[CI]: 5 day maximum for Comprehensive Test flexibility#2676
[CI]: 5 day maximum for Comprehensive Test flexibility#2676ApostaC merged 2 commits intoLMCache:devfrom
Conversation
Signed-off-by: Samuel Shen <slshen@uchciago.edu>
Summary of ChangesHello, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request introduces an adaptive mechanism for performance test thresholds within the comprehensive CI suite. By dynamically adjusting the acceptable performance degradation based on the number of available historical baselines, the system gains flexibility. This prevents tests from failing too strictly when baseline data is sparse, ensuring more reliable and less flaky CI results without compromising overall performance monitoring. Highlights
🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console. Changelog
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
There was a problem hiding this comment.
Code Review
This pull request introduces an adaptive threshold for performance tests, making them more flexible when fewer baselines are available. This is a sensible improvement. My review includes suggestions to refactor the implementation for better maintainability by reducing code duplication and to improve conciseness and efficiency in the shell script.
| local threshold | ||
| threshold=$(awk -v n="$num_baselines" 'BEGIN { | ||
| gap = (5 - n); if (gap < 0) gap = 0 | ||
| printf "%.1f", 1.2 + gap * 0.1 | ||
| }') | ||
| local pct | ||
| pct=$(awk -v t="$threshold" 'BEGIN { printf "%d", (t - 1) * 100 }') |
There was a problem hiding this comment.
The calculation of threshold and pct currently uses two separate awk commands, which creates two separate processes. This can be optimized by combining them into a single awk call that calculates and outputs both values. This change would make the code more concise and slightly more efficient.
| local threshold | |
| threshold=$(awk -v n="$num_baselines" 'BEGIN { | |
| gap = (5 - n); if (gap < 0) gap = 0 | |
| printf "%.1f", 1.2 + gap * 0.1 | |
| }') | |
| local pct | |
| pct=$(awk -v t="$threshold" 'BEGIN { printf "%d", (t - 1) * 100 }') | |
| local threshold pct | |
| read -r threshold pct <<< "$(awk -v n=\"$num_baselines\" 'BEGIN { | |
| gap = (5 - n); if (gap < 0) gap = 0 | |
| t = 1.2 + gap * 0.1 | |
| p = (t - 1) * 100 | |
| printf \"%.1f %d\", t, p | |
| }')" |
| if [[ "$check_ttft" == "true" ]]; then | ||
| echo "Worst-case baseline query ttft per prompt: $expected_query_ttft" | ||
| echo "Actual query ttft per prompt: $query_ttft_per_prompt" | ||
| awk -v expected="$expected_query_ttft" -v actual="$query_ttft_per_prompt" 'BEGIN { | ||
| if (actual > expected * 1.2) { print "FAIL: Query ttft per prompt >20% worse than rolling baseline"; exit 1 } | ||
| awk -v expected="$expected_query_ttft" -v actual="$query_ttft_per_prompt" -v t="$threshold" -v pct="$pct" 'BEGIN { | ||
| if (actual > expected * t) { printf "FAIL: Query ttft per prompt >%d%% worse than rolling baseline\n", pct; exit 1 } | ||
| else { print "PASS: Query ttft per prompt within threshold" } | ||
| }' | ||
| fi | ||
|
|
||
| if [[ "$check_round" == "true" ]]; then | ||
| echo "Worst-case baseline query round time per prompt: $expected_query_round" | ||
| echo "Actual query round time per prompt: $query_round_time_per_prompt" | ||
| awk -v expected="$expected_query_round" -v actual="$query_round_time_per_prompt" 'BEGIN { | ||
| if (actual > expected * 1.2) { print "FAIL: Query round time per prompt >20% worse than rolling baseline"; exit 1 } | ||
| awk -v expected="$expected_query_round" -v actual="$query_round_time_per_prompt" -v t="$threshold" -v pct="$pct" 'BEGIN { | ||
| if (actual > expected * t) { printf "FAIL: Query round time per prompt >%d%% worse than rolling baseline\n", pct; exit 1 } | ||
| else { print "PASS: Query round time per prompt within threshold" } | ||
| }' | ||
| fi | ||
|
|
||
| if [[ "$check_warmup" == "true" ]]; then | ||
| echo "Worst-case baseline warmup round time per prompt: $expected_warmup" | ||
| echo "Actual warmup round time per prompt: $warmup_round_time_per_prompt" | ||
| awk -v expected="$expected_warmup" -v actual="$warmup_round_time_per_prompt" 'BEGIN { | ||
| if (actual > expected * 1.2) { print "FAIL: Warmup round time per prompt >20% worse than rolling baseline"; exit 1 } | ||
| awk -v expected="$expected_warmup" -v actual="$warmup_round_time_per_prompt" -v t="$threshold" -v pct="$pct" 'BEGIN { | ||
| if (actual > expected * t) { printf "FAIL: Warmup round time per prompt >%d%% worse than rolling baseline\n", pct; exit 1 } | ||
| else { print "PASS: Warmup round time per prompt within threshold" } | ||
| }' | ||
| fi |
There was a problem hiding this comment.
