1
0

chore: rename repository state file to STATE.md

This commit is contained in:
2026-04-26 02:35:38 +02:00
parent c138d9201e
commit 76101e77c9
4 changed files with 139 additions and 166 deletions
+83
View File
@@ -0,0 +1,83 @@
# Repository state
## Project summary
- Name: `netupgrade`
- Language: Bash
- Entry point: `bin/netupgrade`
- Purpose: interactive CLI to run maintenance and upgrade actions on multiple remote hosts over SSH
- Config model: Bash configuration files sourced from `~/.config/netupgrade/*.cfg`
## Repository structure
- `bin/netupgrade`: main executable script containing CLI parsing, node selection, remote execution, and logging
- `config/netupgrade/*.cfg`: sample configuration files defining host groups and action sequences
- `README.md`: installation and usage documentation
- `TODO.md`: prioritized next actions and follow-up work
- `docs/`: project documentation
## Current behavior
- Loads a Bash configuration file
- Expects a `NODES` array with entries formatted as `host;display-name;action1;action2;...`
- Displays an interactive multi-select checklist using `whiptail`
- Executes selected actions sequentially over SSH
- Uses `root@host` by default, or `SSH_USER@host` when configured
- Writes execution logs to `~/netupgrade.log`
- Opens the log with `$EDITOR`, then `nano`, `vi`, or `less`; if none is available, prints the log path and continues with the existing optional cleanup prompt
## Supported actions
- `apt`
- `yum`
- `pkg`
- `pacman`
- `apk`
- `reboot`
- `cmd:<remote command>`
- `docker-stacks:<directory>`
## Architecture and constraints
- The project is intentionally lightweight and shell-based
- The main execution flow still lives mostly in a single Bash script
- Configuration is code-driven rather than declarative, because config files are sourced as shell files
- Config files are trusted inputs and imply arbitrary code execution
- Remote operations are sequential, not parallel
- Logging is file-based and tightly coupled to command execution
- Backward compatibility for existing config files should be preserved where possible
- Changes to SSH command construction require extra care because quoting regressions are easy to introduce
## Notable implementation details
- SSH execution is centralized through a `runSSH` helper
- `SSH_USER` is configurable and defaults to `root`
- `NODES` parsing was hardened to preserve spaces in action values such as `cmd:...`
- `cmd:<...>` is executed through a remote shell to support shell operators, but remains intentionally powerful and unsafe
- `reboot` uses a detached remote command to reduce false failures caused by expected SSH disconnects
- `docker-stacks` runs through a remote shell script sent over SSH stdin to improve path handling and quoting
- Startup dependency checks cover required local commands, including `mv` for log-summary rewriting
## Sensitive areas for future changes
- SSH command construction and shell quoting behavior
- `cmd:<...>` execution semantics
- `docker-stacks:<...>` path handling and remote shell behavior
- Package-manager cleanup commands and confirmation flags
- Error propagation consistency across action types
- Documentation and runtime behavior alignment
- Sample configuration accuracy to avoid operator mistakes
## Recent meaningful changes
- README and CLI help were aligned with current behavior, including the meaning of `-f`
- Startup dependency checks were added and updated
- Log viewer fallback was made more flexible
- `NODES` parsing was hardened to preserve spaces in action values
- SSH execution was centralized and made configurable through `SSH_USER`
- Remote command construction was improved for `pacman`, `docker-stacks`, and `reboot`
- Log summary generation was rewritten to avoid `sed -i` interpolation and now replaces the log atomically
- `apk` and `apt` cleanup behavior was corrected to better match real package-manager semantics
- The sample configuration in `config/netupgrade/hypervisor-01.cfg` now keeps the alternate `docker-01` reboot example commented out rather than active by default
## Change guidance
- `STATE.md` should stay focused on current state, architecture, and constraints
- `TODO.md` should hold prioritized follow-up work and next implementation steps
- Prefer small, reviewable hardening changes over broad rewrites
- Keep the tool simple and admin-friendly
- Treat documentation consistency as part of functional correctness
- Be cautious with remote command construction, because quoting changes can easily introduce regressions
- Avoid introducing a global `set -euo pipefail` baseline in one step without first documenting and validating the intended failure semantics
- Focus future testing first on parsing, command construction, and result reporting