11 Best Airflow Alternatives for 2026

Apache Airflow has been a go-to choice for workflow orchestration, but as data pipelines grow more complex, teams face challenges with setup, scaling, and maintenance. 

In 2026, organizations need tools that offer simpler management, better scalability, and modern features like cloud-native execution, no-code pipelines, and improved observability.

This article highlights 11 of the best Apache Airflow alternatives, comparing their strengths, unique features, and real-world use cases. 

Whether you need a fully managed ELT platform, Kubernetes-native orchestration, or a developer-friendly workflow engine, these tools can help streamline data operations. Let’s get started!

What Is Apache Airflow?

Apache Airflow is an open-source platform for developing, scheduling, and monitoring batch data workflows. Pipelines are defined in Python and managed through a web interface. 

Workflows are organized as DAGs, where task dependencies and execution order are defined in code and executed by a scheduler and workers. Its workflows-as-code approach lets teams use Python logic, version control, and modular design to build flexible, production-grade data pipelines.

Key features:

• Modular provider ecosystem: Airflow has a modular provider ecosystem where each provider package adds specific integrations for external systems. You can install providers and extend Airflow with custom or community-built components.
• Extensible operators, hooks, and plugins: Airflow’s architecture lets you build custom operators and hooks beyond the built-ins, giving you precise control over how tasks interact with different systems or APIs.
• Task SDK: The Task SDK provides stable, Python-native interfaces for defining and interacting with Airflow tasks at runtime. It separates task logic from scheduler internals, ensuring consistent development across versions.

Quick Tabular Comparison of the Airflow Competitors

	Hevo	Prefect	Dagster	AWS Step Functions	Luigi
Ease of use	No-code, intuitive UI ✅	Intuitive ✅	Developer-centric	Visual designer UI	CLI-heavy setup ❌
Scalability	Real-time auto-scaling	Auto-scales	Moderate	Auto-scales	Limited distributed scaling
Flexibility	Fully managed, no infra setup	Python-native workflows	Flexible design + typing	Limited ❌	Code-defined tasks ✅
Monitoring	Automated monitoring	Strong observability UI	Rich metadata tracking	Built-in via CloudWatch	Basic task logs
Integration	150+ built-in connectors	API-driven integrations	Works with Snowflake, Spark	Deep AWS integration	Limited native connectors
Cost efficiency	Transparent, predictable pricing ✅	Usage-based pricing ✅	Infra-managed cost ✅	Pay-per-state pricing ❌	Low infra cost ✅

Top 11 Apache Airflow Alternatives to Consider in 2026

1. Hevo Data

Hevo Data is a fully managed, no-code ELT platform that simplifies moving data from multiple sources to data warehouses. It allows teams to build reliable data pipelines without writing code or managing orchestration infrastructure.

Hevo connects to 150+ sources and automatically extracts, loads, and keeps data in sync with your warehouse. Pipelines are created through a guided interface, while Hevo handles scaling, retries, and schema changes. It helps data teams deliver analytics-ready data without ongoing engineering effort.

What makes Hevo unique is its fully managed, fault-tolerant architecture. Pipelines auto-heal from failures, scale automatically, and provide real-time visibility into every data flow. This ensures consistent, trustworthy pipelines without operational overhead.

Key features:

Simple to use: Hevo enables teams to build data pipelines in minutes through a guided, no-code interface. There is no need to write scripts, manage infrastructure, or configure orchestration logic.

Reliable: Hevo is built on a fault-tolerant architecture that ensures pipelines continue to run even when source systems fail. Auto-healing mechanisms and intelligent retries recover from transient issues automatically.

Transparent: Hevo provides complete visibility into pipeline health through real-time dashboards, detailed logs, and lineage tracking. Batch-level checks help ensure data accuracy and consistency across systems.

Predictable pricing: Hevo uses an event-based pricing model that makes costs easy to understand and forecast. There are no hidden infrastructure fees, scaling surprises, or usage-based complexity.

Scalable: Hevo automatically scales to support increasing data volumes and pipeline complexity.

Pros:

• Supports pre-load and post-load transformations to prepare data for analytics.
• Continuously syncs data with minimal latency, enabling faster analytics and reporting.
• End-to-end encryption and role-based access controls protect sensitive data.

