Update cache every run. Add /compiled and /logs. Make key sensitive to matrix. (#71)

Co-authored-by: Rik Huijzer <github@huijzer.xyz>
Co-authored-by: Sascha Mann <git@mail.saschamann.eu>

.github/workflows/CI.yml

@@ -6,10 +6,12 @@ on:
       - main
     paths:
       - 'action.yml'
+      - 'handle_caches.jl'
       - '.github/**'
   pull_request:
     paths:
       - 'action.yml'
+      - 'handle_caches.jl'
       - '.github/**'
 
 jobs:
@@ -48,6 +50,26 @@ jobs:
       shell: 'julia --color=yes {0}'
       run: 'using Pkg; Pkg.add("pandoc_jll")'
 
+  # Do tests with no matrix also given the matrix is auto-included in cache key
+  test-save-nomatrix:
+    needs: generate-key
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+    - name: Save cache
+      id: cache
+      uses: ./
+      with:
+        cache-name: ${{ needs.generate-key.outputs.cache-name }}
+    - name: Check no artifacts dir
+      shell: 'julia --color=yes {0}'
+      run: |
+        dir = joinpath(first(DEPOT_PATH), "artifacts")
+        @assert !isdir(dir)
+    - name: Install a small binary
+      shell: 'julia --color=yes {0}'
+      run: 'using Pkg; Pkg.add("pandoc_jll")'
+
   test-restore:
     needs: [generate-key, test-save]
     strategy:
@@ -72,13 +94,47 @@ jobs:
     - name: Check existance or emptiness of affected dirs
       shell: 'julia --color=yes {0}'
       run: |
-        # Artifacts and Packages should exist as they've been cached
+        # These dirs should exist as they've been cached
         artifacts_dir = joinpath(first(DEPOT_PATH), "artifacts")
         @assert !isempty(readdir(artifacts_dir))
         packages_dir = joinpath(first(DEPOT_PATH), "packages")
         @assert !isempty(readdir(packages_dir))
-
-        # Caching the compiled dir is disabled by default and should not exist after restoring a cache
         compiled_dir = joinpath(first(DEPOT_PATH), "compiled")
-        @assert !isdir(compiled_dir) || isempty(readdir(compiled_dir))
+        @assert !isempty(readdir(compiled_dir))
+        scratchspaces_dir = joinpath(first(DEPOT_PATH), "scratchspaces")
+        @assert !isempty(readdir(scratchspaces_dir))
+        logs_dir = joinpath(first(DEPOT_PATH), "logs")
+        @assert !isempty(readdir(logs_dir))
+
+  test-restore-nomatrix:
+    needs: [generate-key, test-save-nomatrix]
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+    - name: Restore cache
+      id: cache
+      uses: ./
+      with:
+        cache-name: ${{ needs.generate-key.outputs.cache-name }}
+    - name: Test cache-hit output
+      shell: 'julia --color=yes {0}'
+      run: |
+        @show ENV["cache-hit"]
+        @assert ENV["cache-hit"] == "true"
+      env:
+        cache-hit: ${{ steps.cache.outputs.cache-hit }}
+    - name: Check existance or emptiness of affected dirs
+      shell: 'julia --color=yes {0}'
+      run: |
+        # These dirs should exist as they've been cached
+        artifacts_dir = joinpath(first(DEPOT_PATH), "artifacts")
+        @assert !isempty(readdir(artifacts_dir))
+        packages_dir = joinpath(first(DEPOT_PATH), "packages")
+        @assert !isempty(readdir(packages_dir))
+        compiled_dir = joinpath(first(DEPOT_PATH), "compiled")
+        @assert !isempty(readdir(compiled_dir))
+        scratchspaces_dir = joinpath(first(DEPOT_PATH), "scratchspaces")
+        @assert !isempty(readdir(scratchspaces_dir))
+        logs_dir = joinpath(first(DEPOT_PATH), "logs")
+        @assert !isempty(readdir(logs_dir))
 

README.md

@@ -1,6 +1,6 @@
 # julia-actions/cache Action
 
-A shortcut action to cache Julia artifacts, packages and (optionally) registries to reduce GitHub Actions running time.
+A shortcut action to cache Julia depot contents to reduce GitHub Actions running time.
 
