mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-02-04 18:21:52 +08:00

History

chenfeiz0326 e97af45556 [TRTLLM-10300][feat] Upload regression info to artifactory (#10599 ) Signed-off-by: Chenfei Zhang <chenfeiz@nvidia.com>		2026-01-19 10:16:31 +08:00
..
disaggregated	[TRTLLM-8952][feat] Support Multi-Node Disagg Perf Test in CI (#9138 )	2025-12-26 22:50:53 +08:00
perf_regression.py	[TRTLLM-10300][feat] Upload regression info to artifactory (#10599 )	2026-01-19 10:16:31 +08:00
perf_sanity_triage.py	[TRTLLM-8263][feat] Add Aggregated Perf Tests (#10598 )	2026-01-17 13:16:36 +08:00
README.md	[TRTLLM-8263][feat] Add Aggregated Perf Tests (#10598 )	2026-01-17 13:16:36 +08:00

README.md

Perf Sanity Triage

This directory contains perf_sanity_triage.py, a helper script for querying and updating perf sanity data in OpenSearch, and for sending regression summaries to Slack.

Basic Usage

This script is run by the Jenkins pipeline: https://prod.blsm.nvidia.com/sw-tensorrt-top-1/job/LLM/job/TRTLLM-Perf/job/PerfSanityTriage/

Inputs are configured in jenkins/runPerfSanityTriage.groovy:

BRANCH: repo branch to checkout
OPEN_SEARCH_PROJECT_NAME: OpenSearch project name
OPERATION: operation to perform (see Operations below)
QUERY_JOB_NUMBER: number of latest jobs to query (OPERATION = "SLACK BOT SENDS MESSAGE" only)
SLACK_CHANNEL_ID: Slack channel IDs (OPERATION = "SLACK BOT SENDS MESSAGE" only)
SLACK_BOT_TOKEN: Slack bot token (OPERATION = "SLACK BOT SENDS MESSAGE" only)

Operations

1) `SLACK BOT SENDS MESSAGE`

Queries regression data (post-merge only) and sends a formatted summary to Slack. The query filters for:

b_is_valid = true
b_is_post_merge = true
b_is_regression = true
b_is_baseline = false

Format

SLACK BOT SENDS MESSAGE

2) `UPDATE SET ... (WHERE ...)`

Updates fields on existing perf records that match a query scope and posts the updated documents back to OpenSearch.