Cirrus: Improve caching effectiveness

.cirrus.yml

@@ -50,16 +50,15 @@ build_task:
     # all PRs & branches will share caches with other PRs and branches
     # for a given $DEST_BRANCH and vX value.  Adjust vX if cache schema
     # changes.
-    fingerprint_key: "cargo_v2_${DEST_BRANCH}_amd64"
+    fingerprint_script: echo -e "cargo_v3_${DEST_BRANCH}_amd64\n---\n$(<Cargo.lock)\n---\n$(<Cargo.toml)"yy
     # Required to be set explicitly since fingerprint_key is also set
     reupload_on_changes: true
   targets_cache: &targets_cache
-    # Similar to cargo_cache, but holds the actual compiled artifacts. This must
-    # be scoped similar to bin_cache to avoid binary pollution across cache
-    # contexts.  For example, two PRs that happen to coincidentally change
-    # and use cache.  Adjust vX if cache schema changes.
+    # Similar to cargo_cache, but holds the actual compiled dependent artifacts.
+    # This should be scoped to a hash of the dependency-metadata lock file.
+    # Cirrus-CI will automatically use separate caches for PRs and branches.
     folder: "$CARGO_TARGET_DIR"
-    fingerprint_key: "targets_v2_${CIRRUS_TAG}${DEST_BRANCH}${CIRRUS_PR}_amd64" # Cache only within same tag, branch, or PR (branch will be 'pull/#')
+    fingerprint_script: echo -e "targets_v3_${CIRRUS_TAG}${DEST_BRANCH}${CIRRUS_PR}_amd64\n---\n$(<Cargo.lock)\n---\n$(<Cargo.toml)"
     reupload_on_changes: true
   bin_cache: &bin_cache
     # This simply prevents rebuilding bin/netavark for every subsequent task.
@@ -70,6 +69,7 @@ build_task:
     reupload_on_changes: true
   setup_script: &setup "$SCRIPT_BASE/setup.sh"
   main_script: &main "$SCRIPT_BASE/runner.sh $CIRRUS_TASK_NAME"
+  cache_grooming_script: &groom bash "$SCRIPT_BASE/cache_groom.sh"
   upload_caches: [ "cargo", "targets", "bin" ]
 
 
@@ -82,11 +82,15 @@ build_aarch64_task:
     architecture: arm64  # CAUTION: This has to be "arm64", not "aarch64"
   cargo_cache: &cargo_cache_aarch64
     folder: "$CARGO_HOME"
-    fingerprint_key: "cargo_v2_${DEST_BRANCH}_aarch64"
+    # N/B: Should exactly match (except for arch) line from build_task (above).
+    # (No, there isn't an easy way to not duplicate most of this :()
+    fingerprint_script: echo -e "cargo_v3_${DEST_BRANCH}_aarch64\n---\n$(<Cargo.lock)\n---\n$(<Cargo.toml)"yy
     reupload_on_changes: true
   targets_cache: &targets_cache_aarch64
     folder: "$CARGO_TARGET_DIR"
-    fingerprint_key: "targets_v2_${CIRRUS_TAG}${DEST_BRANCH}${CIRRUS_PR}_aarch64" # Cache only within same tag, branch, or PR (branch will be 'pull/#')
+    # N/B: Should exactly match (except for arch) line from build_task (above).
+    # (No, there isn't an easy way to not duplicate most of this :()
+    fingerprint_script: echo -e "targets_v3_${CIRRUS_TAG}${DEST_BRANCH}${CIRRUS_PR}_aarch64\n---\n$(<Cargo.lock)\n---\n$(<Cargo.toml)"
     reupload_on_changes: true
   bin_cache: &bin_cache_aarch64
     # This simply prevents rebuilding bin/netavark for every subsequent task.
@@ -95,6 +99,7 @@ build_aarch64_task:
     reupload_on_changes: true
   setup_script: *setup
   main_script: *main
+  cache_grooming_script: *groom
   upload_caches: [ "cargo", "targets", "bin" ]
   # Downstream CI needs the aarch64 binaries from this CI system.
   # However, we don't want to confuse architectures.

