operator/statement/

ddl.rs

// Copyright 2023 Greptime Team
//
// Licensed 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.

use std::collections::{HashMap, HashSet};
use std::sync::Arc;
use std::time::Duration;

use api::helper::ColumnDataTypeWrapper;
use api::v1::alter_table_expr::Kind;
use api::v1::meta::CreateFlowTask as PbCreateFlowTask;
use api::v1::repartition::Source;
use api::v1::{
    AlterDatabaseExpr, AlterTableExpr, CreateFlowExpr, CreateTableExpr, CreateViewExpr,
    PartitionedSource, Repartition, TargetPartitionColumns, UnpartitionedSource, column_def,
};
#[cfg(feature = "enterprise")]
use api::v1::{
    CreateTriggerExpr as PbCreateTriggerExpr, meta::CreateTriggerTask as PbCreateTriggerTask,
};
use catalog::CatalogManagerRef;
use chrono::Utc;
use common_base::regex_pattern::NAME_PATTERN_REG;
use common_catalog::consts::{
    DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, is_ddl_reserved_table, is_readonly_schema,
    is_readonly_table,
};
use common_catalog::{format_full_flow_name, format_full_table_name};
use common_error::ext::BoxedError;
#[cfg(feature = "enterprise")]
use common_meta::cache_invalidator::CacheInvalidatorRef;
use common_meta::cache_invalidator::Context;
use common_meta::ddl::create_flow::{
    DEFER_ON_MISSING_SOURCE_KEY, FLOW_EXPERIMENTAL_ENABLE_INCREMENTAL_READ_KEY, FlowType,
    INTERNAL_EVAL_OFFSET_KEY, INTERNAL_EVAL_SCHEDULE_KEY,
};
use common_meta::instruction::CacheIdent;
#[cfg(feature = "enterprise")]
use common_meta::key::TableMetadataManagerRef;
use common_meta::key::schema_name::{SchemaName, SchemaNameKey};
#[cfg(feature = "enterprise")]
use common_meta::procedure_executor::ProcedureExecutorRef;
#[cfg(feature = "enterprise")]
use common_meta::rpc::ddl::trigger::CreateTriggerTask;
#[cfg(feature = "enterprise")]
use common_meta::rpc::ddl::trigger::DropTriggerTask;
use common_meta::rpc::ddl::{
    CreateFlowTask, CreatorGrantIntent, DdlTask, DropFlowTask, DropViewTask, SubmitDdlTaskRequest,
    SubmitDdlTaskResponse, TriggerReason,
};
use common_query::Output;
use common_recordbatch::{RecordBatch, RecordBatches};
use common_sql::convert::sql_value_to_value;
use common_telemetry::{debug, info, tracing, warn};
use common_time::{Timestamp, Timezone};
use datafusion_common::tree_node::TreeNodeVisitor;
use datafusion_expr::LogicalPlan;
use datafusion_expr::logical_plan::Distinct;
use datatypes::prelude::ConcreteDataType;
use datatypes::schema::{ColumnSchema, Schema};
use datatypes::value::Value;
use datatypes::vectors::{StringVector, VectorRef};
use humantime::parse_duration;
use partition::expr::{Operand, PartitionExpr, RestrictedOp};
use partition::function::IcebergHash;
use partition::multi_dim::MultiDimPartitionRule;
use query::parser::QueryStatement;
use query::plan::extract_and_rewrite_full_table_names;
use query::query_engine::DefaultSerializer;
use query::sql::create_table_stmt;
use session::context::QueryContextRef;
use session::table_name::table_idents_to_full_name;
use snafu::{OptionExt, ResultExt, ensure};
use sql::parser::{ParseOptions, ParserContext};
use sql::parsers::utils::is_tql;
use sql::statements::OptionMap;
#[cfg(feature = "enterprise")]
use sql::statements::alter::trigger::AlterTrigger;
use sql::statements::alter::{AlterDatabase, AlterTable, AlterTableOperation};
#[cfg(feature = "enterprise")]
use sql::statements::create::trigger::CreateTrigger;
use sql::statements::create::{
    CreateExternalTable, CreateFlow, CreateTable, CreateTableLike, CreateView, Partitions,
};
use sql::statements::statement::Statement;
use sqlparser::ast::{
    Expr, Function, FunctionArg, FunctionArgExpr, FunctionArguments, Ident, UnaryOperator,
    Value as ParserValue,
};
use store_api::metric_engine_consts::{
    LOGICAL_TABLE_METADATA_KEY, METRIC_ENGINE_NAME, PHYSICAL_TABLE_METADATA_KEY,
};
use substrait::{DFLogicalSubstraitConvertor, SubstraitPlan};
use table::TableRef;
use table::dist_table::DistTable;
use table::metadata::{self, TableId, TableInfo, TableMeta, TableType};
use table::requests::{
    AlterKind, AlterTableRequest, COMMENT_KEY, DDL_TIMEOUT, DDL_WAIT, INGEST_ROWS_RATE_LIMIT_KEY,
    TableOptions, validate_and_normalize_annotation_options,
};
use table::table_name::TableName;
use table::table_reference::TableReference;

use crate::error::{
    self, AlterExprToRequestSnafu, BuildDfLogicalPlanSnafu, CatalogSnafu, ColumnDataTypeSnafu,
    ColumnNotFoundSnafu, ConvertSchemaSnafu, CreateLogicalTablesSnafu,
    DeserializePartitionExprSnafu, EmptyDdlExprSnafu, ExternalSnafu, ExtractTableNamesSnafu,
    FlowNotFoundSnafu, InvalidPartitionRuleSnafu, InvalidPartitionSnafu, InvalidSqlSnafu,
    InvalidTableNameSnafu, InvalidViewNameSnafu, InvalidViewStmtSnafu, NotSupportedSnafu,
    PartitionExprToPbSnafu, Result, SchemaInUseSnafu, SchemaNotFoundSnafu, SchemaReadOnlySnafu,
    SerializePartitionExprSnafu, SubstraitCodecSnafu, TableAlreadyExistsSnafu,
    TableDdlReservedSnafu, TableMetadataManagerSnafu, TableNotFoundSnafu, TableReadOnlySnafu,
    UnrecognizedTableOptionSnafu, ViewAlreadyExistsSnafu,
};
use crate::expr_helper::{self, RepartitionRequest, RepartitionSource};
use crate::statement::StatementExecutor;
use crate::statement::show::create_partitions_stmt;
use crate::utils::{to_executor_context, to_executor_context_with_origin_frontend};

/// Options shared by SQL DDL submission paths.
#[derive(Debug, Clone, Copy)]
pub(crate) struct DdlSubmitOptions {
    wait: bool,
    timeout: Duration,
}

impl Default for DdlSubmitOptions {
    fn default() -> Self {
        Self {
            wait: SubmitDdlTaskRequest::default_wait(),
            timeout: SubmitDdlTaskRequest::default_timeout(),
        }
    }
}

#[cfg(not(feature = "enterprise"))]
const ALLOWED_FLOW_OPTIONS: [&str; 2] = [
    DEFER_ON_MISSING_SOURCE_KEY,
    FLOW_EXPERIMENTAL_ENABLE_INCREMENTAL_READ_KEY,
];

fn build_procedure_id_output(procedure_id: &str) -> Result<Output> {
    let vector: VectorRef = Arc::new(StringVector::from(vec![procedure_id]));
    let schema = Arc::new(Schema::new(vec![ColumnSchema::new(
        "Procedure ID",
        vector.data_type(),
        false,
    )]));
    let batch =
        RecordBatch::new(schema.clone(), vec![vector]).context(error::BuildRecordBatchSnafu)?;
    let batches =
        RecordBatches::try_new(schema, vec![batch]).context(error::BuildRecordBatchSnafu)?;
    Ok(Output::new_with_record_batches(batches))
}

/// Parses DDL submission options using the existing request defaults.
pub(crate) fn parse_ddl_options(options: &OptionMap) -> Result<DdlSubmitOptions> {
    let defaults = DdlSubmitOptions::default();
    let wait = match options.get(DDL_WAIT) {
        Some(value) => value.parse::<bool>().map_err(|_| {
            InvalidSqlSnafu {
                err_msg: format!("invalid DDL option '{DDL_WAIT}': '{value}'"),
            }
            .build()
        })?,
        None => defaults.wait,
    };

    let timeout = match options.get(DDL_TIMEOUT) {
        Some(value) => parse_duration(value).map_err(|err| {
            InvalidSqlSnafu {
                err_msg: format!("invalid DDL option '{DDL_TIMEOUT}': '{value}': {err}"),
            }
            .build()
        })?,
        None => defaults.timeout,
    };

    Ok(DdlSubmitOptions { wait, timeout })
}

#[cfg(not(feature = "enterprise"))]
fn supported_flow_options() -> String {
    ALLOWED_FLOW_OPTIONS.join(", ")
}

fn normalize_flow_bool_option(key: &str, value: &str) -> Result<String> {
    value
        .trim()
        .to_ascii_lowercase()
        .parse::<bool>()
        .map(|value| value.to_string())
        .map_err(|_| {
            InvalidSqlSnafu {
                err_msg: format!("invalid flow option '{key}': '{value}'"),
            }
            .build()
        })
}

fn validate_and_normalize_flow_options(
    options: HashMap<String, String>,
    eval_interval: Option<i64>,
) -> Result<HashMap<String, String>> {
    // Reject non-positive eval_interval (zero or negative).
    if let Some(secs) = eval_interval
        && secs <= 0
    {
        return InvalidSqlSnafu {
            err_msg: format!("EVAL INTERVAL must be positive, got {secs} seconds"),
        }
        .fail();
    }

    options
        .into_iter()
        .map(|(key, value)| {
            if matches!(
                key.as_str(),
                FlowType::FLOW_TYPE_KEY | INTERNAL_EVAL_OFFSET_KEY | INTERNAL_EVAL_SCHEDULE_KEY
            ) {
                return InvalidSqlSnafu {
                    err_msg: format!("flow option '{key}' is reserved for internal use"),
                }
                .fail();
            }

            let normalized_value = match key.as_str() {
                DEFER_ON_MISSING_SOURCE_KEY | FLOW_EXPERIMENTAL_ENABLE_INCREMENTAL_READ_KEY => {
                    normalize_flow_bool_option(&key, &value)?
                }
                #[cfg(feature = "enterprise")]
                _ => value,
                #[cfg(not(feature = "enterprise"))]
                _ => {
                    return InvalidSqlSnafu {
                        err_msg: format!(
                            "unknown flow option '{key}', supported options: {}",
                            supported_flow_options()
                        ),
                    }
                    .fail();
                }
            };

            Ok((key, normalized_value))
        })
        .collect()
}

/// Validates `EVAL OFFSET` semantics at the operator boundary: an offset is
/// only legal together with `EVAL INTERVAL` and must lie in
/// `[0, eval_interval)`. Never modulo-normalized.
fn validate_eval_offset(
    eval_offset_secs: Option<i64>,
    eval_interval_secs: Option<i64>,
) -> Result<()> {
    if let Some(offset_secs) = eval_offset_secs {
        let Some(eval_interval_secs) = eval_interval_secs else {
            return InvalidSqlSnafu {
                err_msg: "EVAL OFFSET requires EVAL INTERVAL to be specified".to_string(),
            }
            .fail();
        };
        if !(0..eval_interval_secs).contains(&offset_secs) {
            return InvalidSqlSnafu {
                err_msg: format!(
                    "EVAL OFFSET must be in range [0, EVAL INTERVAL), got {offset_secs} seconds with EVAL INTERVAL {eval_interval_secs} seconds"
                ),
            }
            .fail();
        }
    }
    Ok(())
}

fn determine_flow_type_for_source_state(
    flow_name: &str,
    flow_options: &HashMap<String, String>,
    has_missing_source_table: bool,
    has_instant_ttl_source_table: bool,
    force_batching: bool,
) -> Result<Option<FlowType>> {
    if has_instant_ttl_source_table && force_batching {
        // The batching scheduler cannot read instant-TTL source tables, so
        // reject this combination even when another source table is missing.
        return InvalidSqlSnafu {
            err_msg: format!(
                "flow '{}' with EVAL INTERVAL requires the batching scheduler, but source tables with ttl=instant are not supported under batching mode; use a TTL longer than the flush interval",
                flow_name
            ),
        }
        .fail();
    }

    if has_missing_source_table {
        let defer_on_missing_source = flow_options
            .get(DEFER_ON_MISSING_SOURCE_KEY)
            .is_some_and(|value| value == "true");
        ensure!(
            defer_on_missing_source,
            InvalidSqlSnafu {
                err_msg: format!(
                    "missing source tables for flow '{}'; use WITH ({DEFER_ON_MISSING_SOURCE_KEY} = true) to create a pending flow",
                    flow_name
                )
            }
        );
        info!(
            "Flow `{}` is created as a pending batching flow because source tables are missing and defer_on_missing_source=true",
            flow_name
        );
        return Ok(Some(FlowType::Batching));
    }

    Ok(None)
}

/// The stateless streaming runtime accepts only a single source scan wrapped by
/// projections and filters. Keep this check local to the operator: the flow
/// validator is intentionally private to the flow crate.
fn is_stateless_flow_plan(plan: &LogicalPlan) -> bool {
    let mut scans = 0;
    let mut supported_nodes = true;
    let result = plan.apply_with_subqueries(|node| {
        match node {
            LogicalPlan::TableScan(_) => scans += 1,
            LogicalPlan::Projection(_) | LogicalPlan::Filter(_) => {}
            LogicalPlan::Distinct(Distinct::All(_)) => {}
            _ => supported_nodes = false,
        }
        Ok(datafusion_common::tree_node::TreeNodeRecursion::Continue)
    });

    result.is_ok() && supported_nodes && scans == 1
}

const INSTANT_TTL_FLOW_QUERY_ERROR: &str = "instant-TTL flow sources support only stateless projection/filter queries over one source; non-stateless queries require a persisted source";

const STATELESS_FLOW_QUERY_ERROR: &str =
    "flow streaming supports only stateless projection/filter queries over one source";

impl StatementExecutor {
    pub fn catalog_manager(&self) -> CatalogManagerRef {
        self.catalog_manager.clone()
    }

    #[tracing::instrument(skip_all)]
    pub async fn create_table(&self, stmt: CreateTable, ctx: QueryContextRef) -> Result<TableRef> {
        let mut create_expr = self.prepare_create_table(&stmt, &ctx).await?;
        self.create_table_inner(
            &mut create_expr,
            stmt.partitions,
            ctx,
            TriggerReason::Manual,
        )
        .await
    }

    async fn prepare_create_table(
        &self,
        stmt: &CreateTable,
        ctx: &QueryContextRef,
    ) -> Result<CreateTableExpr> {
        let (catalog, schema, _table) = table_idents_to_full_name(&stmt.name, ctx)
            .map_err(BoxedError::new)
            .context(error::ExternalSnafu)?;

        let schema_options = self
            .table_metadata_manager
            .schema_manager()
            .get(SchemaNameKey {
                catalog: &catalog,
                schema: &schema,
            })
            .await
            .context(TableMetadataManagerSnafu)?
            .map(|v| v.into_inner());

        let mut create_expr = expr_helper::create_to_expr(stmt, ctx)?;
        // TTL and compaction options are handled separately; ingestion quota is database-only.
        if let Some(schema_options) = schema_options {
            for (key, value) in schema_options.extra_options.iter() {
                if key == INGEST_ROWS_RATE_LIMIT_KEY || key.starts_with("compaction.") {
                    continue;
                }
                create_expr
                    .table_options
                    .entry(key.clone())
                    .or_insert(value.clone());
            }
        }

        Ok(create_expr)
    }

    /// Prepares every explicit logical CREATE before submitting one batch procedure.
    pub async fn batch_create_logical_tables(
        &self,
        statements: Vec<CreateTable>,
        ctx: QueryContextRef,
    ) -> Result<()> {
        let mut expressions = Vec::with_capacity(statements.len());
        let mut targets = std::collections::HashSet::new();
        let mut group = None;
        for statement in statements {
            ensure!(
                statement.engine == METRIC_ENGINE_NAME
                    && statement.options.get(LOGICAL_TABLE_METADATA_KEY).is_some()
                    && statement
                        .options
                        .value(PHYSICAL_TABLE_METADATA_KEY)
                        .is_none(),
                InvalidSqlSnafu {
                    err_msg: "batch requires explicit Metric logical CREATE TABLE statements"
                }
            );
            let expr = self.prepare_create_table(&statement, &ctx).await?;
            let key = (
                expr.catalog_name.clone(),
                expr.schema_name.clone(),
                expr.table_options.get(LOGICAL_TABLE_METADATA_KEY).cloned(),
            );
            ensure!(
                group.as_ref().is_none_or(|group| group == &key),
                InvalidSqlSnafu {
                    err_msg: "batch must share catalog, schema and literal on_physical_table"
                }
            );
            group = Some(key);
            ensure!(
                targets.insert(expr.table_name.clone()),
                InvalidSqlSnafu {
                    err_msg: "duplicate target in logical-table batch"
                }
            );
            self.validate_logical_create(&expr, statement.partitions.as_ref(), &ctx)
                .await?;
            expressions.push(expr);
        }
        self.create_logical_tables(&expressions, ctx, TriggerReason::Manual)
            .await?;
        Ok(())
    }

    async fn validate_logical_create(
        &self,
        expr: &CreateTableExpr,
        partitions: Option<&Partitions>,
        ctx: &QueryContextRef,
    ) -> Result<()> {
        ensure_table_definition_writable(&expr.schema_name, &expr.table_name)?;
        if let Some(partitions) = partitions
            && !partitions.exprs.is_empty()
        {
            self.validate_logical_table_partition_rule(expr, partitions, ctx)
                .await?;
        }
        Ok(())
    }

