Merge pull request #19 from logos-co/test-dispersal-resilience

Test/dispersal resilience

README.md

@@ -1,6 +1,6 @@
 # nomos-e2e-tests
 
-Nomos e2e framework used to test various implementations of the Nomos node.
+Nomos E2E framework used to test various implementations of the Nomos node.
 
 ## Setup and contribute
 
@@ -16,12 +16,36 @@ pre-commit install
 (optional) Overwrite default vars from src/env_vars.py via env vars or by adding a .env file
 pytest
 ```
-Set optional environment variable to search logs for errors after each tests:
+
+### Additional instructions for dispersal resilience tests  
+
+1. Build prerequisites
+```sh
+git clone https://github.com/logos-co/nomos-security-tests.git
+cd nomos-security-tests
+git fetch; git switch test-dispersal-resilience
+
+git checkout d8bbc464420ef86337df963c64ac2f7c3fd97008
+docker build --no-cache -f testnet/Dockerfile.debug -t nomos-mod-da-d8bbc46:testnet .
+# (x86_64) docker build --no-cache -f testnet/Dockerfile -t nomos-mod-da-d8bbc46:testnet .
+
+git checkout d19a1f3d8c80f654e6cf6139641519f16fe670ec
+docker build --no-cache -f testnet/Dockerfile.debug -t nomos-executor-mod-da-d19a1f3:testnet . 
+
+git checkout 7f54114b6c320dc32577b0e8bb85c2d543b4bd56
+docker build --no-cache -f testnet/Dockerfile.debug -t nomos-executor-mod-da-7f54114:testnet . 
+
+git checkout 4a58376ac4956d87502b9fd72b64a756396f2a8d
+docker build --no-cache -f testnet/Dockerfile.debug -t nomos-executor-mod-da-4a58376:testnet . 
+```
+
+2. Run tests with `pytest --run-with-mod-da-node tests/dispersal_resilience/test_dispersal_resilience.py`
+
+### Enable node log search with environment variable:
 ```shell
 export CHECK_LOG_ERRORS=True
 ```
 
-
 ## License
 
 Licensed and distributed under either of

cluster_config/scripts/run_nomos_executor_debug.sh

@@ -0,0 +1,14 @@
+#!/bin/sh
+
+set -e
+
+export CFG_FILE_PATH="/config.yaml" \
+       CFG_SERVER_ADDR="http://cfgsync:4400" \
+       CFG_HOST_IP=$(hostname -i) \
+       CFG_HOST_KIND="executor" \
+       CFG_HOST_IDENTIFIER="executor-$(hostname -i)" \
+       LOG_LEVEL="DEBUG" \
+       RISC0_DEV_MODE=true
+
+/usr/bin/cfgsync-client && \
+    exec /usr/bin/nomos-executor /config.yaml

cluster_config/scripts/run_nomos_node_debug.sh

@@ -0,0 +1,13 @@
+#!/bin/sh
+
+set -e
+
+export CFG_FILE_PATH="/config.yaml" \
+       CFG_SERVER_ADDR="http://cfgsync:4400" \
+       CFG_HOST_IP=$(hostname -i) \
+       CFG_HOST_IDENTIFIER="validator-$(hostname -i)" \
+       LOG_LEVEL="DEBUG" \
+       RISC0_DEV_MODE=true
+
+/usr/bin/cfgsync-client && \
+    exec /usr/bin/nomos-node /config.yaml

src/env_vars.py

@@ -14,10 +14,25 @@ def get_env_var(var_name, default=None):
     return env_var
 
 
+def make_mod_da_var(node_type, version, is_image=False):
+    base = "nomos_executor_mod_da" if node_type == "executor" else "nomos_mod_da"
+    value = f"{base}_{version}"
+
+    if is_image:
+        return value.replace("_", "-") + ":testnet"
+    else:
+        return value
+
+
 # Configuration constants. Need to be upercase to appear in reports
 DEFAULT_NOMOS_IMAGE = "ghcr.io/logos-co/nomos:testnet"
 NOMOS_IMAGE = get_env_var("NOMOS_IMAGE", DEFAULT_NOMOS_IMAGE)
 
+NOMOS_MOD_DA_IMAGE_d8bbc46 = make_mod_da_var("validator", "d8bbc46", True)
+NOMOS_EXECUTOR_MOD_DA_IMAGE_d19a1f3 = make_mod_da_var("executor", "d19a1f3", True)
+NOMOS_EXECUTOR_MOD_DA_IMAGE_7f54114 = make_mod_da_var("executor", "7f54114", True)
+NOMOS_EXECUTOR_MOD_DA_IMAGE_4a58376 = make_mod_da_var("executor", "4a58376", True)
+
 DEFAULT_PROXY_IMAGE = "bitnami/configurable-http-proxy:latest"
 HTTP_PROXY_IMAGE = get_env_var("HTTP_PROXY_IMAGE", DEFAULT_PROXY_IMAGE)
 
@@ -26,6 +41,10 @@ NOMOS = "nomos"
 NOMOS_EXECUTOR = "nomos_executor"
 CFGSYNC = "cfgsync"
 
+NOMOS_EXECUTOR_MOD_DA_d19a1f3 = make_mod_da_var("executor", "d19a1f3")
+NOMOS_EXECUTOR_MOD_DA_7f54114 = make_mod_da_var("executor", "7f54114")