The logic for checking performance metrics (TTFT, round time, and warmup time) is repeated three times with only minor differences in variable names and strings. This code duplication can be reduced by refactoring it into a loop. By defining the metrics in an associative array, you can iterate through them and use a single, generalized awk command for all checks. This will make the script more maintainable and easier to extend in the future.
local -A metrics_to_check
metrics_to_check[ttft]="query ttft per prompt:${expected_query_ttft}:${query_ttft_per_prompt}"
metrics_to_check[round]="query round time per prompt:${expected_query_round}:${query_round_time_per_prompt}"
metrics_to_check[warmup]="warmup round time per prompt:${expected_warmup}:${warmup_round_time_per_prompt}"
for check_type in ttft round warmup; do
local check_var="check_${check_type}"
if [[ "${!check_var}" == "true" ]]; then
local metric_name expected_value actual_value
IFS=':' read -r metric_name expected_value actual_value <<< "${metrics_to_check[${check_type}]}"
echo "Worst-case baseline ${metric_name}: ${expected_value}"
echo "Actual ${metric_name}: ${actual_value}"
awk -v expected="${expected_value}" -v actual="${actual_value}" -v t="${threshold}" -v pct="${pct}" -v name="${metric_name}" 'BEGIN {
if (actual > expected * t) {
printf "FAIL: %s >%d%% worse than rolling baseline\n", name, pct;
exit 1;
} else {
printf "PASS: %s within threshold\n", name;
}
}'
fi
done… test redundant compile flag Signed-off-by: Samuel Shen <slshen@uchciago.edu>
|
Merging this to unblock LMCache CI. |
* make flexible comprehensive test Signed-off-by: Samuel Shen <slshen@uchciago.edu> * 2 more fixes: 1. GITHUB_TOKEN for nightly comprehensve 2. correctness test redundant compile flag Signed-off-by: Samuel Shen <slshen@uchciago.edu> Co-authored-by: Samuel Shen <slshen@uchciago.edu>
* make flexible comprehensive test Signed-off-by: Samuel Shen <slshen@uchciago.edu> * 2 more fixes: 1. GITHUB_TOKEN for nightly comprehensve 2. correctness test redundant compile flag Signed-off-by: Samuel Shen <slshen@uchciago.edu> Co-authored-by: Samuel Shen <slshen@uchciago.edu> Signed-off-by: shaoxiawjc <wjc2800@163.com>
* make flexible comprehensive test Signed-off-by: Samuel Shen <slshen@uchciago.edu> * 2 more fixes: 1. GITHUB_TOKEN for nightly comprehensve 2. correctness test redundant compile flag Signed-off-by: Samuel Shen <slshen@uchciago.edu> Co-authored-by: Samuel Shen <slshen@uchciago.edu> Signed-off-by: Aaron Wu <aaron.wu@dell.com>
* make flexible comprehensive test Signed-off-by: Samuel Shen <slshen@uchciago.edu> * 2 more fixes: 1. GITHUB_TOKEN for nightly comprehensve 2. correctness test redundant compile flag Signed-off-by: Samuel Shen <slshen@uchciago.edu> Co-authored-by: Samuel Shen <slshen@uchciago.edu>
* make flexible comprehensive test Signed-off-by: Samuel Shen <slshen@uchciago.edu> * 2 more fixes: 1. GITHUB_TOKEN for nightly comprehensve 2. correctness test redundant compile flag Signed-off-by: Samuel Shen <slshen@uchciago.edu> Co-authored-by: Samuel Shen <slshen@uchciago.edu>
Proposal: If we have fewer than 5 baselines for any reason, we should be slightly more flexible
2 more ad hoc fixes: