Marmot: Multi-writer distributed SQLite based on NATS

Marmot: Multi-writer distributed SQLite based on NATS

Go Report Card

What & Why?

Marmot is a distributed SQLite replicator with leaderless, and eventual consistency. It allows you to build a robust replication
between your nodes by building on top of fault-tolerant NATS JetStream.

So if you are running a read heavy website based on SQLite, you should be easily able to scale it out by adding more SQLite replicated nodes.
SQLite is probably the most ubiquitous DB that exists almost everywhere, Marmot aims to make it even more ubiquitous for server
side applications by building a replication layer on top.

Quick Start

Download latest Marmot and extract package using:

tar vxzf marmot-v*.tar.gz

From extracted directory run examples/run-cluster.sh. Make a change in /tmp/marmot-1.db using:

bash> sqlite3 /tmp/marmot-1.db
sqlite3> INSERT INTO Books (title, author, publication_year) VALUES (‘Pride and Prejudice’, ‘Jane Austen’, 1813);

Now observe changes getting propagated to other database /tmp/marmot-2.db:

bash> sqlite3 /tmp/marmot-2.db
sqlite3> SELECT * FROM Books;

You should be able to make changes interchangeably and see the changes getting propagated.

Out in wild

Here are some official, and community demos/usages showing Marmot out in wild:

2-node HA for edge Kubernetes – Using Marmot
Scaling Isso with Marmot on Fly.io
Scaling PocketBase with Marmot on Fly.io
Scaling PocketBase with Marmot 0.4.x
Scaling Keystone 6 with Marmot 0.4.x

What is the difference from others?

Marmot is essentially a CDC (Change Data Capture) and replication pipeline running top of NATS. It can automatically configure appropriate
JetStreams making sure those streams evenly distribute load over those shards, so scaling simply boils down to adding more nodes, and
re-balancing those JetStreams (auto rebalancing not implemented yet).

There are a few solutions like rqlite, dqlite, and
LiteFS etc. All of them either are layers on top of SQLite (e.g.
rqlite, dqlite) that requires them to sit in the middle with network layer in order to provide
replication; or intercept physical page level writes to stream them off to replicas. In both
cases they require a single primary node where all the writes have to go, and then these
changes are applied to multiple readonly replicas.

Marmot on the other hand is born different. It’s born to act as a side-car to your existing processes:

Instead of requiring single primary, there is no primary! Which means any node can make changes to its local DB.
Marmot will use triggers to capture your changes, and then stream them off to NATS.
Instead of being strongly consistent, Marmot is eventually consistent. Which means no locking, or blocking of nodes.
It does not require any changes to your existing SQLite application logic for reading/writing.

Making these choices has multiple benefits:

You can read, and write to your SQLite database like you normally do. No extension, or VFS changes.
You can write on any node! You don’t have to go to single primary for writing your data.
As long as you start with same copy of database, all the mutations will eventually converge
(hence eventually consistent).

What happens when there is a race condition?

In Marmot every row is uniquely mapped to a JetStream. This guarantees that for any node to publish changes for a row it has to go through
same JetStream as everyone else. If two nodes perform a change to same row in parallel, both of the nodes will compete to publish their
change to JetStream cluster. Due to RAFT quorum
constraint only one of the writer will be able to get its changes published first. Now as these changes are applied (even the publisher applies
its own changes to database) the last writer will always win. This means there is NO serializability guarantee of a transaction
spanning multiple tables. This is a design choice, in order to avoid any sort of global locking, and performance.

Limitations

Right now there are a few limitations on current solution:

Marmot does not support schema changes propagation, so any tables you create or columns you change won’t be reflected.
This feature is being debated and will be available in future
versions of Marmot.
You can’t watch tables selectively on a DB. This is due to various limitations around snapshot and restore mechanism.
WAL mode required – since your DB is going to be processed by multiple processes the only way to have multi-process
changes reliably is via WAL.
Marmot is eventually consistent – This simply means rows can get synced out of order, and SERIALIZABLE assumptions
on transactions might not hold true anymore. However your application can choose to redirect writes to single node
so that your changes are always replayed in order.

Features




Leaderless replication never requiring a single node to handle all write load.

Ability to snapshot and fully recover from those snapshots. Multiple storage options for snapshot:

Built with NATS, abstracting stream distribution and replication.

Support for log entry compression, handling content heavy CMS needs.

Sleep timeout support for serverless scenarios.

Dependencies

Starting 0.8+ Marmot comes with embedded nats-server with JetStream support. This not only reduces
the dependencies/processes that one might have to spin up, but also provides with out-of-box tooling like
nat-cli. You can also use existing libraries to build additional
tooling and scripts due to standard library support. Here is one example using Deno:

deno run –allow-net https://gist.githubusercontent.com/maxpert/d50a49dfb2f307b30b7cae841c9607e1/raw/6d30803c140b0ba602545c1c0878d3394be548c3/watch-marmot-change-logs.ts -u -p -s

The output will look something like this:

Production status

v0.8.x introduced support for embedded NATS. This is recommended version for production.
v0.7.x moves to file based configuration rather than CLI flags, and S3 compatible snapshot storage.
v0.6.x introduces snapshot save/restore. It’s in pre-production state.
v0.5.x introduces change log compression with zstd.
v0.4.x introduces NATS based change log streaming, and continuous multi-directional sync.
v0.3.x is deprecated, and unstable. DO NOT USE IT IN PRODUCTION.

CLI Documentation

Marmot picks simplicity, and lesser knobs to configure by choice. Here are command line options you can use to
configure marmot:

config – Path to a TOML configuration file. Check out config.toml comments for detailed documentation
on various configurable options.
cleanup (default: false) – Just cleanup and exit marmot. Useful for scenarios where you are
performing a cleanup of hooks and change logs.
save-snapshot (default: false Since 0.6.x) – Just snapshot the local database, and upload snapshot
to NATS/S3 server
cluster-addr (default: none Since 0.8.x) – Sets the binding address for cluster, when specifying
this flag at-least two nodes will be required (or replication_log.replicas). It’s a simple
: pair that can be used to bind cluster listening server.

Since v0.8.4 Marmot will automatically expose a leaf server on :. This is
intended to reduce the number for flags. So if you expose cluster on port 4222 the port 4223 will
be automatically a leaf server listener.

cluster-peers (default: none Since 0.8.x) – Comma separated list of nats://:/ peers of
NATS cluster. You can also use (Since version v0.8.4 ) dns://:/ to A/AAAA record lookups.
Marmot will automatically resolve the DNS IPs at boot time to expand the routes with value of
nats://:/ value, where is replaced with all the DNS entries queried. There
are two additional query parameters you can use:

min – forcing Marmot to wait for minimum number of entries (e.g. dns://foo:4222/?min=3 will require
3 DNS entries to be present before embedded NATs server is started)
interval_ms – delay between DNS queries, which will prevent Marmot from flooding DNS server.

leaf-server (default: none Since v0.8.4 )- Comma separated list of nats://:/
or dns://:/ just like cluster-peers can be used to connect to a cluster
as a leaf node.

For more details and internal workings of marmot go to these docs.

FAQs & Community

For FAQs visit this page
For community visit our discord or discussions on GitHub

Our sponsor

Last but not least we would like to thank our sponsors who have been supporting development of this project.


>>> Read full article>>>
Copyright for syndicated content belongs to the linked Source : Hacker News – https://github.com/maxpert/marmot

Exit mobile version