File size: 9,709 Bytes
e3278e4 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 |
#### What this does ####
# identifies least busy deployment
# How is this achieved?
# - Before each call, have the router print the state of requests {"deployment": "requests_in_flight"}
# - use litellm.input_callbacks to log when a request is just about to be made to a model - {"deployment-id": traffic}
# - use litellm.success + failure callbacks to log when a request completed
# - in get_available_deployment, for a given model group name -> pick based on traffic
import random
from typing import Optional
from litellm.caching.caching import DualCache
from litellm.integrations.custom_logger import CustomLogger
class LeastBusyLoggingHandler(CustomLogger):
test_flag: bool = False
logged_success: int = 0
logged_failure: int = 0
def __init__(self, router_cache: DualCache, model_list: list):
self.router_cache = router_cache
self.mapping_deployment_to_id: dict = {}
self.model_list = model_list
def log_pre_api_call(self, model, messages, kwargs):
"""
Log when a model is being used.
Caching based on model group.
"""
try:
if kwargs["litellm_params"].get("metadata") is None:
pass
else:
model_group = kwargs["litellm_params"]["metadata"].get(
"model_group", None
)
id = kwargs["litellm_params"].get("model_info", {}).get("id", None)
if model_group is None or id is None:
return
elif isinstance(id, int):
id = str(id)
request_count_api_key = f"{model_group}_request_count"
# update cache
request_count_dict = (
self.router_cache.get_cache(key=request_count_api_key) or {}
)
request_count_dict[id] = request_count_dict.get(id, 0) + 1
self.router_cache.set_cache(
key=request_count_api_key, value=request_count_dict
)
except Exception:
pass
def log_success_event(self, kwargs, response_obj, start_time, end_time):
try:
if kwargs["litellm_params"].get("metadata") is None:
pass
else:
model_group = kwargs["litellm_params"]["metadata"].get(
"model_group", None
)
id = kwargs["litellm_params"].get("model_info", {}).get("id", None)
if model_group is None or id is None:
return
elif isinstance(id, int):
id = str(id)
request_count_api_key = f"{model_group}_request_count"
# decrement count in cache
request_count_dict = (
self.router_cache.get_cache(key=request_count_api_key) or {}
)
request_count_value: Optional[int] = request_count_dict.get(id, 0)
if request_count_value is None:
return
request_count_dict[id] = request_count_value - 1
self.router_cache.set_cache(
key=request_count_api_key, value=request_count_dict
)
### TESTING ###
if self.test_flag:
self.logged_success += 1
except Exception:
pass
def log_failure_event(self, kwargs, response_obj, start_time, end_time):
try:
if kwargs["litellm_params"].get("metadata") is None:
pass
else:
model_group = kwargs["litellm_params"]["metadata"].get(
"model_group", None
)
id = kwargs["litellm_params"].get("model_info", {}).get("id", None)
if model_group is None or id is None:
return
elif isinstance(id, int):
id = str(id)
request_count_api_key = f"{model_group}_request_count"
# decrement count in cache
request_count_dict = (
self.router_cache.get_cache(key=request_count_api_key) or {}
)
request_count_value: Optional[int] = request_count_dict.get(id, 0)
if request_count_value is None:
return
request_count_dict[id] = request_count_value - 1
self.router_cache.set_cache(
key=request_count_api_key, value=request_count_dict
)
### TESTING ###
if self.test_flag:
self.logged_failure += 1
except Exception:
pass
async def async_log_success_event(self, kwargs, response_obj, start_time, end_time):
try:
if kwargs["litellm_params"].get("metadata") is None:
pass
else:
model_group = kwargs["litellm_params"]["metadata"].get(
"model_group", None
)
id = kwargs["litellm_params"].get("model_info", {}).get("id", None)
if model_group is None or id is None:
return
elif isinstance(id, int):
id = str(id)
request_count_api_key = f"{model_group}_request_count"
# decrement count in cache
request_count_dict = (
await self.router_cache.async_get_cache(key=request_count_api_key)
or {}
)
request_count_value: Optional[int] = request_count_dict.get(id, 0)
if request_count_value is None:
return
request_count_dict[id] = request_count_value - 1
await self.router_cache.async_set_cache(
key=request_count_api_key, value=request_count_dict
)
### TESTING ###
if self.test_flag:
self.logged_success += 1
except Exception:
pass
async def async_log_failure_event(self, kwargs, response_obj, start_time, end_time):
try:
if kwargs["litellm_params"].get("metadata") is None:
pass
else:
model_group = kwargs["litellm_params"]["metadata"].get(
"model_group", None
)
id = kwargs["litellm_params"].get("model_info", {}).get("id", None)
if model_group is None or id is None:
return
elif isinstance(id, int):
id = str(id)
request_count_api_key = f"{model_group}_request_count"
# decrement count in cache
request_count_dict = (
await self.router_cache.async_get_cache(key=request_count_api_key)
or {}
)
request_count_value: Optional[int] = request_count_dict.get(id, 0)
if request_count_value is None:
return
request_count_dict[id] = request_count_value - 1
await self.router_cache.async_set_cache(
key=request_count_api_key, value=request_count_dict
)
### TESTING ###
if self.test_flag:
self.logged_failure += 1
except Exception:
pass
def _get_available_deployments(
self,
healthy_deployments: list,
all_deployments: dict,
):
"""
Helper to get deployments using least busy strategy
"""
for d in healthy_deployments:
## if healthy deployment not yet used
if d["model_info"]["id"] not in all_deployments:
all_deployments[d["model_info"]["id"]] = 0
# map deployment to id
# pick least busy deployment
min_traffic = float("inf")
min_deployment = None
for k, v in all_deployments.items():
if v < min_traffic:
min_traffic = v
min_deployment = k
if min_deployment is not None:
## check if min deployment is a string, if so, cast it to int
for m in healthy_deployments:
if m["model_info"]["id"] == min_deployment:
return m
min_deployment = random.choice(healthy_deployments)
else:
min_deployment = random.choice(healthy_deployments)
return min_deployment
def get_available_deployments(
self,
model_group: str,
healthy_deployments: list,
):
"""
Sync helper to get deployments using least busy strategy
"""
request_count_api_key = f"{model_group}_request_count"
all_deployments = self.router_cache.get_cache(key=request_count_api_key) or {}
return self._get_available_deployments(
healthy_deployments=healthy_deployments,
all_deployments=all_deployments,
)
async def async_get_available_deployments(
self, model_group: str, healthy_deployments: list
):
"""
Async helper to get deployments using least busy strategy
"""
request_count_api_key = f"{model_group}_request_count"
all_deployments = (
await self.router_cache.async_get_cache(key=request_count_api_key) or {}
)
return self._get_available_deployments(
healthy_deployments=healthy_deployments,
all_deployments=all_deployments,
)
|