dynoxide-rs 2.0.0

use crate::actions::helpers;
use crate::errors::{DynoxideError, Result};
use crate::storage_backend::StorageBackend;
use crate::types::{self, AttributeValue, Item};
use serde::{Deserialize, Serialize};
use std::collections::HashMap;

#[derive(Debug, Default, Deserialize)]
pub struct BatchWriteItemRequest {
    #[serde(rename = "RequestItems")]
    pub request_items: HashMap<String, Vec<WriteRequest>>,
    #[serde(rename = "ReturnConsumedCapacity", default)]
    pub return_consumed_capacity: Option<String>,
    #[serde(rename = "ReturnItemCollectionMetrics", default)]
    pub return_item_collection_metrics: Option<String>,
}

#[derive(Debug, Default, Deserialize)]
pub struct WriteRequest {
    #[serde(rename = "PutRequest", default)]
    pub put_request: Option<PutRequest>,
    #[serde(rename = "DeleteRequest", default)]
    pub delete_request: Option<DeleteRequest>,
}

#[derive(Debug, Default, Deserialize)]
pub struct PutRequest {
    #[serde(rename = "Item")]
    pub item: Item,
}

#[derive(Debug, Default, Deserialize)]
pub struct DeleteRequest {
    #[serde(rename = "Key")]
    pub key: HashMap<String, AttributeValue>,
}

#[derive(Debug, Default, Serialize)]
pub struct BatchWriteItemResponse {
    #[serde(rename = "UnprocessedItems")]
    pub unprocessed_items: HashMap<String, serde_json::Value>,
    #[serde(rename = "ConsumedCapacity", skip_serializing_if = "Option::is_none")]
    pub consumed_capacity: Option<Vec<crate::types::ConsumedCapacity>>,
    #[serde(
        rename = "ItemCollectionMetrics",
        skip_serializing_if = "Option::is_none"
    )]
    pub item_collection_metrics: Option<HashMap<String, Vec<crate::types::ItemCollectionMetrics>>>,
}

/// Fold one item's per-index units into the running per-table totals. A batch
/// reports the sum across its items, and an index no item in the batch touched
/// stays absent rather than appearing as a zero.
fn accumulate_index_units(
    totals: &mut HashMap<String, HashMap<String, f64>>,
    table_name: &str,
    units: &HashMap<String, f64>,
) {
    if units.is_empty() {
        return;
    }
    let table_entry = totals.entry(table_name.to_string()).or_default();
    for (index_name, units) in units {
        *table_entry.entry(index_name.clone()).or_insert(0.0) += units;
    }
}

pub async fn execute<S: StorageBackend>(
    storage: &S,
    mut request: BatchWriteItemRequest,
) -> Result<BatchWriteItemResponse> {
    const MAX_REQUEST_SIZE: usize = 16 * 1024 * 1024; // 16MB

    // Validate RequestItems is not empty.
    // AWS routes the empty-map case through a separate parameter-required path
    // rather than the standard "N validation errors detected" envelope.
    //
    // Do not align this with BatchGetItem, which now answers the standard
    // envelope. The two have diverged: as of the 2026-09-02 sweep 31 of the 33
    // regions observed still return this sentence for BatchWriteItem, against
    // 1 of 33 for BatchGetItem. Matching them would break the common case here.
    if request.request_items.is_empty() {
        return Err(DynoxideError::ValidationException(
            "The requestItems parameter is required for BatchWriteItem".to_string(),
        ));
    }

    // Validate each table entry has at least one write request
    for (table_name, wrs) in &request.request_items {
        if wrs.is_empty() {
            return Err(DynoxideError::ValidationException(format!(
                "1 validation error detected: Value at 'requestItems.{table_name}.member' failed to satisfy constraint: Member must have length greater than or equal to 1"
            )));
        }
    }

    // Validate table name format for all tables before checking existence
    for table_name in request.request_items.keys() {
        crate::validation::validate_table_name(table_name)?;
    }

