faucet_sink_mysql/

sink.rs

//! MySQL sink implementation.

use crate::config::{MysqlColumnMapping, MysqlSinkConfig};
use async_trait::async_trait;
use faucet_core::{FaucetError, SchemaEvolution, SqlBaseType, json_schema_base_type};
use serde_json::Value;
use sqlx::mysql::MySqlPoolOptions;
use sqlx::{MySqlConnection, MySqlPool, Row};

/// A sink that writes JSON records to a MySQL table.
pub struct MysqlSink {
    config: MysqlSinkConfig,
    pool: MySqlPool,
}

/// Quote a MySQL identifier using backticks.
///
/// Wraps the name in backticks and escapes any embedded backticks by doubling
/// them, per MySQL convention.
fn quote_ident_mysql(name: &str) -> String {
    format!("`{}`", name.replace('`', "``"))
}

/// Map a [`SqlBaseType`] to the MySQL type keyword used when adding/widening a
/// column during schema evolution (issue #194). Integers always widen to
/// `BIGINT` and floats to `DOUBLE` so a later, wider value never overflows a
/// narrower column. `Text` maps to `LONGTEXT` (the widest text type, so a long
/// value never truncates) and `Json` to MySQL's native `JSON`.
fn mysql_keyword(t: SqlBaseType) -> &'static str {
    match t {
        SqlBaseType::Integer => "BIGINT",
        SqlBaseType::Double => "DOUBLE",
        SqlBaseType::Boolean => "TINYINT(1)",
        SqlBaseType::Text => "LONGTEXT",
        SqlBaseType::Json => "JSON",
    }
}

/// `ALTER TABLE <table> ADD COLUMN `col` <kw>` — column addition.
///
/// MySQL (pre-8.0.x) has no `ADD COLUMN IF NOT EXISTS`, so idempotency is
/// achieved by the caller pre-checking the existing column set and only
/// emitting this for columns not already present. `table` is the already-quoted
/// table reference.
fn build_add_column_sql(table: &str, col: &str, t: SqlBaseType) -> String {
    format!(
        "ALTER TABLE {table} ADD COLUMN {} {}",
        quote_ident_mysql(col),
        mysql_keyword(t)
    )
}

/// `ALTER TABLE <table> MODIFY COLUMN `col` <kw>` — widen an existing column's
/// type. Naturally idempotent (re-running the same MODIFY is a no-op). `table`
/// is the already-quoted table reference.
fn build_modify_column_sql(table: &str, col: &str, t: SqlBaseType) -> String {
    format!(
        "ALTER TABLE {table} MODIFY COLUMN {} {}",
        quote_ident_mysql(col),
        mysql_keyword(t)
    )
}

/// Map a MySQL `INFORMATION_SCHEMA.COLUMNS.DATA_TYPE` value (lowercase, no
/// precision — e.g. `bigint`, `double`, `json`, `varchar`) back to a JSON-Schema
/// type fragment so [`MysqlSink::current_schema`] round-trips with
/// [`faucet_core::diff_schema`]. `nullable` reflects `IS_NULLABLE = 'YES'`.
///
/// Note: `INFORMATION_SCHEMA.DATA_TYPE` returns bare `tinyint` without the
/// `(1)` precision, so a `TINYINT(1)` (conventionally boolean) is
/// indistinguishable from a real `TINYINT` — both map to `integer` for safety.
fn mysql_data_type_to_json_schema(data_type: &str, nullable: bool) -> Value {
    let base = match data_type {
        "bigint" | "int" | "integer" | "smallint" | "mediumint" | "tinyint" => "integer",
        "double" | "float" | "decimal" | "numeric" => "number",
        "json" => "object",
        _ => "string",
    };
    if nullable {
        serde_json::json!({ "type": [base, "null"] })
    } else {
        serde_json::json!({ "type": base })
    }
}

/// Build the `ON DUPLICATE KEY UPDATE …` tail for an upsert INSERT.
///
/// MySQL's `ON DUPLICATE KEY UPDATE` does not name a conflict target — it
/// relies on the table's existing PRIMARY or UNIQUE key. Non-key columns are
/// set from `VALUES(col)`. If every column is a key column there is nothing to
/// update, so a self-assignment no-op on the first key column is emitted to
/// keep the statement syntactically valid.
/// Decide whether the configured upsert/delete `key` exactly corresponds to one
/// of the target table's PRIMARY/UNIQUE indexes.
///
/// MySQL's `INSERT … ON DUPLICATE KEY UPDATE` does not name a conflict target —
/// it resolves on *any* unique index present on the table. The unified
/// write-mode contract, however, treats the configured `key` as the
/// authoritative conflict target (`plan_writes` dedups and routes by exactly
/// that key). If the configured `key` does not match a real unique index, MySQL
/// would silently resolve the conflict on a *different* index, producing wrong
/// upsert results that the user cannot detect (finding F33). This check lets the
/// sink fail fast at construction instead.
///
/// `unique_indexes` is the set of the table's PRIMARY/UNIQUE indexes, each given
/// as the full set of its column names. `key` is the configured key columns.
/// The comparison is **order-insensitive** (a UNIQUE index on `(a, b)` matches a
/// key of `[b, a]`) and requires the **full** column set of some index to match
/// the key exactly — a prefix, subset, or superset does **not** match, because
/// `ON DUPLICATE KEY UPDATE` would then trigger on a broader or narrower index
/// than the one the pipeline deduped on.
fn key_matches_unique_index(
    unique_indexes: &[std::collections::BTreeSet<String>],
    key: &[String],
) -> bool {
    if key.is_empty() {
        return false;
    }
    let key_set: std::collections::BTreeSet<String> = key.iter().cloned().collect();
    unique_indexes.contains(&key_set)
}

