Skip to content

How to implement DataLoader to batch load GraphQL class rows in a single DB query? #358

@paussu

Description

@paussu

Hi,

Any example on how to implement a DataLoader for a GraphQL class field?

For example we have an InvoiceHeader class that returns list of InvoiceRow fields and we would want to batch the keys and execute a single DB query to get the rows.

This probably can be implemented with a future or a coroutine?

Asked AI and it gave a DataLoader class like this:

#pragma once

#include <unordered_map>
#include <vector>
#include <future>
#include <mutex>
#include <functional>
#include <queue>
#include <memory>

template<typename Key, typename Value>
class DataLoader {
public:
    using BatchFn = std::function<std::unordered_map<Key, Value>(const std::vector<Key>&)>;

    explicit DataLoader(BatchFn batchFn)
        : _batchFunction(std::move(batchFn))
    {
    }

    std::future<Value> load(const Key& key) {
        std::lock_guard lock(_mutex);

        // Check cache
        auto it = _cache.find(key);
        if (it != _cache.end()) {
            std::promise<Value> promise;
            promise.set_value(it->second);
            return promise.get_future();
        }
        // Check if already pending
        auto pit = _pending.find(key);
        if (pit != _pending.end())
            return pit->second->get_future();

        // New key: add to queue and pending
        auto promisePtr = std::make_shared<std::promise<Value>>();
        _pending[key] = promisePtr;
        _queue.push_back(key);
        return promisePtr->get_future();
    }

    // Call this after all load()s to resolve the batch
    void dispatch()
    {
        std::vector<Key> keys;
        {
            std::lock_guard lock(_mutex);
            if (_queue.empty()) return;
            keys.swap(_queue);
        }
        auto results = _batchFunction(keys);
        std::lock_guard lock(_mutex);
        for (const auto& key : keys)
        {
            auto pit = _pending.find(key);
            if (pit != _pending.end())
            {
                auto rit = results.find(key);
                if (rit != results.end())
                {
                    pit->second->set_value(rit->second);
                    _cache[key] = rit->second;
                }
                else
                {
                    pit->second->set_exception(std::make_exception_ptr(std::runtime_error("Key not found in batch result")));
                }
                _pending.erase(pit);
            }
        }
    }

    // Optional: clear cache for new request
    void clearCache()
    {
        std::lock_guard lock(_mutex);
        _cache.clear();
    }

private:
    BatchFn _batchFunction;
    std::unordered_map<Key, std::shared_ptr<std::promise<Value>>> _pending;
    std::unordered_map<Key, Value> _cache;
    std::vector<Key> _queue;
    std::mutex _mutex;
};

Invoice row

service::AwaitableObject<std::vector<std::shared_ptr<object::InvoiceRow>>> InvoiceHeader::getRows(service::FieldParams&& params) const
   {
      const auto &key = invoiceSeqNumber.value();
      const auto &loader = _invoiceQueries->getInvoiceRowsLoader();

      return loader->load(key);
   }

Does this look correct and where should the dispatch happen? In query endSelectionSet or somewhere else?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions