isucon-initial-strategy

name	isucon-initial-strategy
description	First 30 minutes strategy checklist for ISUCON competitions — what to do immediately after the contest starts

ISUCON Initial Strategy — First 30 Minutes

A structured approach for the opening phase of an ISUCON competition.

Minute 0-5: Environment Check

1. Verify all services are running:

   exec host="vm1" command="systemctl status isupipe-go nginx mysql pdns --no-pager"
   exec host="vm2" command="systemctl status isupipe-go nginx mysql pdns --no-pager"
   exec host="vm3" command="systemctl status isupipe-go nginx mysql pdns --no-pager"
   

2. Check system resources:

   exec host="vm1" command="free -h && df -h && nproc"
   

3. Get baseline benchmark score:

   benchmark_start
   

Minute 5-15: Profiling Setup

4. Enable slow query log:

   exec host="vm1" command="sudo mysql -u isucon -pisucon -e \"SET GLOBAL slow_query_log = 'ON'; SET GLOBAL long_query_time = 0.01;\""
   

5. Install alp (if available in skill):

   exec host="vm1" command="wget -q https://github.com/tkuchiki/alp/releases/download/v1.0.21/alp_linux_amd64.tar.gz -O /tmp/alp.tar.gz && cd /tmp && tar xzf alp.tar.gz && sudo mv alp /usr/local/bin/"
   

6. Read the application source:

   exec host="vm1" command="ls /home/isucon/isucon13/webapp/go/"
   exec host="vm1" command="wc -l /home/isucon/isucon13/webapp/go/*.go"
   

7. Read the database schema:

   exec host="vm1" command="cat /home/isucon/isucon13/webapp/sql/initdb.d/10_schema.sql"
   

Minute 15-25: Quick Wins

8. Add missing indexes (almost always the #1 quick win):

   • Check existing indexes: SHOW INDEX FROM <table>
   • Add indexes for foreign keys and WHERE clauses

9. Check for obvious N+1 patterns:

   exec host="vm1" command="grep -n 'for.*range\|\.Get\|\.Select\|\.Query' /home/isucon/isucon13/webapp/go/*.go | head -30"
   

10. Run profiling benchmark:

    exec host="vm1" command="sudo truncate -s 0 /var/log/nginx/access.log"
    benchmark_start
    

Minute 25-30: Analyze & Plan

11. Analyze results:

    • alp output → identify slowest endpoints
    • pt-query-digest → identify slowest queries
    • Compare with baseline score

12. Prioritize next steps based on data:

    • Fix the single biggest bottleneck first
    • Always validate with --pretest-only before full benchmark

Anti-Patterns to Avoid

Don't	Do Instead
Optimize without profiling	Always measure first
Change multiple things at once	One change → one benchmark
Skip pretest validation	Always --pretest-only before full run
Forget to persist index changes	Add to schema.sql or init script
Ignore benchmark errors	Fix 5xx/timeout errors before optimizing