Pricing:

• Starts as a free tier with limited connectors up to 1 million events
• Starter: $239/month up to 5 million events
• Professional: $679/month up to 20 million events
• Business: Custom pricing

I really appreciate Hevo Data\'s great customer service and easy interface. People get back to you super fast, and tickets are resolved quickly, which is a big plus for me. The customer support team is also a great help with research because they know the documentation of all APIs really well. The initial setup was easy, which made the transition smooth.

Simon E.

View Review

2. Prefect

Prefect is a modern workflow orchestration platform designed to simplify data pipeline management. Built as a developer-friendly alternative to traditional schedulers, it focuses on reducing orchestration complexity while maintaining flexibility.

Prefect works by defining workflows in Python and executing them through its orchestration engine, with optional managed cloud control planes. It helps data engineers and platform teams build, monitor, and scale workflows without managing heavy infrastructure.

Pros:

• Supports local, on-prem, cloud, and containerized execution for deployment flexibility.
• Prefect Cloud provides detailed run history, state tracking, and logs with a modern UI.
• Supports reactive, event-based orchestration beyond strict time-based scheduling.

Cons:

• Still developer-centric and not ideal for non-technical users.
• Self-hosted setups require infrastructure and container orchestration knowledge.
• Advanced monitoring and orchestration features require Prefect Cloud, which adds cost.

Why Choose Prefect Over Apache Airflow

Simpler workflow design: Prefect uses standard Python functions instead of rigid DAG definitions. Pipelines are easier to write, test, and modify without adapting code to fit scheduler constraints.

Built-in state handling: Prefect’s core engine includes native state management, retries, and failure handling in the open-source version. Tasks transition through clearly defined states, making execution behavior predictable.

Pricing:

Prefect is free to use through its open-source (self-hosted) version. It offers six pricing tiers:

• Hobby: Free up to 5 deployments
• Starter: $100 per month up to 20 deployments
• Team: $100/user per month up to 100 deployments
• Pro: Custom pricing up to 1000 deployments
• Enterprise: Custom pricing, unlimited deployments
• Customer Managed: Self-hosted for maximum control

It is easy to stand up a flow, get UI benefits for orchestration, and test and debug when flows fail. Developing prefect flows in v2/v3 of prefect is also much more straightforward than it was on prefect v1.

Kyle H

Staff Data Scientist

View Review

3. Dagster

Dagster is a data orchestration platform designed to build, run, and monitor reliable data pipelines. It focuses on software engineering best practices such as modular design, strong typing, and built-in testing to improve pipeline quality and maintainability.

Dagster defines pipelines as code using assets, ops, and schedules, making dependencies and data lineage explicit. It helps data engineers manage complex workflows, debug failures faster, and maintain production-grade pipelines with better observability and control.

Pros:

• Provides detailed pipeline run history, logs, asset health, and metadata. 
• Dagster integrates deeply with dbt, orchestrating dbt models as first-class assets.
• Dagster automatically tracks upstream and downstream dependencies between assets.

Cons:

• Not ideal for non-technical users.
• Advanced collaboration, monitoring, and enterprise features require Dagster Cloud.
• Running Dagster in production requires managing containers, compute, and orchestration components.

Why Choose Dagster Over Apache Airflow

Richer data observability: Dagster provides detailed asset lineage, metadata tracking, and health visibility out of the box. Debugging and monitoring are often more structured compared to Airflow’s task-centric view.

Explicit dependency management: Dependencies are defined through asset relationships rather than inferred through DAG structure. This reduces accidental misconfiguration, and pipeline structure remains easier to maintain as complexity grows.

Pricing:

Dagster’s open-source version is free to use. Dagster Cloud follows a usage-based pricing model:

• Solo Plan: $10 per month with 7.5k credits/month
• Starter Plan: $100 per month with 30k credits/month
• Pro Plan: Custom pricing

Dagster stands out for its asset‑centric approach to data orchestration, treating data as first‑class assets rather than just tasks. This makes pipelines more reliable, transparent, and easier to monitor. Users appreciate its strong integrations with modern data tools (like dbt, Snowflake, and Databricks), built‑in observability and monitoring, and a developer‑friendly design that emphasizes testability and maintainability. Its scalability across small and enterprise workflows ensures long‑term relevance, while the clear abstractions reduce complexity compared to older orchestrators.

