Performance Profiler

The Performance Profiler measures and analyzes CodeBuddy’s operational performance — search latency, indexing throughput, memory usage, cache efficiency, and error rates. It provides both real-time alerts and on-demand performance reports.

Metrics tracked

Metric	Window size	What it measures
Search latency	100 samples	Time to complete a vector/hybrid search
Indexing throughput	50 samples	Files processed per second during indexing
Memory usage	20 samples	Heap memory consumed by the extension
Cache hit rate	100 samples	Percentage of queries served from cache
Error rate	100 samples	Percentage of operations that fail

All metrics use a RollingAverage — a fixed-size window that evicts the oldest sample when full.

Performance reports

A performance report includes:

Field	Calculation
avgSearchLatency	Mean of last 100 searches
p95SearchLatency	95th percentile latency
avgIndexingThroughput	Mean files/sec throughput
avgMemoryUsage	Mean heap usage in MB
cacheHitRate	Mean cache hit percentage
errorRate	Mean error percentage

Alerts

The profiler generates alerts when metrics cross thresholds:

Alert type	Severity	Trigger
HIGH_SEARCH_LATENCY	warning	P95 search latency exceeds threshold
HIGH_MEMORY_USAGE	warning	Average memory exceeds system limit
HIGH_ERROR_RATE	critical	Error rate exceeds acceptable threshold
LOW_THROUGHPUT	info	Indexing throughput drops below baseline

Adaptive configuration

The profiler auto-configures operational parameters based on system capabilities:

System tier	Embedding batch	Cache size	Search timeout	Indexing concurrency
High (≥16 GB RAM, ≥8 cores)	20	10,000	10s	8
Medium (8–16 GB, 4–8 cores)	10	5,000	15s	4
Low (<8 GB or <4 cores)	5	2,000	20s	2

System detection uses os.totalmem() and os.cpus().length.

Measurement API

Wrap any operation to automatically record its performance:

const result = await profiler.measure("search", async () => {
  return vectorDb.search(query);
});

This records:

• Duration (milliseconds)
• Memory delta (heap before vs. after)
• Success/failure status

Failed operations are counted toward the error rate metric.