Skip to main content

How we built our data layer on FoundationDB

ยท 11 min read
Himank Chaudhary

Tigris is an open source developer data platform that makes building data-rich serverless applications a breeze. It enables developers to stick to just being developers and not be forced into DevOps.

Tigris uses FoundationDB's transactional key-value interface as its underlying storage engine. In our blog post Skipping the boring parts of building a database using FoundationDB we went into the details of why we chose to build on FoundationDB. To recap, FoundationDB is an ordered, transactional, key-value store with native support for multi-key strictly serializable transactions across its entire keyspace. We leverage FoundationDB to handle the hard problems of durability, replication, sharding, transaction isolation, and load balancing so we can focus on higher-level concerns.

We are starting a series of blog posts that go into the details of how Tigris has been implemented. In the first post of the series, we will share the details of how we have built the multi-model document layer on top of FoundationDB. We will cover the topics of data layout, and schema management.

How we architected Tigris

Data layoutโ€‹

To understand the data layout, the first step is to talk about how data is modeled in Tigris.

Data modelingโ€‹

Tigris stores data records as documents. Documents are analogous to JSON objects but Tigris stores them in an optimized binary format. Documents are grouped together in collections. Collections are grouped together in databases.

{
"field1": 1,
"field2": "string",
"field3": { "field1": "value1", "field2": "value2" },
"field4": [1, 2, 3]
}

You can read more about the data modeling concepts in the Documents section of the docs.

As Tigris is a multi-tenant system, when a user is created they are assigned to a tenant. All of their data is then stored under this tenant. Thus, the hierarchy of data storage looks something like this

Data storage hierarchy in Tigris

Data layoutโ€‹

As Tigris leverages FoundationDB as the storage engine, which exposes a key-value interface, the data has to be stored as key-value pairs. This means there needs to be some translation from a logical layout of storing tenants, databases, collections, documents, and schemas to a physical layout.

Tigris maintains different key layouts depending on the information it stores. Each key layout has a custom encoder-decoder and has a prefix at the start of the key. The encoder adds this prefix; then, the decoder uses it to decode it according to the appropriate structure. The high-level concept of key encoding remains the same for all types of data (users or system data). The following section discusses how user data (JSON documents) are stored inside Tigris.

Key encodingโ€‹

As we have seen in the data modeling section, a collection is identified by a tenant, database, and collection name. This is the minimum information we need in the key to identify a record. However, a collection may have secondary indexes as well. Therefore, an index identifier must also be part of the key.

This key structure is made extensible by having a version component allowing us to add or remove attributes in the future.

To summarize, we need to pack the following information in the key:

tenant | database | collection | index name | index value

Taking a more realistic example, let's say we have a tenant foobar with a database userdb, a collection users, and an id field defined as the primary key of the documents. This translates to the following key structure

["fooApp", "userdb", "users", "pkey", [1]] =>
{"id": 1, "email": "alex@example.com", "phone_number": 12345}

The index values are seen as an array here because Tigris supports composite indexes as well, meaning a collection can have one or more than one field defined as index fields. These index values are packed in a single binary structure.

However, storing this information in every key as-is means unnecessary costs attached to each document. Therefore we implemented key compression.

Key compressionโ€‹

The key compression algorithm that we have implemented replaces these long strings with integers and ensures that the integer assignments are unique. This is accomplished by persisting these mapping in an internal metadata collection and assigning to these strings during creation time.

In order to compress a key, the first step is to assign a unique integer to the container (tenant, database, collection) names. This is done during a Data Definition Language (DDL) operation. Whenever Tigris receives a tenant creation, database creation, or collection creation request, it starts a transaction. In this transaction, a unique integer value is reserved for this container name and then assigned to this container. This mapping is stored in an internal collection called encoding. The value assigned is incremented in a fashion that each mapping is unique. This mapping is immutable so that it can be freely cached on Tigris servers. In other words, all the user metadata like a tenant, database, collection, and index names are uniquely identifiable by their corresponding integer representation. As this is done in a transaction, completing the request means a unique assignment of the integer to the string.

Using integers has the following benefits:

  • Machine-level instruction can be used to perform integer comparison in one cycle
  • Packed as 4bytes; on the other hand, if the string size grows, it will take up a lot of memory. This results in compact keys, therefore, optimizing CPU, memory, and storage usage
  • Even nominally saving a few bytes per string representation does add up when there are billions of records

The encoding collection that has this information and has the following layout:

encoding | version | tenant-name => integer identifier
encoding | version | tenant-id | database-name => integer identifier
encoding | version | tenant-id | database-id | collection-name => integer identifier

Where

  • encoding โ†’ identifier of this metadata collection (key layout)
  • version โ†’ version of this key structure so that we can evolve this layout
  • tenant-name, database-name, collection-name โ†’ The user-facing names

These integer values are then used to form the key. Now with this information, here is how the key structure looks like:

["usr",0x01,0x01,0x02,0x03,0x04,[1]] =>
{"id": 1, "email": "alex@example.com", "phone_number": 12345}

Where

  • usr โ†’ the identifier for this key layout
  • 0x01 โ†’ version of this key layout
  • 0x01 โ†’ tenant-id
  • 0x02 โ†’ database-id
  • 0x03 โ†’ collection-id

Value encodingโ€‹

Internally all user values are stored inside a protobuf message. We don't mutate user payload. Introducing this top-level structure allows us to have metadata of the record attached to it along with raw user content. As an example, this protobuf message, apart from user data, has information such as compression, created time, updated time, the schema version, and other housekeeping fields. Some of this information is then indexed so that we can support time series queries like returning records that are created after Jan 1, 2022.

message ValueWrapper {
int32 ver = 1; // schema version
int32 enc = 2; // encoding of the raw data
int32 comp = 3; // compression of the raw data
Timestamp created_at = 4; // created timestamp of the document
Timestamp updated_at = 5; // updated timestamp of the document
bytes raw = 6; // raw user payload
}

Our wrapper adds a slight overhead to every value, but provides us with much flexibility. For example, we can switch between different compression algorithms and value encodings on the fly with no downtime or backfills required. In addition, unused fields don't have to be encoded at all.

Schemasโ€‹

For Tigris, schemas are an integral part and form the basis of rich features such as automatic indexing and real-time search. Tigris enforces that all documents in a collection must conform to a schema. The schema is defined as part of creating the collection. Tigris, then, provides a lightweight way to modify the schema.

This aligns well with application development flow as schemas provide a way to structure the data according to the application logic and flexibility in modifying the schema allows for it to evolve with the application.

Schema storageโ€‹

To store schemas, there is an internal metadata collection. All operations on this collection are performed using transactions. This allows us to create a single collection, or multiple collections, or evolve a schema of collections atomically.

Tigris supports performing DDLs in an interactive transaction. Once this transaction is committed, Tigris guarantees that this schema will be atomically applied and any new request will see the latest schema. The schema change is applied in an online manner, and there is no downtime, so this operation can be safely done in production during live traffic. Since schema change for Tigris is a metadata operation, the schema change can be completed in milliseconds.

As mentioned before, there is an internal collection that stores the schemas. The key is of the following structure:

schemas | version | tenant-id | database-id | collection-id | revision-id

Where

  • schemas โ†’ identifier for this collection
  • version โ†’ version of this key structure so that we can evolve this layout
  • tenant-id โ†’ tenant integer identifier
  • database-id โ†’ database integer identifier
  • collection-id โ†’ collection integer identifier
  • revision-id โ†’ version of the schema

As an example

["schemas", 0x01, 0x01, 0x02, 0x03, 0x01] =>
{"properties":{"a":{"type":"integer"},"primary_key":["a"]}}

The schemas are cached on the Tigris servers to avoid reloading them from the storage but as we provide transactional schema updates we need to ensure the cache is always consistent and never return stale schema version. The following section talks about how these schemas are propagated atomically and how the cache always returns a consistent view of the schemas.

How are schema changes propagated instantaneously?โ€‹

One of the ways of atomically propagating schema change is for all the servers to block the transactions until all other servers have received the new schema. However, in a production environment, this behavior is highly undesirable.

Tigris has chosen a different route to avoid this coordination but still support atomic schema changes. As Tigris transactionally stores the schema, in the same transaction, it bumps up a metadata version using FoundationDB's versionstamp.

Now any request that arrives must be checked against this version.

How schema propagation happens in Tigris

Checking metadata versions at the start of every transaction is an expensive operation so Tigris has optimized this flow by attaching a FoundationDB futures to it.

FoundationDB provides asynchronous APIs that rather than blocking the calling thread until the result is available, immediately return a future object. Using futures helps us to not penalize the request and let it proceed as usual.

Once the request is processed and is ready to be committed, this future is ready. Before committing the transaction we check if the metadata version has changed. If the version is bumped for a tenant that means a schema change operation has been performed so the server reloads the latest version of the schema for this tenant otherwise the request can proceed as-is.

This guarantees atomicity of schema propagation across the Tigris servers. This is also how Tigris provides the guarantee that once success is returned for a schema change operation, the new schema is always used for subsequent requests.

Schema enforcementโ€‹

As mentioned above, Tigris enforces that all documents in a collection must conform to a schema. Schema defines all the fields that make up the document in a collection, and is required as part of creating the collection.

The schema validation happens during all write requests. Validating during writes means there is no penalty during read operations as the stored data already conforms to the schema. Tigris allows evolving schema but with some restrictions at this time.

Read the following sections in the documentation to learn more about how schemas work in Tigris:

This is the first blog post in a multi-part series where we will be sharing the details of how we have implemented some of the core features in Tigris. Be on the lookout for the next part!


Tigris is the data platform built for developers! Use it as a scalable, ACID transactional, real-time backend for your serverless applications. Build data-rich features without worrying about slow queries or missing indexes. Seamlessly implement search within your applications with its embedded search engine. Connect serverless functions with its event streams to build highly responsive applications that scale automatically.

Tigris at Github

Checkout the quickstart, give us a star on Github, and join us on slack.