fn on_duplicate_clause(key: &[String], all_cols: &[String]) -> String {
    let updates: Vec<String> = all_cols
        .iter()
        .filter(|c| !key.iter().any(|k| k == *c))
        .map(|c| {
            let q = quote_ident_mysql(c);
            format!("{q} = VALUES({q})")
        })
        .collect();
    if updates.is_empty() {
        let q = quote_ident_mysql(&key[0]);
        format!("ON DUPLICATE KEY UPDATE {q} = {q}")
    } else {
        format!("ON DUPLICATE KEY UPDATE {}", updates.join(", "))
    }
}

impl MysqlSink {
    /// Create a new MySQL sink. Establishes a connection pool.
    pub async fn new(config: MysqlSinkConfig) -> Result<Self, FaucetError> {
        config.write.validate()?;
        if !matches!(config.write.write_mode, faucet_core::WriteMode::Append)
            && !matches!(config.column_mapping, MysqlColumnMapping::AutoMap)
        {
            return Err(FaucetError::Config(
                "mysql sink: write_mode upsert/delete requires column_mapping: auto_map \
                 (key columns must be real columns, not inside a JSON blob)"
                    .into(),
            ));
        }

        let pool = MySqlPoolOptions::new()
            .max_connections(config.max_connections)
            .connect(&config.connection_url)
            .await
            .map_err(|e| FaucetError::Sink(format!("MySQL connection failed: {e}")))?;

        let sink = Self { config, pool };

        // For upsert/delete, MySQL's anonymous `ON DUPLICATE KEY UPDATE` /
        // `DELETE … WHERE (key) IN (…)` only resolves correctly when the
        // configured `key` matches a real PRIMARY/UNIQUE index. Assert that here
        // so a silent mismatch (finding F33) fails fast at construction.
        if !matches!(sink.config.write.write_mode, faucet_core::WriteMode::Append) {
            sink.assert_key_is_unique_index().await?;
        }

        Ok(sink)
    }

    /// Read the target table's PRIMARY/UNIQUE indexes from
    /// `INFORMATION_SCHEMA.STATISTICS`, each as the full set of its column names.
    ///
    /// `STATISTICS` lists one row per index column; only non-unique-flag-zero
    /// rows (`NON_UNIQUE = 0`) are unique indexes (the PRIMARY KEY is reported as
    /// an index named `PRIMARY` and is also `NON_UNIQUE = 0`). Returns an empty
    /// `Vec` when the table does not exist or has no unique indexes — the caller
    /// decides what to do with an absent table.
    ///
    /// Thin I/O shim; the pure decision is [`key_matches_unique_index`].
    async fn read_unique_indexes(
        &self,
    ) -> Result<Vec<std::collections::BTreeSet<String>>, FaucetError> {
        // INFORMATION_SCHEMA string columns use a binary collation that sqlx
        // decodes as Vec<u8>; CAST to CHAR so they decode as String.
        let rows = sqlx::query(
            "SELECT CAST(INDEX_NAME AS CHAR) AS INDEX_NAME, \
                    CAST(COLUMN_NAME AS CHAR) AS COLUMN_NAME \
             FROM INFORMATION_SCHEMA.STATISTICS \
             WHERE TABLE_NAME = ? AND TABLE_SCHEMA = DATABASE() AND NON_UNIQUE = 0 \
             ORDER BY INDEX_NAME, SEQ_IN_INDEX",
        )
        .bind(&self.config.table_name)
        .fetch_all(&self.pool)
        .await
        .map_err(|e| FaucetError::Sink(format!("failed to query table indexes: {e}")))?;

        let mut by_index: std::collections::BTreeMap<String, std::collections::BTreeSet<String>> =
            std::collections::BTreeMap::new();
        for row in &rows {
            let index_name: String = row.get("INDEX_NAME");
            let column_name: String = row.get("COLUMN_NAME");
            by_index.entry(index_name).or_default().insert(column_name);
        }
        Ok(by_index.into_values().collect())
    }

    /// Assert that the configured `key` exactly matches a PRIMARY/UNIQUE index on
    /// the target table, or fail with a clear typed [`FaucetError::Config`].
    ///
    /// **Table-absent behaviour:** if the table has no unique indexes — which is
    /// the case when it does not exist yet — the assertion is **skipped** with a
    /// warning, matching the rest of this sink which auto-discovers columns and
    /// lets the first write surface a missing-table error. (`current_schema`
    /// likewise returns `None` for an absent table.) The check is a guard against
    /// a *mismatched* index on an existing table, not a table-existence preflight.
    async fn assert_key_is_unique_index(&self) -> Result<(), FaucetError> {
        let unique_indexes = self.read_unique_indexes().await?;
        if unique_indexes.is_empty() {
            tracing::warn!(
                table = %self.config.table_name,
                "mysql sink: no PRIMARY/UNIQUE index found on target table (it may not exist \
                 yet); skipping upsert key validation — the first write will surface a \
                 missing-table or missing-constraint error"
            );
            return Ok(());
        }
        if !key_matches_unique_index(&unique_indexes, &self.config.write.key) {
            let available: Vec<String> = unique_indexes
                .iter()
                .map(|idx| {
                    let mut cols: Vec<&str> = idx.iter().map(String::as_str).collect();
                    cols.sort_unstable();
                    format!("({})", cols.join(", "))
                })
                .collect();
            return Err(FaucetError::Config(format!(
                "mysql sink: write_mode {} requires `key` {:?} to exactly match a PRIMARY KEY or \
                 UNIQUE index on table '{}' — MySQL's `ON DUPLICATE KEY UPDATE` resolves on the \
                 table's real unique indexes, so an unmatched key would silently upsert on the \
                 wrong index. Existing unique indexes: {}",
                self.config.write.write_mode.as_str(),
                self.config.write.key,
                self.config.table_name,
                available.join(", "),
            )));
        }
        Ok(())
    }

