This is an automated email from the ASF dual-hosted git repository.
baoyuan pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/apisix.git
The following commit(s) were added to refs/heads/master by this push:
new 09f3d215e feat(ai-proxy): add support for pushing logs in ai-proxy
plugins (#12515)
09f3d215e is described below
commit 09f3d215eaece685389e34f74854418d72942316
Author: Ashish Tiwari <[email protected]>
AuthorDate: Wed Aug 27 08:21:21 2025 +0530
feat(ai-proxy): add support for pushing logs in ai-proxy plugins (#12515)
---
apisix/plugins/ai-proxy-multi.lua | 5 +
apisix/plugins/ai-proxy.lua | 5 +
apisix/plugins/ai-proxy/base.lua | 21 ++
apisix/plugins/ai-proxy/schema.lua | 17 ++
apisix/utils/log-util.lua | 9 +
t/plugin/{ai-proxy.t => ai-proxy-kafka-log.t} | 414 ++++++++------------------
t/plugin/ai-proxy-multi.openai-compatible.t | 3 +-
t/plugin/ai-proxy-multi.t | 2 +-
t/plugin/ai-proxy.openai-compatible.t | 2 +-
t/plugin/ai-proxy.t | 2 +-
10 files changed, 194 insertions(+), 286 deletions(-)
diff --git a/apisix/plugins/ai-proxy-multi.lua
b/apisix/plugins/ai-proxy-multi.lua
index 4c2dff582..b162eee96 100644
--- a/apisix/plugins/ai-proxy-multi.lua
+++ b/apisix/plugins/ai-proxy-multi.lua
@@ -357,5 +357,10 @@ end
_M.before_proxy = base.before_proxy
+function _M.log(conf, ctx)
+ if conf.logging then
+ base.set_logging(ctx, conf.logging.summaries, conf.logging.payloads)
+ end
+end
return _M
diff --git a/apisix/plugins/ai-proxy.lua b/apisix/plugins/ai-proxy.lua
index d8a5ac17e..092eb6a08 100644
--- a/apisix/plugins/ai-proxy.lua
+++ b/apisix/plugins/ai-proxy.lua
@@ -54,5 +54,10 @@ end
_M.before_proxy = base.before_proxy
+function _M.log(conf, ctx)
+ if conf.logging then
+ base.set_logging(ctx, conf.logging.summaries, conf.logging.payloads)
+ end
+end
return _M
diff --git a/apisix/plugins/ai-proxy/base.lua b/apisix/plugins/ai-proxy/base.lua
index 4f1102885..0c188f1e4 100644
--- a/apisix/plugins/ai-proxy/base.lua
+++ b/apisix/plugins/ai-proxy/base.lua
@@ -24,6 +24,27 @@ local bad_request = ngx.HTTP_BAD_REQUEST
local _M = {}
+function _M.set_logging(ctx, summaries, payloads)
+ if summaries then
+ ctx.llm_summary = {
+ model = ctx.var.llm_model,
+ duration = ctx.var.llm_time_to_first_token,
+ prompt_tokens = ctx.var.llm_prompt_tokens,
+ completion_tokens = ctx.var.llm_completion_tokens,
+ }
+ end
+ if payloads then
+ ctx.llm_request = {
+ messages = ctx.var.llm_request_body and
ctx.var.llm_request_body.messages,
+ stream = ctx.var.request_type == "ai_stream"
+ }
+ ctx.llm_response_text = {
+ content = ctx.var.llm_response_text
+ }
+ end
+end
+
+
function _M.before_proxy(conf, ctx)
local ai_instance = ctx.picked_ai_instance
local ai_driver = require("apisix.plugins.ai-drivers." ..