    #[tracing::instrument(skip_all)]
    pub async fn create_table_like(
        &self,
        stmt: CreateTableLike,
        ctx: QueryContextRef,
    ) -> Result<TableRef> {
        let (catalog, schema, table) = table_idents_to_full_name(&stmt.source_name, &ctx)
            .map_err(BoxedError::new)
            .context(error::ExternalSnafu)?;
        let table_ref = self
            .catalog_manager
            .table(&catalog, &schema, &table, Some(&ctx))
            .await
            .context(CatalogSnafu)?
            .context(TableNotFoundSnafu { table_name: &table })?;
        let partition_info = self
            .partition_manager
            .find_physical_partition_info(table_ref.table_info().table_id())
            .await
            .context(error::FindTablePartitionRuleSnafu { table_name: table })?;

        // CREATE TABLE LIKE also inherits database level options.
        let schema_options = self
            .table_metadata_manager
            .schema_manager()
            .get(SchemaNameKey {
                catalog: &catalog,
                schema: &schema,
            })
            .await
            .context(TableMetadataManagerSnafu)?
            .map(|v| v.into_inner());

        let quote_style = ctx.quote_style();
        let mut create_stmt =
            create_table_stmt(&table_ref.table_info(), schema_options, quote_style)
                .context(error::ParseQuerySnafu)?;
        create_stmt.name = stmt.table_name;
        create_stmt.if_not_exists = false;

        let table_info = table_ref.table_info();
        let partitions = create_partitions_stmt(&table_info, &partition_info.partitions)?.and_then(
            |mut partitions| {
                if !partitions.column_list.is_empty() {
                    partitions.set_quote(quote_style);
                    Some(partitions)
                } else {
                    None
                }
            },
        );

        let create_expr = &mut expr_helper::create_to_expr(&create_stmt, &ctx)?;
        self.create_table_inner(create_expr, partitions, ctx, TriggerReason::Manual)
            .await
    }

    #[tracing::instrument(skip_all)]
    pub async fn create_external_table(
        &self,
        create_expr: CreateExternalTable,
        ctx: QueryContextRef,
    ) -> Result<TableRef> {
        let create_expr =
            &mut expr_helper::create_external_expr(create_expr, &ctx, &self.local_file_access)
                .await?;
        self.create_table_inner(create_expr, None, ctx, TriggerReason::Manual)
            .await
    }

    /// Creates the declared-edge table with its canonical schema, entering
    /// below the user-DDL guard ([`ensure_table_definition_writable`]).
    /// `create_if_not_exists` makes concurrent first inserts race safely.
    pub async fn create_declared_relationships_table(
        &self,
        catalog: &str,
        query_ctx: QueryContextRef,
    ) -> Result<TableRef> {
        let mut expr = super::semantic_graph::build_declared_relationships_expr(catalog);
        self.create_non_logic_table(&mut expr, None, query_ctx, TriggerReason::AutoCreate)
            .await
    }

    #[tracing::instrument(skip_all)]
    pub async fn create_table_inner(
        &self,
        create_table: &mut CreateTableExpr,
        partitions: Option<Partitions>,
        query_ctx: QueryContextRef,
        trigger_reason: TriggerReason,
    ) -> Result<TableRef> {
        ensure_table_definition_writable(&create_table.schema_name, &create_table.table_name)?;

        if create_table.engine == METRIC_ENGINE_NAME
            && create_table
                .table_options
                .contains_key(LOGICAL_TABLE_METADATA_KEY)
        {
            self.validate_logical_create(create_table, partitions.as_ref(), &query_ctx)
                .await?;
            // Create logical tables
            self.create_logical_tables(
                std::slice::from_ref(create_table),
                query_ctx,
                trigger_reason,
            )
            .await?
            .into_iter()
            .next()
            .context(error::UnexpectedSnafu {
                violated: "expected to create logical tables",
            })
        } else {
            // Create other normal table
            self.create_non_logic_table(create_table, partitions, query_ctx, trigger_reason)
                .await
        }
    }

    #[tracing::instrument(skip_all)]
    pub async fn create_non_logic_table(
        &self,
        create_table: &mut CreateTableExpr,
        partitions: Option<Partitions>,
        query_ctx: QueryContextRef,
        trigger_reason: TriggerReason,
    ) -> Result<TableRef> {
        let _timer = crate::metrics::DIST_CREATE_TABLE.start_timer();

        // Check if schema exists
        let schema = self
            .table_metadata_manager
            .schema_manager()
            .get(SchemaNameKey::new(
                &create_table.catalog_name,
                &create_table.schema_name,
            ))
            .await
            .context(TableMetadataManagerSnafu)?;
        ensure!(
            schema.is_some(),
            SchemaNotFoundSnafu {
                schema_info: &create_table.schema_name,
            }
        );

        // if table exists.
        if let Some(table) = self
            .catalog_manager
            .table(
                &create_table.catalog_name,
                &create_table.schema_name,
                &create_table.table_name,
                Some(&query_ctx),
            )
            .await
            .context(CatalogSnafu)?
        {
            return if create_table.create_if_not_exists {
                Ok(table)
            } else {
                TableAlreadyExistsSnafu {
                    table: format_full_table_name(
                        &create_table.catalog_name,
                        &create_table.schema_name,
                        &create_table.table_name,
                    ),
                }
                .fail()
            };
        }

        ensure!(
            NAME_PATTERN_REG.is_match(&create_table.table_name),
            InvalidTableNameSnafu {
                table_name: &create_table.table_name,
            }
        );

        let table_name = TableName::new(
            &create_table.catalog_name,
            &create_table.schema_name,
            &create_table.table_name,
        );

        let (partitions, partition_cols) = parse_partitions(create_table, partitions, &query_ctx)?;
        let mut table_info = create_table_info(create_table, partition_cols)?;

        let resp = self
            .create_table_procedure(
                create_table.clone(),
                partitions,
                table_info.clone(),
                query_ctx,
                trigger_reason,
            )
            .await?;

        let table_id = resp
            .table_ids
            .into_iter()
            .next()
            .context(error::UnexpectedSnafu {
                violated: "expected table_id",
            })?;
        info!("Successfully created table '{table_name}' with table id {table_id}");

        table_info.ident.table_id = table_id;

        let table_info = Arc::new(table_info);
        create_table.table_id = Some(api::v1::TableId { id: table_id });

        let table = DistTable::table(table_info);

        Ok(table)
    }

    #[tracing::instrument(skip_all)]
    pub async fn create_logical_tables(
        &self,
        create_table_exprs: &[CreateTableExpr],
        query_context: QueryContextRef,
        trigger_reason: TriggerReason,
    ) -> Result<Vec<TableRef>> {
        let _timer = crate::metrics::DIST_CREATE_TABLES.start_timer();
        ensure!(
            !create_table_exprs.is_empty(),
            EmptyDdlExprSnafu {
                name: "create logic tables"
            }
        );

        // Check table names
        for create_table in create_table_exprs {
            ensure!(
                NAME_PATTERN_REG.is_match(&create_table.table_name),
                InvalidTableNameSnafu {
                    table_name: &create_table.table_name,
                }
            );
        }

        let raw_tables_info = create_table_exprs
            .iter()
            .map(|create| create_table_info(create, vec![]))
            .collect::<Result<Vec<_>>>()?;
        let tables_data = create_table_exprs
            .iter()
            .cloned()
            .zip(raw_tables_info.iter().cloned())
            .collect::<Vec<_>>();

        let resp = self
            .create_logical_tables_procedure(tables_data, query_context.clone(), trigger_reason)
            .await?;

        let table_ids = resp.table_ids;
        ensure!(
            table_ids.len() == raw_tables_info.len(),
            CreateLogicalTablesSnafu {
                reason: format!(
                    "The number of tables is inconsistent with the expected number to be created, expected: {}, actual: {}",
                    raw_tables_info.len(),
                    table_ids.len()
                )
            }
        );
        info!("Successfully created logical tables: {:?}", table_ids);

        // Reacquire table infos from catalog so logical tables inherit the latest partition
        // metadata (e.g. partition_key_indices) from their physical tables.
        // And the returned table info also included extra partition columns that are in physical table but not in logical table's create table expr
        let mut tables_info = Vec::with_capacity(table_ids.len());
        for (table_id, create_table) in table_ids.iter().zip(create_table_exprs.iter()) {
            let table = self
                .catalog_manager
                .table(
                    &create_table.catalog_name,
                    &create_table.schema_name,
                    &create_table.table_name,
                    Some(&query_context),
                )
                .await
                .context(CatalogSnafu)?
                .with_context(|| TableNotFoundSnafu {
                    table_name: format_full_table_name(
                        &create_table.catalog_name,
                        &create_table.schema_name,
                        &create_table.table_name,
                    ),
                })?;

            let table_info = table.table_info();
            // Safety check: ensure we are returning the table info that matches the newly created table id.
            ensure!(
                table_info.table_id() == *table_id,
                CreateLogicalTablesSnafu {
                    reason: format!(
                        "Table id mismatch after creation, expected {}, got {} for table {}",
                        table_id,
                        table_info.table_id(),
                        format_full_table_name(
                            &create_table.catalog_name,
                            &create_table.schema_name,
                            &create_table.table_name
                        )
                    )
                }
            );

            tables_info.push(table_info);
        }

        Ok(tables_info.into_iter().map(DistTable::table).collect())
    }

    async fn validate_logical_table_partition_rule(
        &self,
        create_table: &CreateTableExpr,
        partitions: &Partitions,
        query_ctx: &QueryContextRef,
    ) -> Result<()> {
        let (_, mut logical_partition_exprs) =
            parse_partitions_for_logical_validation(create_table, partitions, query_ctx)?;

        let physical_table_name = create_table
            .table_options
            .get(LOGICAL_TABLE_METADATA_KEY)
            .with_context(|| CreateLogicalTablesSnafu {
                reason: format!(
                    "expect `{LOGICAL_TABLE_METADATA_KEY}` option on creating logical table"
                ),
            })?;

        let physical_table = self
            .catalog_manager
            .table(
                &create_table.catalog_name,
                &create_table.schema_name,
                physical_table_name,
                Some(query_ctx),
            )
            .await
            .context(CatalogSnafu)?
            .context(TableNotFoundSnafu {
                table_name: physical_table_name.clone(),
            })?;

        let physical_table_info = physical_table.table_info();
        let (partition_rule, _) = self
            .partition_manager
            .find_table_partition_rule(&physical_table_info)
            .await
            .context(error::FindTablePartitionRuleSnafu {
                table_name: physical_table_name.clone(),
            })?;

        let multi_dim_rule = partition_rule
            .as_ref()
            .as_any()
            .downcast_ref::<MultiDimPartitionRule>()
            .context(InvalidPartitionRuleSnafu {
                reason: "physical table partition rule is not range-based",
            })?;

        // TODO(ruihang): project physical partition exprs to logical partition column
        let mut physical_partition_exprs = multi_dim_rule.exprs().to_vec();
        logical_partition_exprs.sort_unstable();
        physical_partition_exprs.sort_unstable();

        ensure!(
            physical_partition_exprs == logical_partition_exprs,
            InvalidPartitionRuleSnafu {
                reason: format!(
                    "logical table partition rule must match the corresponding physical table's\n logical table partition exprs:\t\t {:?}\n physical table partition exprs:\t {:?}",
                    logical_partition_exprs, physical_partition_exprs
                ),
            }
        );

        Ok(())
    }

    #[cfg(feature = "enterprise")]
    #[tracing::instrument(skip_all)]
    pub async fn create_trigger(
        &self,
        stmt: CreateTrigger,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        let expr = expr_helper::to_create_trigger_task_expr(stmt, &query_context)?;
        self.create_trigger_inner(expr, query_context).await
    }

    #[cfg(feature = "enterprise")]
    pub async fn create_trigger_inner(
        &self,
        expr: PbCreateTriggerExpr,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        self.create_trigger_procedure(expr, query_context).await?;
        Ok(Output::new_with_affected_rows(0))
    }

    #[cfg(feature = "enterprise")]
    async fn create_trigger_procedure(
        &self,
        expr: PbCreateTriggerExpr,
        query_context: QueryContextRef,
    ) -> Result<SubmitDdlTaskResponse> {
        let task = CreateTriggerTask::try_from(PbCreateTriggerTask {
            create_trigger: Some(expr),
        })
        .context(error::InvalidExprSnafu)?;

        let executor_context = to_executor_context(query_context, TriggerReason::Manual);
        let request = SubmitDdlTaskRequest::new(DdlTask::new_create_trigger(task));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    #[tracing::instrument(skip_all)]
    pub async fn create_flow(
        &self,
        stmt: CreateFlow,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        // TODO(ruihang): do some verification
        let eval_offset_secs = stmt.eval_offset;
        let expr = expr_helper::to_create_flow_task_expr(stmt, &query_context)?;

        // The typed `EVAL OFFSET` comes straight from the parsed SQL AST; it is
        // the only trusted source of the offset at this boundary.
        self.create_flow_procedure(expr, eval_offset_secs, query_context)
            .await?;
        Ok(Output::new_with_affected_rows(0))
    }

    /// Direct gRPC entry point for `CREATE FLOW`.
    pub async fn create_flow_inner(
        &self,
        expr: CreateFlowExpr,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        self.create_flow_procedure(expr, None, query_context)
            .await?;
        Ok(Output::new_with_affected_rows(0))
    }

    async fn create_flow_procedure(
        &self,
        mut expr: CreateFlowExpr,
        eval_offset_secs: Option<i64>,
        query_context: QueryContextRef,
    ) -> Result<SubmitDdlTaskResponse> {
        let eval_interval_secs = expr.eval_interval.as_ref().map(|e| e.seconds);

        // Reject non-positive eval_interval (zero or negative).
        if let Some(secs) = eval_interval_secs
            && secs <= 0
        {
            return InvalidSqlSnafu {
                err_msg: format!("EVAL INTERVAL must be positive, got {secs} seconds"),
            }
            .fail();
        }

        // Validate EVAL OFFSET semantics at the operator boundary (defense in
        // depth; the parser already enforces them deterministically). The
        // offset arrives as a typed parameter, never from `flow_options`.
        validate_eval_offset(eval_offset_secs, eval_interval_secs)?;

        // Validate user options. `validate_and_normalize_flow_options` also
        // rejects the internal transport keys, so a spoofed key smuggled into
        // the expr can never be honored here.
        expr.flow_options =
            validate_and_normalize_flow_options(expr.flow_options, eval_interval_secs)?;

        let flow_type = self
            .determine_flow_type(&expr, query_context.clone())
            .await?;
        info!("determined flow={} type: {:#?}", expr.flow_name, flow_type);

        expr.flow_options
            .insert(FlowType::FLOW_TYPE_KEY.to_string(), flow_type.to_string());

        let mut task = CreateFlowTask::try_from(PbCreateFlowTask {
            create_flow: Some(expr),
        })
        .context(error::InvalidExprSnafu)?;
        task.eval_offset_secs = eval_offset_secs;
        let executor_context = to_executor_context(query_context, TriggerReason::Manual);
        let request = SubmitDdlTaskRequest::new(DdlTask::new_create_flow(task));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    /// Determines the flow type from source-table state, schedule requirements,
    /// and SQL shape.
    ///
    /// Aggregates and persisted-source DISTINCT use batching. Plain DISTINCT is request-local only
    /// for instant-TTL sources, where it retains the legacy streaming compatibility behavior.
    async fn determine_flow_type(
        &self,
        expr: &CreateFlowExpr,
        query_ctx: QueryContextRef,
    ) -> Result<FlowType> {
        let has_eval_interval = expr.eval_interval.is_some();

        let mut has_missing_source_table = false;
        let mut has_instant_ttl_source_table = false;

        for src_table_name in &expr.source_table_names {
            let table = self
                .catalog_manager()
                .table(
                    &src_table_name.catalog_name,
                    &src_table_name.schema_name,
                    &src_table_name.table_name,
                    Some(&query_ctx),
                )
                .await
                .map_err(BoxedError::new)
                .context(ExternalSnafu)?;

            let Some(table) = table else {
                has_missing_source_table = true;
                continue;
            };

            if table.table_info().meta.options.ttl == Some(common_time::TimeToLive::Instant) {
                warn!(
                    "Source table `{}` for flow `{}`'s ttl=instant, {}",
                    format_full_table_name(
                        &src_table_name.catalog_name,
                        &src_table_name.schema_name,
                        &src_table_name.table_name
                    ),
                    expr.flow_name,
                    if has_eval_interval {
                        "rejecting flow because EVAL INTERVAL requires the batching scheduler which does not support instant TTL"
                    } else {
                        "fallback to streaming mode"
                    }
                );
                has_instant_ttl_source_table = true;
            }
        }

        if let Some(flow_type) = determine_flow_type_for_source_state(
            &expr.flow_name,
            &expr.flow_options,
            has_missing_source_table,
            has_instant_ttl_source_table,
            has_eval_interval,
        )? {
            return Ok(flow_type);
        }

