#cache #command #exit #utility #output #results #replay

app deja

A simple command line utility to cache output of commands

7 releases

new 0.2.1 Nov 3, 2024
0.2.0 Nov 2, 2024
0.1.4 Oct 17, 2024

#15 in Caching

Download history 53/week @ 2024-10-03 274/week @ 2024-10-10 352/week @ 2024-10-17 5/week @ 2024-10-24 236/week @ 2024-10-31

920 downloads per month

MIT license

59KB
1.5K SLoC

Deja

deja is a utility to cache CLI commands. It captures stdout, stderr, and the exit code of a given command, and replays them when the command is run again. This is particularly useful for commands that are slow and idempotent, or return results with a known lifespan.

Here's a quick example:

$ deja run --cache-for 15m -- aws rds generate-db-auth-token …args
db.example.com:5432/?Action=connect&DBUser=user&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=ABCDEFG%2F20201010%2Feu-west-2%2Frds-db%2Faws4_request&X-Amz-Date=20201010T010101Z&X-Amz-Expires=900&X-Amz-SignedHeaders=host&X-Amz-Signature=abcdefg1234567890

For the next 15 minutes, invoking the same command will re-print the result to stdout. After 15 minutes, the command will be run again, and the new result stored in the cache.

This uses the --cache-for option to limit the cached lifespan, and there are a wide range of other options to control caching, including when environment variables change, if the contents of a file or folder change, or based on arbitary strings.

Examples of how you might use deja:

  • Run a command only once a day:
    deja run --watch-scope "$(date +%Y-%m-%d)" -- command-that-should-only-run-once-a-day
    
  • Run a command only when a file changes:
    deja run --watch-file /path/to/file -- command-that-should-only-run-when-file-changes
    
  • Play around with a slow API:
    alias deja-http="deja run --watch-scope $(uuidgen) -- http "
    deja-http https://api.example.com/slow-endpoint | jq '.[] | .date'
    deja-http https://api.example.com/slow-endpoint | jq '.[] | .name'
    
  • Use the same RDS token for its full 15 minute lifespan:
    deja run --cache-for 15m -- aws rds generate-db-auth-token…
    
  • Reuse a command throughout a build:
    deja run --watch-env BUILD_ID -- cargo audit
    
  • Run tests only when the git HEAD changes:
    deja run --watch-scope "$(git rev-parse HEAD)" -- rake test
    
  • Pre-record CLI output for a demo:
    deja run --record-exit-codes 0,1 -- ./demo.sh
    

Installation

Deja is written in rust. You can install it easily with cargo, using cargo install deja.

How deja works

For each command, deja creates a hash from the command, arguments, and other options (by default the user and working directory). If a fresh result for this hash is found in the cache, it's replayed. If not, the command is run, and when the exit code is 0, the result stored in the cache. When replaying a command, both stdout and stderr are rewritten to the terminal in the same order as recorded. Deja will then exit with the original exit code.

Deja stores cached results in a dedicated directory (by default $HOME/Library/Caches/deja on macOS, or either $XDG_CACHE_HOME/deja or $HOME/.cache/deja on Linux). Stored results are not encrypted, but are stored with permissions so only the user who created the entry can read or write to it.

Options

--cache [path] sets the path to the cache directory. If the directory does not exist, it will be created. By default deja will use $XDG_CACHE_HOME/deja or $HOME/.cache/deja on Linux, or $HOME/Library/Caches/deja on macOS.

--share-cache sets the cache to shared. By default the cache is per-user, and only the user who created the cache can read or write to it. When --share-cache is used, the cache is created with group read/write permissions, allowing other users to read and write to it.

--watch-path [path] returns the cached result until the path contents change (detected via a content hash). Multiple paths can be watched by providing the option multiple times.

  • --watch-path Gemfile.lock - Reuse the result until Gemfile.lock changes
  • --watch-path src - Reuse the result until the contents src changes

--watch-scope [scope] returns the cached result until the scope changes. This accepts any string, and combined with shell substitution can be extremely powerful:

  • --watch-scope "$(date +%Y-%m-%d)" - Reuse the result throughout the day
  • --watch-scope "$(git rev-parse HEAD)" - Reuse the result for the current git commit

As with --watch-path, --watch-scope can be provided multiple times to watch multiple scopes.

--watch-env returns the cached result until the given environment variables change. This option can be provided multiple times to watch multiple different environment variables.

--exclude-pwd removes the working directory from the cache key. Without this flag deja includes the working directory; cached results are only returned when called from the same directory. With this flag, cached results can be returned whatever directory the command is called from, but only if --exclude-pwd was originally used. A result generated without --exclude-pwd will never be returned from a different directory.

--cache-for [duration] limits for how long a cached result is valid. It accepts durations in the form 30s, 5m, 1h, 30d, etc. If a result is stored with --cache-for, it will never be returned after the duration has passed.

--record-exit-codes [codes] expands the list of exit codes deja will cache. It accepts a comma separated list of either individual codes like 0,1, inclusive ranges like 100-200, or open-ended ranges like 0+. By default, deja only caches the result of a command if the exit code is 0. In some cases you may want other exit codes to be cached, for example if grepping a huge file for a string that may or may not be present.

  • --record-exit-codes 0,1 will cache the result if the exit code is 0 or 1.
  • --record-exit-codes 0,10-12,100+ will cache the result if the exit code is 0, 10, 11, or 12, or 100 or greater.

--look-back [duration] limits how far back in time to look for a cached result. It accepts durations in the form 30s, 5m, 1h, 30d, etc. When --look-back is used, deja will only reuse a result if it was generated within the given duration. If no result is found within the period, the command will be run and the result cached.

  • --look-back 30s will return any result generated in the last 30 seconds.

--cache-miss-exit-code (for read subcommand only) returns the given exit status on cache miss.

  • deja read --cache-miss-exit-code 200 -- grep -q needle haystack will return 200 if the cache is missed, and the exit status of grep if the cache is hit.

Subcommands

run is the main subcommand, used to run a command and cache the result.

test takes the same options as run, but never runs the command. Instead it exits with a status code of 0 if a cached result is found, or 1 if not.

read never runs the given command, but will replay a cached result if one exists. If no result is found, deja will exit with a status of 1 (though this can be changed with --cache-miss-exit-code).

force always runs the given command and caches the result.

remove removes any cached result that would have been returned.

explain returns information about the given options including the hash components and the cache result (if any)

hash returns the hash used to cache results

Motivation

This utility was inspired by some code we use at Farillio to speed up our CI builds. We use rake as our main build tool, and have a custom CachedTask class that caches results. deja is an attempt to do this in a more generic, faster, flexible way.

Dependencies

~7–17MB
~247K SLoC