    /// Insert a batch of records using JSON column mode.
    ///
    /// Executes on the provided connection (a bare pool connection for
    /// `write_batch`, or a `&mut *tx` transaction for `write_batch_idempotent`).
    /// Uses a single multi-row INSERT for efficiency.
    async fn insert_json(
        &self,
        conn: &mut MySqlConnection,
        records: &[Value],
        column: &str,
    ) -> Result<usize, FaucetError> {
        if records.is_empty() {
            return Ok(0);
        }

        // Build multi-row INSERT: INSERT INTO t (col) VALUES (?), (?), ...
        let placeholders: Vec<&str> = records.iter().map(|_| "(?)").collect();
        let insert_sql = format!(
            "INSERT INTO {} ({}) VALUES {}",
            quote_ident_mysql(&self.config.table_name),
            quote_ident_mysql(column),
            placeholders.join(", ")
        );

        let mut q = sqlx::query(&insert_sql);
        for record in records {
            let json_str = serde_json::to_string(record)
                .map_err(|e| FaucetError::Sink(format!("failed to serialize record: {e}")))?;
            q = q.bind(json_str);
        }

        q.execute(&mut *conn)
            .await
            .map_err(|e| FaucetError::Sink(format!("MySQL insert failed: {e}")))?;

        Ok(records.len())
    }

    /// Core auto-map insert logic, optionally appending an `ON DUPLICATE KEY
    /// UPDATE …` clause when `conflict_key` is `Some`.
    ///
    /// Discovers column names from `INFORMATION_SCHEMA.COLUMNS` and maps
    /// top-level JSON fields to columns. Executes on the provided connection
    /// (a bare pool connection for `write_batch`, or `&mut *tx` for
    /// transactional paths). Uses sub-chunked multi-row INSERTs.
    ///
    /// When `conflict_key` is `Some(key)`, each sub-chunk's INSERT is given an
    /// `ON DUPLICATE KEY UPDATE …` tail so it upserts by the existing PRIMARY
    /// or UNIQUE key (last-write-wins within the batch is handled by the
    /// planner's dedup, so a single sub-chunk never double-hits the same
    /// conflict target).
    async fn insert_auto_map_with_conflict(
        &self,
        conn: &mut MySqlConnection,
        records: &[Value],
        conflict_key: Option<&[String]>,
    ) -> Result<usize, FaucetError> {
        if records.is_empty() {
            return Ok(0);
        }

        // Get column names from the table.
        let columns: Vec<String> = sqlx::query(
            "SELECT COLUMN_NAME FROM INFORMATION_SCHEMA.COLUMNS WHERE TABLE_NAME = ? AND TABLE_SCHEMA = DATABASE() ORDER BY ORDINAL_POSITION"
        )
        .bind(&self.config.table_name)
        .fetch_all(&mut *conn)
        .await
        .map_err(|e| FaucetError::Sink(format!("failed to query table columns: {e}")))?
        .iter()
        .map(|row| row.get::<String, _>("COLUMN_NAME"))
        .collect();

        if columns.is_empty() {
            return Err(FaucetError::Sink(format!(
                "table '{}' has no columns or does not exist",
                self.config.table_name
            )));
        }

        // Pre-validate all records and collect matched column values. The
        // INSERT column set is the UNION of table columns present in ANY record
        // (in declared table order), not just the first record's keys —
        // otherwise a field present only in a later record of the batch would be
        // silently dropped (audit #146 H1). A row missing a unioned column binds
        // SQL NULL.
        let mut matched_rows: Vec<Vec<(&String, &Value)>> = Vec::with_capacity(records.len());
        let mut used: std::collections::HashSet<&str> = std::collections::HashSet::new();

        for record in records {
            let obj = record
                .as_object()
                .ok_or_else(|| FaucetError::Sink("AutoMap requires JSON object records".into()))?;

            let matching: Vec<(&String, &Value)> = columns
                .iter()
                .filter_map(|col| obj.get(col).map(|v| (col, v)))
                .collect();

            if matching.is_empty() {
                tracing::warn!(
                    record_keys = ?obj.keys().collect::<Vec<_>>(),
                    table_columns = ?columns,
                    "record has no keys matching table columns, skipping"
                );
                continue;
            }

            for (c, _) in &matching {
                used.insert(c.as_str());
            }
            matched_rows.push(matching);
        }

        if matched_rows.is_empty() {
            return Ok(0);
        }

        // Table columns (in declared order) that appear in at least one record.
        let insert_columns: Vec<String> = columns
            .iter()
            .filter(|c| used.contains(c.as_str()))
            .cloned()
            .collect();

        let num_cols = insert_columns.len();
        let num_rows = matched_rows.len();
        let col_names: Vec<String> = insert_columns
            .iter()
            .map(|c| quote_ident_mysql(c))
            .collect();

        // MySQL caps prepared-statement placeholders at 65535. A multi-row
        // INSERT binds `rows × num_cols`, so a wide table at a large batch_size
        // overflows and fails at runtime; split into sub-INSERTs of at most
        // floor(MAX / num_cols) rows (audit #146 H14 — postgres/sqlite/mssql
        // already sub-chunk this way).
        const MAX_MYSQL_PARAMS: usize = 65535;
        let max_rows_per_insert = (MAX_MYSQL_PARAMS / num_cols).max(1);

        for sub in matched_rows.chunks(max_rows_per_insert) {
            // Build multi-row VALUES clause: (?, ?), (?, ?), ...
            let row_placeholder = format!("({})", vec!["?"; num_cols].join(", "));
            let value_tuples: Vec<&str> =
                (0..sub.len()).map(|_| row_placeholder.as_str()).collect();
            let base_query = format!(
                "INSERT INTO {} ({}) VALUES {}",
                quote_ident_mysql(&self.config.table_name),
                col_names.join(", "),
                value_tuples.join(", ")
            );
            let query = match conflict_key {
                Some(key) => format!("{base_query} {}", on_duplicate_clause(key, &insert_columns)),
                None => base_query,
            };

