Fix on_up() destroying healthy pool after replace-with-same-IP

[bitpathfinder]

Summary

When a node is replaced with the same IP, the driver receives both TOPOLOGY_CHANGE NEW_NODE and STATUS_CHANGE UP events. The NEW_NODE handler runs first, replacing the old host and establishing a new pool via on_add. The STATUS_CHANGE UP handler fires later (after a random 0-2s delay) with a stale reference to the old host object. Because Host.__eq__/__hash__ are endpoint-based, the stale on_up() tears down the new host's pool, causing a brief window where queries fail with NoHostAvailable.

Fix

Two early-return guards added at the top of Cluster.on_up():

1. Stale host check: If the host object has been replaced in metadata (different identity, same endpoint) and the new host is already up, skip processing.
2. Healthy pool check: If a non-shutdown pool already exists for this host (established by on_add), mark the host as up and skip the teardown/rebuild cycle.

Fixes: SCYLLADB-833

[Copilot]

Pull request overview

Addresses SCYLLADB-833 where Cluster.on_up() can be invoked with a stale Host instance after a replace-with-same-IP, causing the driver to tear down a healthy pool and briefly fail queries.

Changes:

• Add early-return guards in Cluster.on_up() to skip handling when the Host reference is stale or when a healthy pool already exists.
• Add unit tests covering stale-host and healthy-pool scenarios to prevent regressions.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
cassandra/cluster.py	Adds on_up() guard clauses to avoid tearing down an already-healthy pool when handling stale host references.
tests/unit/test_cluster.py	Adds unit tests validating the new on_up() early-return behavior for stale hosts and pre-existing healthy pools.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[dkropachev]

Blocking: the healthy-pool early return at

python-driver/cassandra/cluster.py

Line 1944 in c9b8e5e

for session in tuple(self.sessions):

is too broad. It exits the entire on_up() handler as soon as any session has a live pool, but on_up() still needs to do per-session reconciliation. In a multi-session cluster, later sessions never get their remove/rebuild or update_created_pools() pass, so a stale UP event can leave the driver half-up.

The new tests only cover a single mock session at

python-driver/tests/unit/test_cluster.py

Line 784 in c9b8e5e

cluster = self._make_cluster(sessions={mock_session})

, so this case stays untested. I’d narrow the skip to the session that already has the healthy pool, or keep the rest of on_up() bookkeeping running.

[mykaul]

@bitpathfinder - are all review comments replied? Can it be re-reviewed?