    // Validate total request count.
    // AWS surfaces this as the standard "1 validation error detected" envelope
    // and echoes the WriteRequest list inside `Value '{<table>=[<dump>]}'`. The
    // conformance suite anchors a regex around the envelope and the constraint
    // phrase but leaves the dump body unconstrained (because the AWS SDK's
    // Java-toString shape adds new AttributeValue fields over time). We emit
    // the table name verbatim and a Rust Debug dump of the WriteRequests so
    // the envelope matches without coupling to a specific SDK version. If a
    // future suite tightens the regex to pin the dump exactly, this site
    // will need a follow-up change.
    let total_requests: usize = request.request_items.values().map(|v| v.len()).sum();
    if total_requests > 25 {
        let empty: Vec<WriteRequest> = Vec::new();
        let (table_name, requests) = request
            .request_items
            .iter()
            .max_by_key(|(_, v)| v.len())
            .map(|(name, v)| (name.as_str(), v))
            .unwrap_or(("", &empty));
        let dump = format!("{requests:?}");
        return Err(DynoxideError::ValidationException(format!(
            "1 validation error detected: Value '{{{table_name}=[{dump}]}}' at 'requestItems' failed to satisfy constraint: Map value must satisfy constraint: [Member must have length less than or equal to 25, Member must have length greater than or equal to 1]"
        )));
    }

    // --- Pre-table validations ---
    // DynamoDB validates attribute values, item size, and empty write requests
    // BEFORE checking table existence.
    for write_requests in request.request_items.values() {
        for wr in write_requests {
            if wr.put_request.is_none() && wr.delete_request.is_none() {
                return Err(DynoxideError::ValidationException(
                    "Supplied AttributeValue has more than one datatypes set, must contain exactly one of the supported datatypes".to_string(),
                ));
            }
            if let Some(ref put_req) = wr.put_request {
                // Validate attribute values (empty strings, empty sets, invalid numbers)
                crate::validation::validate_item_attribute_values(&put_req.item)?;

                // Validate item size before table lookup
                let size = types::item_size(&put_req.item);
                if size > types::MAX_ITEM_SIZE {
                    return Err(DynoxideError::ValidationException(
                        "Item size has exceeded the maximum allowed size".to_string(),
                    ));
                }
            }
            if let Some(ref del_req) = wr.delete_request {
                crate::validation::validate_item_attribute_values(&del_req.key)?;
            }
        }
    }

    // Validate aggregate request size
    let total_size: usize = request
        .request_items
        .values()
        .flat_map(|wrs| wrs.iter())
        .map(|wr| {
            if let Some(ref put_req) = wr.put_request {
                types::item_size(&put_req.item)
            } else if let Some(ref del_req) = wr.delete_request {
                types::item_size(&del_req.key)
            } else {
                0
            }
        })
        .sum();
    if total_size > MAX_REQUEST_SIZE {
        return Err(DynoxideError::ValidationException(
            "Item collection too large: aggregate size of items in BatchWriteItem exceeds 16MB limit".to_string(),
        ));
    }

    // Validate: no duplicate keys across all operations
    {
        let mut seen_keys: std::collections::HashSet<(String, String, String)> =
            std::collections::HashSet::new();
        for (table_name, write_requests) in &request.request_items {
            let meta = helpers::require_table_for_item_op(storage, table_name).await?;
            let key_schema = helpers::parse_key_schema(&meta)?;
            for wr in write_requests {
                // Validate keys BEFORE extract_key_strings: that helper returns
                // InternalServerError (HTTP 500) for a missing partition or sort
                // key, but a key-less request is client input and must surface as
                // a 400 ValidationException. Mirrors put_item.rs, which validates
                // before extracting.
                let key_item = if let Some(ref put) = wr.put_request {
                    helpers::validate_item_keys_for_batch(&put.item, &key_schema, &meta)?;
                    &put.item
                } else if let Some(ref del) = wr.delete_request {
                    helpers::validate_key_only(&del.key, &key_schema)?;
                    &del.key
                } else {
                    continue;
                };
                let (pk, sk) = helpers::extract_key_strings(key_item, &key_schema)?;
                let key = (table_name.clone(), pk, sk);
                if !seen_keys.insert(key) {
                    return Err(DynoxideError::ValidationException(
                        "Provided list of item keys contains duplicates".to_string(),
                    ));
                }
            }
        }
    }