        // A flow with `EVAL INTERVAL` (and therefore possibly `EVAL OFFSET`)
        // always uses the batching scheduler: the fixed epoch-phase schedule is
        // only honored by the batching scheduler, regardless of the SQL shape.
        // Non-aggregate SQL is supported as an explicit full-query flow (the
        // batching engine requires a time-window expression or EVAL INTERVAL).
        if has_eval_interval {
            return Ok(FlowType::Batching);
        }

        let engine = &self.query_engine;
        let stmts = ParserContext::create_with_dialect(
            &expr.sql,
            query_ctx.sql_dialect(),
            ParseOptions::default(),
        )
        .map_err(BoxedError::new)
        .context(ExternalSnafu)?;

        ensure!(
            stmts.len() == 1,
            InvalidSqlSnafu {
                err_msg: format!("Expect only one statement, found {}", stmts.len())
            }
        );
        let stmt = &stmts[0];

        let is_tql_query = is_tql(query_ctx.sql_dialect(), &expr.sql)
            .map_err(BoxedError::new)
            .context(ExternalSnafu)?;
        if is_tql_query {
            if has_instant_ttl_source_table {
                return InvalidSqlSnafu {
                    err_msg: INSTANT_TTL_FLOW_QUERY_ERROR.to_string(),
                }
                .fail();
            }
            return Ok(FlowType::Batching);
        }

        // Plan before selecting a mode. In particular, instant-TTL sources
        // must not bypass validation of the stateless streaming subset.
        let plan = match stmt {
            Statement::Tql(_) => {
                if has_instant_ttl_source_table {
                    return InvalidSqlSnafu {
                        err_msg: INSTANT_TTL_FLOW_QUERY_ERROR.to_string(),
                    }
                    .fail();
                }
                return Ok(FlowType::Batching);
            }
            _ => engine
                .planner()
                .plan(&QueryStatement::Sql(stmt.clone()), query_ctx)
                .await
                .map_err(BoxedError::new)
                .context(ExternalSnafu)?,
        };

        /// Visitor to classify aggregation and DISTINCT separately. Plain DISTINCT is request
        /// local only for instant-TTL sources; persisted-source DISTINCT keeps batching semantics.
        struct FindQueryShape {
            is_aggr: bool,
            is_distinct: bool,
            has_distinct_on: bool,
        }

        impl TreeNodeVisitor<'_> for FindQueryShape {
            type Node = LogicalPlan;
            fn f_down(
                &mut self,
                node: &Self::Node,
            ) -> datafusion_common::Result<datafusion_common::tree_node::TreeNodeRecursion>
            {
                match node {
                    LogicalPlan::Aggregate(_) => self.is_aggr = true,
                    LogicalPlan::Distinct(Distinct::All(_)) => self.is_distinct = true,
                    LogicalPlan::Distinct(Distinct::On(_)) => self.has_distinct_on = true,
                    _ => (),
                }
                Ok(datafusion_common::tree_node::TreeNodeRecursion::Continue)
            }
        }

        let mut query_shape = FindQueryShape {
            is_aggr: false,
            is_distinct: false,
            has_distinct_on: false,
        };

        plan.visit_with_subqueries(&mut query_shape)
            .context(BuildDfLogicalPlanSnafu)?;
        if query_shape.has_distinct_on {
            return InvalidSqlSnafu {
                err_msg: if has_instant_ttl_source_table {
                    INSTANT_TTL_FLOW_QUERY_ERROR.to_string()
                } else {
                    STATELESS_FLOW_QUERY_ERROR.to_string()
                },
            }
            .fail();
        }
        if query_shape.is_aggr {
            return if has_instant_ttl_source_table {
                InvalidSqlSnafu {
                    err_msg: INSTANT_TTL_FLOW_QUERY_ERROR.to_string(),
                }
                .fail()
            } else {
                Ok(FlowType::Batching)
            };
        }
        if query_shape.is_distinct && !has_instant_ttl_source_table {
            return Ok(FlowType::Batching);
        }

        ensure!(
            is_stateless_flow_plan(&plan),
            InvalidSqlSnafu {
                err_msg: if has_instant_ttl_source_table {
                    INSTANT_TTL_FLOW_QUERY_ERROR.to_string()
                } else {
                    STATELESS_FLOW_QUERY_ERROR.to_string()
                },
            }
        );
        Ok(FlowType::Streaming)
    }

    #[tracing::instrument(skip_all)]
    pub async fn create_view(
        &self,
        create_view: CreateView,
        ctx: QueryContextRef,
    ) -> Result<TableRef> {
        // convert input into logical plan
        let logical_plan = match &*create_view.query {
            Statement::Query(query) => {
                self.plan(
                    &QueryStatement::Sql(Statement::Query(query.clone())),
                    ctx.clone(),
                )
                .await?
            }
            Statement::Tql(query) => self.plan_tql(query.clone(), &ctx).await?,
            _ => {
                return InvalidViewStmtSnafu {}.fail();
            }
        };
        // Save the definition for `show create view`.
        let definition = create_view.to_string();

        // Save the columns in plan, it may changed when the schemas of tables in plan
        // are altered.
        let schema: Schema = logical_plan
            .schema()
            .clone()
            .try_into()
            .context(ConvertSchemaSnafu)?;
        let plan_columns: Vec<_> = schema
            .column_schemas()
            .iter()
            .map(|c| c.name.clone())
            .collect();

        let columns: Vec<_> = create_view
            .columns
            .iter()
            .map(|ident| ident.to_string())
            .collect();

        // Validate columns
        if !columns.is_empty() {
            ensure!(
                columns.len() == plan_columns.len(),
                error::ViewColumnsMismatchSnafu {
                    view_name: create_view.name.to_string(),
                    expected: plan_columns.len(),
                    actual: columns.len(),
                }
            );
        }

        // Extract the table names from the original plan
        // and rewrite them as fully qualified names.
        let (table_names, plan) = extract_and_rewrite_full_table_names(logical_plan, ctx.clone())
            .context(ExtractTableNamesSnafu)?;

        let table_names = table_names.into_iter().map(|t| t.into()).collect();

        // TODO(dennis): we don't save the optimized plan yet,
        // because there are some serialization issue with our own defined plan node (such as `MergeScanLogicalPlan`).
        // When the issues are fixed, we can use the `optimized_plan` instead.
        // let optimized_plan = self.optimize_logical_plan(logical_plan)?.unwrap_df_plan();

        // encode logical plan
        let encoded_plan = DFLogicalSubstraitConvertor
            .encode(&plan, DefaultSerializer)
            .context(SubstraitCodecSnafu)?;

        let expr = expr_helper::to_create_view_expr(
            create_view,
            encoded_plan.to_vec(),
            table_names,
            columns,
            plan_columns,
            definition,
            ctx.clone(),
        )?;

        // TODO(dennis): validate the logical plan
        self.create_view_by_expr(expr, ctx).await
    }

    pub async fn create_view_by_expr(
        &self,
        expr: CreateViewExpr,
        ctx: QueryContextRef,
    ) -> Result<TableRef> {
        // A view could otherwise squat a reserved name and block the canonical
        // table's first-write creation.
        ensure_table_definition_writable(&expr.schema_name, &expr.view_name)?;
        ensure! {
            !(expr.create_if_not_exists & expr.or_replace),
            InvalidSqlSnafu {
                err_msg: "syntax error Create Or Replace and If Not Exist cannot be used together",
            }
        };
        let _timer = crate::metrics::DIST_CREATE_VIEW.start_timer();

        let schema_exists = self
            .table_metadata_manager
            .schema_manager()
            .exists(SchemaNameKey::new(&expr.catalog_name, &expr.schema_name))
            .await
            .context(TableMetadataManagerSnafu)?;

        ensure!(
            schema_exists,
            SchemaNotFoundSnafu {
                schema_info: &expr.schema_name,
            }
        );

        // if view or table exists.
        if let Some(table) = self
            .catalog_manager
            .table(
                &expr.catalog_name,
                &expr.schema_name,
                &expr.view_name,
                Some(&ctx),
            )
            .await
            .context(CatalogSnafu)?
        {
            let table_type = table.table_info().table_type;

            match (table_type, expr.create_if_not_exists, expr.or_replace) {
                (TableType::View, true, false) => {
                    return Ok(table);
                }
                (TableType::View, false, false) => {
                    return ViewAlreadyExistsSnafu {
                        name: format_full_table_name(
                            &expr.catalog_name,
                            &expr.schema_name,
                            &expr.view_name,
                        ),
                    }
                    .fail();
                }
                (TableType::View, _, true) => {
                    // Try to replace an exists view
                }
                _ => {
                    return TableAlreadyExistsSnafu {
                        table: format_full_table_name(
                            &expr.catalog_name,
                            &expr.schema_name,
                            &expr.view_name,
                        ),
                    }
                    .fail();
                }
            }
        }

        ensure!(
            NAME_PATTERN_REG.is_match(&expr.view_name),
            InvalidViewNameSnafu {
                name: expr.view_name.clone(),
            }
        );

        let view_name = TableName::new(&expr.catalog_name, &expr.schema_name, &expr.view_name);

        let mut view_info = TableInfo {
            ident: metadata::TableIdent {
                // The view id of distributed table is assigned by Meta, set "0" here as a placeholder.
                table_id: 0,
                version: 0,
            },
            name: expr.view_name.clone(),
            desc: None,
            catalog_name: expr.catalog_name.clone(),
            schema_name: expr.schema_name.clone(),
            // The meta doesn't make sense for views, so using a default one.
            meta: TableMeta::empty(),
            table_type: TableType::View,
        };

        let executor_context = to_executor_context(ctx, TriggerReason::Manual);
        let request = SubmitDdlTaskRequest::new(DdlTask::new_create_view(expr, view_info.clone()));

        let resp = self
            .procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)?;

        debug!(
            "Submit creating view '{view_name}' task response: {:?}",
            resp
        );

        let view_id = resp
            .table_ids
            .into_iter()
            .next()
            .context(error::UnexpectedSnafu {
                violated: "expected table_id",
            })?;
        info!("Successfully created view '{view_name}' with view id {view_id}");

        view_info.ident.table_id = view_id;

        let view_info = Arc::new(view_info);

        let table = DistTable::table(view_info);

        // Invalidates local cache ASAP.
        self.cache_invalidator
            .invalidate(
                &Context::default(),
                &[
                    CacheIdent::TableId(view_id),
                    CacheIdent::TableName(view_name.clone()),
                ],
            )
            .await
            .context(error::InvalidateTableCacheSnafu)?;

        Ok(table)
    }

    #[tracing::instrument(skip_all)]
    pub async fn drop_flow(
        &self,
        catalog_name: String,
        flow_name: String,
        drop_if_exists: bool,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        if let Some(flow) = self
            .flow_metadata_manager
            .flow_name_manager()
            .get(&catalog_name, &flow_name)
            .await
            .context(error::TableMetadataManagerSnafu)?
        {
            let flow_id = flow.flow_id();
            let task = DropFlowTask {
                catalog_name,
                flow_name,
                flow_id,
                drop_if_exists,
            };
            self.drop_flow_procedure(task, query_context).await?;

            Ok(Output::new_with_affected_rows(0))
        } else if drop_if_exists {
            Ok(Output::new_with_affected_rows(0))
        } else {
            FlowNotFoundSnafu {
                flow_name: format_full_flow_name(&catalog_name, &flow_name),
            }
            .fail()
        }
    }

    async fn drop_flow_procedure(
        &self,
        expr: DropFlowTask,
        query_context: QueryContextRef,
    ) -> Result<SubmitDdlTaskResponse> {
        let executor_context = to_executor_context(query_context, TriggerReason::Manual);
        let request = SubmitDdlTaskRequest::new(DdlTask::new_drop_flow(expr));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    #[cfg(feature = "enterprise")]
    #[tracing::instrument(skip_all)]
    pub(super) async fn drop_trigger(
        &self,
        catalog_name: String,
        trigger_name: String,
        drop_if_exists: bool,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        let task = DropTriggerTask {
            catalog_name,
            trigger_name,
            drop_if_exists,
        };
        self.drop_trigger_procedure(task, query_context).await?;
        Ok(Output::new_with_affected_rows(0))
    }

    #[cfg(feature = "enterprise")]
    async fn drop_trigger_procedure(
        &self,
        expr: DropTriggerTask,
        query_context: QueryContextRef,
    ) -> Result<SubmitDdlTaskResponse> {
        let executor_context = to_executor_context(query_context, TriggerReason::Manual);
        let request = SubmitDdlTaskRequest::new(DdlTask::new_drop_trigger(expr));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    /// Drop a view
    #[tracing::instrument(skip_all)]
    pub async fn drop_view(
        &self,
        catalog: String,
        schema: String,
        view: String,
        drop_if_exists: bool,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        let view_info = if let Some(view) = self
            .catalog_manager
            .table(&catalog, &schema, &view, None)
            .await
            .context(CatalogSnafu)?
        {
            view.table_info()
        } else if drop_if_exists {
            // DROP VIEW IF EXISTS meets view not found - ignored
            return Ok(Output::new_with_affected_rows(0));
        } else {
            return TableNotFoundSnafu {
                table_name: format_full_table_name(&catalog, &schema, &view),
            }
            .fail();
        };

        // Ensure the exists one is view, we can't drop other table types
        ensure!(
            view_info.table_type == TableType::View,
            error::InvalidViewSnafu {
                msg: "not a view",
                view_name: format_full_table_name(&catalog, &schema, &view),
            }
        );

        let view_id = view_info.table_id();
        let view_name = TableName::new(&catalog, &schema, &view);

        let task = DropViewTask {
            catalog,
            schema,
            view,
            view_id,
            drop_if_exists,
        };

        self.drop_view_procedure(task, query_context).await?;

        // Invalidates local cache ASAP.
        self.cache_invalidator
            .invalidate(
                &Context::default(),
                &[
                    CacheIdent::TableId(view_id),
                    CacheIdent::TableName(view_name),
                ],
            )
            .await
            .context(error::InvalidateTableCacheSnafu)?;

        Ok(Output::new_with_affected_rows(0))
    }

    /// Submit [DropViewTask] to procedure executor.
    async fn drop_view_procedure(
        &self,
        expr: DropViewTask,
        query_context: QueryContextRef,
    ) -> Result<SubmitDdlTaskResponse> {
        let executor_context = to_executor_context(query_context, TriggerReason::Manual);
        let request = SubmitDdlTaskRequest::new(DdlTask::new_drop_view(expr));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    #[tracing::instrument(skip_all)]
    pub async fn alter_logical_tables(
        &self,
        alter_table_exprs: Vec<AlterTableExpr>,
        query_context: QueryContextRef,
        trigger_reason: TriggerReason,
    ) -> Result<Output> {
        let _timer = crate::metrics::DIST_ALTER_TABLES.start_timer();
        ensure!(
            !alter_table_exprs.is_empty(),
            EmptyDdlExprSnafu {
                name: "alter logical tables"
            }
        );

        // group by physical table id
        let mut groups: HashMap<TableId, Vec<AlterTableExpr>> = HashMap::new();
        for expr in alter_table_exprs {
            // Get table_id from catalog_manager
            let catalog = if expr.catalog_name.is_empty() {
                query_context.current_catalog()
            } else {
                &expr.catalog_name
            };
            let schema = if expr.schema_name.is_empty() {
                query_context.current_schema()
            } else {
                expr.schema_name.clone()
            };
            let table_name = &expr.table_name;
            let table = self
                .catalog_manager
                .table(catalog, &schema, table_name, Some(&query_context))
                .await
                .context(CatalogSnafu)?
                .with_context(|| TableNotFoundSnafu {
                    table_name: format_full_table_name(catalog, &schema, table_name),
                })?;
            let table_id = table.table_info().ident.table_id;
            let physical_table_id = self
                .table_metadata_manager
                .table_route_manager()
                .get_physical_table_id(table_id)
                .await
                .context(TableMetadataManagerSnafu)?;
            groups.entry(physical_table_id).or_default().push(expr);
        }

        // Submit procedure for each physical table
        let mut handles = Vec::with_capacity(groups.len());
        for (_physical_table_id, exprs) in groups {
            let fut =
                self.alter_logical_tables_procedure(exprs, query_context.clone(), trigger_reason);
            handles.push(fut);
        }
        let _results = futures::future::try_join_all(handles).await?;

