fix(skillfs): ensure SLS ops log written even on SIGPIPE panic

[zhangtaibo]

问题

Fixes #1506

skillfs CLI subcommands list and classify do not write SLS ops log entries when stdout is piped to a consumer that closes early (e.g. head -5). The log_command() call runs after all println! output, but println! panics on broken pipe (SIGPIPE/EPIPE) when the downstream reader has already exited, causing the process to unwind before reaching log_command().

This is a regression from the original SLS ops logging fix (commit 1a07b20, issue #1306). The log_command calls were added to all subcommands, but they are unreachable when the process panics from a broken stdout pipe.

修复

Add an RAII SlsLogGuard that calls sls_ops::log_command() in its Drop implementation. The guard is created before the command runs and dropped on all exit paths — including panic unwinds from broken pipe. This ensures the SLS entry is always written regardless of stdout pipe state.

+struct SlsLogGuard {
+    ops_name: String,
+    start: std::time::Instant,
+    err_reason: std::cell::Cell<Option<String>>,
+    completed: std::cell::Cell<bool>,
+}
+
+impl Drop for SlsLogGuard {
+    fn drop(&mut self) {
+        let reason = if self.completed.get() {
+            self.err_reason.take()
+        } else {
+            Some("panic".to_string())
+        };
+        sls_ops::log_command(&self.ops_name, self.start, reason);
+    }
+}

Each command handler replaces its manual log_command call with guard creation:

let _guard = SlsLogGuard::new("list");
let result = cmd_list(source, enabled_only).await;
_guard.set_reason(err_reason(&result));
result

验证

# ECS fresh clone + apply patch + build
cd /root && rm -rf anolisa && git clone https://github.com/alibaba/anolisa.git && cd anolisa
git apply skillfs-sls-panic-guard.patch
bash scripts/rpm-build.sh skillfs
# → exit 0, RPM 2.0M

# Install patched RPM and test
rpm -U --force scripts/rpmbuild/RPMS/x86_64/skillfs-0.3.3-1.alnx4.x86_64.rpm
touch /var/log/anolisa/sls/ops/skillfs.jsonl
> /var/log/anolisa/sls/ops/skillfs.jsonl

# Before fix: list piped to head -5 → 0 bytes (no SLS entry)
# After fix: list piped to head -5 → 180 bytes (SLS entry written!)
skillfs list /usr/share/anolisa/runtime/skills 2>&1 | head -5
stat -c '%s' /var/log/anolisa/sls/ops/skillfs.jsonl
# 180

# All 3 parametrized tests pass:
pytest tests/test_sls_ops_emission.py::test_skillfs_cli_emits_sls_entry -v
# 3 passed in 10.50s

Co-Authored-By: Claude noreply@anthropic.com

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

zhangtaibo seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}