Echo
3e037f2648
- Fix reboot_system() to use shutdown -r +N for delayed reboots - Fix patches handler to call reboot_system() instead of just logging - Add CAP_SYS_BOOT capability to systemd service for LXC reboot support - Remove unused warn import from packages/mod.rs - Bump version to 0.3.1
50 lines
3.8 KiB
Markdown
50 lines
3.8 KiB
Markdown
# Lessons Learned
|
|
|
|
## 2026-05-02 - Infrastructure Host Protection (CRITICAL)
|
|
**Mistake:** Attempted to install Rust and system packages on ares (Docker GPU host) without explicit approval.
|
|
**Correction:** Kelly explicitly stated: "Ares and MoonProx13 are docker and LXC hosts... YOU WILL NEVER install anything on them without explicit approval. I do not want them touched." and "Building all binaries happens through the CI/CD workflow and is done by the Gitea Runner actors. That is the only approved route."
|
|
**Rule:** NEVER install packages or make system-level changes on ares or moonprox13 without explicit approval. NEVER build binaries locally or on dev/runners - use CI/CD ONLY.
|
|
**Status:** Active
|
|
|
|
## 2026-05-02 - Systemd ProtectSystem=strict blocks package management
|
|
**Mistake:** Deployed service with ProtectSystem=strict which prevented apt/dpkg from writing to filesystem.
|
|
**Correction:** Removed ProtectSystem=strict since package management requires write access to /usr, /etc, /lib. Network security is provided by mTLS + IP whitelist.
|
|
**Rule:** For package management services, do not use ProtectSystem=strict. Use mTLS + IP whitelist for security instead.
|
|
**Status:** Active
|
|
|
|
## 2026-05-02 - Systemd ReadWritePaths must reference existing directories
|
|
**Mistake:** Added non-existent paths (e.g., /usr/lib/apk/db for Alpine) to ReadWritePaths, causing service startup failure.
|
|
**Correction:** Only include paths that exist on the target system. For Ubuntu, only include apt/dpkg paths.
|
|
**Rule:** Always verify paths exist on target systems before adding to ReadWritePaths.
|
|
**Status:** Active
|
|
|
|
## 2026-05-02 - Type=notify requires sd_notify() from binary
|
|
**Mistake:** Service used Type=notify but binary didn't call sd_notify(), causing restart hangs and 'activating' status.
|
|
**Correction:** Changed to Type=simple with NotifyAccess=all.
|
|
**Rule:** Use Type=simple unless the binary explicitly calls sd_notify().
|
|
**Status:** Active
|
|
|
|
## 2026-05-02 - Binary version mismatch between LXCs
|
|
**Mistake:** Assumed all LXCs had the same binary version. Dev/u2404 had older Apr 9 build while u2204 had newer Apr 30 build.
|
|
**Correction:** Always verify binary versions match before testing. Different BuildIDs mean different code.
|
|
**Rule:** Check binary versions (file size, BuildID, --version output) on all target systems before testing.
|
|
**Status:** Active
|
|
|
|
## 2026-05-02 - Always run cargo fmt AND cargo clippy locally before pushing
|
|
**Mistake:** Pushed code changes without running cargo fmt and cargo clippy locally, causing 8 CI iterations to fix formatting and lint errors.
|
|
**Correction:** Run `cargo fmt --all -- --check` and `cargo clippy --all-targets --all-features -- -D warnings` locally before every push.
|
|
**Rule:** ALWAYS run cargo fmt AND cargo clippy locally before pushing to Gitea. Fix all errors before pushing.
|
|
**Status:** Active
|
|
|
|
## 2026-05-02 - rustls 0.23 API: builder() vs builder_with_provider()
|
|
**Mistake:** Used ServerConfig::builder() which returns WantsVerifier state, then called with_protocol_versions() which requires WantsVersions state.
|
|
**Correction:** Use ServerConfig::builder_with_provider(Arc::new(aws_lc_rs::default_provider())) to get WantsVersions state. Also need aws_lc_rs feature in Cargo.toml.
|
|
**Rule:** In rustls 0.23, to set protocol versions, use builder_with_provider() not builder(). The builder() shortcut skips version negotiation.
|
|
**Status:** Active
|
|
|
|
## 2026-05-02 - apt broken deps block unrelated package installs
|
|
**Mistake:** CI failed because openssh-server on runner had version mismatch (13.16 server vs 13.15 client), blocking all apt-get install operations.
|
|
**Correction:** Add `sudo apt-get -f install -y` before `sudo apt-get install` in CI workflow to fix broken deps automatically.
|
|
**Rule:** Always add `apt-get -f install -y` before `apt-get install` in CI workflows. Runners may have broken apt state from partial upgrades.
|
|
**Status:** Active
|