 ## Usage
 
@@ -22,8 +22,7 @@ jobs:
     - uses: julia-actions/julia-runtest@v1
 ```
 
-By default, this caches the files in `~/.julia/artifacts/` and `~/.julia/packages/`.
+By default the majority of the depot is cached. To also cache `~/.julia/registries/`, use
-To also cache `~/.julia/registries/`, use
 
 ```yaml
     - uses: julia-actions/cache@v1
@@ -31,28 +30,62 @@ To also cache `~/.julia/registries/`, use
         cache-registries: "true"
 ```
 
-Note that caching the registries may actually slow down the workflow running time on Windows runners.
+However note that caching the registries may mean that the registry will not be updated each run.
-That is why caching the registries is disabled by default.
 
-### Inputs
+### Optional Inputs
 
-- `cache-name` - Name used as part of the cache keys
+- `cache-name` - The cache key prefix. Defaults to `julia-cache`. The key body automatically includes matrix vars and the OS. Include any other parameters/details in this prefix to ensure one unique cache key per concurrent job type.
-- `cache-artifacts` - Whether to cache `~/.julia/artifacts/`. Enabled by default.
+- `include-matrix` - Whether to include the matrix values when constructing the cache key. Defaults to `true`.
-- `cache-packages` - Whether to cache `~/.julia/packages/`. Enabled by default.
+- `cache-artifacts` - Whether to cache `~/.julia/artifacts/`. Defaults to `true`.
-- `cache-registries` - Whether to cache `~/.julia/registries/`. Disabled by default.
+- `cache-packages` - Whether to cache `~/.julia/packages/`. Defaults to `true`.
-- `cache-compiled` - Whether to cache `~/.julia/compiled/`. Disabled by default. **USE ONLY IF YOU KNOW WHAT YOU'RE DOING!** See [#11](https://github.com/julia-actions/cache/issues/11).
+- `cache-registries` - Whether to cache `~/.julia/registries/`. Defaults to `false`. Disabled to ensure CI gets latest versions.
-- `cache-scratchspaces` - Whether to cache `~/.julia/scratchspaces/`. Enabled by default.
+- `cache-compiled` - Whether to cache `~/.julia/compiled/`. Defaults to `true`.
+- `cache-scratchspaces` - Whether to cache `~/.julia/scratchspaces/`. Defaults to `true`.
+- `cache-log` - Whether to cache `~/.julia/logs/`. Defaults to `true`. Helps auto-`Pkg.gc()` keep the cache small.
+- `delete-old-caches` - Whether to delete old caches for the given key. Defaults to `true`
 
 ### Outputs
 
 - `cache-hit` - A boolean value to indicate an exact match was found for the primary key. Returns \"\" when the key is new. Forwarded from actions/cache.
 
-## How it works
+## How It Works
 
 This action is a wrapper around <https://github.com/actions/cache>.
 In summary, this action stores the files in the aforementioned paths in one compressed file when running for the first time.
-This cached file is then restored upon the second run.
+This cached file is then restored upon the second run, and afterwards resaved under a new key, and the previous cache deleted.
-The benefit of this is that downloading one big file is quicker than downloading many different files from many different locations.
+The benefit of caching is that downloading one big file is quicker than downloading many different files from many different locations
+and precompiling them.
+
+### Cache keys
+
+The cache key that the cache will be saved as is based on:
+- The `cache-name` input
+- All variables in the `matrix` (unless disabled via `include-matrix: 'false'`)
+- The `runner.os` (may be in the matrix too, but included for safety)
+- The run id
+- The run attempt number
+
+> [!NOTE]
+> If in your workflow if you do not use a matrix for concurrency you should make `cache-name` such that it is unique for
+> concurrent jobs, otherwise caching may not be effective.
+
+### Cache Retention
+
