VALUES (N'one; two', [a;]); -- trailing ;\n/* block ; comment */ INSERT INTO dbo.items VALUES (2);\n")
.into_iter()
.map(|statement| statement.sql)
.collect::<Vec<_>>();
assert_eq!(
statements,
vec![
"INSERT INTO dbo.items VALUES (N'one; two', [a;]);",
"-- trailing ;\n/* block ; comment */ INSERT INTO dbo.items VALUES (2);"
]
);
assert!(splitter.finish().is_empty());
}
#[test]
fn streaming_sqlserver_splitter_keeps_split_json_literal_across_parts() {
let part_one = "INSERT INTO [dbo].[VersionValue] ([value]) VALUES (N'{\"materialOrSymbolMate";
let part_two = "rialId\":\"9625a891-3682-4151-acdb-6480db860033\",\"label\":\"B\\\\u1ebb ch\\\\u00e2n/Th\\\\u1eb3ng/\",\"quote\":\"it''s valid\"}');\n";
let expected = format!("{part_one}{part_two}").trim().to_string();
let mut splitter = StreamingSqlFileSplitter::new(Some(DatabaseType::SqlServer), SqlParsingOptions::default());
let mut splitter =
StreamingSqlFileSplitter::new(Some(DatabaseType::SqlServer), SqlParsingOptions::default(), true);
assert!(splitter.push_chunk(part_one).is_empty());
assert!(splitter.push_chunk(part_two).is_empty());
let statements = splitter.finish().into_iter().map(|statement| statement.sql).collect::<Vec<_>>();
let statements = splitter.push_chunk(part_two).into_iter().map(|statement| statement.sql).collect::<Vec<_>>();
assert_eq!(statements, vec![expected]);
assert!(splitter.finish().is_empty());
}
#[test]
fn streaming_sqlserver_splitter_keeps_leading_comment_module_semicolons_until_go() {
let mut splitter =
StreamingSqlFileSplitter::new(Some(DatabaseType::SqlServer), SqlParsingOptions::default(), true);
assert!(splitter
.push_chunk("/* setup */\nCREATE OR ALTER PROCEDURE dbo.demo AS\nBEGIN\n SELECT 1;\n SELECT 2;\nEND\n")
.is_empty());
let statements = splitter.push_chunk("GO\n").into_iter().map(|statement| statement.sql).collect::<Vec<_>>();
assert_eq!(
statements,
vec!["/* setup */\nCREATE OR ALTER PROCEDURE dbo.demo AS\nBEGIN\n SELECT 1;\n SELECT 2;\nEND"]
);
assert!(splitter.finish().is_empty());
}
#[test]
@@ -3219,6 +3496,59 @@ mod tests {
assert_eq!(batches, vec!["CREATE PROCEDURE dbo.demo AS\nBEGIN\nSELECT 1;\nEND", "SELECT 2;"]);
}
#[test]
fn ordinary_sqlserver_import_keeps_declared_variable_batch_scope() {
let mut splitter =
StreamingSqlFileSplitter::new(Some(DatabaseType::SqlServer), SqlParsingOptions::default(), false);
let statements = splitter
.push_chunk("DECLARE @name nvarchar(50);\nSET @name = 'x';\nSELECT * FROM users WHERE name = @name;\nGO\n")
.into_iter()
.map(|statement| statement.sql)
.collect::<Vec<_>>();
assert_eq!(
statements,
vec!["DECLARE @name nvarchar(50);\nSET @name = 'x';\nSELECT * FROM users WHERE name = @name;"]
);
assert!(splitter.finish().is_empty());
}
#[test]
fn ordinary_sqlserver_import_keeps_control_flow_batch_together() {
let mut splitter =
StreamingSqlFileSplitter::new(Some(DatabaseType::SqlServer), SqlParsingOptions::default(), false);
let statements = splitter
.push_chunk(
"IF EXISTS (SELECT 1 FROM sys.tables WHERE name = 'demo')\nBEGIN\n SELECT 1;\n SELECT 2;\nEND\nELSE\nBEGIN\n SELECT 3;\nEND\nGO\n",
)
.into_iter()
.map(|statement| statement.sql)
.collect::<Vec<_>>();
assert_eq!(
statements,
vec!["IF EXISTS (SELECT 1 FROM sys.tables WHERE name = 'demo')\nBEGIN\n SELECT 1;\n SELECT 2;\nEND\nELSE\nBEGIN\n SELECT 3;\nEND"]