ai_instance.provider)
diff --git a/apisix/plugins/ai-proxy/schema.lua
b/apisix/plugins/ai-proxy/schema.lua
index 3510dca69..c6b674ace 100644
--- a/apisix/plugins/ai-proxy/schema.lua
+++ b/apisix/plugins/ai-proxy/schema.lua
@@ -102,6 +102,21 @@ local ai_instance_schema = {
},
}
+local logging_schema = {
+ type = "object",
+ properties = {
+ summaries = {
+ type = "boolean",
+ default = false,
+ description = "Record user request llm model, duration, req/res
token"
+ },
+ payloads = {
+ type = "boolean",
+ default = false,
+ description = "Record user request and response payload"
+ }
+ }
+}
_M.ai_proxy_schema = {
type = "object",
@@ -117,6 +132,7 @@ _M.ai_proxy_schema = {
}, -- add more providers later
},
+ logging = logging_schema,
auth = auth_schema,
options = model_options_schema,
timeout = {
@@ -176,6 +192,7 @@ _M.ai_proxy_multi_schema = {
default = { algorithm = "roundrobin" }
},
instances = ai_instance_schema,
+ logging_schema = logging_schema,
fallback_strategy = {
type = "string",
enum = { "instance_health_and_rate_limiting" },
diff --git a/apisix/utils/log-util.lua b/apisix/utils/log-util.lua
index c9cda1d6a..17a2e0e6d 100644
--- a/apisix/utils/log-util.lua
+++ b/apisix/utils/log-util.lua
@@ -290,6 +290,15 @@ function _M.get_log_entry(plugin_name, conf, ctx)
end
end
+ if ctx.llm_summary then
+ entry.llm_summary = ctx.llm_summary
+ end
+ if ctx.llm_request then
+ entry.llm_request = ctx.llm_request
+ end
+ if ctx.llm_response_text then
+ entry.llm_response_text = ctx.llm_response_text
+ end
return entry, customized
end
diff --git a/t/plugin/ai-proxy.t b/t/plugin/ai-proxy-kafka-log.t
similarity index 62%
copy from t/plugin/ai-proxy.t
copy to t/plugin/ai-proxy-kafka-log.t
index c99a6c11e..aa039987c 100644
--- a/t/plugin/ai-proxy.t
+++ b/t/plugin/ai-proxy-kafka-log.t
@@ -181,65 +181,7 @@ run_tests();
__DATA__
-=== TEST 1: minimal viable configuration
---- config
- location /t {
- content_by_lua_block {
- local plugin = require("apisix.plugins.ai-proxy")
- local ok, err = plugin.check_schema({
- provider = "openai",
- options = {
- model = "gpt-4",
- },
- auth = {
- header = {
- some_header = "some_value"
- }
- }
- })
-
- if not ok then
- ngx.say(err)
- else
- ngx.say("passed")
- end
- }
- }
---- response_body
-passed
-
-
-
-=== TEST 2: unsupported provider
---- config
- location /t {
- content_by_lua_block {
- local plugin = require("apisix.plugins.ai-proxy")
- local ok, err = plugin.check_schema({
- provider = "some-unique",
- options = {
- model = "gpt-4",
- },
- auth = {
- header = {
- some_header = "some_value"
- }
- }
- })
-
- if not ok then
- ngx.say(err)
- else
- ngx.say("passed")
- end
- }
- }
---- response_body eval
-qr/.*property "provider" validation failed: matches none of the enum values.*/
-
-
-
-=== TEST 3: set route with wrong auth header
+=== TEST 1: set route with logging summaries and payloads
--- config
location /t {
content_by_lua_block {
@@ -253,7 +195,7 @@ qr/.*property "provider" validation failed: matches none of
the enum values.*/
"provider": "openai",
"auth": {
"header": {
- "Authorization": "Bearer wrongtoken"
+ "Authorization": "Bearer token"
}
},
"options": {
@@ -264,8 +206,22 @@ qr/.*property "provider" validation failed: matches none
