Checkpoint runtime-committed session projections#733
Conversation
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b09b4834b2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| &mut self, | ||
| _session_snapshot: &[u8], | ||
| ) -> Result<(), CoreExecutorError> { | ||
| Ok(()) |
There was a problem hiding this comment.
Wire checkpoint hook for non-mob executors
When runtime-backed sessions run through the existing REST/MCP/CLI/RPC executors, those executors return CoreApplyOutput from PersistentSessionService::apply_runtime_turn but do not override this new hook, so the runtime loop calls this default Ok(()) after the machine commit and the SessionStore projection is never updated. I checked meerkat-rest/src/lib.rs, meerkat-mcp-server/src/runtime_ingress.rs, meerkat/src/surface/runtime_backed.rs, and meerkat-rpc/src/session_executor.rs; only MobSessionRuntimeExecutor delegates to checkpoint_committed_runtime_session_snapshot, leaving those production surfaces with stale persisted projections after each committed turn.
Useful? React with 👍 / 👎.
Summary
Why
0.6.18 made StoreCheckpointer save-capable for runtime-backed sessions, but the MeerkatMachine-owned runtime loop committed turns through RuntimeStore without invoking the SessionStore projection save path. This wires the projection update after the machine commit succeeds, avoiding a pre-commit split-brain while restoring per-turn SessionStore saves for MobKit/UnifiedRuntime consumers.
Tests