);
assert!(splitter.finish().is_empty());
}
#[test]
fn zip_sqlserver_import_still_keeps_module_body_whole_with_comment_before_keyword() {
let mut splitter =
StreamingSqlFileSplitter::new(Some(DatabaseType::SqlServer), SqlParsingOptions::default(), true);
let statements = splitter
.push_chunk("CREATE /*c*/ PROCEDURE dbo.demo AS\nBEGIN\n SELECT 1;\n SELECT 2;\nEND\nGO\n")
.into_iter()
.map(|statement| statement.sql)
.collect::<Vec<_>>();
assert_eq!(statements, vec!["CREATE /*c*/ PROCEDURE dbo.demo AS\nBEGIN\n SELECT 1;\n SELECT 2;\nEND"]);
assert!(splitter.finish().is_empty());
}
#[tokio::test]
async fn streaming_decoder_detects_gbk_after_ascii_prefix() {
let (encoded, _, _) = encoding_rs::GBK.encode("-- Navicat dump\nINSERT INTO t VALUES ('中文');");
+1 -1
View File
@@ -2209,7 +2209,7 @@ impl SqlImportRowStream {
.await?;
Ok(Self {
decoder,
splitter: Some(StreamingSqlFileSplitter::new(options.sql_dialect, parsing_options)),
splitter: Some(StreamingSqlFileSplitter::new(options.sql_dialect, parsing_options, false)),
family,
target: None,
rows: Vec::new(),
@@ -15,7 +15,7 @@ use dbx_core::connection::AppState;
use dbx_core::database_export::{export_database_sql_core, DatabaseExportOutputCompression, DatabaseExportRequest};
use dbx_core::models::connection::DatabaseType;
use dbx_core::sql::SqlFileRequest;
use dbx_core::sql_file_import::execute_sql_file_paths;
use dbx_core::sql_file_import::execute_sql_file_zip_package_paths;
use std::sync::Arc;
use std::time::Instant;
use tokio_util::sync::CancellationToken;
@@ -225,7 +225,12 @@ async fn manual_e2e_sqlserver_default_split_zip_export_then_import_skip_relation
let file_path_refs: Vec<&std::path::Path> = extracted_paths.iter().map(|path| path.as_path()).collect();
let import_progress_log = std::sync::Mutex::new(Vec::new());
let started_at = Instant::now();
let import_result = execute_sql_file_paths(
// Use the dedicated ZIP-package entry point, exactly as the desktop/web
// dispatch sites now do once `is_zip_package` is correctly derived from
// these extracted part paths (see PR #10632 review: routing previously
// fell through to `execute_sql_file_paths`, silently losing the
// persistent splitter across parts this whole flow is meant to exercise).
let import_result = execute_sql_file_zip_package_paths(
&reimport_state,
&import_request,
&file_path_refs,
@@ -8,6 +8,35 @@ pub const SQL_FILE_ZIP_MAX_MANIFEST_BYTES: u64 = 1024 * 1024;
pub const SQL_FILE_ZIP_MAX_TOTAL_BYTES: u64 = 4 * 1024 * 1024 * 1024;
pub const SQL_FILE_ZIP_MAX_PART_BYTES: u64 = 1024 * 1024 * 1024;
/// Extraction-directory prefix used by the desktop (Tauri) dispatch site,
/// under the OS temp dir: `dbx-sql-package-<uuid>`.
pub const SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_DESKTOP: &str = "dbx-sql-package-";
/// Extraction-directory prefix used by the web dispatch site, under the
/// managed upload tmp dir: `package-<uuid>`.
pub const SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_WEB: &str = "package-";
/// True if `path`'s parent directory name starts with `prefix`, i.e. `path` is
/// one of the parts `extract_sql_file_zip_package` wrote out for a `.zip`
/// upload extracted under that prefix.
