Display TensorRT-LLM AgentX server metrics - #747
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
cquil11
force-pushed
the
agent/capture-trtllm-server-metrics
branch
from
August 17, 2026 15:28
83d2f0b to
4e703c0
Compare
Contributor
Author
|
Closing because the project scope is now limited to porting active, nondeprecated AMD multi-node configurations to srt-slurm. This PR is outside that scope. |
Contributor
Author
|
Reopening: this PR predates the recent out-of-scope configuration-porting detour and was closed by mistake during an overly broad cleanup. Its prior state is being restored. |
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
cquil11
force-pushed
the
agent/capture-trtllm-server-metrics
branch
from
August 27, 2026 19:13
4e703c0 to
1e48786
Compare
The hand-rolled point-detail charts laid their legend out on an equal-width grid (`innerW / itemCount`) with no regard for how long the labels actually are. At the inline 720-unit render the KV-cache chart packs eleven per-engine series into ~58 units each, so "prefill (500e)", "decode (5021)" and the rest overprinted one another into an unreadable smear. Add a shared legend layout helper that estimates each label's advance width from its characters (fullwidth CJK counted at ~1em, Latin by character class, rounded up so we wrap early rather than collide) and packs items greedily into as many rows as they need. Charts grow their viewBox height and bottom padding by the same amount, so the plot area keeps its exact geometry and the extra rows extend the SVG downward — a legend that already fit on one row renders exactly where it did before. Applied to all three charts that shared the pattern: TimeSeriesChart, StackedAreaChart, and the distribution percentile chips. Legend items are now packed left rather than spread across the plot width. Verified at inline (720), narrow, and expanded (1300) sizes with the eleven real KV-cache labels, with CJK labels, and against real trace data for point 439515 inline and in the expanded dialog. 中文：修复 agentic 明细页图表图例重叠问题。 这些手写 SVG 图表此前按 `innerW / 图例项数` 等宽分配图例位置，完全没有 考虑标签实际长度。在 720 单位的内联渲染下，KV cache 图表要放下 11 条 per-engine 曲线，每项只有约 58 单位，导致 "prefill (500e)"、"decode (5021)" 等标签互相叠印，完全无法辨认。 新增共享的图例排版辅助模块：按字符估算每个标签的宽度（CJK 全角字符按约 1em 计算，拉丁字符按字符类别估算，并统一向上取整，宁可提前换行也不重叠）， 再贪心地将图例项分配到所需的行数。图表将 viewBox 高度与底部内边距同步增加 相同数值，因此绘图区几何尺寸完全不变，多出的行向下延伸——原本单行就能放下的 图例，渲染位置与改动前完全一致。 该修复已应用到共用此排版逻辑的全部三个图表：TimeSeriesChart、 StackedAreaChart 以及分布图的分位数图例。图例项现在改为左对齐紧凑排列， 不再横向铺满绘图区宽度。 已在内联（720）、窄容器、展开（1300）三种尺寸下，使用真实的 11 条 KV cache 标签及 CJK 标签验证，并针对 439515 号数据点的真实 trace 数据验证了内联渲染 与展开弹窗。 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 0896009. Configure here.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Follows SemiAnalysisAI/InferenceX#2625 and NVIDIA/TensorRT-LLM#12545.
Validation
Note
Medium Risk
Touches production ETL versioning, disaggregated metric identity, and workflows that write production DB and invalidate cache; chart-series bumps require backfill/recompute for historical TRT runs to match new behavior.
Overview
Adds end-to-end TensorRT-LLM / Dynamo-TRT support in the AgentX data path: a dedicated server-metrics adapter, chart-series ETL (v16–18) for native
trtllm_*and Dynamo rank-aware KV gauges, matching aggregate extraction, and UI that shows a combined chip + CPU cache hit label when TRT reports only the GPU field under offload.CI / ops: ingest can target a named Neon child branch (resolve connection via Neon API, verify host, invalidate a preview URL). New
Recompute Agentic Metricsworkflow onmasterforce-backfills chart series and aggregate stats for one GitHub run, then invalidates production cache. Staging adds a refresh-chart-series step after ingest (scoped by--run-idon backfills).UI: Agentic point-detail charts replace equal-width legend slots with wrapped legends (
layoutChartLegend+ sharedChartLegend) so dense KV-cache labels no longer overlap; docs note the TRT metric preferences and recompute workflow.Reviewed by Cursor Bugbot for commit 0896009. Bugbot is set up for automated code reviews on this repo. Configure here.