modelcontextprotocol / modelcontextprotocol/python-sdk
Simplify and harden session lifecycle with an explicit state machine
Personne n'a encore pris cette issue.
- Langage dominant
- Python
- Étoiles
- 24.3k
- Forks
- 4k
- Merge moyen
- 1 j 1 h
- PR mergées (30 j)
- 31
Description
Description
Summary
Server and client sessions use an internal initialization state machine combined with various flags and manual AsyncExitStack management. Stateless mode complicates this further by setting an "Initialized" state early as a workaround.
This makes the lifecycle hard to reason about and contributes to subtle bugs (e.g., past issues like #756 in stateless mode).
Problems
- Scattered state management: Initialization and teardown logic are spread across methods and files.
- Special cases for stateless mode: Stateless HTTP sets the session as initialized even though no real protocol negotiation has occurred.
- Resource management risk: If initialization fails midway, transport tasks and resources may not be cleaned up reliably.
- Testing difficulty: Tests often bypass parts of the lifecycle using in-memory transports.
Proposal
-
Introduce an explicit session state machine
Represent distinct states as separate types, for example:
UninitializedSessionNegotiatingSessionActiveSessionClosedSessionStatelessSession(for per-request / ephemeral cases)
Each state exposes only the operations that are valid in that state, and transitions return the next state type.
-
Model stateless mode explicitly
- Instead of marking a stateful session as "Initialized" prematurely, model stateless HTTP as a dedicated
StatelessSessiontype with a simpler lifecycle. - Make behavior differences clear in code and docs.
- Instead of marking a stateful session as "Initialized" prematurely, model stateless HTTP as a dedicated
-
Centralize resource cleanup
- Ensure that all lifecycle paths (happy path, error, cancellation) lead through code that:
- cancels outstanding tasks,
- closes transports,
- releases resources stored in
AsyncExitStack.
- Ensure that all lifecycle paths (happy path, error, cancellation) lead through code that:
Why this matters
- Reliability: Fewer edge cases and surprise states where messages can be processed incorrectly.
- Debuggability: Easier to reason about what can happen in each state.
- Extensibility: Adding new lifecycle behavior (e.g., resuming sessions) becomes more manageable.
Acceptance criteria
- Session lifecycle is represented via explicit types or a clearly defined state machine.
- Stateless HTTP mode uses a dedicated path rather than setting "Initialized" as a hack.
- All entry/exit paths of sessions ensure proper cleanup of transports and tasks.
- Tests cover state transitions, including error cases and stateless/stat eful differences.
References
No response
Guide de contribution
Ouvrir le guide de contribution
Par où commencer
- Lisez l'issue en entier, puis le guide de contribution du projet.
- Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
- Forkez le dépôt et travaillez sur une branche.
- Ouvrez une pull request qui référence le numéro de l'issue.
Piste de recherche
Aucun fichier ni point d’entrée n’est nommé ; commencez par localiser l’initialisation de session existante, le chemin HTTP sans état et le nettoyage de AsyncExitStack. Examinez l’issue #756 et les tests du transport en mémoire décrits dans l’issue. Le travail est terminé lorsque des états explicites du cycle de vie, un chemin dédié sans état, un nettoyage fiable sur chaque chemin de sortie et des tests de transition pour les cas d’erreur et avec/sans état sont présents.
Rédigé par le modèle d'indexation à partir du texte de l'issue.
Évaluation
- Stack technique
- python
- Domaine
- api, backend
- Type d'issue
- Refactorisation
- Difficulté
- 5/5
- Temps estimé
- Plus d'une semaine
- Activité
- À l'abandon
- Clarté
- À clarifier
- Accessibilité débutants
- 25/100