contrib/cirrus/cache_groom.sh

@@ -0,0 +1,77 @@
+#!/bin/bash
+#
+# This script is intended to be run from Cirrus-CI to prepare the
+# rust targets cache for re-use during subsequent runs.  This mainly
+# involves removing files and directories which change frequently
+# but are cheap/quick to regenerate - i.e. prevent "cache-flapping".
+# Any other use of this script is not supported and may cause harm.
+
+set -eo pipefail
+
+source $(dirname ${BASH_SOURCE[0]})/lib.sh
+
+if [[ "$CIRRUS_CI" != true ]]; then
+  die "Script is not intended for use outside of Cirrus-CI"
+fi
+
+req_env_vars CARGO_HOME CARGO_TARGET_DIR CIRRUS_BUILD_ID
+
+# Giant-meat-cleaver HACK: It's possible (with a long-running cache key) for
+# the targets and/or cargo cache to grow without-bound (gigabytes). Ref:
+# https://github.com/rust-lang/cargo/issues/5026
+# There isn't a good way to deal with this or account for outdated content
+# in some intelligent way w/o trolling through config and code files.  So,
+# Any time the Cirrus-CI build ID is evenly divisible by some number (chosen
+# arbitrarily) clobber the whole thing and make the next run entirely
+# re-populate cache.  This is ugly, but maybe the best option available :(
+if [[ "$CIRRUS_BRANCH" == "$DEST_BRANCH" ]] && ((CIRRUS_BUILD_ID%15==0)); then
+  msg "It's a cache-clobber build, yay! This build has been randomly selected for"
+  msg "a forced cache-wipe!  Congradulations! This means the next build will be"
+  msg "slow, and nobody will know who to to blame!.  Lucky you!  Hurray!"
+  msg "(This is necessary to prevent branch-level cache from infinitely growing)"
+  cd $CARGO_TARGET_DIR
+  # Could use `cargo clean` for this, but it's easier to just clobber everything.
+  rm -rf ./* ./.??*
+  # In case somebody goes poking around, leave a calling-card hopefully leading
+  # them back to this script.  I don't know of a better way to handle this :S
+  touch CACHE_WAS_CLOBBERED
+
+  cd $CARGO_HOME
+  rm -rf ./* ./.??*
+  touch CACHE_WAS_CLOBBERED
+  exit 0
+fi
+
+# The following applies to both PRs and branch-level cache.  It attempts to remove
+# things which are non-essential and/or may change frequently.  It stops short of
+# trolling through config & code files to determine what is relevant or not.
+# Ref: https://doc.rust-lang.org/nightly/cargo/guide/build-cache.html
+#      https://github.com/Swatinem/rust-cache/tree/master/src
+cd $CARGO_TARGET_DIR
+for targetname in $(find ./ -type d -maxdepth 1 -mindepth 1); do
+  msg "Grooming $CARGO_TARGET_DIR/$targetname..."
+  cd $CARGO_TARGET_DIR/$targetname
+  # Any top-level hidden files or directories
+  showrun rm -rf ./.??*
+  # Example targets
+  showrun rm -rf ./target/debug/examples
+  # Documentation
+  showrun rm -rf ./target/doc
+  # Internal to rust build process
+  showrun rm -rf ./target/debug/deps ./target/debug/incremental ./target/debug/build
+done
+
+# The following only applies to dependent packages (crates).  It follows recommendations
+# Ref: https://doc.rust-lang.org/nightly/cargo/guide/cargo-home.html#caching-the-cargo-home-in-ci
+# and probably shouldn't be extended beyond what's documented.  This cache plays a major
+# role in built-time reduction, but must also be prevented from causing "cache-flapping".
+cd $CARGO_HOME
+for dirname in $(find ./ -type d -maxdepth 2 -mindepth 1); do
+  case "$dirname" in
+    ./bin) ;&  # same steps as next item
+    ./registry/index) ;&
+    ./registry/cache) ;&
+    ./git/db) continue ;;  # Keep
+    *) rm -rf $dirname ;;  # Remove
+  esac
+done

contrib/cirrus/lib.sh

@@ -32,6 +32,12 @@ else  # set default values - see make_cienv() below
     # VM Images are built with this setup
     CARGO_HOME="${CARGO_HOME:-/var/cache/cargo}"
     source $CARGO_HOME/env
+
+    # Make caching more effective - disable incremental compilation,
+    # so that the Rust compiler doesn't waste time creating the
+    # additional artifacts required for incremental builds.
+    # Ref: https://github.com/marketplace/actions/rust-cache#cache-details
+    CARGO_INCREMENTAL=0
 fi
 
 # END Global export of all variables