Eager & Lazy Caching feat. Elixir

At Tigris we offer a number of novel and practical improvements beyond what your typical Object Storage does. We fit these within the existing common APIs or as graceful extensions when necessary. In this post we look at how you can take control of the Tigris caching mechanism if you feel the need.

To try the examples and follow along you can install the flyctl CLI tool and then run flyctl storage create to get credentials. If you use Livebook, the collaborative coding notebook for Elixir, this entire post can be used from Livebook by going here. Or you can copy and paste the examples into a .exs file to run Elixir as a script.

You need to install: ex_aws, ex_aws_s3, hackney, poison and sweet_xml

And you need the following config, similar setup will be needed in your preferred language if you don't do Elixir:

Mix.install([
  {:ex_aws, "~> 2.5"},
  {:ex_aws_s3, "~> 2.5"},
  {:hackney, "~> 1.20"},
  {:poison, "~> 3.0"},
  {:sweet_xml, "~> 0.6"}
])

Application.put_env(:ex_aws, :s3,
  scheme: "https://",
  host: "fly.storage.tigris.dev",
  port: 443
)

# Livebook prefixes env vars with LB_ and
# we strip that out for ex_aws_s3
[
  "AWS_ACCESS_KEY_ID",
  "AWS_ENDPOINT_URL_S3",
  "AWS_REGION",
  "AWS_SECRET_ACCESS_KEY",
  "BUCKET_NAME"
]
|> Enum.each(fn key ->
  val = System.fetch_env!("LB_#{key}")
  System.put_env(key, val)
end)

alias ExAws.S3

bucket = System.fetch_env!("BUCKET_NAME")

Regular PUT and GET

With Tigris we can do the usual things for object storage. PUT things in the bucket and then GET them out of the bucket.

# The PUT
bucket
|> S3.put_object("myfile.txt", "mycontents")
|> ExAws.request!()
|> Map.fetch!(:status_code)

# The GET
bucket
|> S3.get_object("myfile.txt")
|> ExAws.request!()
|> Map.fetch!(:body)

Not quite so usual is that we offer a whole CDN experience along with your bucket. And as with any reasonable CDN we will do adaptive caching based on where requests originate from and try to offer your users the best low-latency experience in that way. The automatic way is very often the best way. As outlined in the caching docs there are also some sound defaults in place for file types that are typically static assets.

Of course you deserve more control for when that is desirable. So you can upload with a particular cache header set which will then be honored by Tigris.

# The PUT with cache headers
bucket
|> S3.put_object("myfile-2.txt", "mycontents")
# a minor kludge to set the custom header, will see if we can PR the library :)
|> then(fn op ->
  headers = Map.put(op.headers, "Cache-Control", "public, max-age=30")
  %{op | headers: headers}
end)
|> ExAws.request!()

# The GET, now with cache headers
bucket
|> S3.get_object("myfile-2.txt")
|> ExAws.request!()
|> Map.fetch!(:headers)

But in many cases you may know your access patterns or have particular plans. You may want to ensure eager caching of uploaded files, where every file uploaded gets spread across a decent chunk of the world. This is possible by setting the bucket accelerate configuration. I would try this for bucket intended to do HTTP Live Streaming (HLS) where latency can really matter or for podcast recordings where you might expect a lot of geographically distributed clients will request the thing at once.

With AWS CLI it looks like this, unfortunately ExAws doesn't expose a way to do this so far:

aws s3api put-bucket-accelerate-configuration \
	--bucket foo-bucket \
	--accelerate-configuration Status=Enabled

Pre-fetching object listings

Another cool way to control the flow of caching and replication is via Eager caching when listing objects. This allows you to tell Tigris that you'd like all the files you are listing to move to your region and be ready and nearby for subsequent fetching. With one header.

bucket
|> S3.list_objects_v2(headers: %{"X-Tigris-Prefetch" => "true"})
|> ExAws.request!()
|> Map.fetch!(:body)
|> Map.fetch!(:contents)
|> Enum.map(& &1.key)

With that all the object data behind the keys you have listed immediately start moving to you across the network. This is incredibly convenient when traversing a large number of files stored and wanting to make fetching efficient for many small files where latency can otherwise ruin your throughput.

Now we've covered how to wrangle eager and lazy caching on Tigris with Elixir. This is just the beginning, we have a lot more to cover.

Check back soon :)