feat: emit exiting{reason:"requested"} on clean shutdown - #164
Merged
Conversation
…x-swift submodule pins mlx-swift-lm was uncommitted-pinned backward past the commit that introduced DualModelMTP/MTPTokenIterator/Gemma4AssistantModel, breaking Gemma4MTPBench and forcing MTP speculative decoding to be commented out in InferenceEngine.swift as a workaround. Repointing both submodules to their current origin/main tips (mlx-swift-lm past the merged DSA stage-2 PR #61, mlx-swift past the matching MLXFast.fromFp8 addition) restores those types and lets the benchmark build again — no source workaround needed. Also adds scripts/bootstrap_local_tests.sh, which mirrors CI's "Install MLX Metal library" step (pip install mlx, copy its bundled metallib into every built .xctest bundle) so `swift test` is runnable locally without the manual cmake+make dance. Addresses the Tier 3 item in #128. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…e Protocol v1)
Aegis-AI's daemon (aegis-daemon, W7 Rust rewrite) needs to tell a clean,
planned shutdown apart from a real crash. Today SwiftLM's SIGTERM/SIGINT
handlers print plain text and exit(0) with no machine-readable signal at
all.
Adds a small shared emitEvent(_:) helper (used by both the existing ready
event and this new one, replacing ready's inline JSONSerialization call)
and emits {"event":"exiting","reason":"requested"} on both signal handlers
before exiting. Consumers should ignore unrecognized future reason values
rather than error on them.
This is Phase 1 of a larger plan (see Aegis-AI's daemon/spec/engine-protocol.md
§3/§4) — self-reporting a SPECIFIC failure reason on a real crash (OOM,
model load failure, etc.) is separate, larger follow-up work, since none
of those paths have any error handling to build on today.
Verified: swift build --target SwiftLM (debug) and swift build -c release
--product SwiftLM both green. Smoke-tested against a real Aegis-AI daemon
build in an isolated sandbox: reached ready normally, and a real SIGTERM
via the daemon's stop endpoint exited the process cleanly with the new
event observed, without the daemon misclassifying it as a crash.
- emitEvent() used to silently swallow any JSON encoding failure with no diagnostic trail, for a function that's now the sole path for a protocol-critical signal. Now logs a message to stderr on failure. - Ignore SIGPIPE before the SIGTERM/SIGINT handlers are wired up: if the daemon's read end of our stdout pipe is already gone by the time a shutdown signal arrives, the exiting-event print/fflush could raise SIGPIPE, whose default disposition kills the process via signal instead of reaching Darwin.exit(0) -- producing exactly the ambiguous "was this a crash?" signature this feature exists to eliminate. - docs/AEGIS_INTEGRATION.md: reworded reason:"requested" to not imply it proves the daemon itself initiated the shutdown (any SIGTERM/SIGINT sender produces the identical event), and added that the event is best-effort (dispatched on the main queue, so a busy queue delays emission) rather than a synchronous guarantee at signal-delivery time. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Aegis-AI's daemon (
aegis-daemon, the W7 Rust rewrite) needs to tell a clean, planned shutdown apart from a real crash. Today SwiftLM's SIGTERM/SIGINT handlers print plain text andexit(0)with no machine-readable signal at all.emitEvent(_:)helper (used by both the existingreadyevent and this new one, replacingready's inlineJSONSerializationcall).{"event":"exiting","reason":"requested"}on both signal handlers before exiting. Consumers should ignore unrecognized futurereasonvalues rather than error on them.docs/AEGIS_INTEGRATION.mdupdated alongside the existingreadyevent documentation.This is Phase 1 of a larger plan (see Aegis-AI's
daemon/spec/engine-protocol.md§3/§4) — self-reporting a specific failure reason on a real crash (OOM, model load failure, etc.) is separate, larger follow-up work, since none of those paths have any error handling to build on today.Test plan
swift build --target SwiftLM(debug) — greenswift build -c release --product SwiftLM— greenreadynormally, and a real SIGTERM via the daemon's stop endpoint exited the process cleanly with the new event observed, without the daemon misclassifying it as a crash.🤖 Generated with Claude Code