Aarsh B

Technical officer

View Review

4. AWS Step Function

AWS Step Functions is a fully managed, serverless orchestration service for building distributed application workflows and state machines on AWS. It lets you define application logic as a series of steps that execute in order while coordinating services and tracking state.

Step Functions works by breaking complex logic into state machines defined in Amazon States Language or via a visual editor. Each step (state) can call AWS services, handle retries, make choices, or run tasks in parallel. Developers can automate multi-step processes, integrate AWS services, and manage event-driven workflows.

Pros:

• The fully serverless architecture handles scaling, availability, and state management.
• Standard Workflows can run for up to one year for durable orchestration.
• Provides a graphical console to view execution paths, input/output data, and failure points.

Cons:

• State transition-based pricing can become expensive for high-frequency workflows.
• External system orchestration often requires Lambda wrappers or custom code.
• Large state machines with multiple transitions can become difficult to manage and maintain.

Why Choose AWS Step Functions Over Apache Airflow

No scheduler management: Step Functions does not rely on a central scheduler process to trigger and manage workflows. Execution is handled by the AWS service itself, eliminating the need to maintain scheduler uptime, database health, or worker coordination.

Native event-driven orchestration: Step Functions integrates directly with AWS event sources such as EventBridge, SQS, and API Gateway. Workflows can start automatically in response to system events.

Pricing:

• Free tier: Includes 4,000 state transitions per month at no cost.
• Standard Workflows pricing: Charged per state transition executed.
• Express Workflows pricing: Charged per request and execution duration.
• No infrastructure costs: No charges for servers or idle capacity.

I liked how AWS Step Functions have a detailed execution flow for each of my other AWS resources., also I liked the feature of how without the help of lambda I can quickly call Dynamodb and do the operation, this helps in reducing resources and saving some bucks.

Debarshi M.

Small-Business

5. Luigi

Luigi is an open-source workflow orchestration tool originally developed at Spotify to manage complex batch data pipelines. The tool focuses on building pipelines as a sequence of tasks with clear dependencies, ensuring each step runs in the correct order.

Luigi defines workflows using Python classes, where each task specifies its inputs, outputs, and dependencies. It helps data engineers automate ETL jobs, data preparation, and batch processing workflows, especially in environments that rely heavily on Python-based data infrastructure.

Pros:

• The web interface shows task dependencies, execution progress, and failures.
• Ensures correct task order and avoids rerunning completed tasks.
• Luigi runs with a central scheduler, without requiring complex orchestration infrastructure.

Cons:

• Lacks advanced observability and lineage tracking.
• Smaller community and fewer new feature updates.
• No native support for event-driven workflows.

Why Choose Luigi Over Apache Airflow

Output-based execution model: Luigi determines task completion by checking the existence of output files or tables. Tasks run only when outputs are missing, which ensures data consistency. Airflow primarily tracks task states instead of verifying actual data outputs.

No continuous DAG parsing: Luigi executes pipelines only when triggered, without constantly scanning workflow definitions. Airflow continuously parses DAG files, which adds background load.

Pricing:

Luigi is completely free & open source.

6. Orchestra

Orchestra is a modern data orchestration platform designed to help teams build, schedule, and monitor data workflows. It focuses on simplifying orchestration for analytics engineering teams by integrating directly with existing tools rather than replacing them.

Orchestra supports orchestration of SQL-based transformations, data ingestion, and external tools through a clean UI and configuration-driven approach. Teams can define workflows declaratively, making pipelines easier to manage, version, and maintain over time.

Pros:

• Git-native orchestration enables version-controlled workflows and CI/CD integration.
• Offers column-level lineage visibility for enhanced data transparency.
• Provides a visual interface to track workflow runs, dependencies, and system health.

Cons:

• Learning curve for teams unfamiliar with Git-based workflows.
• Limited historical adoption and community support.
• Primarily optimized for dbt-centric workflows.

Why Choose Orchestra Over Apache Airflow

Unified orchestration and documentation: Orchestra automatically generates documentation from pipeline metadata and transformations. Teams can understand workflows, dependencies, and data structure without maintaining separate documentation.

Declarative pipeline configuration: Orchestra allows workflows to be defined declaratively instead of writing Python DAG code. Pipelines are easier to maintain, review, and update through configuration files. Airflow requires manual DAG development, which increases engineering effort.

