Skip to content

feat(save): max-size-mb LRU prune — break the fat-blob ratchet #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
emooreatx merged 1 commit into from
Jul 24, 2026
Merged

feat(save): max-size-mb LRU prune — break the fat-blob ratchet #2

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 40 additions & 0 deletions save/action.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,21 @@ inputs:
paths:
description: "Space-separated dirs to cache. `~` and workspace-relative paths are expanded."
default: "~/.cargo/registry/index ~/.cargo/registry/cache ~/.cargo/git/db target"
max-size-mb:
description: >
Optional target/-size budget in MB. When set (and GNU find is
available — i.e. the Linux lanes, where the fat-blob pain lives), the
save prunes the OLDEST-mtime files under target/ until the total is
within budget BEFORE tarring. This breaks the cache RATCHET
(restore(fat) -> build -> save(fat+)): stale artifact generations —
e.g. an entire superseded dependency-version set after a re-pin —
are restored, never rebuilt, and re-saved forever, growing the blob
and every download+extract on every lane. LRU-by-mtime means a
reused-but-ancient artifact may take ONE forced recompile and then
re-enters with a fresh mtime — mild steady-state churn in exchange
for a bounded blob. Empty (default) = disabled. Never touches
~/.cargo (registry/git are lockfile-keyed, not generational).
default: ""
token:
description: >
GHCR token with write:packages. Pass the job's GITHUB_TOKEN (the
Expand Down Expand Up @@ -105,6 +120,31 @@ runs:
# the unconverted RUNNER_TEMP (cd "$TMP" lands in the same physical dir).
TMP="${RUNNER_TEMP}"; FL=""
if [ "${RUNNER_OS:-}" = "Windows" ]; then TMP="$(cygpath -u "${RUNNER_TEMP}")"; FL="--force-local"; fi
# ── LRU prune (see max-size-mb input) ─────────────────────────────
if [ -n "${{ inputs.max-size-mb }}" ] && [ -d "$GITHUB_WORKSPACE/target" ] \
&& find --version >/dev/null 2>&1; then
BUDGET_KB=$(( ${{ inputs.max-size-mb }} * 1024 ))
T="$GITHUB_WORKSPACE/target"
USED_KB=$(du -sk "$T" | cut -f1)
if [ "$USED_KB" -gt "$BUDGET_KB" ]; then
echo "CIRISCache: target/ ${USED_KB}KB over ${BUDGET_KB}KB budget — LRU prune"
# Oldest-mtime first; delete until within budget. %T@ (GNU find)
# + a size column so we can stop exactly at the line.
find "$T" -type f -printf '%T@ %k %p\n' 2>/dev/null | sort -n | \
while read -r _mt kb path; do
[ "$USED_KB" -le "$BUDGET_KB" ] && break
rm -f -- "$path" 2>/dev/null || continue
USED_KB=$(( USED_KB - kb ))
done
# The subshell above cannot export USED_KB; re-measure honestly.
AFTER_KB=$(du -sk "$T" | cut -f1)
echo "CIRISCache: pruned target/ ${USED_KB}KB→${AFTER_KB}KB (budget ${BUDGET_KB}KB)"
else
echo "CIRISCache: target/ ${USED_KB}KB within ${BUDGET_KB}KB budget — no prune"
fi
elif [ -n "${{ inputs.max-size-mb }}" ]; then
echo "CIRISCache: max-size-mb set but GNU find unavailable — prune skipped (fail-open)"
fi
tar $FL -czPf "$TMP/cache.tar.gz" $EXP
cd "$TMP"
_cc_timeout 1200 oras push "$REF" cache.tar.gz:application/gzip
Expand Down
Loading