    // Track per-table secondary index capacity and affected partition keys for
    // deferred metrics
    let mut table_gsi_units: HashMap<String, HashMap<String, f64>> = HashMap::new();
    let mut table_lsi_units: HashMap<String, HashMap<String, f64>> = HashMap::new();
    // Vector replication is billed in bytes, so it accumulates on its own map
    // and never joins the unit totals.
    let mut table_vector_bytes: HashMap<String, HashMap<String, f64>> = HashMap::new();
    // Track per-table WCU (table-level, excludes GSI)
    let mut table_wcu: HashMap<String, f64> = HashMap::new();
    // Collect unique (table, pk_str, pk_attr, pk_value) for deferred metrics computation
    let mut affected_partitions: Vec<(String, String, String, AttributeValue)> = Vec::new();

    for (table_name, write_requests) in &mut request.request_items {
        let meta = helpers::require_table_for_item_op(storage, table_name).await?;
        let key_schema = helpers::parse_key_schema(&meta)?;
        // Parse the index definitions once per table: the maintain helpers'
        // meta-accepting forms deserialise the JSON on every call, which for
        // a 25-item batch against one table would repeat the work per item.
        let gsi_defs = super::gsi::parse_gsi_defs(&meta)?;
        let lsi_defs = super::lsi::parse_lsi_defs(&meta)?;
        let vector_defs = super::vector_index::parse_vector_defs(&meta)?;
        let attr_defs = super::vector_index::parse_attr_defs(&meta)?;

        for wr in write_requests {
            if let Some(ref mut put_req) = wr.put_request {
                // Validate keys
                helpers::validate_item_keys_for_batch(&put_req.item, &key_schema, &meta)?;

                // Validate attribute values (empty strings, empty sets)
                crate::validation::validate_item_attribute_values(&put_req.item)?;

                // Normalize sets (deduplication)
                crate::validation::normalize_item_sets(&mut put_req.item);

                // Validate item size
                let size = types::item_size(&put_req.item);
                if size > types::MAX_ITEM_SIZE {
                    return Err(DynoxideError::ValidationException(
                        "Item size has exceeded the maximum allowed size".to_string(),
                    ));
                }

                // TODO: validation must precede this call -- if reaching this line, caller has already validated keys.
                let (pk, sk) = helpers::extract_key_strings(&put_req.item, &key_schema)?;
                let item_json = serde_json::to_string(&put_req.item)
                    .map_err(|e| DynoxideError::InternalServerError(e.to_string()))?;
                let hash_prefix = put_req
                    .item
                    .get(&key_schema.partition_key)
                    .map(crate::storage::compute_hash_prefix)
                    .unwrap_or_default();
                // Base write + index fan-out + stream are one atomic unit per
                // item: a mid-fan-out failure rolls this item's write back.
                // BatchWriteItem items are independent, so this is one
                // transaction per write request.
                let (old_size, gsi_units, lsi_units, vector_bytes) =
                    helpers::with_write_transaction(storage, async {
                        let old_json = storage
                            .put_item_with_hash(
                                table_name,
                                &pk,
                                &sk,
                                &item_json,
                                size,
                                &hash_prefix,
                            )
                            .await?;
                        let old_item: Option<Item> =
                            old_json.and_then(|j| serde_json::from_str(&j).ok());
                        let target = super::gsi::IndexWrite {
                            table_name,
                            pk: &pk,
                            sk: &sk,
                            pk_attr: &key_schema.partition_key,
                            sk_attr: key_schema.sort_key.as_deref(),
                            old_item: old_item.as_ref(),
                            capacity_mode: request.return_consumed_capacity.as_deref(),
                        };
                        let gsi_units = super::gsi::maintain_gsis_after_write_with_defs(
                            storage,
                            &gsi_defs,
                            &target,
                            &put_req.item,
                        )
                        .await?;
                        let lsi_units = super::lsi::maintain_lsis_after_write_with_defs(
                            storage,
                            &lsi_defs,
                            &target,
                            &put_req.item,
                        )
                        .await?;
                        let vector_bytes =
                            super::vector_index::maintain_vector_indexes_after_write_with_defs(
                                storage,
                                &vector_defs,
                                &attr_defs,
                                &target,
                                &put_req.item,
                                false,
                            )
                            .await?;
                        crate::streams::record_stream_event(
                            storage,
                            &meta,
                            old_item.as_ref(),
                            Some(&put_req.item),
                        )
                        .await?;
                        Ok((
                            old_item.as_ref().map(types::item_size),
                            gsi_units,
                            lsi_units,
                            vector_bytes,
                        ))
                    })
                    .await?;