Pricing:

Orchestra has a capacity-based pricing model with three levels:

• Lean: Free for 3 pipelines
• Scale-up: $600 per month for 10 pipelines
• Enterprise: Custom pricing

Orchestra makes organizing, scheduling and deploying data pipelines incredibly easy. In what is typically a very disjointed and manual process, they bring automation and ease-of-use to give teams time back to focus on their actual jobs.

Daniel N.

Sales Director in Mid-Market

7. Astronomer

Astronomer is a managed platform built to run and scale Apache Airflow in production environments. Provides tooling, infrastructure, and automation to help teams deploy Airflow reliably without managing the underlying systems themselves.

Astronomer handles Airflow deployment, scaling, monitoring, and upgrades through its managed service, Astro. Data teams, platform engineers, and enterprises can operate workflow orchestration at scale, while reducing operational overhead and improving reliability.

Pros:

• Provides built-in tools to monitor DAG runs, track pipeline health, and debug failures.
• Includes CLI tools, templates, and CI/CD integrations that streamline DAG deployment.
• Astronomer supports deployment across major cloud providers.

Cons:

• Users need Python and Airflow expertise to build and manage workflows effectively.
• Platform is designed specifically for Airflow.
• Managed infrastructure and enterprise features increase overall costs.

Why Choose Astronomer Over Apache Airflow

Standardized deployment patterns: Astronomer enforces containerized, version-controlled deployments through its CLI and project structure. This reduces configuration drift and ensures consistent environments across teams.

Optimized Airflow performance tuning: Infrastructure is pre-configured with best practices for scheduler performance, worker scaling, and resource allocation. Teams avoid manual tuning and trial-and-error performance adjustments common in self-managed Airflow.

Pricing:

Astronomer follows a usage-based pricing model:

• Developer: Deployments starting at $0.35/hr.
• Team: Deployments starting at $0.42/hr.
• Business: Custom pricing
• Enterprise: Custom pricing

Easy to implement and maintain. Setup is pretty quick and the painful parts of managing Airflow are removed. Deploying and upgrading are simple. Also very easy to get a local environment up and running, which is a big win. The Docker based deployments make it easy to test out version upgrading locally before deploying. Astro CLI is great for implementing a CI/CD process.

Verified User

Computer Software

8. Apache NiFi

Apache NiFi is an open‑source dataflow automation tool designed to automate the flow of data between systems with reliability and scale. Users can create, manage, and monitor complex data pipelines through a visual web interface rather than code.

NiFi works by building directed graphs of processors that route, transform, and deliver data. It uses flow‑based programming concepts to handle data buffering, prioritization, back‑pressure, and guaranteed delivery. Ideal for data engineers and enterprises that need ingestion and transformation across diverse sources and destinations.

Pros:

• Every data object’s journey is tracked, providing audit trails, and compliance visibility.
• It can handle both real-time and batch data ingestion.
• NiFi’s drag-and-drop interface allows users to design pipelines without coding.

Cons:

• Enterprise-grade support depends on contributors or commercial support.
• NiFi focuses on dataflow management rather than full workflow orchestration.
• Complex transformations sometimes require scripting or custom processors.

Why Choose Apache NiFi Over Apache Airflow

Built-in security and encryption: NiFi includes SSL/TLS, user authentication, and fine-grained access controls out of the box, ensuring secure data transfer between systems. Airflow requires additional setup and plugins for equivalent security.

Automatic flow prioritization: NiFi can dynamically manage queue sizes, apply back-pressure, and prioritize critical flows automatically. Airflow does not natively manage flow-level congestion or prioritize task execution beyond scheduling order.

Pricing:

Apache NiFi is completely free and open-source.

The best thing about Nifi is that the tools bar is located at convenient place for the user to acces the tools. The drag and drop feature comes handy. The grid offers a perfect measure of components. DAG is represented properly by connecting arrows.

Shubham G.

Full Stack Engineer

9. Kedro

Kedro is an open‑source Python framework for building robust, reproducible, and maintainable data and machine learning pipelines. It applies established software engineering principles like modularity, separation of concerns, and configuration management to how you structure code and workflows.

Kedro works by breaking data workflows into nodes (pure Python functions) that are connected into pipelines with clear dependencies. It uses a centralized Data Catalog to manage dataset definitions and supports project templating, versioning, and testing.