            let mut q = sqlx::query(&query);
            for matched in sub {
                for col in &insert_columns {
                    let val = matched.iter().find(|(c, _)| *c == col).map(|(_, v)| *v);
                    // Bind native MySQL types. Binding every value as a JSON string
                    // (the old behaviour) stored `"Bob"` with embedded quotes,
                    // turned `true` into the text "true", and bound the literal
                    // text "null" for absent columns instead of SQL NULL (#78/#4).
                    q = match val {
                        None | Some(Value::Null) => q.bind(None::<String>),
                        Some(Value::Bool(b)) => q.bind(*b),
                        Some(Value::Number(n)) => {
                            if let Some(i) = n.as_i64() {
                                q.bind(i)
                            } else if let Some(f) = n.as_f64() {
                                q.bind(f)
                            } else {
                                // u64 above i64::MAX — preserve exact text.
                                q.bind(n.to_string())
                            }
                        }
                        Some(Value::String(s)) => q.bind(s.clone()),
                        // Arrays/objects have no scalar SQL representation — store
                        // their JSON text (suitable for TEXT / JSON columns).
                        Some(v) => q.bind(v.to_string()),
                    };
                }
            }

            q.execute(&mut *conn)
                .await
                .map_err(|e| FaucetError::Sink(format!("MySQL insert failed: {e}")))?;
        }

        Ok(num_rows)
    }

    /// Auto-map insert with plain append semantics (no `ON DUPLICATE KEY`
    /// clause). Thin wrapper over
    /// [`insert_auto_map_with_conflict`](Self::insert_auto_map_with_conflict)
    /// so the append path and `write_batch_idempotent` keep their original
    /// signature.
    async fn insert_auto_map(
        &self,
        conn: &mut MySqlConnection,
        records: &[Value],
    ) -> Result<usize, FaucetError> {
        self.insert_auto_map_with_conflict(conn, records, None)
            .await
    }

    /// Delete rows whose key columns match any of `deletes`, using
    /// `DELETE FROM t WHERE (k1, …) IN ((?, …), …)`, chunked at MySQL's
    /// 65535-placeholder limit. Runs inside the caller's transaction.
    async fn delete_by_keys(
        &self,
        conn: &mut MySqlConnection,
        deletes: &[faucet_core::KeyTuple],
    ) -> Result<usize, FaucetError> {
        if deletes.is_empty() {
            return Ok(0);
        }
        let key = &self.config.write.key;
        let table_ref = quote_ident_mysql(&self.config.table_name);
        let col_list = key
            .iter()
            .map(|k| quote_ident_mysql(k))
            .collect::<Vec<_>>()
            .join(", ");

        const MAX_MYSQL_PARAMS: usize = 65535;
        let per = (MAX_MYSQL_PARAMS / key.len().max(1)).max(1);
        let mut total = 0usize;

        for chunk in deletes.chunks(per) {
            let tuples: Vec<String> = chunk
                .iter()
                .map(|_| format!("({})", vec!["?"; key.len()].join(", ")))
                .collect();
            let sql = format!(
                "DELETE FROM {table_ref} WHERE ({col_list}) IN ({})",
                tuples.join(", ")
            );
            let mut q = sqlx::query(&sql);
            for kt in chunk {
                for (_, v) in &kt.0 {
                    // Bind native MySQL types — same logic as in the INSERT path.
                    q = match v {
                        Value::Null => q.bind(None::<String>),
                        Value::Bool(b) => q.bind(*b),
                        Value::Number(n) => {
                            if let Some(i) = n.as_i64() {
                                q.bind(i)
                            } else if let Some(f) = n.as_f64() {
                                q.bind(f)
                            } else {
                                q.bind(n.to_string())
                            }
                        }
                        Value::String(s) => q.bind(s.clone()),
                        other => q.bind(other.to_string()),
                    };
                }
            }
            let res = q
                .execute(&mut *conn)
                .await
                .map_err(|e| FaucetError::Sink(format!("MySQL delete failed: {e}")))?;
            total += res.rows_affected() as usize;
        }
        Ok(total)
    }

    /// Apply a planned upsert/delete batch inside one `BEGIN`/`COMMIT`
    /// transaction. Upserts and deletes are wrapped together so they commit
    /// atomically.
    async fn apply_plan(&self, plan: &faucet_core::WritePlan) -> Result<usize, FaucetError> {
        let mut tx = self
            .pool
            .begin()
            .await
            .map_err(|e| FaucetError::Sink(format!("MySQL transaction begin failed: {e}")))?;

        let mut affected = 0usize;
        if !plan.upserts.is_empty() {
            affected += self
                .insert_auto_map_with_conflict(&mut tx, &plan.upserts, Some(&self.config.write.key))
                .await?;
        }
        if !plan.deletes.is_empty() {
            affected += self.delete_by_keys(&mut tx, &plan.deletes).await?;
        }

        tx.commit()
            .await
            .map_err(|e| FaucetError::Sink(format!("MySQL transaction commit failed: {e}")))?;
        Ok(affected)
    }

    /// Read the live destination columns as `(name, data_type, nullable)`
    /// tuples, in declared order, from `INFORMATION_SCHEMA.COLUMNS`. An empty
    /// vector means the table does not exist (or has no columns).
    ///
    /// Shared by [`current_schema`](faucet_core::Sink::current_schema) and
    /// [`evolve_schema`](faucet_core::Sink::evolve_schema) — the latter needs the
    /// current set to make `ADD COLUMN` idempotent (MySQL lacks
    /// `ADD COLUMN IF NOT EXISTS`).
    async fn read_columns(&self) -> Result<Vec<(String, String, bool)>, FaucetError> {
        // MySQL's INFORMATION_SCHEMA string columns are typed as a binary/blob
        // collation, which sqlx decodes as `Vec<u8>` rather than `String`; cast
        // each to CHAR so they decode as `String`.
        let rows = sqlx::query(
            "SELECT CAST(COLUMN_NAME AS CHAR) AS COLUMN_NAME, \
                    CAST(DATA_TYPE AS CHAR) AS DATA_TYPE, \
                    CAST(IS_NULLABLE AS CHAR) AS IS_NULLABLE \
             FROM INFORMATION_SCHEMA.COLUMNS \
             WHERE TABLE_NAME = ? AND TABLE_SCHEMA = DATABASE() ORDER BY ORDINAL_POSITION",
        )
        .bind(&self.config.table_name)
        .fetch_all(&self.pool)
        .await
        .map_err(|e| FaucetError::Sink(format!("failed to query table columns: {e}")))?;