+This action automatically deletes old caches that match the first 4 fields of the above key:
+- The `cache-name` input
+- All variables in the `matrix` (unless disabled via `include-matrix: 'false'`)
+- The `runner.os` (may be in the matrix too, but included for safety)
+
+Which means your caches files will not grow needlessly. Github also deletes cache files after
+[90 days which can be increased in private repos to up to 400 days](https://docs.github.com/en/organizations/managing-organization-settings/configuring-the-retention-period-for-github-actions-artifacts-and-logs-in-your-organization)
+
+To disable deletion set input `delete-old-caches: 'false'`.
+
+### Cache Garbage Collection
+
+Caches are restored and re-saved after every run, retaining the state of the depot throughout runs.
+Their size will be regulated like a local depot automatically by the automatic `Pkg.gc()` functionality that
+clears out old content, which is made possible because the `/log` contents are cached.
 
 ## Third Party Notice
 

action.yml

@@ -8,8 +8,11 @@ branding:
 
 inputs:
   cache-name:
-    description: 'Name used as part of the cache keys'
+    description: 'The cache key prefix. Unless disabled the key body automatically includes matrix vars, and the OS. Include any other parameters/details in this prefix to ensure one unique cache key per concurrent job type.'
     default: 'julia-cache'
+  include-matrix:
+    description: 'Whether to include the matrix values when constructing the cache key'
+    default: 'true'
   cache-artifacts:
     description: 'Whether to cache ~/.julia/artifacts/'
     default: 'true'
@@ -17,14 +20,20 @@ inputs:
     description: 'Whether to cache ~/.julia/packages/'
     default: 'true'
   cache-registries:
-    description: 'Whether to cache ~/.julia/registries/'
+    description: 'Whether to cache ~/.julia/registries/. This is off by default to ensure CI gets latest versions'
     default: 'false'
   cache-compiled:
-    description: 'Whether to cache ~/.julia/compiled. USE WITH CAUTION! See https://github.com/julia-actions/cache/issues/11 for caveats.'
+    description: 'Whether to cache ~/.julia/compiled/'
-    default: 'false'
+    default: 'true'
   cache-scratchspaces:
     description: 'Whether to cache ~/.julia/scratchspaces/'
     default: 'true'
+  cache-logs:
+    description: 'Whether to cache ~/.julia/logs/. This helps automatic Pkg.gc() keep the cache size down'
+    default: 'true'
+  delete-old-caches:
+    description: 'Whether to delete old caches for the given key'
+    default: 'true'
 
 outputs:
   cache-hit:
@@ -43,18 +52,52 @@ runs:
         [ "${{ inputs.cache-registries }}" = "true" ] && R_PATH="~/.julia/registries"
         echo "registries-path=$R_PATH" >> $GITHUB_OUTPUT
         [ "${{ inputs.cache-compiled }}" = "true" ] && PCC_PATH="~/.julia/compiled"
-        echo "precompilation-cache-path=$PCC_PATH" >> $GITHUB_OUTPUT
+        echo "compiled-path=$PCC_PATH" >> $GITHUB_OUTPUT
         [ "${{ inputs.cache-scratchspaces }}" = "true" ] && S_PATH="~/.julia/scratchspaces"
         echo "scratchspaces-path=$S_PATH" >> $GITHUB_OUTPUT
+        [ "${{ inputs.cache-logs }}" = "true" ] && L_PATH="~/.julia/logs"
+        echo "logs-path=$L_PATH" >> $GITHUB_OUTPUT
       shell: bash
 
-    - uses: actions/cache@704facf57e6136b1bc63b828d79edcd491f0ee84
+      # MATRIX_STRING is a join of all matrix variables that helps concurrent runs have a unique cache key.
+      # The underscore at the end of the restore key demarks the end of the restore section. Without this
+      # a runner without a matrix has a restore key that will cause impropper clearing of caches from those
+      # with a matrix.
+    - id: keys
+      run: |
+        MATRIX_STRING="${{ join(matrix.*, '-') }}"