of the enum values.*/
"override": {
"endpoint": "http://localhost:6724";
},
- "ssl_verify": false
- }
+ "ssl_verify": false,
+ "logging": {
+ "summaries": true,
+ "payloads": true
+ }
+ },
+ "kafka-logger": {
+ "broker_list" :
+ {
+ "127.0.0.1":9092
+ },
+ "kafka_topic" : "test2",
+ "key" : "key1",
+ "timeout" : 1,
+ "batch_max_size": 1
+ }
}
}]]
)
@@ -281,17 +237,23 @@ passed
-=== TEST 4: send request
+=== TEST 2: send request
--- request
POST /anything
{ "messages": [ { "role": "system", "content": "You are a mathematician" }, {
"role": "user", "content": "What is 1+1?"} ] }
---- error_code: 401
---- response_body
-Unauthorized
+--- more_headers
+Authorization: Bearer token
+--- error_log
+send data to kafka:
+llm_request
+llm_summary
+You are a mathematician
+gpt-35-turbo-instruct
+llm_response_text
-=== TEST 5: set route with right auth header
+=== TEST 3: set route with logging summary but no payload
--- config
location /t {
content_by_lua_block {
@@ -316,8 +278,22 @@ Unauthorized
"override": {
"endpoint": "http://localhost:6724";
},
- "ssl_verify": false
- }
+ "ssl_verify": false,
+ "logging": {
+ "summaries": true,
+ "payloads": false
+ }
+ },
+ "kafka-logger": {
+ "broker_list" :
+ {
+ "127.0.0.1":9092
+ },
+ "kafka_topic" : "test2",
+ "key" : "key1",
+ "timeout" : 1,
+ "batch_max_size": 1
+ }
}
}]]
)
@@ -333,64 +309,23 @@ passed
-=== TEST 6: send request
---- request
-POST /anything
-{ "messages": [ { "role": "system", "content": "You are a mathematician" }, {
"role": "user", "content": "What is 1+1?"} ] }
---- more_headers
-Authorization: Bearer token
---- error_code: 200
---- response_body eval
-qr/\{ "content": "1 \+ 1 = 2\.", "role": "assistant" \}/
-
-
-
-=== TEST 7: send request with empty body
+=== TEST 4: send request
--- request
POST /anything
---- more_headers
-Authorization: Bearer token
---- error_code: 400
---- response_body_chomp
-failed to get request body: request body is empty
-
-
-
-=== TEST 8: send request with wrong method (GET) should work
---- request
-GET /anything
{ "messages": [ { "role": "system", "content": "You are a mathematician" }, {
"role": "user", "content": "What is 1+1?"} ] }
--- more_headers
Authorization: Bearer token
---- error_code: 200
---- response_body eval
-qr/\{ "content": "1 \+ 1 = 2\.", "role": "assistant" \}/
+--- error_log
+send data to kafka:
+llm_summary
+gpt-35-turbo-instruct
+--- no_error_log
+llm_request
+llm_response_text
-=== TEST 9: wrong JSON in request body should give error
---- request
-GET /anything
-{}"messages": [ { "role": "system", "cont
---- error_code: 400
---- response_body
-{"message":"could not get parse JSON request body: Expected the end but found
T_STRING at character 3"}
-
-
-
-=== TEST 10: content-type should be JSON
---- request
-POST /anything
-prompt%3Dwhat%2520is%25201%2520%252B%25201
---- more_headers
-Content-Type: application/x-www-form-urlencoded
---- error_code: 400
---- response_body chomp
-unsupported content-type: application/x-www-form-urlencoded, only
application/json is supported
-
-
-
-=== TEST 11: model options being merged to request body
+=== TEST 5: set route with no logging summary and payload - default behaviour