///
/// Dispatch sites must derive "is this execution a ZIP package" from the
/// actual paths they are about to execute (via this helper), not from the
/// original upload's file name/extension: the frontend always sends the
/// extracted `.sql` part paths as the execution paths, never the original
/// `.zip` path, so a suffix check on the original request field never
/// matches. Each site must pass only the prefix it extracts under itself:
/// the prefixes are per transport, and `package-` is generic enough that
/// user-chosen directories can share it, so such a path must never be
/// mistaken for an extraction dir and routed through the ZIP importer or its
/// `remove_dir_all` cleanup.
pub fn is_extracted_sql_zip_package_path(path: &Path, prefix: &str) -> bool {
path.parent()
.and_then(|parent| parent.file_name())
.and_then(|name| name.to_str())
.is_some_and(|name| name.starts_with(prefix))
}
#[derive(Debug, Clone, Deserialize)]
#[serde(rename_all = "camelCase")]
struct SqlZipManifest {
@@ -135,6 +164,32 @@ mod tests {
zip.finish().unwrap();
}
#[test]
fn is_extracted_sql_zip_package_path_matches_only_the_requested_prefix() {
let desktop_part = Path::new("/tmp/dbx-sql-package-11111111-1111-1111-1111-111111111111/00001-dump.sql");
let web_part = Path::new("/var/data/tmp/sql_file/package-22222222-2222-2222-2222-222222222222/00001-dump.sql");
assert!(is_extracted_sql_zip_package_path(desktop_part, SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_DESKTOP));
assert!(!is_extracted_sql_zip_package_path(desktop_part, SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_WEB));
assert!(is_extracted_sql_zip_package_path(web_part, SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_WEB));
assert!(!is_extracted_sql_zip_package_path(web_part, SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_DESKTOP));
assert!(!is_extracted_sql_zip_package_path(
Path::new("/tmp/upload-33333333.sql"),
SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_DESKTOP
));
assert!(!is_extracted_sql_zip_package_path(
Path::new("/tmp/upload-33333333.sql"),
SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_WEB
));
assert!(!is_extracted_sql_zip_package_path(
Path::new("/var/data/tmp/sql_file/restore-44444444/backup.sql"),
SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_WEB
));
assert!(!is_extracted_sql_zip_package_path(
Path::new("standalone.sql"),
SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_DESKTOP
));
}
#[test]
fn preserves_manifest_order_when_extracting_parts() {
let dir = tempfile::tempdir().unwrap();
+53 -8
View File
@@ -1364,7 +1364,7 @@ fn starts_with_sqlserver_control_flow_batch(sql: &str) -> bool {
&& tokens.iter().any(|token| token.eq_ignore_ascii_case("END"))
}
fn starts_with_sqlserver_module_ddl(sql: &str) -> bool {
pub fn starts_with_sqlserver_module_ddl(sql: &str) -> bool {
let tokens = first_sql_tokens(sql, 4);
if tokens.len() >= 4
&& tokens[0].eq_ignore_ascii_case("CREATE")
@@ -1617,15 +1617,46 @@ fn mysql_routine_tokens(sql: &str) -> Vec<String> {
fn first_sql_tokens(sql: &str, limit: usize) -> Vec<String> {
let mut tokens = Vec::new();
for token in sql.split(|ch: char| !ch.is_ascii_alphanumeric() && ch != '_') {
if token.is_empty() {
continue;
let mut token = String::new();
let mut chars = sql.chars().peekable();
while let Some(ch) = chars.next() {
if ch == '-' && chars.peek() == Some(&'-') {
chars.next();
if !token.is_empty() {
tokens.push(std::mem::take(&mut token));
}
for comment_char in chars.by_ref() {
if comment_char == '\n' {
break;
}
}
} else if ch == '/' && chars.peek() == Some(&'*') {
chars.next();
if !token.is_empty() {
tokens.push(std::mem::take(&mut token));
}
let mut previous = None;
for comment_char in chars.by_ref() {
if previous == Some('*') && comment_char == '/' {
