danvics/haiku.rag
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Say what the sandbox's connection lock guards

Browse source 
The lock claimed to serialize a whole read, while `_documents` resolves the
owners under it and then reads through them outside. That is right: resolving
the owners is an operation on the shared connection, reading through them is
not, and each owner is a session of its own. Holding the lock across a read per
database would serialize them against the capability's searches to guard state
none of them touch.
This commit is contained in:
Yiorgis Gozadinos 2026-08-27 17:13:07 +03:00
parent b6bc54c69e
commit 18c22f1ddb
No known key found for this signature in database
1 changed files with 10 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
haiku_rag_slim/haiku/rag/sandbox/sandbox.py
View file

@ -225,13 +225,17 @@ class Sandbox:
async def _connection( async def _connection(
self, owner: "HaikuRAG | None" = None self, owner: "HaikuRAG | None" = None
) -> "AsyncIterator[HaikuRAG]": ) -> "AsyncIterator[HaikuRAG]":
"""Yield the shared connection (serialized by the lock), or an ephemeral """Yield the shared connection, or an ephemeral read-only one.
read-only one. The lock guards the whole block so a read's awaits cannot
interleave with another task's operation on the same connection. The lock, where there is one, is held for the block: what a caller does
to the shared connection is serialized against the capability's own tool
calls, since both hold that client. It guards that connection and not
the databases underneath, which have their own locks.
`owner` is the client holding one document, for the reads addressed to a `owner` is the client holding one document, for the reads addressed to a
single document. The shared connection covers a set of databases and has single document. The shared connection covers a set of databases and has
no repositories of its own, so those reads have to name their owner. no repositories of its own, so those reads have to name their owner. An
owner is a session of its own, so it is yielded unserialized.
""" """
connection = owner if owner is not None else self._rag connection = owner if owner is not None else self._rag
if connection is not None: if connection is not None:
@ -272,6 +276,8 @@ class Sandbox:
docs = await rag.list_documents(filter=self._context.filter) docs = await rag.list_documents(filter=self._context.filter)
return docs, {} return docs, {}
owners = await rag.clients_covering(self._context.sources) owners = await rag.clients_covering(self._context.sources)
# Resolve owners under the shared-client lock; owner sessions perform
# reads independently.
groups = await asyncio.gather( groups = await asyncio.gather(
*(owner.list_documents(filter=self._context.filter) for owner in owners) *(owner.list_documents(filter=self._context.filter) for owner in owners)
) )