From 47e90c749e6ce0b71d02b056be1e885ebd679724 Mon Sep 17 00:00:00 2001
From: HusseinAdeiza <HusseinAdeiza@users.noreply.github.com>
Date: Wed, 14 Oct 2026 11:26:54 +0200
Subject: [PATCH] fix(consensus): guard against zero total stake in VRF
 validator selection

get_validators_for_transaction divided each validator's stake by the total
stake without checking for zero. When all validators have stake=0 (e.g.
internally-created validators in protocol_rpc/endpoints.py), this raised
ZeroDivisionError and crashed the consensus worker, halting transaction
processing.

Fall back to a uniform probability distribution when the total stake is
zero; non-zero totals are unaffected. Adds a unit test covering the
zero-stake path.

Closes #1732

Signed-off-by: HusseinAdeiza <HusseinAdeiza@users.noreply.github.com>
---
 backend/consensus/vrf.py |  8 +++++++-
 tests/unit/test_vrf.py   | 26 ++++++++++++++++++++++++++
 2 files changed, 33 insertions(+), 1 deletion(-)

diff --git a/backend/consensus/vrf.py b/backend/consensus/vrf.py
index 55c953600..c13daafcb 100644
--- a/backend/consensus/vrf.py
+++ b/backend/consensus/vrf.py
@@ -18,7 +18,13 @@ def get_validators_for_transaction(
     num_validators = min(num_validators, len(nodes))
 
     total_stake = sum(validator["stake"] for validator in nodes)
-    probabilities = [validator["stake"] / total_stake for validator in nodes]
+    if total_stake == 0:
+        # All validators have zero stake (e.g. internally-created validators in
+        # protocol_rpc/endpoints.py). Dividing by zero crashed the consensus
+        # worker, so fall back to a uniform distribution across equal-stake nodes.
+        probabilities = [1 / len(nodes)] * len(nodes)
+    else:
+        probabilities = [validator["stake"] / total_stake for validator in nodes]
 
     selected_validators = rng.choice(
         nodes,
diff --git a/tests/unit/test_vrf.py b/tests/unit/test_vrf.py
index 95473770e..9c472438e 100644
--- a/tests/unit/test_vrf.py
+++ b/tests/unit/test_vrf.py
@@ -1,6 +1,8 @@
 from backend.consensus.vrf import get_validators_for_transaction
 from unittest.mock import Mock
 
+import pytest
+
 
 def list_of_dicts_to_set(list_of_dicts: list[dict]) -> set:
     return set(map(lambda x: tuple(x.items()), list_of_dicts))
@@ -66,3 +68,27 @@ def choice_mock(a, p, size, replace):
 
     rng.choice.assert_called_once()
     assert validators == [{"stake": 3}, {"stake": 2}, {"stake": 1}]
+
+
+def test_get_validators_for_transaction_all_zero_stake():
+    """
+    Tests that a uniform probability is used when the total stake is zero.
+
+    Previously this divided by zero and crashed the consensus worker
+    (ZeroDivisionError). Internally-created validators with stake=0 make this a
+    reachable path, so equal-stake nodes must fall back to a uniform distribution.
+    """
+    nodes = [{"stake": 0}, {"stake": 0}, {"stake": 0}]
+
+    def choice_mock(a, p, size, replace):
+        assert p == [1 / 3, 1 / 3, 1 / 3]
+        assert sum(p) == pytest.approx(1.0)
+        return list(a)
+
+    rng = Mock()
+    rng.choice.side_effect = choice_mock
+
+    validators = get_validators_for_transaction(nodes, 3, rng=rng)
+
+    rng.choice.assert_called_once()
+    assert validators == nodes