30 lines
4.3 KiB
Markdown
30 lines
4.3 KiB
Markdown
# Development guide
|
||
|
||
Run commands from the repository root. Use a Python 3.11+ virtual environment and requirements.txt. Start with `python -m server_buddy.main`. Assets are resolved relative to the module, independent of the user's working directory; tests use repository-relative schema/fixture paths.
|
||
|
||
## Phase 1 behavior
|
||
|
||
Qt widgets stay on the GUI thread. CollectionJob loads the bundled JSON in a QThread and delivers stage/result/error signals. Analyze and import buttons are disabled until the worker finishes; closing an active job is deferred to avoid destroying a running thread. Demo loading is finite local I/O; production timeout/cancellation and process lifecycle management are Phase 3 requirements. No subprocesses are launched in Phase 1.
|
||
|
||
Tables use Qt proxy filtering across all columns with native numeric sorting. Session history records repeated fixture loads without changing fixture timestamps. Dashboard counts unique hostnames from fixture/import history, while history counts runs. High-risk findings refer to the current assessment. Two OS-specific fixtures supply the demo; user input cannot relabel synthetic evidence as a live server assessment.
|
||
|
||
Risk rules live in analyzers/assessment.py. Each finding contains classification, evidence, severity and points. Score is the sum capped at 100: 0 informational, 1–29 low, 30–59 medium, 60–89 high, 90–100 critical. These preliminary rules are demonstration logic, not a validated production risk model. Severity is an individual finding's severity, while overall risk is the aggregate. Observation limits always contribute uncertainty.
|
||
|
||
Collection JSON uses schemas/inventory-v1.schema.json. Export JSON wraps the collection as inventory_result and includes the derived assessment. Do not treat the report wrapper as collection JSON. Live collection must enforce the schema at the boundary and tighten field schemas as PowerShell contracts are implemented. Never replace failed checks with synthetic values.
|
||
|
||
## Verification in this development environment
|
||
|
||
Python 3.12.3 is available. `python3 -m unittest discover -s tests -v`: four tests passed; schema and Qt smoke tests skipped because jsonschema and PySide6 are absent. An isolated installation under /tmp failed due to package server DNS/network resolution. GUI launch and visual appearance have not been verified here. Run the documented tests after installing dependencies on Linux Mint; also test a Windows desktop before distribution.
|
||
|
||
No PowerShell collection scripts or production collector exist yet, so there is no real PowerShell output to validate. Phase 1 requires no remote accounts, remoting configuration, or administrative privilege.
|
||
|
||
## Follow-up verification
|
||
|
||
After dependencies were installed in the project virtual environment: `.venv/bin/python -m unittest discover -s tests -v` passed all 12 tests, including two offscreen Qt tests, schema validation, Linux inventory import/NFS evidence, sensitive-field exclusion and startup diagnostics. Source compilation passed. Normal X11 startup is blocked by the confirmed missing `libxcb-cursor.so.0`; the launcher prints Mint installation instructions. Windows execution and real RHEL collection remain untested/unimplemented as described in PLATFORMS.md.
|
||
|
||
## Local Linux collector
|
||
|
||
`collectors/linux.py` executes fixed argument lists from system binary directories with a restricted environment, no shell or user hostname, DEVNULL stderr and no sudo. Each process is polled for cancellation, an eight-second time limit and a 10 MiB output limit; the process is killed and reaped on interruption/failure. Overall collection is bounded to approximately 60 seconds for command execution. Local /proc and /etc/os-release reads do not access mounted remote filesystems. There is no arbitrary command UI. Collection source is included in server metadata; JSON import strips that source field and treats imported evidence as unverified. GUI workers retain the previous assessment on cancellation. New local checks intentionally return partial/missing statuses rather than asserting full coverage.
|
||
|
||
Verification: 21 unit/UI tests passed, plus a real local read-only smoke collection validated against the JSON schema with HTML/JSON generation. Tests use both RPM and dpkg fixtures, check service/IPv6 parsing and unknown-service classification, command timeout/reaping, cancellation, and source labels. Real RHEL and Windows execution is still pending.
|