Show HN: Greenmask 0.2 – Database anonymization tool (github.com)
Before I describe Greenmask’s features, I want to share the story of how and why I started implementing it.
Everyone strives to have their staging environment resemble production as closely as possible because it plays a critical role in ensuring quality and improving metrics like time to delivery. To achieve this, many teams have started migrating databases and data from production to staging environments. Obviously this requires anonymizing the data, and for this people use either custom scripts or existing anonymization software.
Having worked as a database engineer for 8 years, I frequently struggled with routine tasks like setting up development environments—this was a common request. Initially, I used custom scripts to handle this, but things became increasingly complex as the number of services grew, especially with the rise of microservices architecture.
When I began exploring tools to solve this issue, I listed my key expectations for such software: documentation; type safety (the tool should validate any changes to the data); streaming (I want the ability to stream the data while transformations are being applied); consistency (transformations must maintain constraints, functional dependencies, and more); reliability; customizability; interactivity and usability; simplicity.
I found a few options, but none fully met my expectations. Two interesting tools I discovered were pganonymizer and replibyte. I liked the architecture of Replibyte, but when I tried it, it failed due to architectural limitations.
With these thoughts in mind, I began developing Greenmask in mid-2023. My goal was to create a tool that meets all of these requirements, based on the design principles I laid out. Here are some key highlights:
* It is a single utility - Greenmask delegates the schema dump to vendor utilities and takes responsibility only for data dumps and transformations.
* Database Subset (https://docs.greenmask.io/latest/database_subset) - specify the subset condition and scale down size. We did a deep research in graph algorithms and now we can subset almost any complexity of database.
* Database type safety - it uses the DBMS driver to decode and encode data into real types (such as int, float, etc.) in the stream. This guarantees consistency and almost eliminates the chance of corrupted dumps.
* Deterministic engine (https://docs.greenmask.io/latest/built_in_transformers/trans...) - generate data using the hash engine that produces consistent output for the same input.
* Dynamic parameters for transformers (https://docs.greenmask.io/latest/built_in_transformers/dynam...) - imagine having created_at and updated_at dates with functional dependencies. Dynamic parameters ensure these dates are generated correctly.
We are actively maintaining the current project and continuously improving it—our public roadmap at https://github.com/...
19 comments
[ 2.2 ms ] story [ 66.1 ms ] threadI can absolutely speak to the pain of having a dozen pg_dump --exclude-table-data arguments and having a developer experience that makes it difficult to reproduce bugs due to drift between production data and test fixtures (even if they share the same schema, assumptions can change massively!).
Secure and robust database cloning also enables preview apps that actually answer the stakeholder question "can I see/play with what the new code would do, if applied to the actual [document/record/product listing] that motivated the feature/bugfix?" Subsetting and PII masking are both critical for this, and it's amazing to see that you've thought about them as integral parts of the same product.
I really want to see a product like this succeed! The easier the tool is to use, the harder it might be to monetize... but there are so many applications of a tool like this, including ones that can materially improve security at organizations large and small (https://nabeelqu.substack.com/i/150188028/secrets just posted here earlier today remarks on this!) that I'm sure you'll find the right niche!
> The easier the tool is to use, the harder it might be to monetize... but there are so many applications of a tool like this
I am in favour with you! That’s exactly why we've started developing our Dynamic Staging Environments platform. It will integrate seamlessly with CI/CD systems, enabling to generate, create and maintain stateful components of your services more efficiently.
One key insight we’ve gained is that this type of software introduces new responsibilities across the Dev, Business, and Security teams within a company. This can result in complex processes and interactions. For example, changes to schemas often require approvals from both the security team and service owners, which can slow down workflows. That's why I really appreciate how Bytebase (https://www.bytebase.com/) tackled this issue—by providing a platform that combines GitOps for stateful resources with business logic in a streamlined way. We should be looking in this direction.
Important Remark:
Current design of Greenmask fits well with future plathform. We remain committed to focusing on simple utilities, as they are essential—especially for small businesses. These tools serve a critical purpose by solving problems quickly for smaller companies and for individuals who need a solution that works ASAP.