        Ok(rows
            .iter()
            .map(|row| {
                (
                    row.get::<String, _>("COLUMN_NAME"),
                    // DATA_TYPE is already lowercase per the SQL standard, but
                    // normalize defensively.
                    row.get::<String, _>("DATA_TYPE").to_ascii_lowercase(),
                    row.get::<String, _>("IS_NULLABLE")
                        .eq_ignore_ascii_case("YES"),
                )
            })
            .collect())
    }

    /// Create the commit-token watermark table if it does not yet exist.
    ///
    /// The table holds one row per pipeline scope (state key). MySQL requires a
    /// fixed-length column as the primary key, so `scope` is `VARCHAR(255)` and
    /// `token` is `VARCHAR(32)` (tokens are 20-char ulids, 32 chars is ample).
    async fn ensure_commit_table(&self) -> Result<(), FaucetError> {
        let sql = format!(
            "CREATE TABLE IF NOT EXISTS {t} ({s} VARCHAR(255) PRIMARY KEY, {k} VARCHAR(32) NOT NULL, updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP)",
            t = quote_ident_mysql(faucet_core::idempotency::COMMIT_TOKEN_TABLE),
            s = quote_ident_mysql(faucet_core::idempotency::COMMIT_TOKEN_SCOPE_COL),
            k = quote_ident_mysql(faucet_core::idempotency::COMMIT_TOKEN_TOKEN_COL),
        );
        sqlx::query(&sql)
            .execute(&self.pool)
            .await
            .map_err(|e| FaucetError::Sink(format!("MySQL commit-table create failed: {e}")))?;
        Ok(())
    }
}

#[async_trait]
impl faucet_core::Sink for MysqlSink {
    fn config_schema(&self) -> serde_json::Value {
        serde_json::to_value(faucet_core::schema_for!(MysqlSinkConfig))
            .expect("schema serialization")
    }

    fn dataset_uri(&self) -> String {
        format!(
            "{}?table={}",
            faucet_core::redact_uri_credentials(&self.config.connection_url),
            self.config.table_name
        )
    }

    /// Preflight connectivity probe (`faucet doctor`).
    ///
    /// Acquires a connection from the existing pool and runs `SELECT 1`. This
    /// is non-mutating and idempotent — it validates that the database is
    /// reachable and the credentials are accepted without writing anything.
    async fn check(
        &self,
        ctx: &faucet_core::check::CheckContext,
    ) -> Result<faucet_core::check::CheckReport, FaucetError> {
        use faucet_core::check::{CheckReport, Probe};

        let started = std::time::Instant::now();
        let probe =
            match tokio::time::timeout(ctx.timeout, sqlx::query("SELECT 1").execute(&self.pool))
                .await
            {
                Ok(Ok(_)) => Probe::pass("auth", started.elapsed()),
                Ok(Err(e)) => Probe::fail_hint(
                    "auth",
                    started.elapsed(),
                    e.to_string(),
                    "check connection_url / credentials / that the database is reachable",
                ),
                Err(_) => Probe::fail_hint(
                    "auth",
                    started.elapsed(),
                    "timed out",
                    "check connection_url / credentials / that the database is reachable",
                ),
            };
        Ok(CheckReport::single(probe))
    }

    fn supported_write_modes(&self) -> &'static [faucet_core::WriteMode] {
        &[
            faucet_core::WriteMode::Append,
            faucet_core::WriteMode::Upsert,
            faucet_core::WriteMode::Delete,
        ]
    }

    fn dedups_by_key(&self) -> bool {
        self.config.write.dedups_by_key()
    }

    fn supports_schema_evolution(&self) -> bool {
        true
    }

    /// Read the live destination schema from `INFORMATION_SCHEMA.COLUMNS` as an
    /// `infer_schema`-shaped object (`{"type":"object","properties":{…}}`), or
    /// `None` when the target table does not exist yet (issue #194).
    ///
    /// The MySQL database is implicit (`DATABASE()`), so there is no schema
    /// qualifier to thread. `DATA_TYPE` / `IS_NULLABLE` round-trip through
    /// `mysql_data_type_to_json_schema`.
    async fn current_schema(&self) -> Result<Option<serde_json::Value>, FaucetError> {
        let columns = self.read_columns().await?;
        if columns.is_empty() {
            return Ok(None); // table does not exist yet
        }

        let mut props = serde_json::Map::new();
        for (name, data_type, nullable) in columns {
            props.insert(name, mysql_data_type_to_json_schema(&data_type, nullable));
        }
        Ok(Some(
            serde_json::json!({ "type": "object", "properties": props }),
        ))
    }

    /// Apply an additive schema evolution (new columns, lossless widenings,
    /// nullability relaxations) to the destination table (issue #194).
    ///
    /// MySQL has no `ADD COLUMN IF NOT EXISTS` (pre-8.0.x), so the current
    /// column set is read first and an `ADD COLUMN` is emitted only for names
    /// not already present — making re-runs idempotent. Widenings use
    /// `MODIFY COLUMN` (re-running the same MODIFY is a no-op); nullability
    /// relaxations re-emit the column as its current mapped type with an
    /// explicit `NULL`.
    async fn evolve_schema(&self, evolution: &SchemaEvolution) -> Result<(), FaucetError> {
        let table_ref = quote_ident_mysql(&self.config.table_name);

