File size: 10,270 Bytes
9aa5185 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 | """Integration tests for gateway AIAgent caching.
Verifies that the agent cache correctly:
- Reuses agents across messages (same config → same instance)
- Rebuilds agents when config changes (model, provider, toolsets)
- Updates reasoning_config in-place without rebuilding
- Evicts on session reset
- Evicts on fallback activation
- Preserves frozen system prompt across turns
"""
import hashlib
import json
import threading
from unittest.mock import MagicMock, patch
import pytest
def _make_runner():
"""Create a minimal GatewayRunner with just the cache infrastructure."""
from gateway.run import GatewayRunner
runner = GatewayRunner.__new__(GatewayRunner)
runner._agent_cache = {}
runner._agent_cache_lock = threading.Lock()
return runner
class TestAgentConfigSignature:
"""Config signature produces stable, distinct keys."""
def test_same_config_same_signature(self):
from gateway.run import GatewayRunner
runtime = {"api_key": "sk-test12345678", "base_url": "https://openrouter.ai/api/v1",
"provider": "openrouter", "api_mode": "chat_completions"}
sig1 = GatewayRunner._agent_config_signature("claude-sonnet-4", runtime, ["hermes-telegram"], "")
sig2 = GatewayRunner._agent_config_signature("claude-sonnet-4", runtime, ["hermes-telegram"], "")
assert sig1 == sig2
def test_model_change_different_signature(self):
from gateway.run import GatewayRunner
runtime = {"api_key": "sk-test12345678", "base_url": "https://openrouter.ai/api/v1",
"provider": "openrouter"}
sig1 = GatewayRunner._agent_config_signature("claude-sonnet-4", runtime, ["hermes-telegram"], "")
sig2 = GatewayRunner._agent_config_signature("claude-opus-4.6", runtime, ["hermes-telegram"], "")
assert sig1 != sig2
def test_provider_change_different_signature(self):
from gateway.run import GatewayRunner
rt1 = {"api_key": "sk-test12345678", "base_url": "https://openrouter.ai/api/v1", "provider": "openrouter"}
rt2 = {"api_key": "sk-test12345678", "base_url": "https://api.anthropic.com", "provider": "anthropic"}
sig1 = GatewayRunner._agent_config_signature("claude-sonnet-4", rt1, ["hermes-telegram"], "")
sig2 = GatewayRunner._agent_config_signature("claude-sonnet-4", rt2, ["hermes-telegram"], "")
assert sig1 != sig2
def test_toolset_change_different_signature(self):
from gateway.run import GatewayRunner
runtime = {"api_key": "sk-test12345678", "base_url": "https://openrouter.ai/api/v1", "provider": "openrouter"}
sig1 = GatewayRunner._agent_config_signature("claude-sonnet-4", runtime, ["hermes-telegram"], "")
sig2 = GatewayRunner._agent_config_signature("claude-sonnet-4", runtime, ["hermes-discord"], "")
assert sig1 != sig2
def test_reasoning_not_in_signature(self):
"""Reasoning config is set per-message, not part of the signature."""
from gateway.run import GatewayRunner
runtime = {"api_key": "sk-test12345678", "base_url": "https://openrouter.ai/api/v1", "provider": "openrouter"}
# Same config — signature should be identical regardless of what
# reasoning_config the caller might have (it's not passed in)
sig1 = GatewayRunner._agent_config_signature("claude-sonnet-4", runtime, ["hermes-telegram"], "")
sig2 = GatewayRunner._agent_config_signature("claude-sonnet-4", runtime, ["hermes-telegram"], "")
assert sig1 == sig2
class TestAgentCacheLifecycle:
"""End-to-end cache behavior with real AIAgent construction."""
def test_cache_hit_returns_same_agent(self):
"""Second message with same config reuses the cached agent instance."""
from run_agent import AIAgent
runner = _make_runner()
session_key = "telegram:12345"
runtime = {"api_key": "test", "base_url": "https://openrouter.ai/api/v1",
"provider": "openrouter", "api_mode": "chat_completions"}
sig = runner._agent_config_signature("anthropic/claude-sonnet-4", runtime, ["hermes-telegram"], "")
# First message — create and cache
agent1 = AIAgent(
model="anthropic/claude-sonnet-4", api_key="test",
base_url="https://openrouter.ai/api/v1", provider="openrouter",
max_iterations=5, quiet_mode=True, skip_context_files=True,
skip_memory=True, platform="telegram",
)
with runner._agent_cache_lock:
runner._agent_cache[session_key] = (agent1, sig)
# Second message — cache hit
with runner._agent_cache_lock:
cached = runner._agent_cache.get(session_key)
assert cached is not None
assert cached[1] == sig
assert cached[0] is agent1 # same instance
def test_cache_miss_on_model_change(self):
"""Model change produces different signature → cache miss."""