And I wonder if it would be worth talking to folks who do SOC 2 auditing; if you could say "we can provide a framework that allows you to continue to be SOC 2 certified while letting your developers access real-world data" that would be tremendously valuable.
> provide GUI and governance controls on top of that?
Exactly. Data doesn’t exist in isolation. Databases are dependencies of services, and schemas evolve throughout the software lifecycle, often managed by different data migration tools. In large organisations, regular developers usually don’t have direct access to the data sources, and masking rules along with real data sources are often restricted. Schema changes must be validated by the responsible data governance teams to ensure compliance and accuracy.
That’s why we implemented the validate command even in this standalone tool, which checks for schema differences and prevents running a dump if any schema changes are detected with detailed warnings. https://docs.greenmask.io/latest/commands/validate
I once presented Greenmask at an event organized by Percona in Cyprus, and one of the questions raised was: “What if we have a staging database, but instead of cleaning up the database and data, we want to add something to the existing dataset?” At the time, I didn’t have an immediate answer. However, this question inspired me to think, and eventually, I found a solution that at least partially covers this case:
You can restore data in topological order by preserving references and ensuring proper dependency handling (https://docs.greenmask.io/latest/commands/restore/#restorati...)
You can exclude non-critical errors to streamline the process without disrupting key operations (https://docs.greenmask.io/latest/configuration/#restoration-...)
I want to emphasize that this type of software must be flexible and adaptable to meet the ever-evolving needs of businesses… Otherwise, the project is as good as dead.
——————————
> And I wonder if it would be worth talking to folks who do SOC 2 auditing
I’ve had discussions with professionals from Information Security, including those working in SOCs, and you're absolutely pointing in the right direction. At the moment, I’m actively exploring solutions and building a concept. I believe that by 2025, we’ll be able to showcase something new.
I was recently researching ways of anonymizing production data for staging, and I also found existing tools either cumbersome to setup or lacking in features.
I stumbled upon clickhouse-obfuscator[1], and really liked that it worked on standalone dump formats (CSV, Parquet, etc.) rather than any specific DBMS. I think that's a great approach for this, since it keeps things simple and generic, and it can be conveniently added as a middle step in the backup-restore pipeline. Unfortunately, the tool is quite barebones, and has issues maintaining referential integrity, so we had to abandon it.
This is still an unsolved problem in our team, so I'll keep an eye on your tool. We would need support for ClickHouse as well, so it's good you're planning support for other DBMSs. Good luck!
[1]: https://clickhouse.com/docs/en/operations/utilities/clickhou...
The other tool in this space to look at is neosync: https://www.neosync.dev/
However, I've found that these traditional methods often fall short in preserving the intricate relationships and structures within the data. That's why I was excited to discover synthetic data generation, which has been a game-changer for me. By creating artificial data that mirrors the statistical properties of the original, I've been able to share data with others without compromising sensitive information. I've used synthetic data generation for various projects and it's been a valuable tool in my toolkit.
Anyway, I will definitely try this. It looks real good!
https://docs.greenmask.io/latest/commands/validate/
Within 48 hours of needing to do some masking (exporting data from a database to CSV to import it to a less trusted domain's database for testing/benchmarking), I ran across your HN post and tool.
A pair of us had started with a different masking approach I knew was dumb but would work, but I mentioned Greenmask to the guy working with me doing masking as something to perhaps look at to see if it'd help us later/next-time.
He apparently took me up on it more aggressively than I anticipated and today he indicated he had looked at it and found it hard to understand how to use it (documentation-wise). I.e. he tried it but didn't really get anywhere.
I'm not sure precisely what he meant but my perception from whatever he said (I wasn't trying to gather feedback for you at the time) was he didn't have/see any example code showing the different things he could do.
Having skimmed your documentation for 3 minutes just now, I am inclined to agree. Additionally I would observe that your Show HN bullet explanation of your product is actually clearer/tighter than your official documentation page, where my impatient eye tends to glaze over after a few bullets of stuff I am not sure if I care about...
This was my first masking in recent times, but I may have half a dozen more times in the next year if/as we do more POCs with full data volumes to figure out price/performance trafeoffs...