        Ok(Output::new_with_affected_rows(0))
    }

    #[tracing::instrument(skip_all)]
    pub async fn drop_table(
        &self,
        table_name: TableName,
        drop_if_exists: bool,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        // Reserved for grpc call
        self.drop_tables(&[table_name], drop_if_exists, query_context)
            .await
    }

    #[tracing::instrument(skip_all)]
    pub async fn drop_tables(
        &self,
        table_names: &[TableName],
        drop_if_exists: bool,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        let mut tables = Vec::with_capacity(table_names.len());
        for table_name in table_names {
            ensure_table_writable(&table_name.schema_name, &table_name.table_name)?;

            if let Some(table) = self
                .catalog_manager
                .table(
                    &table_name.catalog_name,
                    &table_name.schema_name,
                    &table_name.table_name,
                    Some(&query_context),
                )
                .await
                .context(CatalogSnafu)?
            {
                tables.push(table.table_info().table_id());
            } else if drop_if_exists {
                // DROP TABLE IF EXISTS meets table not found - ignored
                continue;
            } else {
                return TableNotFoundSnafu {
                    table_name: table_name.to_string(),
                }
                .fail();
            }
        }

        for (table_name, table_id) in table_names.iter().zip(tables.into_iter()) {
            self.drop_table_procedure(table_name, table_id, drop_if_exists, query_context.clone())
                .await?;

            // Invalidates local cache ASAP.
            self.cache_invalidator
                .invalidate(
                    &Context::default(),
                    &[
                        CacheIdent::TableId(table_id),
                        CacheIdent::TableName(table_name.clone()),
                    ],
                )
                .await
                .context(error::InvalidateTableCacheSnafu)?;
        }
        Ok(Output::new_with_affected_rows(0))
    }

    #[cfg(feature = "enterprise")]
    #[tracing::instrument(skip_all)]
    pub async fn undrop_table(
        &self,
        table_name: TableName,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        execute_undrop_table(
            &self.table_metadata_manager,
            &self.procedure_executor,
            &self.cache_invalidator,
            table_name,
            query_context,
        )
        .await
    }

    #[tracing::instrument(skip_all)]
    pub async fn drop_database(
        &self,
        catalog: String,
        schema: String,
        drop_if_exists: bool,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        ensure!(
            !is_readonly_schema(&schema),
            SchemaReadOnlySnafu { name: schema }
        );

        if self
            .catalog_manager
            .schema_exists(&catalog, &schema, None)
            .await
            .context(CatalogSnafu)?
        {
            if schema == query_context.current_schema() {
                SchemaInUseSnafu { name: schema }.fail()
            } else {
                self.drop_database_procedure(catalog, schema, drop_if_exists, query_context)
                    .await?;

                Ok(Output::new_with_affected_rows(0))
            }
        } else if drop_if_exists {
            // DROP TABLE IF EXISTS meets table not found - ignored
            Ok(Output::new_with_affected_rows(0))
        } else {
            SchemaNotFoundSnafu {
                schema_info: schema,
            }
            .fail()
        }
    }

    pub async fn truncate_table(
        &self,
        table_name: TableName,
        time_ranges: Vec<(Timestamp, Timestamp)>,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        self.truncate_table_with_options(
            table_name,
            time_ranges,
            DdlSubmitOptions::default(),
            query_context,
        )
        .await
    }

    /// Truncates a table, optionally waiting for the submitted procedure to finish.
    #[tracing::instrument(name = "truncate_table", skip_all)]
    pub(crate) async fn truncate_table_with_options(
        &self,
        table_name: TableName,
        time_ranges: Vec<(Timestamp, Timestamp)>,
        ddl_options: DdlSubmitOptions,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        ensure_table_writable(&table_name.schema_name, &table_name.table_name)?;

        let table = self
            .catalog_manager
            .table(
                &table_name.catalog_name,
                &table_name.schema_name,
                &table_name.table_name,
                Some(&query_context),
            )
            .await
            .context(CatalogSnafu)?
            .with_context(|| TableNotFoundSnafu {
                table_name: table_name.to_string(),
            })?;
        let table_id = table.table_info().table_id();
        let response = self
            .truncate_table_procedure(
                &table_name,
                table_id,
                time_ranges,
                ddl_options,
                query_context.clone(),
            )
            .await?;

        if !ddl_options.wait {
            let procedure_id = String::from_utf8_lossy(&response.key).into_owned();
            let output = build_procedure_id_output(&procedure_id)?;
            query_context.set_last_procedure_id(procedure_id);
            return Ok(output);
        }

        Ok(Output::new_with_affected_rows(0))
    }

    #[tracing::instrument(skip_all)]
    pub async fn alter_table(
        &self,
        alter_table: AlterTable,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        if matches!(
            alter_table.alter_operation(),
            AlterTableOperation::Repartition { .. } | AlterTableOperation::Partition { .. }
        ) {
            let request = expr_helper::to_repartition_request(alter_table, &query_context)?;
            return self.repartition_table(request, &query_context).await;
        }

        let expr = expr_helper::to_alter_table_expr(alter_table, &query_context)?;
        self.alter_table_inner(expr, query_context, TriggerReason::Manual)
            .await
    }

    #[tracing::instrument(skip_all)]
    pub async fn repartition_table(
        &self,
        request: RepartitionRequest,
        query_context: &QueryContextRef,
    ) -> Result<Output> {
        // Check if the schema is read-only.
        ensure_table_definition_writable(&request.schema_name, &request.table_name)?;

        let table_ref = TableReference::full(
            &request.catalog_name,
            &request.schema_name,
            &request.table_name,
        );
        // Get the table from the catalog.
        let table = self
            .catalog_manager
            .table(
                &request.catalog_name,
                &request.schema_name,
                &request.table_name,
                Some(query_context),
            )
            .await
            .context(CatalogSnafu)?
            .with_context(|| TableNotFoundSnafu {
                table_name: table_ref.to_string(),
            })?;
        let table_id = table.table_info().ident.table_id;
        // Get existing partition expressions from the table route.
        let (physical_table_id, physical_table_route) = self
            .table_metadata_manager
            .table_route_manager()
            .get_physical_table_route(table_id)
            .await
            .context(TableMetadataManagerSnafu)?;

        ensure!(
            physical_table_id == table_id,
            NotSupportedSnafu {
                feat: "REPARTITION on logical tables"
            }
        );

        let table_info = table.table_info();
        let existing_partition_columns = table_info.meta.partition_columns().collect::<Vec<_>>();
        let column_schemas = table_info.meta.schema.column_schemas();
        // `REPARTITION ... ON COLUMNS` uses overwrite semantics: the provided
        // columns are the full target partition columns, not an extension of the
        // current ones. Therefore source expressions are converted with the
        // existing partition columns, while target expressions and the final
        // partition rule are validated against this effective target column set.
        let target_partition_columns = match &request.source {
            RepartitionSource::Partitions {
                target_partition_columns,
                ..
            } => {
                ensure!(
                    !existing_partition_columns.is_empty(),
                    InvalidPartitionRuleSnafu {
                        reason: format!(
                            "table {} does not have partition columns, cannot repartition",
                            table_ref
                        )
                    }
                );

                if let Some(target_partition_columns) = target_partition_columns {
                    ensure!(
                        !target_partition_columns.is_empty(),
                        InvalidPartitionRuleSnafu {
                            reason: "ON COLUMNS requires at least one partition column"
                        }
                    );
                    validate_and_collect_partition_columns(
                        target_partition_columns,
                        column_schemas,
                    )?
                } else {
                    existing_partition_columns.clone()
                }
            }
            RepartitionSource::Unpartitioned { partition_columns } => {
                ensure!(
                    !partition_columns.is_empty(),
                    InvalidPartitionRuleSnafu {
                        reason: "PARTITION ON COLUMNS requires at least one partition column"
                    }
                );
                ensure!(
                    existing_partition_columns.is_empty(),
                    InvalidPartitionRuleSnafu {
                        reason: format!("table {} already has partition columns", table_ref)
                    }
                );
                partition_columns
                    .iter()
                    .map(|column_name| {
                        column_schemas
                            .iter()
                            .find(|column| &column.name == column_name)
                            .with_context(|| ColumnNotFoundSnafu { msg: column_name })
                    })
                    .collect::<Result<Vec<_>>>()?
            }
        };

        let from_column_name_and_type = column_name_and_type(&existing_partition_columns);
        let target_column_name_and_type = column_name_and_type(&target_partition_columns);
        let target_partition_column_names = target_partition_columns
            .iter()
            .map(|column| column.name.clone())
            .collect::<Vec<_>>();
        let timezone = query_context.timezone();
        // Convert SQL Exprs to PartitionExprs.
        let from_partition_exprs = match &request.source {
            RepartitionSource::Partitions { from_exprs, .. } => from_exprs
                .iter()
                .map(|expr| convert_one_expr(expr, &from_column_name_and_type, &timezone))
                .collect::<Result<Vec<_>>>()?,
            RepartitionSource::Unpartitioned { .. } => vec![],
        };

        let mut into_partition_exprs = request
            .into_exprs
            .iter()
            .map(|expr| convert_one_expr(expr, &target_column_name_and_type, &timezone))
            .collect::<Result<Vec<_>>>()?;

        // `MERGE PARTITION` (and some `REPARTITION`) generates a single `OR` expression from
        // multiple source partitions; try to simplify it for better readability and stability.
        if matches!(&request.source, RepartitionSource::Partitions { .. })
            && from_partition_exprs.len() > 1
            && into_partition_exprs.len() == 1
            && let Some(expr) = into_partition_exprs.pop()
        {
            into_partition_exprs.push(partition::simplify::simplify_merged_partition_expr(expr));
        }

        // Parse existing partition expressions from region routes.
        let mut existing_partition_exprs =
            Vec::with_capacity(physical_table_route.region_routes.len());
        for route in &physical_table_route.region_routes {
            let expr_json = route.region.partition_expr();
            if !expr_json.is_empty() {
                match PartitionExpr::from_json_str(&expr_json) {
                    Ok(Some(expr)) => existing_partition_exprs.push(expr),
                    Ok(None) => {
                        // Empty
                    }
                    Err(e) => {
                        return Err(e).context(DeserializePartitionExprSnafu);
                    }
                }
            }
        }

        // Validate that from_partition_exprs are a subset of existing partition exprs.
        // We compare PartitionExpr directly since it implements Eq.
        if matches!(&request.source, RepartitionSource::Partitions { .. }) {
            for from_expr in &from_partition_exprs {
                ensure!(
                    existing_partition_exprs.contains(from_expr),
                    InvalidPartitionRuleSnafu {
                        reason: format!(
                            "partition expression '{}' does not exist in table {}",
                            from_expr, table_ref
                        )
                    }
                );
            }
        }

        // Build the new partition expressions:
        // new_exprs = existing_exprs - from_exprs + into_exprs
        let new_partition_exprs: Vec<PartitionExpr> = match &request.source {
            RepartitionSource::Partitions { .. } => existing_partition_exprs
                .into_iter()
                .filter(|expr| !from_partition_exprs.contains(expr))
                .chain(into_partition_exprs.clone().into_iter())
                .collect(),
            RepartitionSource::Unpartitioned { .. } => into_partition_exprs.clone(),
        };
        ensure_partition_expr_columns_in_target(
            &new_partition_exprs,
            &target_partition_column_names.iter().collect(),
        )?;
        let new_partition_exprs_len = new_partition_exprs.len();
        let from_partition_exprs_len = from_partition_exprs.len();

        // Validate the new partition expressions using MultiDimPartitionRule and PartitionChecker.
        let _ = MultiDimPartitionRule::try_new(
            target_partition_column_names,
            vec![],
            new_partition_exprs,
            true,
        )
        .context(InvalidPartitionSnafu)?;

        let ddl_options = parse_ddl_options(&request.options)?;
        let serialize_exprs = |exprs: Vec<PartitionExpr>| -> Result<Vec<String>> {
            let mut json_exprs = Vec::with_capacity(exprs.len());
            for expr in exprs {
                json_exprs.push(expr.as_json_str().context(SerializePartitionExprSnafu)?);
            }
            Ok(json_exprs)
        };
        let from_partition_exprs_json = serialize_exprs(from_partition_exprs)?;
        let into_partition_exprs_json = serialize_exprs(into_partition_exprs)?;
        let source = match &request.source {
            RepartitionSource::Partitions {
                target_partition_columns,
                ..
            } => Source::PartitionExprs(PartitionedSource {
                exprs: from_partition_exprs_json,
                target_partition_columns: target_partition_columns
                    .clone()
                    .map(|columns| TargetPartitionColumns { columns }),
            }),
            RepartitionSource::Unpartitioned { partition_columns } => {
                Source::Unpartitioned(UnpartitionedSource {
                    partition_columns: partition_columns.clone(),
                })
            }
        };
        let repartition = Repartition {
            into_partition_exprs: into_partition_exprs_json,
            source: Some(source),
            ..Default::default()
        };
        let executor_context = to_executor_context(query_context.clone(), TriggerReason::Manual);
        let mut req = SubmitDdlTaskRequest::new(DdlTask::new_alter_table(AlterTableExpr {
            catalog_name: request.catalog_name.clone(),
            schema_name: request.schema_name.clone(),
            table_name: request.table_name.clone(),
            kind: Some(Kind::Repartition(repartition)),
        }));
        req.wait = ddl_options.wait;
        req.timeout = ddl_options.timeout;

        info!(
            "Submitting repartition task for table {} (table_id={}), from {} to {} partitions, timeout: {:?}, wait: {}",
            table_ref,
            table_id,
            from_partition_exprs_len,
            new_partition_exprs_len,
            ddl_options.timeout,
            ddl_options.wait
        );

        let response = self
            .procedure_executor
            .submit_ddl_task(executor_context, req)
            .await
            .context(error::ExecuteDdlSnafu)?;

        if !ddl_options.wait {
            let procedure_id = String::from_utf8_lossy(&response.key).into_owned();
            let output = build_procedure_id_output(&procedure_id)?;
            query_context.set_last_procedure_id(procedure_id);
            return Ok(output);
        }

        // Only invalidate cache if wait is true.
        let invalidate_keys = vec![
            CacheIdent::TableId(table_id),
            CacheIdent::TableName(TableName::new(
                request.catalog_name,
                request.schema_name,
                request.table_name,
            )),
        ];

        // Invalidates local cache ASAP.
        self.cache_invalidator
            .invalidate(&Context::default(), &invalidate_keys)
            .await
            .context(error::InvalidateTableCacheSnafu)?;

        Ok(Output::new_with_affected_rows(0))
    }

    #[tracing::instrument(skip_all)]
    pub async fn alter_table_inner(
        &self,
        expr: AlterTableExpr,
        query_context: QueryContextRef,
        trigger_reason: TriggerReason,
    ) -> Result<Output> {
        ensure_table_definition_writable(&expr.schema_name, &expr.table_name)?;

        let catalog_name = if expr.catalog_name.is_empty() {
            DEFAULT_CATALOG_NAME.to_string()
        } else {
            expr.catalog_name.clone()
        };

        let schema_name = if expr.schema_name.is_empty() {
            DEFAULT_SCHEMA_NAME.to_string()
        } else {
            expr.schema_name.clone()
        };

        let table_name = expr.table_name.clone();

        let table = self
            .catalog_manager
            .table(
                &catalog_name,
                &schema_name,
                &table_name,
                Some(&query_context),
            )
            .await
            .context(CatalogSnafu)?
            .with_context(|| TableNotFoundSnafu {
                table_name: format_full_table_name(&catalog_name, &schema_name, &table_name),
            })?;

        let table_id = table.table_info().ident.table_id;
        let need_alter = verify_alter(table_id, table.table_info(), expr.clone())?;
        if !need_alter {
            return Ok(Output::new_with_affected_rows(0));
        }
        info!(
            "Table info before alter is {:?}, expr: {:?}",
            table.table_info(),
            expr
        );

        let physical_table_id = self
            .table_metadata_manager
            .table_route_manager()
            .get_physical_table_id(table_id)
            .await
            .context(TableMetadataManagerSnafu)?;

        let executor_context = to_executor_context(query_context, trigger_reason);

        let (req, invalidate_keys) = if physical_table_id == table_id {
            // This is physical table
            let req = SubmitDdlTaskRequest::new(DdlTask::new_alter_table(expr));

            let invalidate_keys = vec![
                CacheIdent::TableId(table_id),
                CacheIdent::TableName(TableName::new(catalog_name, schema_name, table_name)),
            ];

            (req, invalidate_keys)
        } else {
            // This is logical table. Annotation alters only rewrite its own
            // metadata; `AlterLogicalTablesProcedure` only handles column adds.
            let annotation_alter = match expr.kind.as_ref() {
                Some(kind) => common_grpc_expr::annotation_alter_family(kind)
                    .context(AlterExprToRequestSnafu)?
                    .is_some_and(|family| family.allows_logical_tables()),
                None => false,
            };
            let task = if annotation_alter {
                DdlTask::new_alter_table(expr)
            } else {
                DdlTask::new_alter_logical_tables(vec![expr])
            };
            let req = SubmitDdlTaskRequest::new(task);

            let mut invalidate_keys = vec![
                CacheIdent::TableId(physical_table_id),
                CacheIdent::TableId(table_id),
                CacheIdent::TableName(TableName::new(catalog_name, schema_name, table_name)),
            ];

            let physical_table = self
                .table_metadata_manager
                .table_info_manager()
                .get(physical_table_id)
                .await
                .context(TableMetadataManagerSnafu)?
                .map(|x| x.into_inner());
            if let Some(physical_table) = physical_table {
                let physical_table_name = TableName::new(
                    physical_table.table_info.catalog_name,
                    physical_table.table_info.schema_name,
                    physical_table.table_info.name,
                );
                invalidate_keys.push(CacheIdent::TableName(physical_table_name));
            }

            (req, invalidate_keys)
        };

        self.procedure_executor
            .submit_ddl_task(executor_context, req)
            .await
            .context(error::ExecuteDdlSnafu)?;

        // Invalidates local cache ASAP.
        self.cache_invalidator
            .invalidate(&Context::default(), &invalidate_keys)
            .await
            .context(error::InvalidateTableCacheSnafu)?;

        Ok(Output::new_with_affected_rows(0))
    }

    #[cfg(feature = "enterprise")]
    #[tracing::instrument(skip_all)]
    pub async fn alter_trigger(
        &self,
        _alter_expr: AlterTrigger,
        _query_context: QueryContextRef,
    ) -> Result<Output> {
        crate::error::NotSupportedSnafu {
            feat: "alter trigger",
        }
        .fail()
    }

    #[tracing::instrument(skip_all)]
    pub async fn alter_database(
        &self,
        alter_expr: AlterDatabase,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        let alter_expr = expr_helper::to_alter_database_expr(alter_expr, &query_context)?;
        self.alter_database_inner(alter_expr, query_context).await
    }

    #[tracing::instrument(skip_all)]
    pub async fn alter_database_inner(
        &self,
        alter_expr: AlterDatabaseExpr,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        ensure!(
            !is_readonly_schema(&alter_expr.schema_name),
            SchemaReadOnlySnafu {
                name: query_context.current_schema().clone()
            }
        );

        let exists = self
            .catalog_manager
            .schema_exists(&alter_expr.catalog_name, &alter_expr.schema_name, None)
            .await
            .context(CatalogSnafu)?;
        ensure!(
            exists,
            SchemaNotFoundSnafu {
                schema_info: alter_expr.schema_name,
            }
        );

        let cache_ident = [CacheIdent::SchemaName(SchemaName {
            catalog_name: alter_expr.catalog_name.clone(),
            schema_name: alter_expr.schema_name.clone(),
        })];

        self.alter_database_procedure(alter_expr, query_context)
            .await?;

        // Invalidates local cache ASAP.
        self.cache_invalidator
            .invalidate(&Context::default(), &cache_ident)
            .await
            .context(error::InvalidateTableCacheSnafu)?;

        Ok(Output::new_with_affected_rows(0))
    }

    async fn create_table_procedure(
        &self,
        create_table: CreateTableExpr,
        partitions: Vec<PartitionExpr>,
        table_info: TableInfo,
        query_context: QueryContextRef,
        trigger_reason: TriggerReason,
    ) -> Result<SubmitDdlTaskResponse> {
        let partitions = partitions
            .into_iter()
            .map(|expr| expr.as_pb_partition().context(PartitionExprToPbSnafu))
            .collect::<Result<Vec<_>>>()?;

        let executor_context = to_executor_context_with_origin_frontend(
            query_context,
            &self.origin_frontend_addr,
            trigger_reason,
        );
        let request = SubmitDdlTaskRequest::new(DdlTask::new_create_table(
            create_table,
            partitions,
            table_info,
        ));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    async fn create_logical_tables_procedure(
        &self,
        tables_data: Vec<(CreateTableExpr, TableInfo)>,
        query_context: QueryContextRef,
        trigger_reason: TriggerReason,
    ) -> Result<SubmitDdlTaskResponse> {
        let executor_context = to_executor_context_with_origin_frontend(
            query_context,
            &self.origin_frontend_addr,
            trigger_reason,
        );
        let request = SubmitDdlTaskRequest::new(DdlTask::new_create_logical_tables(tables_data));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    async fn alter_logical_tables_procedure(
        &self,
        tables_data: Vec<AlterTableExpr>,
        query_context: QueryContextRef,
        trigger_reason: TriggerReason,
    ) -> Result<SubmitDdlTaskResponse> {
        let executor_context = to_executor_context(query_context, trigger_reason);
        let request = SubmitDdlTaskRequest::new(DdlTask::new_alter_logical_tables(tables_data));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    async fn drop_table_procedure(
        &self,
        table_name: &TableName,
        table_id: TableId,
        drop_if_exists: bool,
        query_context: QueryContextRef,
    ) -> Result<SubmitDdlTaskResponse> {
        let executor_context = to_executor_context(query_context, TriggerReason::Manual);
        let request = SubmitDdlTaskRequest::new(DdlTask::new_drop_table(
            table_name.catalog_name.clone(),
            table_name.schema_name.clone(),
            table_name.table_name.clone(),
            table_id,
            drop_if_exists,
        ));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    async fn drop_database_procedure(
        &self,
        catalog: String,
        schema: String,
        drop_if_exists: bool,
        query_context: QueryContextRef,
    ) -> Result<SubmitDdlTaskResponse> {
        let executor_context = to_executor_context(query_context, TriggerReason::Manual);
        let request =
            SubmitDdlTaskRequest::new(DdlTask::new_drop_database(catalog, schema, drop_if_exists));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    async fn alter_database_procedure(
        &self,
        alter_expr: AlterDatabaseExpr,
        query_context: QueryContextRef,
    ) -> Result<SubmitDdlTaskResponse> {
        let executor_context = to_executor_context(query_context, TriggerReason::Manual);
        let request = SubmitDdlTaskRequest::new(DdlTask::new_alter_database(alter_expr));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    async fn truncate_table_procedure(
        &self,
        table_name: &TableName,
        table_id: TableId,
        time_ranges: Vec<(Timestamp, Timestamp)>,
        ddl_options: DdlSubmitOptions,
        query_context: QueryContextRef,
    ) -> Result<SubmitDdlTaskResponse> {
        let executor_context = to_executor_context(query_context, TriggerReason::Manual);
        let mut request = SubmitDdlTaskRequest::new(DdlTask::new_truncate_table(
            table_name.catalog_name.clone(),
            table_name.schema_name.clone(),
            table_name.table_name.clone(),
            table_id,
            time_ranges,
        ));
        request.wait = ddl_options.wait;
        request.timeout = ddl_options.timeout;

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }

    #[tracing::instrument(skip_all)]
    pub async fn create_database(
        &self,
        database: &str,
        create_if_not_exists: bool,
        options: HashMap<String, String>,
        query_context: QueryContextRef,
    ) -> Result<Output> {
        let catalog = query_context.current_catalog();
        ensure!(
            NAME_PATTERN_REG.is_match(catalog),
            error::UnexpectedSnafu {
                violated: format!("Invalid catalog name: {}", catalog)
            }
        );

        ensure!(
            NAME_PATTERN_REG.is_match(database),
            error::UnexpectedSnafu {
                violated: format!("Invalid database name: {}", database)
            }
        );

        #[cfg(feature = "enterprise")]
        let creator = match &self.create_database_handler {
            Some(handler) => handler.creator(&query_context).context(ExternalSnafu)?,
            None => None,
        };
        #[cfg(not(feature = "enterprise"))]
        let creator = None;

        let output = if !self
            .catalog_manager
            .schema_exists(catalog, database, None)
            .await
            .context(CatalogSnafu)?
            && !self.catalog_manager.is_reserved_schema_name(database)
        {
            self.create_database_procedure(
                catalog.to_string(),
                database.to_string(),
                create_if_not_exists,
                options,
                query_context.clone(),
                creator,
            )
            .await?;

            Output::new_with_affected_rows(1)
        } else if create_if_not_exists {
            Output::new_with_affected_rows(1)
        } else {
            return error::SchemaExistsSnafu { name: database }.fail();
        };

        #[cfg(feature = "enterprise")]
        if let Some(handler) = &self.create_database_handler {
            handler
                .refresh_current_user(&query_context)
                .await
                .context(ExternalSnafu)?;
        }

        Ok(output)
    }

    async fn create_database_procedure(
        &self,
        catalog: String,
        database: String,
        create_if_not_exists: bool,
        options: HashMap<String, String>,
        query_context: QueryContextRef,
        creator: Option<CreatorGrantIntent>,
    ) -> Result<SubmitDdlTaskResponse> {
        let executor_context = to_executor_context(query_context, TriggerReason::Manual);
        let request = SubmitDdlTaskRequest::new(DdlTask::new_create_database(
            catalog,
            database,
            create_if_not_exists,
            options,
            creator,
        ));

        self.procedure_executor
            .submit_ddl_task(executor_context, request)
            .await
            .context(error::ExecuteDdlSnafu)
    }
}

/// Parse partition statement [Partitions] into [PartitionExpr] and partition columns.
pub fn parse_partitions(
    create_table: &CreateTableExpr,
    partitions: Option<Partitions>,
    query_ctx: &QueryContextRef,
) -> Result<(Vec<PartitionExpr>, Vec<String>)> {
    // If partitions are not defined by user, use the timestamp column (which has to be existed) as
    // the partition column, and create only one partition.
    let partition_columns = find_partition_columns(&partitions)?;
    let partition_exprs =
        find_partition_entries(create_table, &partitions, &partition_columns, query_ctx)?;

    // Validates partition
    let exprs = partition_exprs.clone();
    MultiDimPartitionRule::try_new(partition_columns.clone(), vec![], exprs, true)
        .context(InvalidPartitionSnafu)?;

    Ok((partition_exprs, partition_columns))
}

fn parse_partitions_for_logical_validation(
    create_table: &CreateTableExpr,
    partitions: &Partitions,
    query_ctx: &QueryContextRef,
) -> Result<(Vec<String>, Vec<PartitionExpr>)> {
    let partition_columns = partitions
        .column_list
        .iter()
        .map(|ident| ident.value.clone())
        .collect::<Vec<_>>();

    let column_name_and_type = partition_columns
        .iter()
        .map(|pc| {
            let column = create_table
                .column_defs
                .iter()
                .find(|c| &c.name == pc)
                .context(ColumnNotFoundSnafu { msg: pc.clone() })?;
            let column_name = &column.name;
            let data_type = ConcreteDataType::from(
                ColumnDataTypeWrapper::try_new(column.data_type, column.datatype_extension.clone())
                    .context(ColumnDataTypeSnafu)?,
            );
            Ok((column_name, data_type))
        })
        .collect::<Result<HashMap<_, _>>>()?;

    let mut partition_exprs = Vec::with_capacity(partitions.exprs.len());
    for expr in &partitions.exprs {
        let partition_expr = convert_one_expr(expr, &column_name_and_type, &query_ctx.timezone())?;
        partition_exprs.push(partition_expr);
    }

    MultiDimPartitionRule::try_new(
        partition_columns.clone(),
        vec![],
        partition_exprs.clone(),
        true,
    )
    .context(InvalidPartitionSnafu)?;

    Ok((partition_columns, partition_exprs))
}

/// Verifies an alter and returns whether it is necessary to perform the alter.
///
/// # Returns
///
/// Returns true if the alter need to be porformed; otherwise, it returns false.
pub fn verify_alter(
    table_id: TableId,
    table_info: Arc<TableInfo>,
    expr: AlterTableExpr,
) -> Result<bool> {
    let request: AlterTableRequest =
        common_grpc_expr::alter_expr_to_request(table_id, expr, Some(&table_info.meta))
            .context(AlterExprToRequestSnafu)?;

    let AlterTableRequest {
        table_name,
        alter_kind,
        ..
    } = &request;

    if let AlterKind::RenameTable { new_table_name } = alter_kind {
        ensure!(
            NAME_PATTERN_REG.is_match(new_table_name),
            error::UnexpectedSnafu {
                violated: format!("Invalid table name: {}", new_table_name)
            }
        );
        // Renaming INTO a computed graph table's name would let the overlay
        // shadow the renamed physical table, orphaning its data.
        ensure_table_definition_writable(&table_info.schema_name, new_table_name)?;
    } else if let AlterKind::AddColumns { columns } = alter_kind {
        // If all the columns are marked as add_if_not_exists and they already exist in the table,
        // there is no need to perform the alter.
        let column_names: HashSet<_> = table_info
            .meta
            .schema
            .column_schemas()
            .iter()
            .map(|schema| &schema.name)
            .collect();
        if columns.iter().all(|column| {
            column_names.contains(&column.column_schema.name) && column.add_if_not_exists
        }) {
            return Ok(false);
        }
    }

    let new_meta = table_info
        .meta
        .builder_with_alter_kind(table_name, &request.alter_kind)
        .context(error::TableSnafu)?
        .build()
        .context(error::BuildTableMetaSnafu { table_name })?;

    validate_json2_primary_keys(&new_meta.schema, &new_meta.primary_key_indices)?;

    Ok(true)
}

pub fn create_table_info(
    create_table: &CreateTableExpr,
    partition_columns: Vec<String>,
) -> Result<TableInfo> {
    let mut column_schemas = Vec::with_capacity(create_table.column_defs.len());
    let mut column_name_to_index_map = HashMap::new();

    for (idx, column) in create_table.column_defs.iter().enumerate() {
        let schema =
            column_def::try_as_column_schema(column).context(error::InvalidColumnDefSnafu {
                column: &column.name,
            })?;
        let schema = schema.with_time_index(column.name == create_table.time_index);

        column_schemas.push(schema);
        let _ = column_name_to_index_map.insert(column.name.clone(), idx);
    }

    let next_column_id = column_schemas.len() as u32;
    let schema = Arc::new(Schema::try_new(column_schemas).context(ConvertSchemaSnafu)?);

    let primary_key_indices = create_table
        .primary_keys
        .iter()
        .map(|name| {
            column_name_to_index_map
                .get(name)
                .cloned()
                .context(ColumnNotFoundSnafu { msg: name })
        })
        .collect::<Result<Vec<_>>>()?;

    let partition_key_indices = partition_columns
        .into_iter()
        .map(|col_name| {
            column_name_to_index_map
                .get(&col_name)
                .cloned()
                .context(ColumnNotFoundSnafu { msg: col_name })
        })
        .collect::<Result<Vec<_>>>()?;

    let mut table_options = TableOptions::try_from_iter(&create_table.table_options)
        .context(UnrecognizedTableOptionSnafu)?;

    validate_json2_primary_keys(&schema, &primary_key_indices)?;

    validate_and_normalize_annotations(&mut table_options, &schema, &partition_key_indices)?;

    let meta = TableMeta {
        schema,
        primary_key_indices,
        value_indices: vec![],
        engine: create_table.engine.clone(),
        next_column_id,
        options: table_options,
        created_on: Utc::now(),
        updated_on: Utc::now(),
        partition_key_indices,
        column_ids: vec![],
    };

    let desc = if create_table.desc.is_empty() {
        create_table.table_options.get(COMMENT_KEY).cloned()
    } else {
        Some(create_table.desc.clone())
    };

    let table_info = TableInfo {
        ident: metadata::TableIdent {
            // The table id of distributed table is assigned by Meta, set "0" here as a placeholder.
            table_id: 0,
            version: 0,
        },
        name: create_table.table_name.clone(),
        desc,
        catalog_name: create_table.catalog_name.clone(),
        schema_name: create_table.schema_name.clone(),
        meta,
        table_type: TableType::Base,
    };
    Ok(table_info)
}

fn validate_json2_primary_keys(schema: &Schema, primary_key_indices: &[usize]) -> Result<()> {
    for &index in primary_key_indices {
        let column = &schema.column_schemas()[index];
        ensure!(
            !column.data_type.is_json2(),
            InvalidSqlSnafu {
                err_msg: format!("JSON2 column `{}` cannot be a primary key", column.name),
            }
        );
    }
    Ok(())
}

/// Rejects DDL against read-only schemas and computed entity-graph tables.
fn ensure_table_writable(schema: &str, table: &str) -> Result<()> {
    ensure!(
        !is_readonly_schema(schema),
        SchemaReadOnlySnafu {
            name: schema.to_string()
        }
    );
    ensure!(
        !is_readonly_table(schema, table),
        TableReadOnlySnafu {
            name: table.to_string()
        }
    );
    Ok(())
}

/// [`ensure_table_writable`] plus the definition guard for system-defined
/// tables: user CREATE, ALTER and RENAME-into are rejected so the canonical
/// schema cannot be squatted or mutated. DROP and TRUNCATE stay allowed — the
/// next INSERT recreates the table canonically, which is also the recovery
/// path if the canonical definition changes across an upgrade.
fn ensure_table_definition_writable(schema: &str, table: &str) -> Result<()> {
    ensure_table_writable(schema, table)?;
    ensure!(
        !is_ddl_reserved_table(schema, table),
        TableDdlReservedSnafu {
            name: table.to_string()
        }
    );
    Ok(())
}

/// CREATE-side annotation validation: one rule source in the table crate,
/// mapped onto this crate's existing error variants so client-visible codes
/// and messages stay put.
fn validate_and_normalize_annotations(
    options: &mut TableOptions,
    schema: &Schema,
    partition_key_indices: &[usize],
) -> Result<()> {
    use table::requests::AnnotationValidationError as CheckError;
    let handle_error = |e: CheckError| match e {
        CheckError::ColumnNotFound { column } => ColumnNotFoundSnafu { msg: column }.build(),
        e @ (CheckError::UnknownKey { .. }
        | CheckError::InvalidValue { .. }
        | CheckError::InvalidPartitionNumHint { .. }
        | CheckError::ColumnNotStringForm { .. }) => InvalidSqlSnafu {
            err_msg: e.to_string(),
        }
        .build(),
        e @ (CheckError::NotSingleColumn
        | CheckError::PartitionMetadataConflict
        | CheckError::TimeIndexConflict) => InvalidPartitionRuleSnafu {
            reason: e.to_string(),
        }
        .build(),
    };
    validate_and_normalize_annotation_options(options, schema, partition_key_indices)
        .map_err(handle_error)
}

fn find_partition_columns(partitions: &Option<Partitions>) -> Result<Vec<String>> {
    let columns = if let Some(partitions) = partitions {
        partitions
            .column_list
            .iter()
            .map(|x| x.value.clone())
            .collect::<Vec<_>>()
    } else {
        vec![]
    };
    Ok(columns)
}

/// Parse [Partitions] into a group of partition entries.
///
/// Returns a list of [PartitionExpr], each of which defines a partition.
fn find_partition_entries(
    create_table: &CreateTableExpr,
    partitions: &Option<Partitions>,
    partition_columns: &[String],
    query_ctx: &QueryContextRef,
) -> Result<Vec<PartitionExpr>> {
    let Some(partitions) = partitions else {
        return Ok(vec![]);
    };

    // extract concrete data type of partition columns
    let column_name_and_type = partition_columns
        .iter()
        .map(|pc| {
            let column = create_table
                .column_defs
                .iter()
                .find(|c| &c.name == pc)
                // unwrap is safe here because we have checked that partition columns are defined
                .unwrap();
            let column_name = &column.name;
            let data_type = ConcreteDataType::from(
                ColumnDataTypeWrapper::try_new(column.data_type, column.datatype_extension.clone())
                    .context(ColumnDataTypeSnafu)?,
            );
            Ok((column_name, data_type))
        })
        .collect::<Result<HashMap<_, _>>>()?;

    // Transform parser expr to partition expr
    let mut partition_exprs = Vec::with_capacity(partitions.exprs.len());
    for partition in &partitions.exprs {
        let partition_expr =
            convert_one_expr(partition, &column_name_and_type, &query_ctx.timezone())?;
        partition_exprs.push(partition_expr);
    }

    Ok(partition_exprs)
}

fn column_name_and_type<'a>(
    partition_columns: &'a [&'a ColumnSchema],
) -> HashMap<&'a String, ConcreteDataType> {
    partition_columns
        .iter()
        .map(|column| (&column.name, column.data_type.clone()))
        .collect()
}

fn validate_and_collect_partition_columns<'a>(
    column_names: &[String],
    column_schemas: &'a [ColumnSchema],
) -> Result<Vec<&'a ColumnSchema>> {
    let mut seen = HashSet::with_capacity(column_names.len());
    column_names
        .iter()
        .map(|column_name| {
            ensure!(
                seen.insert(column_name),
                InvalidPartitionRuleSnafu {
                    reason: format!("duplicate partition column '{}'", column_name)
                }
            );
            column_schemas
                .iter()
                .find(|column| &column.name == column_name)
                .with_context(|| ColumnNotFoundSnafu { msg: column_name })
        })
        .collect()
}

fn ensure_partition_expr_columns_in_target(
    partition_exprs: &[PartitionExpr],
    target_partition_columns: &HashSet<&String>,
) -> Result<()> {
    for expr in partition_exprs {
        ensure_partition_operand_columns_in_target(&expr.lhs, target_partition_columns)?;
        ensure_partition_operand_columns_in_target(&expr.rhs, target_partition_columns)?;
    }

    Ok(())
}

fn ensure_partition_operand_columns_in_target(
    operand: &Operand,
    target_partition_columns: &HashSet<&String>,
) -> Result<()> {
    match operand {
        Operand::Column(column) => ensure!(
            target_partition_columns.contains(column),
            InvalidPartitionRuleSnafu {
                reason: format!(
                    "partition expression references column '{}' that is not in target partition columns",
                    column
                )
            }
        ),
        Operand::Expr(expr) => {
            ensure_partition_operand_columns_in_target(&expr.lhs, target_partition_columns)?;
            ensure_partition_operand_columns_in_target(&expr.rhs, target_partition_columns)?;
        }
        Operand::IcebergHash(hash) => ensure!(
            target_partition_columns.contains(&hash.column().to_string()),
            InvalidPartitionRuleSnafu {
                reason: format!(
                    "partition expression references column '{}' that is not in target partition columns",
                    hash.column()
                )
            }
        ),
        Operand::Value(_) => {}
    }

    Ok(())
}

fn convert_one_expr(
    expr: &Expr,
    column_name_and_type: &HashMap<&String, ConcreteDataType>,
    timezone: &Timezone,
) -> Result<PartitionExpr> {
    let Expr::BinaryOp { left, op, right } = expr else {
        return InvalidPartitionRuleSnafu {
            reason: "partition rule must be a binary expression",
        }
        .fail();
    };

    let op =
        RestrictedOp::try_from_parser(&op.clone()).with_context(|| InvalidPartitionRuleSnafu {
            reason: format!("unsupported operator in partition expr {op}"),
        })?;

    // convert leaf node.
    let (lhs, op, rhs) = match (left.as_ref(), right.as_ref()) {
        // col, val
        (Expr::Identifier(ident), Expr::Value(value)) => {
            let (column_name, data_type) = convert_identifier(ident, column_name_and_type)?;
            let value = convert_value(&value.value, data_type, timezone, None)?;
            (Operand::Column(column_name), op, Operand::Value(value))
        }
        (Expr::Identifier(ident), Expr::UnaryOp { op: unary_op, expr })
            if let Expr::Value(v) = &**expr =>
        {
            let (column_name, data_type) = convert_identifier(ident, column_name_and_type)?;
            let value = convert_value(&v.value, data_type, timezone, Some(*unary_op))?;
            (Operand::Column(column_name), op, Operand::Value(value))
        }
        // val, col
        (Expr::Value(value), Expr::Identifier(ident)) => {
            let (column_name, data_type) = convert_identifier(ident, column_name_and_type)?;
            let value = convert_value(&value.value, data_type, timezone, None)?;
            (Operand::Value(value), op, Operand::Column(column_name))
        }
        (Expr::UnaryOp { op: unary_op, expr }, Expr::Identifier(ident))
            if let Expr::Value(v) = &**expr =>
        {
            let (column_name, data_type) = convert_identifier(ident, column_name_and_type)?;
            let value = convert_value(&v.value, data_type, timezone, Some(*unary_op))?;
            (Operand::Value(value), op, Operand::Column(column_name))
        }
        // func, val
        (Expr::Function(function), literal)
            if !matches!(op, RestrictedOp::And | RestrictedOp::Or) =>
        {
            let hash = convert_iceberg_hash(function, column_name_and_type)?;
            let value = convert_literal(literal, IcebergHash::output_type(), timezone)?;
            (Operand::IcebergHash(hash), op, Operand::Value(value))
        }
        // val, func
        (literal, Expr::Function(function))
            if !matches!(op, RestrictedOp::And | RestrictedOp::Or) =>
        {
            let hash = convert_iceberg_hash(function, column_name_and_type)?;
            let value = convert_literal(literal, IcebergHash::output_type(), timezone)?;
            (Operand::Value(value), op, Operand::IcebergHash(hash))
        }
        (Expr::BinaryOp { .. }, Expr::BinaryOp { .. }) => {
            // sub-expr must against another sub-expr
            let lhs = convert_one_expr(left, column_name_and_type, timezone)?;
            let rhs = convert_one_expr(right, column_name_and_type, timezone)?;
            (Operand::Expr(lhs), op, Operand::Expr(rhs))
        }
        _ => {
            return InvalidPartitionRuleSnafu {
                reason: format!("invalid partition expr {expr}"),
            }
            .fail();
        }
    };

    Ok(PartitionExpr::new(lhs, op, rhs))
}

/// Converts an `iceberg_hash` call, which takes exactly one partition column.
fn convert_iceberg_hash(
    function: &Function,
    column_name_and_type: &HashMap<&String, ConcreteDataType>,
) -> Result<IcebergHash> {
    let invalid = || {
        InvalidPartitionRuleSnafu {
            reason: format!("unsupported partition function call {function}"),
        }
        .build()
    };
    let FunctionArguments::List(args) = &function.args else {
        return Err(invalid());
    };
    let [FunctionArg::Unnamed(FunctionArgExpr::Expr(Expr::Identifier(ident)))] =
        args.args.as_slice()
    else {
        return Err(invalid());
    };
    if function.filter.is_some()
        || function.over.is_some()
        || function.null_treatment.is_some()
        || !function.within_group.is_empty()
        || !matches!(function.parameters, FunctionArguments::None)
        || args.duplicate_treatment.is_some()
        || !args.clauses.is_empty()
    {
        return Err(invalid());
    }
    ensure!(
        function
            .name
            .to_string()
            .eq_ignore_ascii_case(IcebergHash::NAME),
        InvalidPartitionRuleSnafu {
            reason: format!("unsupported partition function {}", function.name),
        }
    );
    let (column_name, data_type) = convert_identifier(ident, column_name_and_type)?;
    ensure!(
        IcebergHash::accepts(&data_type),
        InvalidPartitionRuleSnafu {
            reason: format!(
                "partition function {} does not support column type {data_type}",
                IcebergHash::NAME
            ),
        }
    );
    Ok(IcebergHash::new(column_name))
}

fn convert_literal(expr: &Expr, data_type: ConcreteDataType, timezone: &Timezone) -> Result<Value> {
    match expr {
        Expr::Value(value) => convert_value(&value.value, data_type, timezone, None),
        Expr::UnaryOp { op, expr } if let Expr::Value(value) = &**expr => {
            convert_value(&value.value, data_type, timezone, Some(*op))
        }
        _ => InvalidPartitionRuleSnafu {
            reason: format!("expect a literal in partition expr, got {expr}"),
        }
        .fail(),
    }
}

fn convert_identifier(
    ident: &Ident,
    column_name_and_type: &HashMap<&String, ConcreteDataType>,
) -> Result<(String, ConcreteDataType)> {
    let column_name = ident.value.clone();
    let data_type = column_name_and_type
        .get(&column_name)
        .cloned()
        .with_context(|| ColumnNotFoundSnafu { msg: &column_name })?;
    Ok((column_name, data_type))
}

fn convert_value(
    value: &ParserValue,
    data_type: ConcreteDataType,
    timezone: &Timezone,
    unary_op: Option<UnaryOperator>,
) -> Result<Value> {
    sql_value_to_value(
        &ColumnSchema::new("<NONAME>", data_type, true),
        value,
        Some(timezone),
        unary_op,
        false,
    )
    .context(error::SqlCommonSnafu)
}

#[cfg(feature = "enterprise")]
async fn execute_undrop_table(
    table_metadata_manager: &TableMetadataManagerRef,
    procedure_executor: &ProcedureExecutorRef,
    cache_invalidator: &CacheInvalidatorRef,
    table_name: TableName,
    query_context: QueryContextRef,
) -> Result<Output> {
    // Undropping restores a table definition and could resurrect a
    // pre-canonical shape of a DDL-reserved table; rejected like CREATE.
    ensure_table_definition_writable(&table_name.schema_name, &table_name.table_name)?;

    let dropped = table_metadata_manager
        .get_dropped_table(&table_name)
        .await
        .context(TableMetadataManagerSnafu)?
        .with_context(|| TableNotFoundSnafu {
            table_name: table_name.to_string(),
        })?;

    let executor_context = to_executor_context(query_context, TriggerReason::Manual);
    let request = SubmitDdlTaskRequest::new(DdlTask::new_undrop_table(dropped.table_id));
    procedure_executor
        .submit_ddl_task(executor_context, request)
        .await
        .context(error::ExecuteDdlSnafu)?;

    if let Err(err) = cache_invalidator
        .invalidate(
            &Context::default(),
            &[
                CacheIdent::TableId(dropped.table_id),
                CacheIdent::TableName(table_name),
            ],
        )
        .await
    {
        warn!(
            "Failed to invalidate cache after restoring table '{}' (id={}): {}",
            dropped.table_name, dropped.table_id, err
        );
    }

    Ok(Output::new_with_affected_rows(0))
}

#[cfg(test)]
mod test {
    #[cfg(feature = "enterprise")]
    use std::sync::{Arc, Mutex};
    use std::time::Duration;

    #[cfg(feature = "enterprise")]
    use api::v1::meta::{ProcedureDetailResponse, ReconcileRequest, ReconcileResponse};
    use arrow::datatypes::{DataType as ArrowDataType, Field as ArrowField};
    #[cfg(feature = "enterprise")]
    use common_meta::cache_invalidator::{CacheInvalidator, CacheInvalidatorRef};
    #[cfg(feature = "enterprise")]
    use common_meta::instruction::CacheIdent;
    #[cfg(feature = "enterprise")]
    use common_meta::key::table_route::TableRouteValue;
    #[cfg(feature = "enterprise")]
    use common_meta::key::test_utils::new_test_table_info_with_name;
    #[cfg(feature = "enterprise")]
    use common_meta::key::{TableMetadataManager, TableMetadataManagerRef};
    #[cfg(feature = "enterprise")]
    use common_meta::kv_backend::memory::MemoryKvBackend;
    #[cfg(feature = "enterprise")]
    use common_meta::procedure_executor::{
        ExecutorContext, ProcedureExecutor, ProcedureExecutorRef,
    };
    #[cfg(feature = "enterprise")]
    use common_meta::rpc::ddl::{DdlTask, SubmitDdlTaskRequest, SubmitDdlTaskResponse};
    #[cfg(feature = "enterprise")]
    use common_meta::rpc::procedure::{
        MigrateRegionRequest, MigrateRegionResponse, ProcedureStateResponse,
    };
    use datafusion::functions_aggregate::expr_fn::count;
    use datafusion::logical_expr::builder::LogicalTableSource;
    use datafusion::logical_expr::{LogicalPlanBuilder, col};
    use session::context::QueryContext;
    use sql::dialect::GreptimeDbDialect;
    use sql::parser::{ParseOptions, ParserContext};
    use sql::statements::statement::Statement;
    use sqlparser::parser::Parser;
    use table::requests::REPARTITION_COLUMN_HINT_KEY;

    use super::*;
    use crate::expr_helper;

    #[cfg(feature = "enterprise")]
    #[derive(Default)]
    struct RecordingProcedureExecutor {
        requests: Mutex<Vec<SubmitDdlTaskRequest>>,
        contexts: Mutex<Vec<ExecutorContext>>,
        fail: bool,
    }

    #[cfg(feature = "enterprise")]
    #[async_trait::async_trait]
    impl ProcedureExecutor for RecordingProcedureExecutor {
        async fn submit_ddl_task(
            &self,
            ctx: ExecutorContext,
            request: SubmitDdlTaskRequest,
        ) -> common_meta::error::Result<SubmitDdlTaskResponse> {
            self.contexts.lock().unwrap().push(ctx);
            self.requests.lock().unwrap().push(request);
            if self.fail {
                return common_meta::error::UnsupportedSnafu {
                    operation: "test submit failure",
                }
                .fail();
            }
            Ok(SubmitDdlTaskResponse::default())
        }

        async fn migrate_region(
            &self,
            _: &ExecutorContext,
            _: MigrateRegionRequest,
        ) -> common_meta::error::Result<MigrateRegionResponse> {
            unimplemented!()
        }
        async fn reconcile(
            &self,
            _: &ExecutorContext,
            _: ReconcileRequest,
        ) -> common_meta::error::Result<ReconcileResponse> {
            unimplemented!()
        }
        async fn query_procedure_state(
            &self,
            _: &ExecutorContext,
            _: &str,
        ) -> common_meta::error::Result<ProcedureStateResponse> {
            unimplemented!()
        }
        async fn list_procedures(
            &self,
            _: &ExecutorContext,
        ) -> common_meta::error::Result<ProcedureDetailResponse> {
            unimplemented!()
        }
    }

    #[cfg(feature = "enterprise")]
    #[derive(Default)]
    struct RecordingCacheInvalidator {
        invalidations: Mutex<Vec<Vec<CacheIdent>>>,
        fail: bool,
    }

    #[cfg(feature = "enterprise")]
    #[async_trait::async_trait]
    impl CacheInvalidator for RecordingCacheInvalidator {
        async fn invalidate(
            &self,
            _: &Context,
            caches: &[CacheIdent],
        ) -> common_meta::error::Result<()> {
            self.invalidations.lock().unwrap().push(caches.to_vec());
            if self.fail {
                return common_meta::error::UnsupportedSnafu {
                    operation: "test cache failure",
                }
                .fail();
            }
            Ok(())
        }

        fn invalidate_all(&self) -> common_meta::error::Result<()> {
            Ok(())
        }
    }

    #[cfg(feature = "enterprise")]
    async fn dropped_table_manager(table_id: TableId, name: &TableName) -> TableMetadataManagerRef {
        let backend = Arc::new(MemoryKvBackend::default());
        let manager = Arc::new(TableMetadataManager::new(backend));
        let mut info = new_test_table_info_with_name(table_id, &name.table_name);
        info.catalog_name = name.catalog_name.clone();
        info.schema_name = name.schema_name.clone();
        let route = TableRouteValue::physical(vec![]);
        manager
            .create_table_metadata(info, route.clone(), HashMap::new())
            .await
            .unwrap();
        manager
            .delete_table_metadata(table_id, name, &route, &HashMap::new(), None)
            .await
            .unwrap();
        manager
    }

    #[cfg(feature = "enterprise")]
    #[tokio::test]
    async fn test_undrop_table_execution_path() {
        let name = TableName::new("greptime", "public", "metrics");
        let manager = dropped_table_manager(42, &name).await;
        // A same-name live table must not redirect restoration to its ID.
        let mut live = new_test_table_info_with_name(99, "metrics");
        live.catalog_name = "greptime".to_string();
        live.schema_name = "public".to_string();
        manager
            .create_table_metadata(live, TableRouteValue::physical(vec![]), HashMap::new())
            .await
            .unwrap();
        let procedure = Arc::new(RecordingProcedureExecutor::default());
        let cache = Arc::new(RecordingCacheInvalidator::default());

        execute_undrop_table(
            &manager,
            &(procedure.clone() as ProcedureExecutorRef),
            &(cache.clone() as CacheInvalidatorRef),
            name.clone(),
            QueryContext::arc(),
        )
        .await
        .unwrap();

        let requests = procedure.requests.lock().unwrap();
        assert!(matches!(&requests[0].task, DdlTask::UndropTable(task) if task.table_id == 42));
        let contexts = procedure.contexts.lock().unwrap();
        assert_eq!(
            contexts[0].event_input.as_ref().map(|input| input.reason),
            Some(TriggerReason::Manual)
        );
        assert_eq!(
            cache.invalidations.lock().unwrap()[0],
            vec![CacheIdent::TableId(42), CacheIdent::TableName(name)]
        );
    }

    #[cfg(feature = "enterprise")]
    #[tokio::test]
    async fn test_undrop_table_missing_tombstone_submits_nothing() {
        let manager = Arc::new(TableMetadataManager::new(Arc::new(
            MemoryKvBackend::default(),
        )));
        let procedure = Arc::new(RecordingProcedureExecutor::default());
        let cache = Arc::new(RecordingCacheInvalidator::default());
        let err = execute_undrop_table(
            &manager,
            &(procedure.clone() as ProcedureExecutorRef),
            &(cache as CacheInvalidatorRef),
            TableName::new("greptime", "public", "missing"),
            QueryContext::arc(),
        )
        .await
        .unwrap_err();
        assert!(matches!(err, crate::error::Error::TableNotFound { .. }));
        assert!(procedure.requests.lock().unwrap().is_empty());
    }

    #[cfg(feature = "enterprise")]
    #[tokio::test]
    async fn test_undrop_table_submit_failure_does_not_invalidate() {
        let name = TableName::new("greptime", "public", "metrics");
        let manager = dropped_table_manager(42, &name).await;
        let procedure = Arc::new(RecordingProcedureExecutor {
            fail: true,
            ..Default::default()
        });
        let cache = Arc::new(RecordingCacheInvalidator::default());
        execute_undrop_table(
            &manager,
            &(procedure as ProcedureExecutorRef),
            &(cache.clone() as CacheInvalidatorRef),
            name,
            QueryContext::arc(),
        )
        .await
        .unwrap_err();
        assert!(cache.invalidations.lock().unwrap().is_empty());
    }

    #[cfg(feature = "enterprise")]
    #[tokio::test]
    async fn test_undrop_table_cache_failure_returns_success_after_submit() {
        let name = TableName::new("greptime", "public", "metrics");
        let manager = dropped_table_manager(42, &name).await;
        let procedure = Arc::new(RecordingProcedureExecutor::default());
        let cache = Arc::new(RecordingCacheInvalidator {
            fail: true,
            ..Default::default()
        });
        execute_undrop_table(
            &manager,
            &(procedure.clone() as ProcedureExecutorRef),
            &(cache.clone() as CacheInvalidatorRef),
            name.clone(),
            QueryContext::arc(),
        )
        .await
        .unwrap();
        assert_eq!(procedure.requests.lock().unwrap().len(), 1);
        assert_eq!(
            cache.invalidations.lock().unwrap()[0],
            vec![CacheIdent::TableId(42), CacheIdent::TableName(name)]
        );
    }

    #[test]
    fn test_parse_ddl_options() {
        let defaults = parse_ddl_options(&OptionMap::default()).unwrap();
        assert!(defaults.wait);
        assert_eq!(defaults.timeout, SubmitDdlTaskRequest::default_timeout());

        let options = OptionMap::from([
            ("timeout".to_string(), "5m".to_string()),
            ("wait".to_string(), "false".to_string()),
        ]);
        let ddl_options = parse_ddl_options(&options).unwrap();
        assert!(!ddl_options.wait);
        assert_eq!(Duration::from_secs(300), ddl_options.timeout);

        let options = OptionMap::from([("timeout".to_string(), "1500ms".to_string())]);
        assert_eq!(
            parse_ddl_options(&options).unwrap().timeout,
            Duration::from_millis(1500)
        );
        for (key, value) in [
            ("wait", "yes"),
            ("timeout", "not-a-duration"),
            ("timeout", "-1s"),
        ] {
            let options = OptionMap::from([(key.to_string(), value.to_string())]);
            assert!(matches!(
                parse_ddl_options(&options),
                Err(error::Error::InvalidSql { .. })
            ));
        }
    }

    #[test]
    fn test_validate_and_normalize_annotations() {
        let schema = Schema::new(vec![
            ColumnSchema::new("service_name", ConcreteDataType::string_datatype(), true),
            ColumnSchema::new("host_id", ConcreteDataType::string_datatype(), true),
            ColumnSchema::new("value", ConcreteDataType::float64_datatype(), true),
            ColumnSchema::new("payload", ConcreteDataType::binary_datatype(), true),
        ]);
        let opts = |pairs: &[(&str, &str)]| TableOptions {
            extra_options: pairs
                .iter()
                .map(|(k, v)| (k.to_string(), v.to_string()))
                .collect(),
            ..Default::default()
        };
        let check = |pairs: &[(&str, &str)]| {
            let mut options = opts(pairs);
            validate_and_normalize_annotations(&mut options, &schema, &[]).map(|()| options)
        };

        // Any existing column with a string form may be an id, tag or field.
        for pairs in [
            &[(
                "greptime.semantic.entity.process.id",
                "service_name,host_id",
            )][..],
            &[("greptime.semantic.entity.service.id", "value")][..],
        ] {
            assert!(check(pairs).is_ok());
        }

        // Missing columns keep this crate's error variant and status code.
        let missing = check(&[("greptime.semantic.entity.service.id", "nope")]).unwrap_err();
        assert!(matches!(missing, error::Error::ColumnNotFound { .. }));
        assert_eq!(
            common_error::status_code::StatusCode::InvalidArguments,
            common_error::ext::ErrorExt::status_code(&missing)
        );

        let binary_id = check(&[("greptime.semantic.entity.service.id", "payload")]).unwrap_err();
        assert!(matches!(binary_id, error::Error::InvalidSql { .. }));

        // The SQL parser checks keys and value domains, but gRPC expressions
        // bypass it.
        let bad_value = check(&[("greptime.semantic.signal_type", "garbage")]).unwrap_err();
        assert!(matches!(bad_value, error::Error::InvalidSql { .. }));

        let unknown_key = check(&[("greptime.semantic.nonsense", "x")]).unwrap_err();
        assert!(matches!(unknown_key, error::Error::InvalidSql { .. }));
    }

    #[test]
    fn test_validate_and_normalize_flow_options_empty() {
        assert!(
            validate_and_normalize_flow_options(HashMap::new(), None)
                .unwrap()
                .is_empty()
        );
    }

    #[test]
    fn test_validate_and_normalize_flow_options_valid() {
        let options = HashMap::from([
            (DEFER_ON_MISSING_SOURCE_KEY.to_string(), "TRUE".to_string()),
            (
                FLOW_EXPERIMENTAL_ENABLE_INCREMENTAL_READ_KEY.to_string(),
                "FALSE".to_string(),
            ),
        ]);

        assert_eq!(
            validate_and_normalize_flow_options(options, None).unwrap(),
            HashMap::from([
                (DEFER_ON_MISSING_SOURCE_KEY.to_string(), "true".to_string(),),
                (
                    FLOW_EXPERIMENTAL_ENABLE_INCREMENTAL_READ_KEY.to_string(),
                    "false".to_string(),
                )
            ])
        );
    }

    #[cfg(not(feature = "enterprise"))]
    #[test]
    fn test_validate_and_normalize_flow_options_unknown_option() {
        let err = validate_and_normalize_flow_options(
            HashMap::from([("foo".to_string(), "bar".to_string())]),
            None,
        )
        .unwrap_err();

        assert!(
            err.to_string()
                .contains("unknown flow option 'foo', supported options:")
        );
    }

    #[test]
    fn test_validate_and_normalize_flow_options_reserved_option() {
        let err = validate_and_normalize_flow_options(
            HashMap::from([(
                FlowType::FLOW_TYPE_KEY.to_string(),
                FlowType::BATCHING.to_string(),
            )]),
            None,
        )
        .unwrap_err();

        assert!(
            err.to_string()
                .contains("flow option 'flow_type' is reserved for internal use")
        );
    }

    #[test]
    fn test_validate_and_normalize_flow_options_invalid_bool() {
        let err = validate_and_normalize_flow_options(
            HashMap::from([(
                DEFER_ON_MISSING_SOURCE_KEY.to_string(),
                "not-a-bool".to_string(),
            )]),
            None,
        )
        .unwrap_err();

        assert!(
            err.to_string()
                .contains("invalid flow option 'defer_on_missing_source': 'not-a-bool'")
        );
    }

    #[test]
    fn test_parser_flow_option_access_key_id() {
        let sql = r"
CREATE FLOW task_6
SINK TO schema_1.table_1
WITH (access_key_id = [true])
AS
SELECT max(c1), min(c2) FROM schema_2.table_2;";
        let stmt =
            ParserContext::create_with_dialect(sql, &GreptimeDbDialect {}, ParseOptions::default())
                .unwrap()
                .pop()
                .unwrap();

        let Statement::CreateFlow(create_flow) = stmt else {
            unreachable!()
        };
        let expr =
            expr_helper::to_create_flow_task_expr(create_flow, &QueryContext::arc()).unwrap();

        #[cfg(not(feature = "enterprise"))]
        {
            let err = validate_and_normalize_flow_options(expr.flow_options, None).unwrap_err();
            assert!(
                err.to_string()
                    .contains("unknown flow option 'access_key_id'")
            );
        }
        #[cfg(feature = "enterprise")]
        assert_eq!(
            validate_and_normalize_flow_options(expr.flow_options, None).unwrap(),
            HashMap::from([("access_key_id".to_string(), "['true']".to_string())])
        );
    }

    fn stateless_test_scan(name: &str) -> LogicalPlan {
        let schema = arrow::datatypes::Schema::new(vec![ArrowField::new(
            "value",
            ArrowDataType::Int32,
            true,
        )]);
        LogicalPlanBuilder::scan(
            name,
            Arc::new(LogicalTableSource::new(Arc::new(schema))),
            None,
        )
        .unwrap()
        .build()
        .unwrap()
    }

    #[test]
    fn test_is_stateless_flow_plan_accepts_scan_projection_and_filter() {
        let scan = stateless_test_scan("source");
        assert!(is_stateless_flow_plan(&scan));

        let projection = LogicalPlanBuilder::from(scan.clone())
            .project(vec![col("value")])
            .unwrap()
            .build()
            .unwrap();
        assert!(is_stateless_flow_plan(&projection));

        let filter = LogicalPlanBuilder::from(projection)
            .filter(col("value").gt(datafusion_expr::lit(0)))
            .unwrap()
            .build()
            .unwrap();
        assert!(is_stateless_flow_plan(&filter));
    }

    #[test]
    fn test_is_stateless_flow_plan_rejects_aggregate_and_multiple_scans() {
        let scan = stateless_test_scan("source");
        let aggregate = LogicalPlanBuilder::from(scan.clone())
            .aggregate(
                Vec::<datafusion_expr::Expr>::new(),
                vec![count(col("value"))],
            )
            .unwrap()
            .build()
            .unwrap();
        assert!(!is_stateless_flow_plan(&aggregate));

        let distinct = LogicalPlanBuilder::from(scan.clone())
            .distinct()
            .unwrap()
            .build()
            .unwrap();
        assert!(is_stateless_flow_plan(&distinct));

        let distinct_on_aggregate = LogicalPlanBuilder::from(aggregate)
            .distinct()
            .unwrap()
            .build()
            .unwrap();
        assert!(!is_stateless_flow_plan(&distinct_on_aggregate));

        let multiple_scans = LogicalPlanBuilder::from(stateless_test_scan("left"))
            .cross_join(stateless_test_scan("right"))
            .unwrap()
            .build()
            .unwrap();
        assert!(!is_stateless_flow_plan(&multiple_scans));
    }

    // --- Schedule option tests ---

    #[test]
    fn test_eval_interval_rejected_non_positive() {
        // Zero eval_interval should be rejected.
        let err = validate_and_normalize_flow_options(HashMap::new(), Some(0)).unwrap_err();
        assert!(err.to_string().contains("EVAL INTERVAL must be positive"));

        // Negative eval_interval should be rejected.
        let err = validate_and_normalize_flow_options(HashMap::new(), Some(-5)).unwrap_err();
        assert!(err.to_string().contains("EVAL INTERVAL must be positive"));

        // Positive eval_interval should be accepted.
        let result = validate_and_normalize_flow_options(HashMap::new(), Some(300));
        assert!(result.is_ok());
    }

    #[test]
    fn test_internal_schedule_key_rejected() {
        let err = validate_and_normalize_flow_options(
            HashMap::from([(INTERNAL_EVAL_SCHEDULE_KEY.to_string(), "value".to_string())]),
            Some(300),
        )
        .unwrap_err();
        assert!(err.to_string().contains("reserved for internal use"));
    }

    #[cfg(not(feature = "enterprise"))]
    #[test]
    fn test_schedule_keys_rejected_as_unknown_options() {
        for key in [
            "eval_interval_anchor",
            "eval_interval_start",
            "eval_interval_missed_tick_policy",
            "eval_interval_catchup_max_runs",
            "eval_interval_catchup_max_lag",
        ] {
            let err = validate_and_normalize_flow_options(
                HashMap::from([(key.to_string(), "value".to_string())]),
                Some(300),
            )
            .unwrap_err();
            assert!(
                err.to_string()
                    .contains(&format!("unknown flow option '{key}'"))
            );
        }
    }

    #[cfg(feature = "enterprise")]
    #[test]
    fn test_schedule_keys_preserved() {
        let options = HashMap::from([
            ("eval_interval_anchor".to_string(), "anchor".to_string()),
            ("eval_interval_start".to_string(), "start".to_string()),
            (
                "eval_interval_missed_tick_policy".to_string(),
                "missed".to_string(),
            ),
            (
                "eval_interval_catchup_max_runs".to_string(),
                "runs".to_string(),
            ),
            (
                "eval_interval_catchup_max_lag".to_string(),
                "lag".to_string(),
            ),
        ]);
        assert_eq!(
            validate_and_normalize_flow_options(options.clone(), Some(300)).unwrap(),
            options
        );
    }

    #[test]
    fn test_internal_transport_keys_rejected_as_reserved() {
        for key in [
            "__greptime_internal_eval_schedule",
            "__greptime_internal_eval_offset_secs",
        ] {
            let err = validate_and_normalize_flow_options(
                HashMap::from([(key.to_string(), "120".to_string())]),
                Some(300),
            )
            .unwrap_err();

            assert!(
                err.to_string()
                    .contains(&format!("flow option '{key}' is reserved for internal use")),
                "unexpected error for {key}: {err}"
            );
        }
    }

    #[test]
    fn test_determine_flow_type_for_source_state_missing_sources_require_opt_in() {
        let err =
            determine_flow_type_for_source_state("my_flow", &HashMap::new(), true, false, false)
                .unwrap_err();

        assert!(err.to_string().contains(
            "missing source tables for flow 'my_flow'; use WITH (defer_on_missing_source = true) to create a pending flow"
        ));
    }

    #[test]
    fn test_determine_flow_type_for_source_state_missing_sources_prefer_batching() {
        let flow_options =
            HashMap::from([(DEFER_ON_MISSING_SOURCE_KEY.to_string(), "true".to_string())]);

        assert_eq!(
            determine_flow_type_for_source_state("my_flow", &flow_options, true, true, false)
                .unwrap(),
            Some(FlowType::Batching)
        );
        let err = determine_flow_type_for_source_state("my_flow", &flow_options, true, true, true)
            .unwrap_err();
        assert!(err.to_string().contains(
            "flow 'my_flow' with EVAL INTERVAL requires the batching scheduler, but source tables with ttl=instant are not supported under batching mode"
        ));
    }

    #[test]
    fn test_determine_flow_type_for_source_state_existing_sources_require_plan() {
        assert_eq!(
            determine_flow_type_for_source_state("my_flow", &HashMap::new(), false, true, false)
                .unwrap(),
            None
        );
    }

    #[test]
    fn test_determine_flow_type_for_source_state_instant_ttl_with_eval_interval_is_rejected() {
        // A flow with `EVAL INTERVAL` is forced onto the batching scheduler,
        // which cannot read instant-TTL source tables: reject instead of
        // silently falling back to streaming (where the schedule/offset would
        // be ignored).
        let err =
            determine_flow_type_for_source_state("my_flow", &HashMap::new(), false, true, true)
                .unwrap_err();
        assert!(
            err.to_string().contains(
                "flow 'my_flow' with EVAL INTERVAL requires the batching scheduler, but source tables with ttl=instant are not supported under batching mode"
            ),
            "unexpected error: {err}"
        );
    }

    #[test]
    fn test_name_is_match() {
        assert!(!NAME_PATTERN_REG.is_match("/adaf"));
        assert!(!NAME_PATTERN_REG.is_match("🈲"));
        assert!(NAME_PATTERN_REG.is_match("hello"));
        assert!(NAME_PATTERN_REG.is_match("test@"));
        assert!(!NAME_PATTERN_REG.is_match("@test"));
        assert!(NAME_PATTERN_REG.is_match("test#"));
        assert!(!NAME_PATTERN_REG.is_match("#test"));
        assert!(!NAME_PATTERN_REG.is_match("@"));
        assert!(!NAME_PATTERN_REG.is_match("#"));
    }

    #[test]
    fn test_convert_iceberg_hash() {
        let names = ["host", "count", "device", "measure", "host-name"].map(String::from);
        let column_name_and_type = HashMap::from([
            (&names[0], ConcreteDataType::string_datatype()),
            (&names[1], ConcreteDataType::int16_datatype()),
            (&names[2], ConcreteDataType::uint64_datatype()),
            (&names[3], ConcreteDataType::float64_datatype()),
            (&names[4], ConcreteDataType::string_datatype()),
        ]);
        let timezone = Timezone::from_tz_string("UTC").unwrap();
        let dialect = GreptimeDbDialect {};
        let parse = |sql: &str| {
            Parser::new(&dialect)
                .try_with_sql(sql)
                .unwrap()
                .parse_expr()
                .unwrap()
        };

        for sql in [
            "iceberg_hash(host) < 0",
            "ICEBERG_HASH(device) >= -1073741824",
            "0 > iceberg_hash(count)",
            "iceberg_hash(host) < 100 AND host < 'm'",
            r#"iceberg_hash("host-name") < 0 AND "host-name" < 'm'"#,
        ] {
            let expr = convert_one_expr(&parse(sql), &column_name_and_type, &timezone).unwrap();
            // SHOW CREATE TABLE prints the rule back as SQL, quoting identifiers.
            let mut partitions = Partitions {
                column_list: vec![],
                exprs: vec![expr.to_parser_expr()],
            };
            partitions.set_quote('"');
            let restored = parse(&partitions.exprs[0].to_string());
            assert_eq!(
                convert_one_expr(&restored, &column_name_and_type, &timezone).unwrap(),
                expr,
                "{sql}"
            );
        }

        for sql in [
            "iceberg_hash(measure) < 0",
            "iceberg_hash(host, count) < 0",
            "iceberg_hash('a') < 0",
            "iceberg_hash(missing) < 0",
            "iceberg_hash(DISTINCT host) < 0",
            "md5(host) < 0",
            "iceberg_hash(host) < host",
            "iceberg_hash(host) < 'abc'",
            "iceberg_hash(host) < 2147483648",
        ] {
            assert!(
                convert_one_expr(&parse(sql), &column_name_and_type, &timezone).is_err(),
                "{sql}"
            );
        }
    }

    #[test]
    fn test_partition_expr_equivalence_with_swapped_operands() {
        let column_name = "device_id".to_string();
        let column_name_and_type =
            HashMap::from([(&column_name, ConcreteDataType::int32_datatype())]);
        let timezone = Timezone::from_tz_string("UTC").unwrap();
        let dialect = GreptimeDbDialect {};

        let mut parser = Parser::new(&dialect)
            .try_with_sql("device_id < 100")
            .unwrap();
        let expr_left = parser.parse_expr().unwrap();

        let mut parser = Parser::new(&dialect)
            .try_with_sql("100 > device_id")
            .unwrap();
        let expr_right = parser.parse_expr().unwrap();

        let partition_left =
            convert_one_expr(&expr_left, &column_name_and_type, &timezone).unwrap();
        let partition_right =
            convert_one_expr(&expr_right, &column_name_and_type, &timezone).unwrap();

        assert_eq!(partition_left, partition_right);
        assert!([partition_left.clone()].contains(&partition_right));

        let mut physical_partition_exprs = vec![partition_left];
        let mut logical_partition_exprs = vec![partition_right];
        physical_partition_exprs.sort_unstable();
        logical_partition_exprs.sort_unstable();
        assert_eq!(physical_partition_exprs, logical_partition_exprs);
    }

    #[test]
    fn test_repartition_target_partition_columns_are_overwrite_context() {
        let device_id = ColumnSchema::new("device_id", ConcreteDataType::int32_datatype(), true);
        let area = ColumnSchema::new("area", ConcreteDataType::string_datatype(), true);
        let existing_partition_columns = vec![&device_id];
        let target_partition_columns = vec![&device_id, &area];
        let existing_column_name_and_type = column_name_and_type(&existing_partition_columns);
        let target_column_name_and_type = column_name_and_type(&target_partition_columns);
        let timezone = Timezone::from_tz_string("UTC").unwrap();
        let dialect = GreptimeDbDialect {};

        let mut parser = Parser::new(&dialect)
            .try_with_sql("device_id < 100 AND area < 'South'")
            .unwrap();
        let expr = parser.parse_expr().unwrap();

        let err = convert_one_expr(&expr, &existing_column_name_and_type, &timezone).unwrap_err();
        assert!(err.to_string().contains("area"));

        let partition_expr = convert_one_expr(&expr, &target_column_name_and_type, &timezone)
            .expect("target columns should overwrite the conversion context");
        let partition_expr = partition_expr.to_string();
        assert!(partition_expr.contains("device_id"));
        assert!(partition_expr.contains("area"));
        assert!(partition_expr.contains("South"));
    }

    #[test]
    fn test_repartition_rejects_remaining_expr_outside_target_columns() {
        let device_id = "device_id".to_string();
        let area = "area".to_string();
        let timezone = Timezone::from_tz_string("UTC").unwrap();
        let column_name_and_type = HashMap::from([
            (&device_id, ConcreteDataType::int32_datatype()),
            (&area, ConcreteDataType::string_datatype()),
        ]);
        let dialect = GreptimeDbDialect {};
        let target_partition_columns = HashSet::from([&area]);
        for sql in ["device_id >= 100", "iceberg_hash(device_id) >= 0"] {
            let mut parser = Parser::new(&dialect).try_with_sql(sql).unwrap();
            let remaining_old_expr = convert_one_expr(
                &parser.parse_expr().unwrap(),
                &column_name_and_type,
                &timezone,
            )
            .unwrap();

            let err = ensure_partition_expr_columns_in_target(
                &[remaining_old_expr],
                &target_partition_columns,
            )
            .unwrap_err();

            assert!(err.to_string().contains("device_id"), "{sql}");
            assert!(
                err.to_string().contains("target partition columns"),
                "{sql}"
            );
        }
    }

    #[test]
    fn test_repartition_rejects_duplicate_target_partition_columns() {
        let device_id = ColumnSchema::new("device_id", ConcreteDataType::int32_datatype(), true);
        let column_schemas = vec![device_id];
        let target_partition_columns = vec!["device_id".to_string(), "device_id".to_string()];

        let err =
            validate_and_collect_partition_columns(&target_partition_columns, &column_schemas)
                .unwrap_err();

        assert!(err.to_string().contains("duplicate partition column"));
        assert!(err.to_string().contains("device_id"));
    }

    fn create_expr_from_sql(sql: &str) -> CreateTableExpr {
        let result =
            ParserContext::create_with_dialect(sql, &GreptimeDbDialect {}, ParseOptions::default())
                .unwrap();

        match &result[0] {
            Statement::CreateTable(create) => {
                expr_helper::create_to_expr(create, &QueryContext::arc()).unwrap()
            }
            _ => unreachable!(),
        }
    }

    #[test]
    fn test_create_table_with_repartition_partition_num_hint() {
        let key = table::requests::REPARTITION_PARTITION_NUM_HINT_KEY;
        for (value, expected) in [
            (" 8 ", Some("8")),
            ("0", None),
            ("-1", None),
            ("4294967296", None),
        ] {
            let expr = create_expr_from_sql(&format!(
                "CREATE TABLE metrics (host STRING, ts TIMESTAMP TIME INDEX, PRIMARY KEY(host)) WITH ('{key}' = '{value}')"
            ));
            let result = create_table_info(&expr, vec![]);
            if let Some(expected) = expected {
                let info = result.unwrap();
                assert_eq!(
                    info.meta.options.extra_options.get(key).map(String::as_str),
                    Some(expected)
                );
            } else {
                assert!(
                    result
                        .unwrap_err()
                        .to_string()
                        .contains("expects a positive integer")
                );
            }
        }
    }

    #[test]
    fn test_create_table_with_repartition_column_hint() {
        let expr = create_expr_from_sql(
            r"
CREATE TABLE metrics (
  host STRING,
  ts TIMESTAMP TIME INDEX,
  cpu DOUBLE,
  PRIMARY KEY(host)
)
WITH ('repartition.column.hint' = ' host ')",
        );

        let table_info = create_table_info(&expr, vec![]).unwrap();
        assert_eq!(
            table_info
                .meta
                .options
                .extra_options
                .get(REPARTITION_COLUMN_HINT_KEY),
            Some(&"host".to_string())
        );
    }

    #[test]
    fn test_create_table_info_rejects_non_timestamp_time_index() {
        let expr = CreateTableExpr {
            catalog_name: "greptime".to_string(),
            schema_name: "public".to_string(),
            table_name: "demo".to_string(),
            desc: String::new(),
            column_defs: vec![api::v1::ColumnDef {
                name: "host".to_string(),
                data_type: api::v1::ColumnDataType::String as i32,
                is_nullable: true,
                default_constraint: vec![],
                semantic_type: 0,
                comment: String::new(),
                datatype_extension: None,
                options: None,
            }],
            time_index: "host".to_string(),
            primary_keys: vec![],
            create_if_not_exists: false,
            table_options: HashMap::new(),
            table_id: None,
            engine: "mito".to_string(),
        };

        let err = create_table_info(&expr, vec![]).unwrap_err();
        assert_eq!(
            common_error::ext::ErrorExt::status_code(&err),
            common_error::status_code::StatusCode::InvalidArguments
        );
    }

    #[test]
    fn test_verify_alter_guards_entity_column_types() {
        let expr = create_expr_from_sql(
            "CREATE TABLE t (svc STRING, ts TIMESTAMP TIME INDEX) \
             WITH ('greptime.semantic.entity.service.id'='svc');",
        );
        let table_info = Arc::new(create_table_info(&expr, vec![]).unwrap());
        let alter = |kind| AlterTableExpr {
            catalog_name: "greptime".to_string(),
            schema_name: "public".to_string(),
            table_name: "t".to_string(),
            kind: Some(kind),
        };

        let modify = alter(Kind::ModifyColumnTypes(api::v1::ModifyColumnTypes {
            modify_column_types: vec![api::v1::ModifyColumnType {
                column_name: "svc".to_string(),
                target_type: api::v1::ColumnDataType::Binary as i32,
                target_type_extension: None,
            }],
        }));
        let err = verify_alter(1, table_info.clone(), modify).unwrap_err();
        let msg = common_error::ext::ErrorExt::output_msg(&err);
        assert!(
            msg.contains("must keep a type that renders as a string"),
            "{msg}"
        );

        // Dropping a declared column stays allowed: the derivation skips the
        // stale declaration.
        let drop = alter(Kind::DropColumns(api::v1::DropColumns {
            drop_columns: vec![api::v1::DropColumn {
                name: "svc".to_string(),
            }],
        }));
        assert!(verify_alter(1, table_info, drop).unwrap());
    }

    #[test]
    fn test_json2_supports_non_append_mode() -> std::result::Result<(), Box<dyn std::error::Error>>
    {
        for options in [
            "",
            "WITH (append_mode='false')",
            "WITH (append_mode='true')",
            "WITH (merge_mode='last_non_null')",
        ] {
            let expr = create_expr_from_sql(&format!(
                "CREATE TABLE monitor (payload JSON2, ts TIMESTAMP TIME INDEX) {options};"
            ));
            create_table_info(&expr, vec![])?;
        }
        Ok(())
    }

    #[test]
    fn test_json2_cannot_be_primary_key() -> std::result::Result<(), Box<dyn std::error::Error>> {
        let expr = create_expr_from_sql(
            "CREATE TABLE monitor (payload JSON2, ts TIMESTAMP TIME INDEX, PRIMARY KEY(payload));",
        );
        let err = create_table_info(&expr, vec![])
            .err()
            .ok_or("expected JSON2 primary key rejection")?;
        assert!(
            err.to_string()
                .contains("JSON2 column `payload` cannot be a primary key"),
            "{err}"
        );
        Ok(())
    }

    #[test]
    fn test_create_table_with_empty_repartition_column_hint() {
        let expr = create_expr_from_sql(
            r"
CREATE TABLE metrics (
  host STRING,
  ts TIMESTAMP TIME INDEX,
  cpu DOUBLE,
  PRIMARY KEY(host)
)
WITH ('repartition.column.hint' = '')",
        );

        let err = create_table_info(&expr, vec![]).unwrap_err();
        assert!(
            err.to_string()
                .contains("repartition.column.hint expects exactly one column name")
        );
    }

    #[test]
    fn test_create_table_with_multiple_repartition_column_hints() {
        let expr = create_expr_from_sql(
            r"
CREATE TABLE metrics (
  host STRING,
  region_id STRING,
  ts TIMESTAMP TIME INDEX,
  cpu DOUBLE,
  PRIMARY KEY(host)
)
WITH ('repartition.column.hint' = 'host,region_id')",
        );

        let err = create_table_info(&expr, vec![]).unwrap_err();
        assert!(
            err.to_string()
                .contains("repartition.column.hint expects exactly one column name")
        );
    }

    #[test]
    fn test_create_table_with_missing_repartition_column_hint() {
        let expr = create_expr_from_sql(
            r"
CREATE TABLE metrics (
  host STRING,
  ts TIMESTAMP TIME INDEX,
  cpu DOUBLE,
  PRIMARY KEY(host)
)
WITH ('repartition.column.hint' = 'region_id')",
        );

        let err = create_table_info(&expr, vec![]).unwrap_err();
        assert!(
            err.to_string()
                .contains("Cannot find column by name: region")
        );
    }

    #[test]
    fn test_create_table_with_time_index_repartition_column_hint() {
        let expr = create_expr_from_sql(
            r"
CREATE TABLE metrics (
  host STRING,
  ts TIMESTAMP TIME INDEX,
  cpu DOUBLE,
  PRIMARY KEY(host)
)
WITH ('repartition.column.hint' = 'ts')",
        );

        let err = create_table_info(&expr, vec![]).unwrap_err();
        assert!(
            err.to_string()
                .contains("cannot set repartition.column.hint to the time index column")
        );
    }

    #[test]
    fn test_create_partitioned_table_with_repartition_column_hint() {
        let expr = create_expr_from_sql(
            r"
CREATE TABLE metrics (
  host STRING,
  ts TIMESTAMP TIME INDEX,
  cpu DOUBLE,
  PRIMARY KEY(host)
)
WITH ('repartition.column.hint' = 'host')",
        );

        let err = create_table_info(&expr, vec!["host".to_string()]).unwrap_err();
        assert!(
            err.to_string()
                .contains("cannot set repartition.column.hint on a table with partition metadata")
        );
    }
}