        // Read the current columns up front: needed for ADD-COLUMN idempotency
        // (pre-check by name) and to derive a column's existing type when
        // relaxing nullability.
        let current = self.read_columns().await?;
        let existing: std::collections::HashSet<&str> =
            current.iter().map(|(n, _, _)| n.as_str()).collect();

        let mut conn = self
            .pool
            .acquire()
            .await
            .map_err(|e| FaucetError::Sink(format!("MySQL evolve acquire failed: {e}")))?;

        for c in &evolution.additions {
            // Idempotency: MySQL lacks ADD COLUMN IF NOT EXISTS, so skip a
            // column that already exists rather than erroring on a re-run.
            if existing.contains(c.name.as_str()) {
                continue;
            }
            let t = json_schema_base_type(&c.to).unwrap_or(SqlBaseType::Text);
            sqlx::query(&build_add_column_sql(&table_ref, &c.name, t))
                .execute(&mut *conn)
                .await
                .map_err(|e| {
                    FaucetError::Sink(format!("MySQL ADD COLUMN {} failed: {e}", c.name))
                })?;
        }

        for c in &evolution.widenings {
            let t = json_schema_base_type(&c.to).unwrap_or(SqlBaseType::Text);
            sqlx::query(&build_modify_column_sql(&table_ref, &c.name, t))
                .execute(&mut *conn)
                .await
                .map_err(|e| {
                    FaucetError::Sink(format!("MySQL MODIFY COLUMN {} failed: {e}", c.name))
                })?;
        }

        for col in &evolution.relax_nullability {
            // Re-emit the column as its CURRENT type but explicitly nullable.
            // MySQL's MODIFY COLUMN requires the full type spec, so map the
            // column's existing DATA_TYPE back to a base type and re-emit it.
            let existing_type = current
                .iter()
                .find(|(n, _, _)| n == col)
                .map(|(_, dt, nullable)| {
                    let fragment = mysql_data_type_to_json_schema(dt, *nullable);
                    json_schema_base_type(&fragment).unwrap_or(SqlBaseType::Text)
                })
                .unwrap_or(SqlBaseType::Text);
            let sql = format!(
                "ALTER TABLE {table_ref} MODIFY COLUMN {} {} NULL",
                quote_ident_mysql(col),
                mysql_keyword(existing_type)
            );
            sqlx::query(&sql)
                .execute(&mut *conn)
                .await
                .map_err(|e| FaucetError::Sink(format!("MySQL DROP NOT NULL {col} failed: {e}")))?;
        }

        Ok(())
    }

    /// Write records to MySQL.
    ///
    /// When `config.batch_size > 0` and the input slice is larger than
    /// `batch_size`, the slice is split into chunks of `batch_size` rows and
    /// each chunk is sent as a separate multi-row `INSERT`. When
    /// `config.batch_size == 0`, the entire slice is sent in a single
    /// multi-row `INSERT` — useful when upstream `StreamPage`s are already
    /// sized for MySQL's `max_allowed_packet` limit.
    ///
    /// Acquires a single connection from the pool and sends every chunk
    /// through it under autocommit (no explicit transaction), preserving the
    /// pre-refactor observable behaviour.
    async fn write_batch(&self, records: &[Value]) -> Result<usize, FaucetError> {
        if records.is_empty() {
            return Ok(0);
        }

        // Non-append modes: plan the writes and apply atomically.
        if !matches!(self.config.write.write_mode, faucet_core::WriteMode::Append) {
            let plan = faucet_core::plan_writes(records, &self.config.write);
            if let Some((idx, msg)) = plan.failed.first() {
                return Err(FaucetError::Sink(format!(
                    "mysql {}: row {idx}: {msg}",
                    self.config.write.write_mode.as_str()
                )));
            }
            return self.apply_plan(&plan).await;
        }

        let mut conn = self
            .pool
            .acquire()
            .await
            .map_err(|e| FaucetError::Sink(format!("MySQL pool acquire failed: {e}")))?;

        let chunks: Vec<&[Value]> = if self.config.batch_size == 0 {
            // Sentinel: pass the entire upstream page through in a single
            // multi-row INSERT. Subject to MySQL's max_allowed_packet
            // (default 64MB).
            vec![records]
        } else {
            records.chunks(self.config.batch_size).collect()
        };

        let mut total = 0;
        for chunk in chunks {
            total += match &self.config.column_mapping {
                MysqlColumnMapping::Json { column } => {
                    self.insert_json(&mut conn, chunk, column).await?
                }
                MysqlColumnMapping::AutoMap => self.insert_auto_map(&mut conn, chunk).await?,
            };
        }

        tracing::info!(
            table = %self.config.table_name,
            rows = total,
            "MySQL write complete"
        );
        Ok(total)
    }

    /// Write a batch and report per-row outcomes.
    ///
    /// In append mode this delegates to [`write_batch`](faucet_core::Sink::write_batch) and
    /// maps a single success onto an all-`Ok(())` vector (the trait default).
    /// In upsert/delete mode the good rows are applied (upserts + deletes), and
    /// only the rows whose key could not be extracted (missing / null key) are
    /// reported as `Err` so the pipeline routes them to the DLQ per-row instead
    /// of sending the whole page.
    async fn write_batch_partial(
        &self,
        records: &[Value],
    ) -> Result<Vec<faucet_core::RowOutcome>, FaucetError> {
        if matches!(self.config.write.write_mode, faucet_core::WriteMode::Append) {
            self.write_batch(records).await?;
            return Ok(records.iter().map(|_| Ok(())).collect());
        }

        let plan = faucet_core::plan_writes(records, &self.config.write);
        self.apply_plan(&plan).await?;