--- config
location /t {
content_by_lua_block {
@@ -408,114 +343,58 @@ unsupported content-type:
application/x-www-form-urlencoded, only application/js
}
},
"options": {
- "model": "some-model",
- "foo": "bar",
+ "model": "gpt-35-turbo-instruct",
+ "max_tokens": 512,
"temperature": 1.0
},
"override": {
"endpoint": "http://localhost:6724";
},
- "ssl_verify": false
- }
+ "ssl_verify": false,
+ "logging": {
+ "summaries": false,
+ "payloads": false
+ }
+ },
+ "kafka-logger": {
+ "broker_list" :
+ {
+ "127.0.0.1":9092
+ },
+ "kafka_topic" : "test2",
+ "key" : "key1",
+ "timeout" : 1,
+ "batch_max_size": 1
+ }
}
}]]
)
if code >= 300 then
ngx.status = code
- ngx.say(body)
- return
end
-
- local code, body, actual_body = t("/anything",
- ngx.HTTP_POST,
- [[{
- "messages": [
- { "role": "system", "content": "You are a
mathematician" },
- { "role": "user", "content": "What is 1+1?" }
- ]
- }]],
- nil,
- {
- ["test-type"] = "options",
- ["Content-Type"] = "application/json",
- }
- )
-
- ngx.status = code
- ngx.say(actual_body)
-
+ ngx.say(body)
}
}
---- error_code: 200
---- response_body_chomp
-options_works
-
-
-
-=== TEST 12: override path
---- config
- location /t {
- content_by_lua_block {
- local t = require("lib.test_admin").test
- local code, body = t('/apisix/admin/routes/1',
- ngx.HTTP_PUT,
- [[{
- "uri": "/anything",
- "plugins": {
- "ai-proxy": {
- "provider": "openai",
- "model": "some-model",
- "auth": {
- "header": {
- "Authorization": "Bearer token"
- }
- },
- "options": {
- "foo": "bar",
- "temperature": 1.0
- },
- "override": {
- "endpoint": "http://localhost:6724/random";
- },
- "ssl_verify": false
- }
- }
- }]]
- )
-
- if code >= 300 then
- ngx.status = code
- ngx.say(body)
- return
- end
+--- response_body
+passed
- local code, body, actual_body = t("/anything",
- ngx.HTTP_POST,
- [[{
- "messages": [
- { "role": "system", "content": "You are a
mathematician" },
- { "role": "user", "content": "What is 1+1?" }
- ]
- }]],
- nil,
- {
- ["test-type"] = "path",
- ["Content-Type"] = "application/json",
- }
- )
- ngx.status = code
- ngx.say(actual_body)
- }
- }
---- response_body_chomp
-path override works
+=== TEST 6: send request
+--- request
+POST /anything
+{ "messages": [ { "role": "system", "content": "You are a mathematician" }, {
"role": "user", "content": "What is 1+1?"} ] }
+--- more_headers
+Authorization: Bearer token
+--- no_error_log
+llm_request
+llm_response_text
+llm_summary
-=== TEST 13: set route with stream = true (SSE)
+=== TEST 7: set route with stream = true (SSE) with ai-proxy-multi plugin
--- config
location /t {
content_by_lua_block {
@@ -525,26 +404,46 @@ path override works
[[{
"uri": "/anything",
"plugins": {
- "ai-proxy": {
- "provider": "openai",
- "auth": {
- "header": {
- "Authorization": "Bearer token"
+ "ai-proxy-multi": {
+ "instances": [
+ {
+ "name": "self-hosted",
+ "provider": "openai-compatible",
+ "weight": 1,
+ "auth": {
+ "header": {
+ "Authorization": "Bearer token"
+ }
+ },
+ "options": {
+ "model": "custom-instruct",
+ "max_tokens": 512,
+ "temperature": 1.0,
+ "stream": true
+ },
+ "override": {
+ "endpoint":
"http://localhost:7737/v1/chat/completions";
+ }
}
- },
- "options": {
