cf-node-bootstrap/extras/model-cache-redirect.install

#!/bin/bash
# Model/dev-tool cache redirect installer
#
# Nodes with a fast secondary local disk mounted at /devl (the convention
# across CF's rack machines — heimdall, navi, sif) fill up their small root
# partition fast once pip/torch/vllm/whisper/clip/huggingface caches land in
# ~/.cache. This redirects those caches to /devl/user-cache and leaves
# symlinks at the old ~/.cache paths so tools that hardcode ~/.cache (rather
# than respecting $XDG_CACHE_HOME) still land on the fast disk.
#
# Deliberately does NOT touch $HF_HOME — that's the curated model store
# convention (/Library/Assets/LLM, NFS) and is unrelated to this cache.
#
# Only applicable on nodes with a real, separate /devl mount. Skips cleanly
# (not an error) on nodes without one, since it's an optional extra.

scriptname="${BASH_SOURCE[0]##*/}"
script_title="${grn}Cache${blu}Redirect${dfl}"
rundir="${BASH_SOURCE[0]%/*}"

# Source PRbL Functions locally or retrieve from online
if [ ! -z $prbl_functions ] ; then
    source $prbl_functions
else
    if [ -f ${rundir}/functions ] ; then
        source ${rundir}/functions
    else
        if command -v curl >/dev/null 2>&1; then
            source <(curl -ks 'https://raw.githubusercontent.com/pyr0ball/PRbL/main/functions')
        elif command -v wget >/dev/null 2>&1; then
            source <(wget -qO- 'https://raw.githubusercontent.com/pyr0ball/PRbL/main/functions')
        elif command -v fetch >/dev/null 2>&1; then
            source <(fetch -qo- 'https://raw.githubusercontent.com/pyr0ball/PRbL/main/functions')
        else
            echo "Error: curl, wget, and fetch commands are not available. Please install one to retrieve PRbL functions."
            exit 1
        fi
    fi
fi

logfile="${rundir}/${pretty_date}_${scriptname}.log"

cache_root="/devl/user-cache"
redirected_tools=(pip whisper clip huggingface)

# Real, separate mount at /devl (not just a directory that happens to exist
# on root) — mirrors how sif/heimdall/navi are actually laid out.
_devl-is-separate-mount() {
    [ -d /devl ] && mountpoint -q /devl
}

check-cache-redirect(){
    if [[ -n "$XDG_CACHE_HOME" ]] && grep -q "XDG_CACHE_HOME=${cache_root}" "$HOME/.bashrc" 2>/dev/null; then
        return 0
    fi
    return 1
}

install(){
    if ! _devl-is-separate-mount ; then
        warn "No separate /devl mount found on this node — skipping cache redirect (nothing to do)"
        return 0
    fi

    if check-cache-redirect ; then
        boxline "Cache redirect already configured (XDG_CACHE_HOME=${cache_root})"
        return 0
    fi

    boxborder "${lyl}Redirecting dev-tool caches to ${cache_root}${dfl}"

    # run() executes its argument directly when not in dry-run (and only
    # boxlines it otherwise) — fine for single literal commands like this.
    run "mkdir -p ${cache_root}"

    if [[ $dry_run == true ]] ; then
        boxline "DryRun: append 'export XDG_CACHE_HOME=${cache_root}' to $HOME/.bashrc"
    else
        take-backup "$HOME/.bashrc"
        echo "" >> "$HOME/.bashrc"
        echo "# Dev-tool caches redirected to fast local disk (see model-cache-redirect extra)" >> "$HOME/.bashrc"
        echo "export XDG_CACHE_HOME=${cache_root}" >> "$HOME/.bashrc"
    fi

    # Tools that hardcode ~/.cache/<name> and ignore XDG_CACHE_HOME: move
    # existing contents onto the fast disk, then symlink the old path back.
    # This is multi-step (move-if-present, then symlink) so it's driven by
    # $dry_run directly rather than run() (which only handles single literal
    # commands).
    for tool in "${redirected_tools[@]}"; do
        local old_path="$HOME/.cache/${tool}"
        local new_path="${cache_root}/${tool}"

        if [ -L "$old_path" ]; then
            boxline "  ${tool}: already redirected"
            continue
        fi

        if [[ $dry_run == true ]] ; then
            boxline "DryRun: move ${old_path} (if present) to ${new_path}, symlink ${old_path} -> ${new_path}"
            continue
        fi

        mkdir -p "$new_path"
        if [ -d "$old_path" ]; then
            # Move existing cache contents onto the fast disk rather than
            # discarding them — avoids a cold-cache redownload.
            cp -a "$old_path/." "$new_path/" 2>/dev/null
            rm -rf "$old_path"
        fi
        ln -s "$new_path" "$old_path"
        installed_files+=("$old_path -> $new_path")
        boxline "  ${tool}: ${gry}~/.cache/${tool} -> ${new_path}${dfl}"
    done

    boxline ""
    boxline "${gry}Note: HF_HOME (curated model store) is untouched by design.${dfl}"
    boxline "${gry}Re-login (or 'source ~/.bashrc') for XDG_CACHE_HOME to take effect.${dfl}"
}

remove(){
    warn "Remove is manual for this extra — restoring ~/.cache/<tool> as real dirs"
    warn "risks re-duplicating data already moved onto ${cache_root}. Not automated."
}

dry-run-report(){
    box-rounded
    boxborder "${grn}Dry-run Report:${dfl}"
    box-norm
    boxborder \
    "installed_files= " \
    "${installed_files[@]}"
}

usage(){
    boxborder \
    "${script_title} usage:" \
    "  -i | --install     Redirect dev-tool caches to /devl/user-cache" \
    "  -D | --dry-run      Show what would change without doing it" \
    "  -r | --remove       (manual only, see warning)" \
    "  -h | --help          This menu"
}

case $1 in
    -i | --install)
        install && success " [${script_title} ${lyl}Installed${dfl}]"
        ;;
    -r | --remove)
        remove
        ;;
    -D | --dry-run)
        export dry_run=true
        install
        dry-run-report
        usage
        unset dry_run
        success "${script_title} ${lyl}Installer Dry-Run Complete!${dfl}"
        ;;
    -h | --help)
        usage
        ;;
    *)
        warn "Invalid argument $@"
        usage
        ;;
esac