🖇🍀 Composable Systems

Desirable Properties

Deterministic operations

  • A given input will always produce the same output
  • Easily cacheable
  • Verifiable by multiple actors
  • Integers
    • e.g. Multiply two integers
    • e.g. Adding two integers
    • e.g. Finding the maximum of two integers
  • Strings
    • e.g. Converting strings to lowercase or uppercase
    • e.g. Concatenating strings
    • e.g. Substring of another string
  • List
    • e.g. Sorting a list of items
    • e.g. Removing duplicates from a list of items
  • Binary data
    • e.g. SHA256 hash digest (RFC 6234)
    • e.g. Base64 decoding (RFC 4648)
    • e.g. Gzip decoding (RFC 1952) 
# Converting to lowercase
> echo "AbC" | tr "[:upper:]" "[:lower:]"
abc

# SHA256 hash digest
> echo "abc" | shasum -a 256
edeaaff3f1774ad2888673770c6d64097e391bc362d7d6fb34982ddf0efd18cb  -

# Base64 encoding
> echo "abc" | base64
YWJjCg==

# Base64 decoding
> echo "YWJjCg==" | base64 -d
abc

# Gzip decompressing
> echo "H4sIADFoHGAAA0tMSuYCAE6BiEcEAAAA" | base64 --decode | gzip --decompress
abc

Unique values

  • Will never clash with existing values.
  • Choice: Unique within a context or globally unique?
  • Warning: just because a value is unique does not mean that it is unpredictable.
  • e.g. Auto-incrementing SQL primary key
    • Con: might be predictable as it is deterministic.
  • e.g. UUID (RFC 4122)
  • e.g. A id attribute used for linking a <label> to its <input>
  • e.g. A key for a React element within an array
    • It only has to be unique within the array.
  • e.g. SHA256 hash digest (RFC 6234) 
# UUID
> uuidgen
5F36D0E2-F524-46B8-870B-9AA70128F8AF

# Cryptographically random source
> openssl rand -base64 20
lrtijvqi3Dz4YrHZMQpcdfqTkJ4=

# Auto-incrementing SQL primary key
> sqlite3
sqlite> create table items (id integer primary key);
sqlite> insert into items values (?);
sqlite> insert into items values (?);
sqlite> insert into items values (?);
sqlite> select id from items;
1
2
3
sqlite>

Immutable values

  • Can never be modified.
  • Changes to a data structure work on a copy, preserving the original.
  • Can help with caching: given I have an resource’s ID, the contents will always be the same.
    • e.g. Tweets on Twitter cannot be edited, only deleted.
    • e.g. Uploaded videos on YouTube cannot be edited, only deleted.
  • Can help with syncing: retrieve only the values I do not yet have.
    • e.g. Committed files in Git become immutable objects stored in .git/objects/. When you git fetch new objects are downloaded from your remote.

Idempotent operations

  • The same result is produced if run once, twice, or a thousand times
  • e.g. Adding an item to a set
  • e.g. Removing an item from a set
  • e.g. Sorting a list of items repeatedly
  • e.g. HTTP PUT request
  • e.g. Consumers of an at-least-once event delivery system
  • e.g. Stripe charges
  • Tip: To get idempotent behavior, you could generate a random identifier for each request, and have the receiver record which requests have been processed, and then skip whenever one is repeated.
  • See also: Distributed Systems Shibboleths

Versioned interfaces