Pros:

• Centralized dataset management ensures reproducible workflows.
• Nodes and pipelines can be unit-tested independently for better CI/CD practices.
• Every Kedro project follows a consistent directory layout and naming conventions.

Cons:

• Deploying Kedro pipelines to production often requires additional infrastructure.
• Kedro has minimal built-in monitoring; users must rely on external tools for pipeline tracking.
• Has a steeper leaning curve.

Why Choose Kedro Over Apache Airflow

Pipeline modularity and reusability: Kedro’s nodes and pipelines are fully modular, making it easy to reuse code across projects. Unlike Airflow, which tightly couples task orchestration with workflow logic, Kedro encourages clean separation of computation and orchestration.

ML and analytics-ready: Kedro is designed for machine learning and data science workflows, supporting experiment tracking, data versioning, and reproducible models. Airflow focuses on general task scheduling and requires extensions for ML-specific pipelines.

Pricing:

Kedro is open-source and free.

10. Azure Data Factory

Azure Data Factory (ADF) is Microsoft’s cloud‑based data integration and orchestration service that lets teams build, schedule, and manage data workflows at scale. It is designed for hybrid ETL and ELT scenarios, connecting disparate data sources and destinations across cloud and on‑premises environments.

Data Factory handles execution infrastructure through its serverless architecture, helping data engineers, analytics teams, and enterprise IT teams automate complex workflows without managing servers.

Pros:

• Supports both cloud and on-premises data sources.
• Automatically scales compute resources for data movement and transformation.
• Integration with Azure DevOps, Git, and Azure Monitor enables pipeline versioning, deployment automation, and operational visibility.

Cons:

• Advanced features, triggers, and mapping data flows require significant learning.
• Primarily batch-oriented; real-time data processing requires additional services.
• Debugging can be cumbersome.

Why Choose Azure Data Factory Over Apache Airflow

Prebuilt transformations: ADF includes built-in transformation activities and mapping data flows, letting teams perform common ETL/ELT tasks without custom coding. Airflow leaves transformations mostly to external scripts or separate compute environments.

Advanced scheduling and triggers: ADF supports time-based triggers, event-based triggers, tumbling windows, and dependency-based execution out of the box. Airflow requires additional configuration and custom DAG logic for similar event-driven or windowed scheduling.

Pricing:

Azure Data Factory follows a pay-as-you-go pricing model, which includes:

• Pipeline Execution: Charged per activity run (starting at $0.00025 per activity)
• Data Flow Execution: Charged based on vCore-hour usage (starting at $0.193 per vCore-hour)
• Data Movement & Connectivity: Charged per integration runtime and data transfer volume
• Monitoring & Debugging: Additional charges for logging, debugging, and execution metrics

What I like best about Azure Data Factory is its robust and versatile data integration capabilities. It offers a wide range of connectors and tools to efficiently manage and transform data from various sources. Its user-friendly interface, combined with the flexibility to handle complex workflows, makes it an excellent choice for orchestrating data pipelines. The seamless integration with other Azure services also enhances its functionality, making it a powerful tool for data engineering tasks.

Sowjanya G.

Digital Education Student Ambassador

View Review

11. Argo Workflows

Argo Workflows is an open‑source, Kubernetes‑native workflow engine used to orchestrate complex, container‑based pipelines on Kubernetes clusters. Workflows are defined as Kubernetes, where each step runs in its own container, enabling scalable parallel execution and dependency management.

Argo runs workflows by submitting container‑based tasks to Kubernetes, where each task becomes a pod, with DAG or step‑based execution handled by the Argo controller. It integrates with Kubernetes scheduling, RBAC, storage, and logs, making it ideal for CI/CD and machine learning teams.

Pros:

• Runs entirely on Kubernetes; no extra orchestration infrastructure is needed.
• Supports DAGs, step-based, and looping workflows, plus templates for reusability.
• Integrates with CI/CD and ML pipelines for enhanced versatility.

Cons:

• Cannot run outside Kubernetes, making it unsuitable for teams without a Kubernetes cluster.
• The web UI is minimal, mainly for workflow visualization and basic logs.
• No native non-container support.

Why Choose Argo Workflows Over Apache Airflow

Lightweight, serverless orchestration: Argo has no central scheduler or heavy server component. Each workflow is managed by the Kubernetes controller, reducing operational overhead compared to Airflow’s scheduler and web server setup.