        let mut outcomes: Vec<faucet_core::RowOutcome> = records.iter().map(|_| Ok(())).collect();
        for (idx, msg) in &plan.failed {
            outcomes[*idx] = Err(FaucetError::Sink(format!(
                "mysql {}: {msg}",
                self.config.write.write_mode.as_str()
            )));
        }
        Ok(outcomes)
    }

    fn supports_idempotent_writes(&self) -> bool {
        true
    }

    async fn last_committed_token(&self, scope: &str) -> Result<Option<String>, FaucetError> {
        self.ensure_commit_table().await?;
        let sql = format!(
            "SELECT {k} FROM {t} WHERE {s} = ?",
            t = quote_ident_mysql(faucet_core::idempotency::COMMIT_TOKEN_TABLE),
            k = quote_ident_mysql(faucet_core::idempotency::COMMIT_TOKEN_TOKEN_COL),
            s = quote_ident_mysql(faucet_core::idempotency::COMMIT_TOKEN_SCOPE_COL),
        );
        let row = sqlx::query(&sql)
            .bind(scope)
            .fetch_optional(&self.pool)
            .await
            .map_err(|e| FaucetError::Sink(format!("MySQL token read failed: {e}")))?;
        Ok(row.map(|r| r.get::<String, _>(0)))
    }

    async fn write_batch_idempotent(
        &self,
        records: &[Value],
        scope: &str,
        token: &str,
    ) -> Result<usize, FaucetError> {
        self.ensure_commit_table().await?;

        // For upsert/delete modes, plan the page before opening the transaction
        // so a key-extraction failure aborts without leaving an open tx.
        let plan = if matches!(self.config.write.write_mode, faucet_core::WriteMode::Append) {
            None
        } else {
            let plan = faucet_core::plan_writes(records, &self.config.write);
            if let Some((idx, msg)) = plan.failed.first() {
                return Err(FaucetError::Sink(format!(
                    "mysql {}: row {idx}: {msg}",
                    self.config.write.write_mode.as_str()
                )));
            }
            Some(plan)
        };

        let mut tx = self
            .pool
            .begin()
            .await
            .map_err(|e| FaucetError::Sink(format!("MySQL transaction begin failed: {e}")))?;

        // Data write and the commit-token upsert share ONE transaction so the
        // page is committed atomically with its watermark. For upsert/delete the
        // planned upserts/deletes commit together with the watermark in this same
        // tx (no nested tx — the helpers run on this transaction's connection).
        let written = match &plan {
            Some(plan) => {
                let mut affected = 0usize;
                if !plan.upserts.is_empty() {
                    affected += self
                        .insert_auto_map_with_conflict(
                            &mut tx,
                            &plan.upserts,
                            Some(&self.config.write.key),
                        )
                        .await?;
                }
                if !plan.deletes.is_empty() {
                    affected += self.delete_by_keys(&mut tx, &plan.deletes).await?;
                }
                affected
            }
            None => match &self.config.column_mapping {
                MysqlColumnMapping::Json { column } => {
                    self.insert_json(&mut tx, records, column).await?
                }
                MysqlColumnMapping::AutoMap => self.insert_auto_map(&mut tx, records).await?,
            },
        };

        let upsert = format!(
            "INSERT INTO {t} ({s}, {k}) VALUES (?, ?) ON DUPLICATE KEY UPDATE {k} = VALUES({k})",
            t = quote_ident_mysql(faucet_core::idempotency::COMMIT_TOKEN_TABLE),
            s = quote_ident_mysql(faucet_core::idempotency::COMMIT_TOKEN_SCOPE_COL),
            k = quote_ident_mysql(faucet_core::idempotency::COMMIT_TOKEN_TOKEN_COL),
        );
        sqlx::query(&upsert)
            .bind(scope)
            .bind(token)
            .execute(&mut *tx)
            .await
            .map_err(|e| FaucetError::Sink(format!("MySQL token upsert failed: {e}")))?;

        tx.commit()
            .await
            .map_err(|e| FaucetError::Sink(format!("MySQL transaction commit failed: {e}")))?;

        Ok(written)
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    // dataset_uri test is skipped: MysqlSink::new() requires a live pool
    // (connects to MySQL in new()), and no offline constructor exists.

    #[test]
    fn commit_token_table_is_the_shared_constant() {
        assert_eq!(
            faucet_core::idempotency::COMMIT_TOKEN_TABLE,
            "_faucet_commit_token"
        );
    }

    #[test]
    fn quote_ident_mysql_simple() {
        assert_eq!(quote_ident_mysql("my_table"), "`my_table`");
    }

    #[test]
    fn quote_ident_mysql_with_backtick() {
        assert_eq!(quote_ident_mysql("has`tick"), "`has``tick`");
    }

    #[test]
    fn quote_ident_mysql_empty() {
        assert_eq!(quote_ident_mysql(""), "``");
    }

    #[test]
    fn quote_ident_mysql_special_chars() {
        assert_eq!(quote_ident_mysql("table; DROP"), "`table; DROP`");
    }

    #[test]
    fn mysql_on_duplicate_clause() {
        let clause =
            on_duplicate_clause(&["id".to_string()], &["id".to_string(), "name".to_string()]);
        assert_eq!(clause, "ON DUPLICATE KEY UPDATE `name` = VALUES(`name`)");
    }

    #[test]
    fn mysql_on_duplicate_all_keys_self_assign() {
        let clause = on_duplicate_clause(&["id".to_string()], &["id".to_string()]);
        assert_eq!(clause, "ON DUPLICATE KEY UPDATE `id` = `id`");
    }

    #[test]
    fn mysql_on_duplicate_composite_key_partial_update() {
        let clause = on_duplicate_clause(
            &["a".to_string(), "b".to_string()],
            &["a".to_string(), "b".to_string(), "v".to_string()],
        );
        assert_eq!(clause, "ON DUPLICATE KEY UPDATE `v` = VALUES(`v`)");
    }

    #[test]
    fn mysql_add_column_ddl() {
        let sql = build_add_column_sql("`t`", "email", SqlBaseType::Text);
        assert_eq!(sql, "ALTER TABLE `t` ADD COLUMN `email` LONGTEXT");

        // A backtick in the column name is doubled (SQL-injection safety).
        let sql = build_add_column_sql("`t`", "ev`il", SqlBaseType::Integer);
        assert_eq!(sql, "ALTER TABLE `t` ADD COLUMN `ev``il` BIGINT");
    }

    #[test]
    fn mysql_modify_column_ddl() {
        let sql = build_modify_column_sql("`t`", "score", SqlBaseType::Double);
        assert_eq!(sql, "ALTER TABLE `t` MODIFY COLUMN `score` DOUBLE");

        let sql = build_modify_column_sql("`t`", "flag", SqlBaseType::Boolean);
        assert_eq!(sql, "ALTER TABLE `t` MODIFY COLUMN `flag` TINYINT(1)");
    }

    #[test]
    fn mysql_keyword_mapping() {
        assert_eq!(mysql_keyword(SqlBaseType::Integer), "BIGINT");
        assert_eq!(mysql_keyword(SqlBaseType::Double), "DOUBLE");
        assert_eq!(mysql_keyword(SqlBaseType::Boolean), "TINYINT(1)");
        assert_eq!(mysql_keyword(SqlBaseType::Text), "LONGTEXT");
        assert_eq!(mysql_keyword(SqlBaseType::Json), "JSON");
    }

    fn idx(cols: &[&str]) -> std::collections::BTreeSet<String> {
        cols.iter().map(|s| s.to_string()).collect()
    }

    fn keyvec(cols: &[&str]) -> Vec<String> {
        cols.iter().map(|s| s.to_string()).collect()
    }

    #[test]
    fn key_matches_single_column_index() {
        let indexes = vec![idx(&["id"])];
        assert!(key_matches_unique_index(&indexes, &keyvec(&["id"])));
    }

    #[test]
    fn key_matches_composite_index_reordered() {
        // A UNIQUE index on (a, b) matches a configured key of [b, a].
        let indexes = vec![idx(&["a", "b"])];
        assert!(key_matches_unique_index(&indexes, &keyvec(&["b", "a"])));
    }

    #[test]
    fn key_does_not_match_subset_of_index() {
        // Index is (a, b); key [a] is a prefix/subset — not an exact match.
        let indexes = vec![idx(&["a", "b"])];
        assert!(!key_matches_unique_index(&indexes, &keyvec(&["a"])));
    }

    #[test]
    fn key_does_not_match_superset_of_index() {
        // Index is (a); key [a, b] is a superset — not an exact match.
        let indexes = vec![idx(&["a"])];
        assert!(!key_matches_unique_index(&indexes, &keyvec(&["a", "b"])));
    }

    #[test]
    fn key_does_not_match_disjoint_index() {
        let indexes = vec![idx(&["id"])];
        assert!(!key_matches_unique_index(&indexes, &keyvec(&["other"])));
    }

    #[test]
    fn key_matches_one_of_multiple_indexes() {
        // Table has a PRIMARY (id) and a UNIQUE (email); key on email matches.
        let indexes = vec![idx(&["id"]), idx(&["email"])];
        assert!(key_matches_unique_index(&indexes, &keyvec(&["email"])));
        assert!(key_matches_unique_index(&indexes, &keyvec(&["id"])));
        // A key on neither matches.
        assert!(!key_matches_unique_index(&indexes, &keyvec(&["name"])));
        // A key spanning two distinct indexes is not itself an index.
        assert!(!key_matches_unique_index(
            &indexes,
            &keyvec(&["id", "email"])
        ));
    }

    #[test]
    fn key_does_not_match_empty_index_set() {
        // No unique indexes (e.g. table absent / no constraints) → never matches.
        let indexes: Vec<std::collections::BTreeSet<String>> = vec![];
        assert!(!key_matches_unique_index(&indexes, &keyvec(&["id"])));
    }

    #[test]
    fn empty_key_never_matches() {
        let indexes = vec![idx(&["id"])];
        assert!(!key_matches_unique_index(&indexes, &keyvec(&[])));
        // Even against an (impossible) empty index, an empty key is rejected.
        let empty_idx = vec![idx(&[])];
        assert!(!key_matches_unique_index(&empty_idx, &keyvec(&[])));
    }

    #[test]
    fn key_matches_composite_index_among_several() {
        let indexes = vec![idx(&["id"]), idx(&["tenant", "slug"])];
        assert!(key_matches_unique_index(
            &indexes,
            &keyvec(&["slug", "tenant"])
        ));
        assert!(!key_matches_unique_index(&indexes, &keyvec(&["tenant"])));
    }

    #[test]
    fn mysql_data_type_round_trips_to_json_schema() {
        use serde_json::json;
        assert_eq!(
            mysql_data_type_to_json_schema("bigint", false),
            json!({"type":"integer"})
        );
        assert_eq!(
            mysql_data_type_to_json_schema("int", false),
            json!({"type":"integer"})
        );
        // tinyint maps to integer (precision is not exposed by DATA_TYPE, so we
        // never guess boolean for safety).
        assert_eq!(
            mysql_data_type_to_json_schema("tinyint", false),
            json!({"type":"integer"})
        );
        assert_eq!(
            mysql_data_type_to_json_schema("double", false),
            json!({"type":"number"})
        );
        assert_eq!(
            mysql_data_type_to_json_schema("decimal", false),
            json!({"type":"number"})
        );
        assert_eq!(
            mysql_data_type_to_json_schema("json", false),
            json!({"type":"object"})
        );
        assert_eq!(
            mysql_data_type_to_json_schema("varchar", false),
            json!({"type":"string"})
        );
        // Unknown types fall back to string; nullable widens the type array.
        assert_eq!(
            mysql_data_type_to_json_schema("datetime", true),
            json!({"type":["string","null"]})
        );
    }
}