from run_agent import AIAgent
runner = _make_runner()
session_key = "telegram:12345"
runtime = {"api_key": "test", "base_url": "https://openrouter.ai/api/v1",
"provider": "openrouter", "api_mode": "chat_completions"}
old_sig = runner._agent_config_signature("anthropic/claude-sonnet-4", runtime, ["hermes-telegram"], "")
agent1 = AIAgent(
model="anthropic/claude-sonnet-4", api_key="test",
base_url="https://openrouter.ai/api/v1", provider="openrouter",
max_iterations=5, quiet_mode=True, skip_context_files=True,
skip_memory=True, platform="telegram",
)
with runner._agent_cache_lock:
runner._agent_cache[session_key] = (agent1, old_sig)
# New model → different signature
new_sig = runner._agent_config_signature("anthropic/claude-opus-4.6", runtime, ["hermes-telegram"], "")
assert new_sig != old_sig
with runner._agent_cache_lock:
cached = runner._agent_cache.get(session_key)
assert cached[1] != new_sig # signature mismatch → would create new agent
def test_evict_on_session_reset(self):
"""_evict_cached_agent removes the entry."""
from run_agent import AIAgent
runner = _make_runner()
session_key = "telegram:12345"
agent = AIAgent(
model="anthropic/claude-sonnet-4", api_key="test",
base_url="https://openrouter.ai/api/v1", provider="openrouter",
max_iterations=5, quiet_mode=True, skip_context_files=True,
skip_memory=True,
)
with runner._agent_cache_lock:
runner._agent_cache[session_key] = (agent, "sig123")
runner._evict_cached_agent(session_key)
with runner._agent_cache_lock:
assert session_key not in runner._agent_cache
def test_evict_does_not_affect_other_sessions(self):
"""Evicting one session leaves other sessions cached."""
runner = _make_runner()
with runner._agent_cache_lock:
runner._agent_cache["session-A"] = ("agent-A", "sig-A")
runner._agent_cache["session-B"] = ("agent-B", "sig-B")
runner._evict_cached_agent("session-A")
with runner._agent_cache_lock:
assert "session-A" not in runner._agent_cache
assert "session-B" in runner._agent_cache
def test_reasoning_config_updates_in_place(self):
"""Reasoning config can be set on a cached agent without eviction."""
from run_agent import AIAgent
agent = AIAgent(
model="anthropic/claude-sonnet-4", api_key="test",
base_url="https://openrouter.ai/api/v1", provider="openrouter",
max_iterations=5, quiet_mode=True, skip_context_files=True,
skip_memory=True,
reasoning_config={"enabled": True, "effort": "medium"},
)
# Simulate per-message reasoning update
agent.reasoning_config = {"enabled": True, "effort": "high"}
assert agent.reasoning_config["effort"] == "high"
# System prompt should not be affected by reasoning change
prompt1 = agent._build_system_prompt()
agent._cached_system_prompt = prompt1 # simulate run_conversation caching
agent.reasoning_config = {"enabled": True, "effort": "low"}
prompt2 = agent._cached_system_prompt
assert prompt1 is prompt2 # same object — not invalidated by reasoning change
def test_system_prompt_frozen_across_cache_reuse(self):
"""The cached agent's system prompt stays identical across turns."""
from run_agent import AIAgent
agent = AIAgent(
model="anthropic/claude-sonnet-4", api_key="test",
base_url="https://openrouter.ai/api/v1", provider="openrouter",
max_iterations=5, quiet_mode=True, skip_context_files=True,
skip_memory=True, platform="telegram",
)
# Build system prompt (simulates first run_conversation)
prompt1 = agent._build_system_prompt()
agent._cached_system_prompt = prompt1
# Simulate second turn — prompt should be frozen
prompt2 = agent._cached_system_prompt
assert prompt1 is prompt2 # same object, not rebuilt
def test_callbacks_update_without_cache_eviction(self):
"""Per-message callbacks can be set on cached agent."""
from run_agent import AIAgent
agent = AIAgent(
model="anthropic/claude-sonnet-4", api_key="test",
base_url="https://openrouter.ai/api/v1", provider="openrouter",
max_iterations=5, quiet_mode=True, skip_context_files=True,
skip_memory=True,
)
# Set callbacks like the gateway does per-message
cb1 = lambda *a: None
cb2 = lambda *a: None
agent.tool_progress_callback = cb1
agent.step_callback = cb2
agent.stream_delta_callback = None
agent.status_callback = None
assert agent.tool_progress_callback is cb1
assert agent.step_callback is cb2
# Update for next message
cb3 = lambda *a: None
agent.tool_progress_callback = cb3
assert agent.tool_progress_callback is cb3
|