break;
}
previous = Some(comment_char);
}
} else if ch.is_ascii_alphanumeric() || ch == '_' {
token.push(ch);
} else if !token.is_empty() {
tokens.push(std::mem::take(&mut token));
}
tokens.push(token.to_string());
if tokens.len() >= limit {
break;
}
}
if tokens.len() < limit && !token.is_empty() {
tokens.push(token);
}
tokens
}
@@ -1799,6 +1830,20 @@ pub fn optimize_sql_file_import_statements(
statements: &[String],
db_type: Option<DatabaseType>,
driver_profile: Option<&str>,
) -> Vec<SqlFileImportStatement> {
optimize_sql_file_import_statements_with_max_insert_batch_statements(
statements,
db_type,
driver_profile,
SQL_FILE_INSERT_BATCH_MAX_STATEMENTS,
)
}
pub fn optimize_sql_file_import_statements_with_max_insert_batch_statements(
statements: &[String],
db_type: Option<DatabaseType>,
driver_profile: Option<&str>,
max_insert_batch_statements: usize,
) -> Vec<SqlFileImportStatement> {
let mut optimized = Vec::new();
let mut pending_insert: Option<PendingInsertBatch> = None;
@@ -1833,7 +1878,7 @@ pub fn optimize_sql_file_import_statements(
.flatten();
if let Some(insert) = mergeable_insert {
match pending_insert.as_mut() {
Some(batch) if batch.can_accept(&insert) => batch.push(insert),
Some(batch) if batch.can_accept(&insert, max_insert_batch_statements) => batch.push(insert),
Some(_) => {
flush_pending_insert(&mut optimized, &mut pending_insert);
pending_insert = Some(PendingInsertBatch::new(insert));
@@ -1904,9 +1949,9 @@ impl PendingInsertBatch {
}
}
fn can_accept(&self, insert: &MergeableInsert) -> bool {
fn can_accept(&self, insert: &MergeableInsert, max_insert_batch_statements: usize) -> bool {
self.prefix_key == insert.prefix_key
&& self.source_statement_count < SQL_FILE_INSERT_BATCH_MAX_STATEMENTS
&& self.source_statement_count < max_insert_batch_statements
&& self.byte_len + insert.values.len() + 3 <= SQL_FILE_INSERT_BATCH_MAX_BYTES
}
+70 -12
View File
@@ -8,8 +8,8 @@ use axum::response::sse::{Event, Sse};
use axum::Json;
use dbx_core::sql::{self, SqlFileProgress, SqlFileRequest, SqlFileStatus};
use dbx_core::sql_file_import::{
execute_sql_file_paths, sql_file_error_progress, sql_file_progress as build_sql_file_progress,
SqlFileProgressEmitter,
execute_sql_file_paths, execute_sql_file_zip_package_paths, sql_file_error_progress,
sql_file_progress as build_sql_file_progress, SqlFileProgressEmitter,
};
use futures::stream::Stream;
use serde::Deserialize;
@@ -223,7 +223,11 @@ async fn preview_uploaded_sql_file(
let tmp_dir = file_path.parent().ok_or_else(|| AppError::from("Invalid SQL file path".to_string()))?;
if file_name.to_ascii_lowercase().ends_with(".zip") {
let extraction_dir = tmp_dir.join(format!("package-{}", Uuid::new_v4()));
let extraction_dir = tmp_dir.join(format!(
"{}{}",
dbx_core::sql_file_zip_package::SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_WEB,
Uuid::new_v4()
));
let package = dbx_core::sql_file_zip_package::extract_sql_file_zip_package(file_path, &extraction_dir)
.map_err(AppError::from)?;
let paths = dbx_core::sql_file_zip_package::extracted_sql_zip_paths(&extraction_dir, &package)
@@ -272,6 +276,7 @@ pub async fn execute_sql_file(
.iter()
.map(|file_path| validated_uploaded_sql_path(&state.data_dir, file_path))
.collect::<Result<Vec<_>, _>>()?;
let is_zip_package = execution_targets_zip_package(&file_paths);
let managed_previews = state.managed_sql_previews.claim(&file_paths)?;
// Fast-fail: reject early if the connection is read-only (individual statements are also checked in do_execute)
@@ -344,10 +349,17 @@ pub async fn execute_sql_file(
let file_path_refs: Vec<&Path> = file_paths.iter().map(PathBuf::as_path).collect();
// The core executor emits exactly one terminal Error with the latest
// cumulative counters before returning Err.
