relative interpreter links, lock holder reporting

1 files changed, 63 insertions, 6 deletions

diff --git a/producer.sh b/producer.sh
index 884d3ac..cab10dd 100755
--- a/producer.sh
+++ b/producer.sh
@@ -62,6 +62,27 @@ USAGE
 log() { printf '[producer] %s\n' "$*" >&2; }
 fail() { printf '[producer] Error: %s\n' "$*" >&2; exit 1; }
 
+# Print the PIDs of processes holding an open descriptor for the given file
+# (a leading space each, e.g. " 1234 5678"). A plain /proc scan avoids a
+# dependency on fuser or lsof, which may not be installed. Only processes
+# visible in the current PID namespace can be found.
+lock_holders() {
+    local file="$1" id p fd target base
+    id="$(stat -c '%d:%i' -- "$file" 2>/dev/null)" || return 0
+    base="${file##*/}"
+    for p in /proc/[0-9]*; do
+        [[ "${p#/proc/}" == "$$" ]] && continue
+        for fd in "$p"/fd/*; do
+            target="$(readlink -- "$fd" 2>/dev/null)" || continue
+            [[ "$target" == *"$base" ]] || continue
+            if [[ "$(stat -c '%d:%i' -L -- "$fd" 2>/dev/null)" == "$id" ]]; then
+                printf ' %s' "${p#/proc/}"
+            fi
+        done
+    done
+    return 0
+}
+
 case "${1:-}" in
     -h|--help) usage; exit 0 ;;
 esac
@@ -130,13 +151,17 @@ export UV_PYTHON_INSTALL_DIR="$RUNTIME/uv/python" XDG_CACHE_HOME="$RUNTIME/cache
 export TMPDIR="$RUNTIME/tmp" PRODUCER_HOME="$RUNTIME"
 
 marker_current() {
-    # The marker must name this exact path, and the interpreter must exist:
-    # a failed or interrupted rebuild leaves neither, forcing a repair.
+    # Only the version suffix decides; the recorded path is informational.
+    # The runtime is shared between environments that mount this directory
+    # at different paths (a container sees /workspace, the host does not),
+    # and a broken move is caught anyway by the interpreter check below.
+    # A failed or interrupted rebuild leaves no valid marker, forcing a
+    # repair.
     [[ -f "$READY" ]] \
-        && [[ "$(cat -- "$READY" 2>/dev/null)" == "$RUNTIME v$RUNTIME_VERSION" ]] \
+        && [[ "$(cat -- "$READY" 2>/dev/null)" == *" v$RUNTIME_VERSION" ]] \
         && [[ -x "$APP/bin/python" ]] \
         && { [[ "$WANT_GUI" -eq 0 ]] || { [[ -f "$GUI_READY" ]] \
-            && [[ "$(cat -- "$GUI_READY" 2>/dev/null)" == "$RUNTIME v$RUNTIME_VERSION" ]]; }; }
+            && [[ "$(cat -- "$GUI_READY" 2>/dev/null)" == *" v$RUNTIME_VERSION" ]]; }; }
 }
 
 bootstrap() {
@@ -146,7 +171,17 @@ bootstrap() {
     # the whole destructive window below. It is released before the
     # application starts.
     exec 9>"$RUNTIME/bootstrap.lock"
-    flock 9
+    if ! flock -n 9; then
+        # Never block silently: another launch is mid-bootstrap.
+        local holders9=""
+        holders9="$(lock_holders "$RUNTIME/bootstrap.lock")"
+        if [[ -n "$holders9" ]]; then
+            log "Waiting for another producer launch to finish preparing the runtime (held by PID(s):$holders9)"
+        else
+            log "Waiting for another producer launch to finish preparing the runtime"
+        fi
+        flock 9
+    fi
     if [[ "$REBUILD" -eq 0 ]] && marker_current; then
         exec 9>&-
         # Runtime lock (fd 8): held shared for the application's lifetime so
@@ -159,7 +194,17 @@ bootstrap() {
     # deleted, so a rebuild can never clobber a live environment.
     exec 8>"$RUNTIME/app.lock"
     if ! flock -n 8; then
-        log "Waiting for running producer jobs to finish before building the runtime"
+        # Point at the concrete holder: usually an earlier GUI or processing
+        # job that is still running, not a leftover lock file (flock locks
+        # disappear with their process).
+        local holders=""
+        holders="$(lock_holders "$RUNTIME/app.lock")"
+        if [[ -n "${holders//[[:space:]]/}" ]]; then
+            log "Waiting for running producer jobs to finish before building the runtime (held by PID(s):$holders)"
+            log "Stop them first, e.g. 'kill$holders' (a running 'gui' server counts as a job)."
+        else
+            log "Waiting for running producer jobs to finish before building the runtime"
+        fi
         flock 8
     fi
     # Invalidate the marker before any destructive work: if this bootstrap
@@ -198,6 +243,18 @@ bootstrap() {
     # The AI worker keeps its own pinned Python 3.11 (glibc only).
     "$UV" venv --python 3.13 --managed-python --relocatable --allow-existing "$APP" \
         || fail "Creating the managed Python 3.13 environment failed; check network access and free space."
+    # uv creates the interpreter links with absolute targets. Make them
+    # relative so one runtime serves every environment that mounts this
+    # directory at a different path (host and containers included).
+    local link target
+    for link in "$APP"/bin/python "$APP"/bin/python3 "$APP"/bin/python3.*; do
+        [[ -L "$link" ]] || continue
+        target="$(readlink -- "$link")"
+        case "$target" in "$RUNTIME"/*)
+            ln -sfn -- "$(realpath -m --relative-to="$APP/bin" -- "$target")" "$link"
+            ;;
+        esac
+    done
     # The bundled FFmpeg wheel is preferred; where it does not exist (for
     # example musl systems), fall back to a system FFmpeg at runtime.
     local bundled="$PROJECT[bundled-ffmpeg]" core="$PROJECT"