+        [ -n "$MATRIX_STRING" ] && MATRIX_STRING="-${MATRIX_STRING}"
+        RESTORE_KEY="${{ inputs.cache-name }}-${{ runner.os }}${MATRIX_STRING}_"
+        echo "restore-key=${RESTORE_KEY}" >> $GITHUB_OUTPUT
+        echo "key=${RESTORE_KEY}${{ github.run_id }}-${{ github.run_attempt }}" >> $GITHUB_OUTPUT
+      shell: bash
+
+    - uses: actions/cache@4d4ae6ae148a43d0fd1eda1800170683e9882738
       id: cache
       with:
-        path: "${{ format('{0}\n{1}\n{2}\n{3}\n{4}', steps.paths.outputs.artifacts-path, steps.paths.outputs.packages-path, steps.paths.outputs.registries-path, steps.paths.outputs.precompilation-cache-path, steps.paths.outputs.scratchspaces-path) }}"
+        path: |
-        key: ${{ runner.os }}-${{ inputs.cache-name }}-${{ hashFiles('**/Project.toml') }}
+          ${{ steps.paths.outputs.artifacts-path }}
-        restore-keys: |
+          ${{ steps.paths.outputs.packages-path }}
-          ${{ runner.os }}-${{ inputs.cache-name }}-
+          ${{ steps.paths.outputs.registries-path }}
+          ${{ steps.paths.outputs.scratchspaces-path }}
+          ${{ steps.paths.outputs.compiled-path }}
+          ${{ steps.paths.outputs.logs-path }}
+
+        key: ${{ steps.keys.outputs.key }}
+        restore-keys: ${{ steps.keys.outputs.restore-key }}
+        enableCrossOsArchive: false
+
+    # github and actions/cache doesn't provide a way to update a cache at a given key, so we delete any
+    # that match the restore key just before saving the new cache
+    - uses: pyTooling/Actions/with-post-step@adef08d3bdef092282614f3b683897cefae82ee3
+      if: ${{ inputs.delete-old-caches == 'true' }}
+      with:
+        # seems like there has to be a `main` step in this action. Could list caches for info if we wanted
+        # main:  julia ${{ github.action_path }}/handle_caches.jl "${{ github.repository }}" "list"
+        main: du -shc ~/.julia/* || true
+        post: julia ${{ github.action_path }}/handle_caches.jl "${{ github.repository }}" "rm" "${{ steps.keys.outputs.restore-key }}"
+      env:
+        GH_TOKEN: ${{ github.token }}
 
     - id: hit
       run: echo "cache-hit=$CACHE_HIT" >> $GITHUB_OUTPUT

handle_caches.jl

@@ -0,0 +1,45 @@
+using Pkg, Dates
+function handle_caches()
+    repo = ARGS[1]
+    func = ARGS[2]
+    restore_key = get(ARGS, 3, "")
+
+    if func == "list"
+        println("Listing existing caches")
+        run(`gh cache list --limit 100 --repo $repo`)
+    elseif func == "rm"
+        caches = String[]
+        for _ in 1:5 # limit to avoid accidental rate limiting
+            hits = split(strip(read(`gh cache list --limit 100 --repo $repo`, String)), keepempty=false)
+            search_again = length(hits) == 100
+            filter!(contains(restore_key), hits)
+            isempty(hits) && break
+            # We can delete everything that matches the restore key because the new cache is saved later.
+            for c in hits
+                try
+                    run(`gh cache delete $(split(c)[1]) --repo $repo`)
+                catch e
+                    @error e
+                end
+            end
+            append!(caches, hits)
+            search_again || break
+        end
+        if isempty(caches)
+            println("No existing caches found for restore key `$restore_key`")
+        else
+            println("$(length(caches)) existing caches deleted that match restore key `$restore_key`:")
+            println.(caches)
+        end
+    else
+        throw(ArgumentError("Unexpected second argument: $func"))
+    end
+end
+
+try
+    # do a gc with the standard 7-day delay
+    Pkg.gc()
+    handle_caches()
+catch e
+    @error "An error occurred while managing existing caches" e
+end