- "model": "gpt-35-turbo-instruct",
- "max_tokens": 512,
- "temperature": 1.0,
- "stream": true
- },
- "override": {
- "endpoint": "http://localhost:7737";
- },
- "ssl_verify": false
- }
+ ],
+ "ssl_verify": false,
+ "logging": {
+ "summaries": true,
+ "payloads": true
+ }
+ },
+ "kafka-logger": {
+ "broker_list" :
+ {
+ "127.0.0.1":9092
+ },
+ "kafka_topic" : "test2",
+ "key" : "key1",
+ "timeout" : 1,
+ "batch_max_size": 1
+ }
}
- }]]
+ }]]
)
if code >= 300 then
@@ -558,7 +457,7 @@ passed
-=== TEST 14: test is SSE works as expected
+=== TEST 8: test is SSE works as expected
--- config
location /t {
content_by_lua_block {
@@ -585,6 +484,7 @@ passed
},
path = "/anything",
body = [[{
+ "stream": true,
"messages": [
{ "role": "system", "content": "some content" }
]
@@ -614,60 +514,10 @@ passed
ngx.print(#final_res .. final_res[6])
}
}
---- response_body_like eval
+--- response_body_eval
qr/6data: \[DONE\]\n\n/
-
-
-
-=== TEST 15: proxy embedding endpoint
---- config
- location /t {
- content_by_lua_block {
- local t = require("lib.test_admin").test
- local code, body = t('/apisix/admin/routes/1',
- ngx.HTTP_PUT,
- [[{
- "uri": "/embeddings",
- "plugins": {
- "ai-proxy": {
- "provider": "openai",
- "auth": {
- "header": {
- "Authorization": "Bearer token"
- }
- },
- "options": {
- "model": "text-embedding-ada-002",
- "encoding_format": "float"
- },
- "override": {
- "endpoint":
"http://localhost:6724/v1/embeddings";
- }
- }
- }
- }]]
- )
-
- if code >= 300 then
- ngx.status = code
- ngx.say(body)
- return
- end
-
- ngx.say("passed")
- }
- }
---- response_body
-passed
-
-
-
-=== TEST 16: send request to embedding api
---- request
-POST /embeddings
-{
- "input": "The food was delicious and the waiter..."
-}
---- error_code: 200
---- response_body_like eval
-qr/.*text-embedding-ada-002*/
+--- error_log
+send data to kafka:
+llm_request
+llm_summary
+some content
diff --git a/t/plugin/ai-proxy-multi.openai-compatible.t
b/t/plugin/ai-proxy-multi.openai-compatible.t
index fe34b4f82..19a123a1d 100644
--- a/t/plugin/ai-proxy-multi.openai-compatible.t
+++ b/t/plugin/ai-proxy-multi.openai-compatible.t
@@ -266,7 +266,8 @@ passed
body = [[{
"messages": [
{ "role": "system", "content": "some content" }
- ]
+ ],
+ "stream": true
}]],
}
diff --git a/t/plugin/ai-proxy-multi.t b/t/plugin/ai-proxy-multi.t
index f557f173f..5434f7699 100644
--- a/t/plugin/ai-proxy-multi.t
+++ b/t/plugin/ai-proxy-multi.t
@@ -603,5 +603,5 @@ passed
ngx.print(#final_res .. final_res[6])
}
}
---- response_body_like eval
+--- response_body_eval
qr/6data: \[DONE\]\n\n/
diff --git a/t/plugin/ai-proxy.openai-compatible.t
b/t/plugin/ai-proxy.openai-compatible.t
index 9168816fd..efeec6eeb 100644
--- a/t/plugin/ai-proxy.openai-compatible.t
+++ b/t/plugin/ai-proxy.openai-compatible.t
@@ -336,5 +336,5 @@ passed
ngx.print(#final_res .. final_res[6])
}
}
---- response_body_like eval
+--- response_body_eval
qr/6data: \[DONE\]\n\n/
diff --git a/t/plugin/ai-proxy.t b/t/plugin/ai-proxy.t
index c99a6c11e..e8f4e1173 100644
--- a/t/plugin/ai-proxy.t
+++ b/t/plugin/ai-proxy.t
@@ -614,7 +614,7 @@ passed
ngx.print(#final_res .. final_res[6])
}
}
---- response_body_like eval
+--- response_body_eval
qr/6data: \[DONE\]\n\n/