let _ = execute_sql_file_paths(&app, &req, &file_path_refs, token, started_at, |progress| {
progress_emitter.emit(progress);
})
.await;
let _ = if is_zip_package {
execute_sql_file_zip_package_paths(&app, &req, &file_path_refs, token, started_at, |progress| {
progress_emitter.emit(progress);
})
.await
} else {
execute_sql_file_paths(&app, &req, &file_path_refs, token, started_at, |progress| {
progress_emitter.emit(progress);
})
.await
};
cleanup_sql_file_package_paths(&file_paths);
cleanup_sql_file_execution(&state_clone, &req.execution_id).await;
@@ -365,11 +377,11 @@ fn send_sql_file_progress(tx: &broadcast::Sender<String>, progress: SqlFileProgr
fn cleanup_sql_file_package_paths(file_paths: &[PathBuf]) {
let mut directories = std::collections::HashSet::new();
for path in file_paths {
let Some(parent) = path.parent() else {
continue;
};
if parent.file_name().and_then(|name| name.to_str()).is_some_and(|name| name.starts_with("package-")) {
directories.insert(parent.to_path_buf());
if dbx_core::sql_file_zip_package::is_extracted_sql_zip_package_path(
path,
dbx_core::sql_file_zip_package::SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_WEB,
) {
directories.insert(path.parent().expect("checked by is_extracted_sql_zip_package_path").to_path_buf());
}
}
for directory in directories {
@@ -436,6 +448,23 @@ fn cleanup_sql_file_uploads_except(tmp_dir: &Path, max_age: Duration, active: &H
}
}
/// The frontend always sends the extracted `.sql` part paths as the execution
/// paths for a ZIP upload, never the original `.zip` path (that stays only in
/// `SqlFileRequest::file_path`), so routing to the ZIP-package incremental
/// -flush importer must be derived from those actual execution paths, not
/// from `req.file_path`'s extension. Web only ever extracts under the
/// `package-` prefix, and every path reaching here has already passed
/// `validated_uploaded_sql_path`, which confines it to the managed upload
/// tmp dir.
fn execution_targets_zip_package(file_paths: &[PathBuf]) -> bool {
file_paths.iter().any(|path| {
dbx_core::sql_file_zip_package::is_extracted_sql_zip_package_path(
path,
dbx_core::sql_file_zip_package::SQL_FILE_ZIP_EXTRACTION_DIR_PREFIX_WEB,
)
})
}
fn validated_uploaded_sql_path(data_dir: &Path, file_path: &str) -> Result<PathBuf, AppError> {
let path = PathBuf::from(file_path);
if !path.is_absolute() {
@@ -633,6 +662,35 @@ mod tests {
assert!(state.sql_file_executions.read().await.contains_key("existing-job"));
}
#[test]
fn zip_package_routing_is_selected_from_extracted_part_paths_not_the_original_zip_name() {
// Regression for PR #10632 review: the frontend sends the extracted
// `.sql` part paths as execution paths, never the original `.zip`
// upload path, so routing must key off the actual paths, not
// `req.file_path`'s extension.
let zip_upload_part_paths = vec![
PathBuf::from("/data/tmp/sql_file/package-11111111-1111-1111-1111-111111111111/00001-dump.sql"),
PathBuf::from("/data/tmp/sql_file/package-11111111-1111-1111-1111-111111111111/00002-dump.sql"),
];
assert!(execution_targets_zip_package(&zip_upload_part_paths));
let ordinary_upload_paths = vec![PathBuf::from("/data/tmp/sql_file/upload-22222222.sql")];
assert!(!execution_targets_zip_package(&ordinary_upload_paths));
// A single-file (non-multi-file) request always executes exactly the
// `req.file_path` value as its only path, so this also exercises the
// no-`file_paths`-supplied call shape from `execute_sql_file`.
let single_zip_part_as_sole_path = vec![PathBuf::from("/data/tmp/sql_file/package-33333333/00001-dump.sql")];
assert!(execution_targets_zip_package(&single_zip_part_as_sole_path));
// Desktop extraction dirs (`dbx-sql-package-`, under the OS temp dir)
// are not this site's to match: web only ever extracts `package-*`
// dirs, and the generic name must not pull user directories into the
// ZIP importer or its cleanup.
let desktop_style_paths = vec![PathBuf::from("/tmp/dbx-sql-package-44444444/00001-dump.sql")];
assert!(!execution_targets_zip_package(&desktop_style_paths));
}
#[test]
fn sql_file_upload_limit_defaults_and_clamps() {
assert_eq!(sql_file_upload_max_bytes_from_value(None), SQL_FILE_UPLOAD_MAX_BYTES);