Artificial Intelligence Recommendation:Log configuration

Debug log configuration

After a service is published, you can collect various information to analyze its online performance. Although you can print logs for offline debugging, this method is not flexible enough for an online service.

This configuration collects online debug information. You can enable it to print information to the console, output it to DataHub for detailed analysis, such as analyzing the effect of each recall, or write it to a local file for other calls. Currently, this configuration collects item data after the recall, filtering, and coarse-ranking stages.

The debug configuration corresponds to `DebugConfs` in the configuration overview. `DebugConfs` is a `Map[string]object` structure where the key represents the scenario. This lets you create isolated configurations for different scenarios.

{
    "DebugConfs": {
        "${scene_name}": {
            "OutputType": "datahub",
            "Rate": 0,
            "DebugUsers": [
                "1001"
            ],
            "DatahubName": "dh_debug_log",
            "KafkaName": "pairec_debug_log",
            "FilePath": "/Users/username/pairec/debug_log/",
            "MaxFileNum": 20
        }
    }
}

The output information includes the following fields:

• request_id: A unique ID for each recommendation request.

• module: The module that generated the log. Current modules include `recall`, `filter`, and `general_rank`.

• scene_id: The scenario ID.

• exp_id: The experiment ID.

• request_time: The request UNIX timestamp in seconds.

• uid: The user ID.

• retrieveid: The recall ID.

• items: The list of items. The format is `"item1:score1:{'dbmtl_prob_click':'0.03'},item2:score2:{'dbmtl_prob_click':'0.04'}"`. If `module` is `recall` or `filter`, `score` is the recall score.

  If `module` is `general_rank`, `rank`, or `sort`, the format is `"item1:score1:{'dbmtl_prob_click':'0.03'},item2:score2:{'dbmtl_prob_click':'0.03'}"`. In this case, `score` is the score from the model service during the general rank or rank stage. `dbmtl_prob_click` is one of the model's target scores. If the model service has multiple targets, the braces contain all target scores. The actual names of the targets depend on the model service, and `dbmtl_prob_click` is just an example.

Currently, within a module, items are output separately for each `retrieveid`. For example, if there are five recall sources, five records are generated when `module` is `recall`. Each record corresponds to one recall source. The `filter` and `general_rank` modules follow the same logging logic.

Datahub configuration

For the Datahub configuration, you do not need to create a topic. The engine automatically creates one if you specify the topic and schemas.

{
    "DatahubConfs": {
        "dh_debug_log": {
            "Endpoint": "http://dh-cn-beijing-int-vpc.aliyuncs.com",
            "ProjectName": "project_test",
            "TopicName": "pairec_debug_log",
            "Schemas": [
                {
                    "Field": "request_id",
                    "Type": "string"
                },
                {
                    "Field": "module",
                    "Type": "string"
                },
                {
                    "Field": "scene_id",
                    "Type": "string"
                },
                {
                    "Field": "request_time",
                    "Type": "integer"
                },
                {
                    "Field": "exp_id",
                    "Type": "string"
                },
                {
                    "Field": "items",
                    "Type": "string"
                },
                {
                    "Field": "retrieveid",
                    "Type": "string"
                },
                {
                    "Field": "uid",
                    "Type": "string"
                }
            ]
        }
    }
}

When the service runs, the engine automatically creates the `pairec_debug_log` topic and applies the specified schema. After you see log output in the DataHub console, you can create a subscription relationship between MaxCompute and DataHub. This saves the DataHub data to a MaxCompute table.

Kafka configuration

For the Kafka configuration, configure the instance name (which corresponds to `KafkaName` in `DebugConfs`), the `BootstrapServers` ingest endpoint, and the name of the pre-existing topic.

{
    "KafkaConfs": {
        "pairec_debug_log": {
            "BootstrapServers": "alikafka-post-cn-xxxxx-1.alikafka.aliyuncs.com:9093,alikafka-post-cn-xxxxx-2.alikafka.aliyuncs.com:9093,alikafka-post-cn-xxxxx-3.alikafka.aliyuncs.com:9093",
            "Topic": "debug_log"
        }
    }
}

Feature log configuration

This configuration is used to collect online user-side and item-side features. These features are then output to DataHub for detailed analysis.

The feature log configuration is defined in `FeatureLogConfs` in the configuration overview. `FeatureLogConfs` is a `Map[string]object` structure. The key represents the scenario, which lets you isolate the configuration for each scenario.

{
    "FeatureLogConfs": {
        "${scene_name}": {
            "OutputType": "datahub",
            "DatahubName": "",
            "UserFeatures": "",
            "ItemFeatures": ""
        }
    }
}

Datahub configuration

You do not need to create a topic in the Datahub configuration. The engine automatically creates a topic if you specify the topic and schemas.

{
    "DatahubConfs": {
        "dh_feature_log": {
            "Endpoint": "http://dh-cn-beijing-int-vpc.aliyuncs.com",
            "ProjectName": "",
            "TopicName": "pairec_feature_log",
            "Schemas": [
                {
                    "Field": "request_id",
                    "Type": "string"
                },
                {
                    "Field": "scene_id",
                    "Type": "string"
                },
                {
                    "Field": "exp_id",
                    "Type": "string"
                },
                {
                    "Field": "request_time",
                    "Type": "integer"
                },
                {
                    "Field": "user_id",
                    "Type": "string"
                },
                {
                    "Field": "user_features",
                    "Type": "string"
                },
                {
                    "Field": "item_id",
                    "Type": "string"
                },
                {
                    "Field": "position",
                    "Type": "string"
                },
                {
                    "Field": "item_features",
                    "Type": "string"
                }
            ]
        }
    }
}

When the service runs, the engine automatically creates the `pairec_feature_log` topic and applies the specified schema.