CPU Overload Diagnosis — top · htop · perf · uptime

# Check Load Average (1-min / 5-min / 15-min averages)
uptime
# Example: load average: 4.20, 3.80, 2.10
# Overload if higher than the number of CPU cores
nproc                  # Check CPU core count

# Real-time CPU status
vmstat 2 10            # 2-second interval, 10 samples

# Key columns
# us: user space, sy: kernel, wa: I/O wait, id: idle
# si/so: swap in/out (high values indicate memory shortage)

# After launching top
top
# P key: sort by CPU
# M key: sort by memory
# k key: kill process by PID
# 1 key: show per-core CPU

# One-shot snapshot (sorted)
ps aux --sort=-%cpu | head -15

# Track CPU for a specific process
pidstat -u -p 1234 2 5   # PID 1234, 2-second interval, 5 samples

# strace — trace system calls
strace -p 1234 -c       # System call statistics summary
strace -p 1234 -e trace=read,write  # Specific calls only

# perf — performance counter profiling
sudo apt install -y linux-tools-common linux-tools-$(uname -r)

# Collect CPU events for 10 seconds
sudo perf record -g -p 1234 -- sleep 10

# Analyze results
sudo perf report

# System-wide hotspots (5 seconds)
sudo perf top

# lsof — check how many files the process has open
sudo lsof -p 1234 | wc -l

# Suspect infinite loops or inefficient algorithms
# Check stack traces (Python)
sudo py-spy top --pid 1234

# Java
sudo jstack 1234 | grep -A 5 "RUNNABLE"

# Node.js
kill -USR1 1234   # Enable node --prof mode

# Suspect excessive system calls
strace -p 1234 -c -S calls | head -20

# Check interrupts
cat /proc/interrupts | sort -k2 -rn | head -10

# I/O bottleneck (see the dedicated troubleshooting guide)
iostat -x 2 5
iotop -o

# Limit CPU usage of a specific process (cgroups)
sudo apt install -y cpulimit
cpulimit -p 1234 -l 50    # Limit PID 1234 to 50% CPU

# Lower priority with nice
renice -n 15 -p 1234

# Adjust OOM Killer score (protect important processes)
echo -1000 > /proc/1234/oom_score_adj