+NOMOS_EXECUTOR_MOD_DA_4a58376 = make_mod_da_var("executor", "4a58376")
+
 NODE_1 = get_env_var("NODE_1", NOMOS)
 NODE_2 = get_env_var("NODE_2", NOMOS_EXECUTOR)
 NODE_3 = get_env_var("NODE_3", CFGSYNC)

src/node/node_vars.py

@@ -1,6 +1,36 @@
-from src.env_vars import NOMOS_IMAGE
+from src.env_vars import (
+    NOMOS_IMAGE,
+    NOMOS_MOD_DA_IMAGE_d8bbc46,
+    NOMOS_EXECUTOR_MOD_DA_IMAGE_d19a1f3,
+    NOMOS_EXECUTOR_MOD_DA_IMAGE_7f54114,
+    NOMOS_EXECUTOR_MOD_DA_IMAGE_4a58376,
+)
 
 nomos_nodes = {
+    "nomos_mod_da_d8bbc46": {
+        "image": NOMOS_MOD_DA_IMAGE_d8bbc46,
+        "volumes": ["cluster_config:/etc/nomos", "./kzgrs/kzgrs_test_params:/kzgrs_test_params:z"],
+        "ports": ["3000/udp", "18080/tcp"],
+        "entrypoint": "/etc/nomos/scripts/run_nomos_node_debug.sh",
+    },
+    "nomos_executor_mod_da_d19a1f3": {
+        "image": NOMOS_EXECUTOR_MOD_DA_IMAGE_d19a1f3,
+        "volumes": ["cluster_config:/etc/nomos", "./kzgrs/kzgrs_test_params:/kzgrs_test_params:z"],
+        "ports": ["3000/udp", "18080/tcp"],
+        "entrypoint": "/etc/nomos/scripts/run_nomos_executor_debug.sh",
+    },
+    "nomos_executor_mod_da_7f54114": {
+        "image": NOMOS_EXECUTOR_MOD_DA_IMAGE_7f54114,
+        "volumes": ["cluster_config:/etc/nomos", "./kzgrs/kzgrs_test_params:/kzgrs_test_params:z"],
+        "ports": ["3000/udp", "18080/tcp"],
+        "entrypoint": "/etc/nomos/scripts/run_nomos_executor_debug.sh",
+    },
+    "nomos_executor_mod_da_4a58376": {
+        "image": NOMOS_EXECUTOR_MOD_DA_IMAGE_4a58376,
+        "volumes": ["cluster_config:/etc/nomos", "./kzgrs/kzgrs_test_params:/kzgrs_test_params:z"],
+        "ports": ["3000/udp", "18080/tcp"],
+        "entrypoint": "/etc/nomos/scripts/run_nomos_executor_debug.sh",
+    },
     "nomos_custom": {
         "image": NOMOS_IMAGE,
         "volumes": ["cluster_config:/etc/nomos", "./kzgrs/kzgrs_test_params:/kzgrs_test_params:z"],

src/steps/common.py

@@ -4,7 +4,7 @@ import os
 import pytest
 
 from src.client.proxy_client import ProxyClient
-from src.env_vars import CFGSYNC, NOMOS, NOMOS_EXECUTOR, CONSENSUS_SLOT_TIME
+from src.env_vars import CFGSYNC, NOMOS, NOMOS_EXECUTOR, CONSENSUS_SLOT_TIME, make_mod_da_var
 from src.libs.common import delay
 from src.libs.custom_logger import get_custom_logger
 from src.node.nomos_node import NomosNode
@@ -112,3 +112,27 @@ class StepsCommon:
     @pytest.fixture(params=["setup_2_node_cluster", "setup_4_node_cluster"])
     def setup_cluster_variant(self, request):
         return request.getfixturevalue(request.param)
+
+    @pytest.fixture(scope="function")
+    def setup_2_node_mod_da_cluster(self, request):
+        logger.debug(f"Running fixture setup: {inspect.currentframe().f_code.co_name}")
+
+        subnet_size = get_param_or_default(request, "subnet_size", 2)
+        dispersal_factor = get_param_or_default(request, "dispersal_factor", 2)
+        min_dispersal_peers = get_param_or_default(request, "min_dispersal_peers", 1)
+        prepare_cluster_config(2, subnet_size, dispersal_factor, min_dispersal_peers)
+
+        validator_version = get_param_or_default(request, "validator_version", "")
+        validator = make_mod_da_var("validator", validator_version)
+
+        executor_version = get_param_or_default(request, "executor_version", "")
+        executor = make_mod_da_var("executor", executor_version)
+
+        self.node1 = NomosNode(CFGSYNC, "cfgsync")
+        self.node2 = NomosNode(validator, "nomos_node_0")
+        self.node3 = NomosNode(executor, "nomos_node_1")
+        self.main_nodes.extend([self.node1, self.node2, self.node3])
+        start_nodes(self.main_nodes)
+        ensure_nodes_ready(self.main_nodes[1:])
+
+        delay(CONSENSUS_SLOT_TIME)

src/steps/da.py

@@ -48,7 +48,7 @@ class StepsDataAvailability(StepsCommon):
     def find_executor_node(self):