Fine-grained resource control: Argo leverages Kubernetes’ native CPU, memory, and GPU allocation per task. Teams can optimize resource usage per workflow step, which is harder to manage with Airflow’s generic executor settings.

Pricing:

Argo Workflows is free to use.

Why Are People Moving Away From Apache Airflow?

Here are the top reasons people are moving away from Apache Airflow based on real feedback:

1. Difficult setup

Users find Airflow’s initial installation and infrastructure setup complex and time-consuming. Configuring the scheduler, web server, database, and workers often requires strong DevOps skills, which can be a barrier for smaller teams.

Airflow can be a bit challenging to set up and configure initially, especially when deploying in production with multiple workers and schedulers. Resource management and scaling sometimes require additional tuning, and debugging can be tricky for new users.

Aditya R

Sofware Development Engineer

View Review

2. Steep learning curve

Users must understand core concepts like DAG structure, task dependencies, executors, operators, scheduling semantics, and backfills before they can build reliable pipelines. For teams without strong Python skills, onboarding can be slow and resource-intensive.

The learning curve is pretty steep, particularly when configuring the scheduler and managing task dependencies. Sometimes Airflow’s web UI feels sluggish, and troubleshooting issues can get complicated with complex DAGs. Also, while there are a lot of integrations, keeping dependencies compatible during upgrades isn’t always smooth.

Bikash S

DevOps Engineer

View Review

3. Outdated user interface

Although the UI provides visibility into DAG runs, task states, and logs, many reviewers feel it lacks the polish and usability of modern data tools. Navigating complex workflows or investigating failures can feel clunky, especially when managing a large number of DAGs.

Sometimes setting it up in the start feels a bit confusing, and if something breaks it can take time to figure out why. The UI also feels a bit old, wish it was more simple to use.

Atin K

Senior Analyst (Planning and Replenishment)

View Review

4. Fragmented logging and debugging challenges

Users often point out that debugging failed workflows can be tedious. Logs are task-specific and sometimes require navigating multiple screens to trace the root cause of an issue. In distributed setups, centralized logging must be configured separately, adding more operational overhead.

The only complaint I have with the actual coding is that Jinja is hard to learn and debugging it can be a nightmare. That being said, if you stay within the straight-forward use cases, you shouldn\'t have any issues.

Verified User in Financial Services This reviewer\'s identity has been verified by our review moderation team. They have asked not to show their name, job title, or picture.

View Review

5. Limited real-time support

Airflow is primarily built for batch scheduling, which limits its suitability for real-time or event-driven use cases. While workarounds exist, they can add complexity and reduce efficiency. Teams that need instant triggers or real-time workflows have to rely on alternative tools built specifically for those use cases.

Key Factors for Choosing an Apache Airflow Alternative

When evaluating an Apache Airflow alternative, focus on the factors that directly impact performance, usability, and long-term sustainability:

1. Scalability and reliability

As workloads grow, your orchestration tool must handle distributed execution without performance bottlenecks. Look for built-in fault tolerance, automatic retries, and strong failure recovery mechanisms to ensure workflows run consistently at scale.

2. Ease of use

A strong alternative should reduce operational friction. Clean UI, intuitive workflow design, simpler configuration, and better debugging capabilities can significantly lower onboarding time and ongoing maintenance effort.

3. Integration and ecosystem compatibility

Seamless integration with your existing cloud platform, databases, and third-party tools is critical. Native connectors and deep cloud integrations reduce the need for custom development and accelerate deployment.

4. Cost and operational overhead

Beyond licensing, consider infrastructure, maintenance, and DevOps effort. Some tools may appear affordable initially but require heavy management or scaling costs over time. Evaluate the total cost of ownership before committing and go for a tool that offers transparent, predictable pricing.

Conclusion

The right Apache Airflow alternative depends on your scalability, ease of use, and integration needs. If you need flexibility, tools like Dagster or Prefect work well, while AWS Step Functions and Azure Data Factory suit cloud-native workflows.

Looking for a simpler, no-code way to automate data pipelines? Try Hevo Data—a fully managed solution that lets you seamlessly integrate and transform data without the complexity. Sign up for a 14-day free trial. 

Frequently Asked Questions