                // Accumulate WCU on the larger of the displaced and new images
                *table_wcu.entry(table_name.clone()).or_insert(0.0) +=
                    types::table_write_capacity_units(old_size, Some(size));

                // Accumulate secondary index units per table
                accumulate_index_units(&mut table_gsi_units, table_name, &gsi_units);
                accumulate_index_units(&mut table_lsi_units, table_name, &lsi_units);
                accumulate_index_units(&mut table_vector_bytes, table_name, &vector_bytes);

                // Track affected partition for deferred metrics
                if let Some(pk_val) = put_req.item.get(&key_schema.partition_key) {
                    affected_partitions.push((
                        table_name.clone(),
                        pk.clone(),
                        key_schema.partition_key.clone(),
                        pk_val.clone(),
                    ));
                }
            } else if let Some(ref del_req) = wr.delete_request {
                helpers::validate_key_only(&del_req.key, &key_schema)?;
                // TODO: validation must precede this call -- if reaching this line, caller has already validated keys.
                let (pk, sk) = helpers::extract_key_strings(&del_req.key, &key_schema)?;

                // Base delete + index fan-out + stream are one atomic unit per item.
                let (old_item, gsi_units, lsi_units, vector_bytes) =
                    helpers::with_write_transaction(storage, async {
                        let old_json = storage.delete_item(table_name, &pk, &sk).await?;
                        let old_item: Option<Item> =
                            old_json.as_ref().and_then(|j| serde_json::from_str(j).ok());
                        let target = super::gsi::IndexWrite {
                            table_name,
                            pk: &pk,
                            sk: &sk,
                            pk_attr: &key_schema.partition_key,
                            sk_attr: key_schema.sort_key.as_deref(),
                            old_item: old_item.as_ref(),
                            capacity_mode: request.return_consumed_capacity.as_deref(),
                        };
                        let gsi_units = super::gsi::maintain_gsis_after_delete_with_defs(
                            storage, &gsi_defs, &target,
                        )
                        .await?;
                        let lsi_units = super::lsi::maintain_lsis_after_delete_with_defs(
                            storage, &lsi_defs, &target,
                        )
                        .await?;
                        let vector_bytes =
                            super::vector_index::maintain_vector_indexes_after_delete_with_defs(
                                storage,
                                &vector_defs,
                                &attr_defs,
                                &target,
                            )
                            .await?;
                        if old_item.is_some() {
                            crate::streams::record_stream_event(
                                storage,
                                &meta,
                                old_item.as_ref(),
                                None,
                            )
                            .await?;
                        }
                        Ok((old_item, gsi_units, lsi_units, vector_bytes))
                    })
                    .await?;

                // Accumulate WCU: based on old item size if it existed, else 1 WCU
                *table_wcu.entry(table_name.clone()).or_insert(0.0) +=
                    types::table_write_capacity_units(
                        old_item.as_ref().map(types::item_size),
                        None,
                    );

                // Accumulate secondary index units per table
                accumulate_index_units(&mut table_gsi_units, table_name, &gsi_units);
                accumulate_index_units(&mut table_lsi_units, table_name, &lsi_units);
                accumulate_index_units(&mut table_vector_bytes, table_name, &vector_bytes);

                // Track affected partition for deferred metrics
                if let Some(pk_val) = del_req.key.get(&key_schema.partition_key) {
                    affected_partitions.push((
                        table_name.clone(),
                        pk.clone(),
                        key_schema.partition_key.clone(),
                        pk_val.clone(),
                    ));
                }
            } else {
                return Err(DynoxideError::ValidationException(
                    "WriteRequest must contain either PutRequest or DeleteRequest".to_string(),
                ));
            }
        }
    }

    // Build consumed capacity per table using pre-tracked WCU
    let consumed_capacity = if matches!(
        request.return_consumed_capacity.as_deref(),
        Some("TOTAL") | Some("INDEXES")
    ) {
        let mut caps = Vec::new();
        for table_name in request.request_items.keys() {
            let total_wcu = table_wcu.get(table_name).copied().unwrap_or(0.0);
            let gsi_units = table_gsi_units.get(table_name).cloned().unwrap_or_default();
            let lsi_units = table_lsi_units.get(table_name).cloned().unwrap_or_default();
            let vector_bytes = table_vector_bytes
                .get(table_name)
                .cloned()
                .unwrap_or_default();
            if let Some(cc) = crate::types::consumed_capacity_with_vector_indexes(
                table_name,
                total_wcu,
                &gsi_units,
                &lsi_units,
                &vector_bytes,
                &request.return_consumed_capacity,
            ) {
                caps.push(cc);
            }
        }
        Some(caps)
    } else {
        None
    };

    // Item collection metrics are computed once per unique (table, pk), deferred from the write loop
    let mut all_item_collection_metrics: HashMap<String, Vec<crate::types::ItemCollectionMetrics>> =
        HashMap::new();
    if matches!(
        request.return_item_collection_metrics.as_deref(),
        Some("SIZE")
    ) {
        // Deduplicate by (table, pk) to avoid redundant queries
        let mut seen = std::collections::HashSet::new();
        for (tbl, pk_str, pk_attr, pk_val) in &affected_partitions {
            let key = (tbl.as_str(), pk_str.as_str());
            if !seen.insert(key) {
                continue;
            }
            let meta = helpers::require_table(storage, tbl).await?;
            if let Some(icm) = helpers::build_item_collection_metrics(
                storage,
                &meta,
                tbl,
                pk_str,
                pk_attr,
                pk_val,
                &request.return_item_collection_metrics,
            )
            .await?
            {
                all_item_collection_metrics
                    .entry(tbl.clone())
                    .or_default()
                    .push(icm);
            }
        }
    }
    let item_collection_metrics = if all_item_collection_metrics.is_empty() {
        None
    } else {
        Some(all_item_collection_metrics)
    };

    Ok(BatchWriteItemResponse {
        unprocessed_items: HashMap::new(),
        consumed_capacity,
        item_collection_metrics,
    })
}

#[cfg(test)]
mod tests {
    use crate::actions::{batch_write_item, create_table};
    use crate::storage::Storage;
    use crate::storage_backend::StorageBackend;

    /// Each batch put is atomic with its own GSI fan-out: a mid-fan-out failure
    /// rolls that item's base write back rather than leaving a torn index.
    #[test]
    fn batch_put_rolls_back_base_write_when_gsi_fan_out_fails() {
        let storage = Storage::memory().unwrap();

        let create = serde_json::from_value(serde_json::json!({
            "TableName": "Orders",
            "KeySchema": [{"AttributeName": "UserId", "KeyType": "HASH"}],
            "AttributeDefinitions": [
                {"AttributeName": "UserId", "AttributeType": "S"},
                {"AttributeName": "Status", "AttributeType": "S"},
                {"AttributeName": "Priority", "AttributeType": "S"}
            ],
            "GlobalSecondaryIndexes": [
                {"IndexName": "StatusIndex", "KeySchema": [{"AttributeName": "Status", "KeyType": "HASH"}], "Projection": {"ProjectionType": "ALL"}},
                {"IndexName": "PriorityIndex", "KeySchema": [{"AttributeName": "Priority", "KeyType": "HASH"}], "Projection": {"ProjectionType": "ALL"}}
            ]
        }))
        .unwrap();
        pollster::block_on(create_table::execute(&storage, create)).unwrap();

        // Break the second GSI's fan-out by dropping its physical table.
        storage.drop_gsi_table("Orders", "PriorityIndex").unwrap();

        let batch = serde_json::from_value(serde_json::json!({
            "RequestItems": {
                "Orders": [
                    {"PutRequest": {"Item": {"UserId": {"S": "u1"}, "Status": {"S": "SHIPPED"}, "Priority": {"S": "HIGH"}}}}
                ]
            }
        }))
        .unwrap();
        let res = pollster::block_on(batch_write_item::execute(&storage, batch));
        assert!(
            res.is_err(),
            "a mid-fan-out failure must surface as an error"
        );

        // The base write must roll back: no item landed.
        let count =
            pollster::block_on(<Storage as StorageBackend>::count_items(&storage, "Orders"))
                .unwrap();
        assert_eq!(
            count, 0,
            "batch put base write must roll back when fan-out fails"
        );
    }
}