GitHub user weiqingy edited a comment on the discussion: [Feature]
Per-Event-Type Configurable Log Levels for Event Log
If max-length is a total size cap, the question becomes: how do we allocate the
budget across fields? Proportional allocation, equal split, largest-first
greedy - each has trade-offs, and the result is hard for users to predict. The
config looks simple (one number), but the behavior is opaque. I'm rethinking
about this - per-field thresholds seems actually the cleaner model. Each
threshold independently controls one truncation strategy, so the behavior is
predictable, e.g.,:
```
event-log.standard.max-string-length: 2000
event-log.standard.max-array-elements: 20
event-log.standard.max-depth: 5
...
```
These compose by walking the object graph - array trimming reduces element
count, then string truncation caps leaf string values within the retained
structure, and depth collapsing handles deeply nested objects. For example,
given a ChatRequestEvent with 50 messages and a 10K-char response:
- max-array-elements: 20 trims messages from 50 to 20 elements
- max-string-length: 2000 truncates the content field inside each retained
message (e.g., a 5000-char content → 2000 chars), and truncates the 10K-char
response to 2000 chars
- Short fields like model, role, requestId are untouched
Truncated fields use the wrapper format you suggested for parsability:
```
{
"messages": {
"truncatedList": [
{"role": "system", "content": "You are a helpful assistant..."},
{"role": "user", "content": {"truncatedString": "Analyze this doc...",
"omittedChars": 1000}},
{"role": "assistant", "content": {"truncatedString": "Based on my...",
"omittedChars": 3000}}
],
"omittedElements": 30
},
"response": {"truncatedString": "Here is my detailed answer...",
"omittedChars": 8000},
"model": "gpt-4",
"tools": ["add", "multiply"]
}
```
Each threshold has clear, independent behavior - no opaque budget allocation.
GitHub link:
https://github.com/apache/flink-agents/discussions/552#discussioncomment-15981219
----
This is an automatically sent email for [email protected].
To unsubscribe, please send an email to: [email protected]