         executor = {}
         for node in self.main_nodes:
-            if node.node_type() == NOMOS_EXECUTOR:
+            if "nomos_executor" in node.node_type():
                 executor = node
         return executor
 

tests/conftest.py

@@ -17,6 +17,21 @@ from src.data_storage import DS
 logger = get_custom_logger(__name__)
 
 
+def pytest_addoption(parser):
+    parser.addoption("--run-with-mod-da-node", action="store_true", default=False, help="Run tests requiring nodes with modified da layer")
+
+
+def pytest_configure(config):
+    config.addinivalue_line("markers", "mod_da_node: Mark test as requiring --run-with-mod-da-node")
+
+
+def pytest_collection_modifyitems(config, items):
+    run_mod_da_node = config.getoption("--run-with-mod-da-node")
+    for item in items:
+        if "mod_da_node" in item.keywords and not run_mod_da_node:
+            item.add_marker(pytest.mark.skip(reason="Requires --run-with-mod-da-node option"))
+
+
 # See https://docs.pytest.org/en/latest/example/simple.html#making-test-result-information-available-in-fixtures
 @pytest.hookimpl(hookwrapper=True, tryfirst=True)
 def pytest_runtest_makereport(item):

tests/dispersal_resilience/test_dispersal_resilience.py

@@ -0,0 +1,48 @@
+import pytest
+
+from src.env_vars import CONSENSUS_SLOT_TIME
+from src.libs.common import to_app_id, to_index, delay
+from src.steps.da import StepsDataAvailability
+from src.test_data import DATA_TO_DISPERSE
+
+
+@pytest.mark.usefixtures("setup_2_node_mod_da_cluster")
+@pytest.mark.mod_da_node
+class TestDispersalResilience(StepsDataAvailability):
+    main_nodes = []
+
+    @pytest.mark.parametrize("setup_2_node_mod_da_cluster", [{"validator_version": "d8bbc46", "executor_version": "d19a1f3"}], indirect=True)
+    def test_integrity_kzg_commitments(self):
+        # Confirm validator node has rejected dispersal request from executor - there is a mismatch between
+        # column data and proofs.
+        self.disperse_data(DATA_TO_DISPERSE[3], to_app_id(1), to_index(0))
+        delay(CONSENSUS_SLOT_TIME)
+        try:
+            rcv_data = self.get_data_range(self.node2, to_app_id(1), to_index(0), to_index(5))
+        except AssertionError as ae:
+            assert "Get data range response is empty" in str(ae), "Get data range response should be empty"
+
+        if rcv_data:
+            raise AssertionError("Get data range response should be empty")
+
+    @pytest.mark.parametrize("setup_2_node_mod_da_cluster", [{"validator_version": "d8bbc46", "executor_version": "7f54114"}], indirect=True)
+    def test_chunkification_robustness_different_chunk_size(self):
+        # Confirm validator node has rejected dispersal request from executor with different data alignment
+        try:
+            self.disperse_data(DATA_TO_DISPERSE[4], to_app_id(1), to_index(0), timeout_duration=0)
+        except Exception as e:
+            assert "does not match destination slice length" in str(e), "Send dispersal request with different data alignment should fail"
+            return
+
+        raise AssertionError("Send dispersal request with different data alignment should fail")
+
+    @pytest.mark.parametrize("setup_2_node_mod_da_cluster", [{"validator_version": "d8bbc46", "executor_version": "4a58376"}], indirect=True)
+    def test_rs_encoding_resistance_to_manipulation(self):
+        # Confirm validator node has rejected dispersal request from executor with inconsistent RS encoding
+        try:
+            self.disperse_data(DATA_TO_DISPERSE[5], to_app_id(1), to_index(0), timeout_duration=0)
+        except Exception as e:
+            assert "blob sampling timed out for" in str(e), "Send dispersal request with inconsistent RS encoding should fail"
+            return
+
+        raise AssertionError("Send dispersal request with inconsistent RS encoding should fail")