Skip to content

README-dev-stop.md

Latest commit

 

History

History
48 lines (42 loc) · 3.17 KB

README-dev-stop.md

File metadata and controls

48 lines (42 loc) · 3.17 KB
Raw

Worktree-safe Aspire lifecycle helpers

Use the repository wrappers rather than an unqualified aspire start, aspire ps, or aspire stop --all:

scripts/dev-start.sh
scripts/dev-status.sh
scripts/dev-stop.sh
scripts/aspire-readiness-smoke.sh

All commands resolve the canonical AppHost path for the checkout that contains the script. dev-start.sh always supplies --apphost, --isolated, and --non-interactive, so ports and Aspire user secrets do not collide with another worktree. It launches the CLI in a captured session and, on any CLI/registration/status failure or signal, identity-checks and removes survivors from that session independently of Aspire's registry before attempting the normal scoped stop path. Because --non-interactive rules out a prompt and --isolated rules out the shared user-secrets store, dev-start.sh also seeds the AppHost's required node-sqlite-key parameter itself: it mints a per-checkout, owner-only XE-Local-AI-Engine.AppHost/.data/node.key (base64 32 bytes, .gitignored) on first use, reuses it afterwards so encrypted dev data stays readable, and passes it as the environment variable Parameters__node-sqlite-key — never on a command line. dev-status.sh selects the same exact AppHost from aspire ps --format Json, then emits only resource name/type/state/health and endpoint URLs with query strings removed. It never prints resource environment, properties, connection strings, or the dashboard login token.

dev-stop.sh sends an AppHost-qualified stop request and never uses --all. Before stopping, it snapshots the selected AppHost, its exact-path Aspire ancestor, the Aspire 13.4 DCP sibling whose command line contains the exact --monitor <AppHost PID> token pair, and their complete descendant closure regardless of executable name. This preserves ownership across DCP children that use separate sessions/process groups without sweeping another worktree merely because it shares a login/session SID. Every selected PID is paired with its /proc start time and revalidated immediately before TERM/KILL, so PID reuse cannot redirect cleanup. Aspire 13.4 survivors are terminated only from that snapshot. A descendant llama-server is therefore provably owned and cleaned. Processes outside the selected graph—including another worktree's DCP or a pre-existing managed llama-server—are untouched and do not make scoped teardown fail; success proves only that the selected graph and exact registration are gone. Query failures or malformed Aspire JSON are errors, never interpreted as "not running"; start, status, readiness, and stop all fail closed. --dry-run sends no stop request.

aspire-readiness-smoke.sh refuses to reuse an already-running instance, starts a new isolated instance, delegates readiness to aspire wait app, and traps every exit path through the scoped stop helper. Override its default 180-second readiness budget with XE_ASPIRE_SMOKE_TIMEOUT_SECONDS.

Set XE_ASPIRE_APPHOST only when intentionally targeting another explicit AppHost. The lifecycle scripts never terminate a managed llama-server merely because it shares a per-user installation directory; ownership must be present in the pre-stop ancestry snapshot.